Qlogic Fast Fabric User Manual

Page 1
Q
Simplify
Fast Fabric
Users Guide
D000006-000 Rev. A Page i
Page 2
Fast Fabric Users Guide
Q
Page 3
Fast Fabric Users Guide
Q
Information furnished in this manual is believed to be accurate and reliable. However, QLogic Corporation assumes no responsibility for its use, nor for any infringements of patents or other rights of third parties which may result from its use. QLogic Corporation reserves the right to change product specifications at any time without notice. Applications described in this document for any of these products are for illustrative purposes only. QLogic Corporation makes no representation nor warranty that such applications are suitable for the specified use without further testing or modification. QLogic Corporation assumes no responsibility for any errors that may appear in this document.
No part of this document may be copied nor reproduced by any means, nor translated nor transmitted to any magnetic medium without the express written consent of QLogic Corporation.
Linux is a registered trademark of Linus Torvalds.
Microsoft and Windows are registered trademarks and Windows Server is a trademark of Microsoft Corporation.
Red Hat and all Red Hat-based trademarks are trademarks or registered trademarks of Red Hat, Inc.
SUSE is a registered trademark of Novell, Inc.
All other brand and product names are trademarks or registered trademarks of their respective owners.
Document Revision History
Rev. A, 01/08/08
D000006-000 Rev. A Page iii
Page 4
Fast Fabric Users Guide
0
Q
© 2008 QLogic Corporation. All rights reserved worldwide.
QLogic Corporation, 26650 Aliso Viejo Parkway, Aliso Viejo, CA 92656, (800) 662-4471 or (949) 389-600
First Published: March, 2007
Printed in U.S.A.
Page 5
Q
Fast Fabric Users Guide
Section 1 Introduction
1.1 Intended Audience . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1-1
1.2 License Agreements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1-1
1.3 Technical Support. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1-2
1.3.1 Availability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1-2
1.3.2 Contact Information . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1-2
Section 2 Fast Fabric Overview
2.1 Feature Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-1
2.2 Fast Fabric Architecture . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-2
2.2.1 How Fast Fabric Works . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-3
Section 3 Getting Started
3.1 Design the Fabric . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-1
3.2 Set Up the Fabric . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-2
3.3 Using Fast Fabric . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-4
3.4 Installing and Verifying Firmware on the SilverStorm IB Chassis. . . . . . . 3-6
3.5 Installing and Configuring the Subnet Manager . . . . . . . . . . . . . . . . . . . . 3-9
3.6 Installing and Verifying Firmware on the IB Switches. . . . . . . . . . . . . . . . 3-10
3.7 Installing InfiniBand on the Remaining Servers . . . . . . . . . . . . . . . . . . . . 3-12
3.8 Verifying InfiniBand on the Remaining Servers . . . . . . . . . . . . . . . . . . . . 3-16
3.9 Complete Installation of additional IB Management Nodes . . . . . . . . . . . 3-18
3.10 Configure and Initialize Health Check Tools. . . . . . . . . . . . . . . . . . . . . . . 3-19
3.11 Running HPL . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-20
3.12 Upgrading IB software . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-21
Section 4 Fast Fabric TUI Menu
4.1 Host Setup via Fast Fabric . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-3
4.1.1 Edit Configuration and Select/Edit Hosts Files . . . . . . . . . . . . . . . . . . 4-4
4.1.2 Verify Hosts via Ethernet ping . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-4
4.1.3 Verify RSH/RCP Configured . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-4
4.1.4 Setup Password-less SSH/SCP . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-5
4.1.5 Copy /etc/hosts to all hosts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-5
4.1.6 Show uname -a for all hosts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-5
4.1.7 Install/Upgrade QuickSilver Software . . . . . . . . . . . . . . . . . . . . . . . . . 4-5
4.1.8 Configure IPoIB IP Address . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-6
4.1.9 Build MPI Test Apps and Copy to Hosts . . . . . . . . . . . . . . . . . . . . . . . 4-6
4.1.10 Reboot Hosts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-6
4.1.11 Refresh SSH Known Hosts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-6
4.1.12 Rebuild MPI Library and Tools . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-6
D000006-000 Rev. A Page v
Page 6
Fast Fabric Users Guide
4.1.13 Run a command on all hosts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-7
4.1.14 Copy a file to all hosts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-7
4.1.15 View ibtest result files . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-7
4.2 Host Admin via Fast Fabric. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-8
4.2.1 Edit Config and Select/Edit Hosts Files . . . . . . . . . . . . . . . . . . . . . . . . 4-8
4.2.2 Verify Hosts via Ethernet Ping . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-8
4.2.3 Summary of Fabric Components . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-9
4.2.4 Show Status of Host IB Ports . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-9
4.2.5 Verify Hosts see each other . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-9
4.2.6 Verify Hosts ping via IPoIB . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-9
4.2.7 Refresh SSH Known Hosts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-9
4.2.8 Check MPI Performance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-10
4.2.9 Generate all Hosts Problem Report Info . . . . . . . . . . . . . . . . . . . . . . . 4-10
4.2.10 Run a command on all hosts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-10
4.2.11 View ibtest result files . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-10
4.3 QLogic IB Chassis Admin via Fast Fabric . . . . . . . . . . . . . . . . . . . . . . . . 4-11
4.3.1 Edit the Configuration and Select/Edit Chassis Files . . . . . . . . . . . . . . 4-11
4.3.2 Verify Chassis via Ethernet Ping . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-11
4.3.3 Update Chassis Firmware . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-12
4.3.4 Show Status of Chassis IB Ports . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-12
4.3.5 Reboot Chassis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-13
4.3.6 Generate all Chassis Problem Report Information . . . . . . . . . . . . . . . 4-13
4.3.7 Run a command on all chassis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-13
4.3.8 View ibtest results files . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-13
4.4 SilverStorm Externally Managed IB Switch Administration via Fast Fabric 4-14
4.4.1 Edit Config and Select/Edit Chassis Files . . . . . . . . . . . . . . . . . . . . . . 4-14
4.4.2 Verify Switch via Firmware Dump . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-14
4.4.3 Update Switch Firmware . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-14
4.4.4 Reboot Switch . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-15
4.4.5 View ibtest result files . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-15
Q
Section 5 Detailed Descriptions of Command LineTools
5.1 Common Tool Options . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5-1
5.1.1 -? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5-1
5.1.2 -p . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5-1
5.1.3 -S . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5-2
5.1.4 -C . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5-2
5.1.5 -n or -I . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5-2
5.1.6 Selection of Hosts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5-3
5.1.7 Selection of Chassis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5-4
Page 7
Q
Fast Fabric Users Guide
5.1.8 Selection of Switches . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5-7
5.1.9 Selection of local Ports (subnets) . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5-9
5.2 Basic Setup and Administration Tools . . . . . . . . . . . . . . . . . . . . . . . . . . . 5-11
5.2.1 pingall . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5-11
5.2.2 check_rsh . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5-12
5.2.3 setup_ssh . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5-13
5.2.4 cmdall . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5-15
5.2.5 captureall . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5-17
5.3 File Management Tools . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5-19
5.3.1 scpall . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5-19
5.3.2 uploadall . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5-21
5.3.3 downloadall . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5-22
5.3.4 Simplified Editing of Node-Specific Files . . . . . . . . . . . . . . . . . . . . . . . 5-24
5.3.5 Simplified Setup of Node-Generic Files . . . . . . . . . . . . . . . . . . . . . . . . 5-24
5.4 Fabric Analysis Tools . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5-25
5.4.1 Fabric_info . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5-25
5.4.2 showallports . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5-26
5.4.3 iba_report . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5-28
5.4.4 saquery . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5-56
5.5 Advanced Initialization and Verification - ibtest . . . . . . . . . . . . . . . . . . . . 5-60
5.5.1 ibtest Host Operations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5-63
5.5.2 ibtest Chassis Operations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5-66
5.5.3 ibtest Switch Operations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5-67
5.5.4 Interpreting the ibtest log files . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5-68
5.6 Health Check and Baselining Tools . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5-69
5.6.1 Usage Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5-69
5.6.2 Common Operations and Options . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5-70
5.6.3 fabric_analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5-72
5.6.4 chassis_analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5-76
5.6.5 hostsm_analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5-81
5.6.6 esm_analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5-83
5.6.7 all_analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5-86
5.6.8 Manual and Automated Usage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5-87
Section 6 MPI Sample Applications
6.1 OSU Latency . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6-2
6.2 OSU Latency2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6-2
6.3 OSU Bandwidth . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6-2
6.4 OSU Bandwidth2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6-3
6.5 OSU Bidirectional Bandwidth . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6-3
D000006-000 Rev. A Page vii
Page 8
Fast Fabric Users Guide
6.6 High Performance Linpack (HPL) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6-3
6.7 Pallas . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6-4
Q
Appendix A Fast Fabric Quick Install Checklist
A.1 Setup The Fabric . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . A-1
A.2 Installing and verifying Firmware on the IB Chassis . . . . . . . . . . . . . . . . A-2
A.3 Installing and Configuring the Subnet Manager . . . . . . . . . . . . . . . . . . . . A-2
A.4 Installing and Verifying Firmware on the IB Switches . . . . . . . . . . . . . . . A-2
A.5 Install Infiniband on the Remaining Servers . . . . . . . . . . . . . . . . . . . . . . A-2
A.6 Verifying Infiniband on the Remaining Servers . . . . . . . . . . . . . . . . . . . . A-3
A.7 Complete Installation of additional IB Management Nodes . . . . . . . . . . . A-3
A.8 Configure and initialize health check tools . . . . . . . . . . . . . . . . . . . . . . . . A-4
Appendix B Fast Fabric Configuration Files
B.1 fastfabric.conf . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . B-1
B.2 iba_mon.conf . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . B-6
B.3 Host List Files . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . B-9
B.4 Chassis List Files . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . B-9
B.5 Selection of slots within a chassis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . B-10
B.6 Switch List Files . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . B-11
B.7 Port List Files . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . B-12
Appendix C Configuration of IPoIB Name Mapping
Appendix D Multi-Subnet Fabrics
D.1 Primarily Independent Subnets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . D-1
D.2 Overlapping Subnets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . D-3
Page 9
Section 1
Introduction
This manual describes installation, configuration and administration task information for the Fast Fabric Toolset.
This manual is organized as follows:
Section 1 describes the intended audience and technical support.
Section 2 describes the Fast Fabric Toolset.
Section 3 describes getting started with Fast Fabric.
Section 4 describes the Fast Fabric Textual User Interface (TUI) menu.
Section 5 describes the Fast Fabric command tools and test tools.
Section 6 describes MPI Sample Applications.
Appendix A presents the Fast Fabric Quick Install Checklist.
Appendix B describes the Fast Fabric Configuration Files.
Appendix C provides information on the configuration of IPoIB name mapping.
Appendix D provides information on configuring Multi-Subnet Fabrics.
1.1
Intended Audience
This manual is intended to provide network administrators and other qualified personnel a reference for installation, configuration and administration task information for the Fast Fabric toolset.
1.2
License Agreements
Refer to the QLogic Software End User License Agreement for a complete listing of all license agreements affecting this product.
D000006-000 Rev. A 1-1
Page 10
1 – Introduction Technical Support
1.3
Technical Support
Customers should contact their authorized maintenance provider for technical support of their QLogic products. QLogic-direct customers may contact QLogic Technical Support; others will be redirected to their authorized maintenance provider.
Visit the QLogic support Web site listed in Contact Information for the latest firmware and software updates.
1.3.1
Availability
QLogic Technical Support for products under warranty is available during local standard working hours excluding QLogic Observed Holidays.
1.3.2
Contact Information
Q
Support Headquarters QLogic Corporation
4601 Dean Lakes Blvd Shakopee, MN 55379 USA
QLogic Web Site www.qlogic.com
Technical Support Web Site support.qlogic.com
Technical Support Email support@qlogic.com
Technical Training Email tech.training@qlogic.com
North American Region
Email support@qlogic.com
Phone +1-952-932-4040
Fax +1 952-974-4910
All other regions of the world
QLogic Web Site www.qlogic.com
Page 11
2.1
Feature Overview
The Fast Fabric Toolset is designed to both simplify and expedite common InfiniBand (IB) cluster management tasks. Fast Fabric can assist in generic management tasks as well as InfiniBand installation, upgrade, configuration and verification tasks.
Fast Fabric has the following key capabilities:
Accelerates initial fabric installation
Verify host management network connectivity
Verify host OS levels
Sets up ssh keys
Performs initial InfiniBand software installation
Configures Internet Protocol over InfiniBand (IPoIB) IP addresses
Section 2
Fast Fabric Overview
Performs InfiniBand driver upgrades or the installation of additional InfiniBand
drivers
Verifies key fabric installation metrics:
Components in fabric
Link error counters
Link widths and speeds
IB and PCI bus bandwidth
IB end-to-end latency
IPoIB connectivity
Subnet Agent (SA) visibility of all nodes
IB connectivity of all switches and nodes
Aids in diagnosis of fabric problems
Fabric error isolation
Fabric topology analysis
Fabric route analysis
D000006-000 Rev A 2-1
Page 12
2 – Fast Fabric Overview Fast Fabric Architecture
Aids in ongoing fabric status and configuration monitoring
Automated fabric health checks and configuration baseline compare
Automated chassis health checks and configuration baseline compare
Automated SM health checks and configuration baseline compare
Provides tools to accelerate common host administration tasks
Executes commands across many hosts
Copies files to and from many hosts
Edits host-specific files across many hosts
Provides tools to accelerate common chassis and switch administration tasks
Manage firmware levels on switches and chassis
Execute commands across many chassis
Assists in the initial benchmarking and tuning of High Performance Computing
(HPC) fabrics.
Q
Fast Fabric includes both a Textual User Interface (TUI) menu system as well as command line tools. The TUI presents the menus in a typical order of execution for a new fabric install, hence simplifying fabric installation for new users. All operations available in the TUI can also be accomplished via the command line. The command line tools are designed to permit customer specific scripts to invoke the command line tools.
2.2
Fast Fabric Architecture
Figure 2-1. Fast Fabric Architecture
2-2 D000006-000 Rev A
Page 13
Q
2 – Fast Fabric Overview
Fast Fabric is typically installed on one or more IB Management Nodes. The IB Management Node must be connected to the rest of the cluster via both InfiniBand and a management network. The management network may be the primary InfiniBand network (IPoIB) or Ethernet. The management network will be used for Fast Fabric host setup and administration tasks. It may also be used for other aspects of server administration or operation.
Depending on cluster size and design, the IB Management node may also be used as the master node for starting MPI jobs. It may also be used to run a QLogic Host SM and other management software. Consult the QLogic SM documentation for details and what combinations are valid.
Note: When InfiniBand is used as the management network, Fast Fabric will not be able to install host IB software nor configure IPoIB, however it will be able to support host IB software upgrades, verification and all the other features of Fast Fabric.
If remote access to Fast Fabric is desired, setup remote access to the IB Management Node via ssh, telnet, X windows, VNC or any other mechanism which will allow the remote user to access a Linux Command Line shell. Typically Fast Fabric is only used by cluster administrators.
2.2.1
How Fast Fabric Works
Fast Fabric consists of a variety of tools to administer hosts, chassis and externally managed switches. Depending on the tool, the method of accessing and administering the target devices may differ.
The following methods are used by Fast Fabric:
Login via management network Host setup and installation,
MPI job startup (can be inband
or via management network)
Typically tools which login to other hosts will do such in a password-less manner using ssh or telnet (configurable). Tools which login to internally managed chassis can use ssh or telnet (configurable). Chassis tools can prompt for a single password for all chassis or can be preconfigured with the password. These approaches permit
Table 2-1. Fast Fabric Methods
Method Examples
Inband access via IB Fabric topology reports, SA
database queries, fabric error
and link speed analysis, tools
for externally managed
switches, etc
tools for internally managed
chassis, etc
Verify MPI performance,
running sample MPI
benchmarks
D000006-000 Rev A 2-3
Page 14
2 – Fast Fabric Overview Fast Fabric Architecture
the tools to operate with minimal user interaction and hence reduce the time to perform operations against many hosts or chassis.
After initial installation, Fast Fabric can be configured to use IPoIB instead of the management network.
NOTE: Any reconfigurations that affect IPoIB or involve installing new IB hosts
Q
will not be able to use IPoIB.
2-4 D000006-000 Rev A
Page 15
Section 3
Getting Started
Before using the Fast Fabric toolset, the Site Implementation Engineer must perform the tasks described in the sections which follow. To aid in keeping track of steps performed a checklist is provided (see appendix A). During the setup procedure, the Fast Fabric configuration files which must be edited or created are described throughout the procedure. For more information about the configuration files used by Fast Fabric see appendix B.
The instructions below describe the basic fabric installation and verification sequence for a typical single IB subnet fabric. For more information on installation and verification of multiple IB subnet fabrics, see appendix D.
Some of the tasks are only applicable when Linux is being used. They will be marked with (Linux). Similarly some of the tasks are only applicable when QuickSilver Linux IB software is being used on the hosts. Those will be marked with (Host). All tasks which are applicable only when SilverStorm IB Switches or SilverStorm IB Chassis are being used will be marked with (Switch). All remaining tasks are generally applicable to all environments and will be marked with (All).
NOTE: Some of the Linux steps may be applicable to other Unix-like operating
systems if it is desired to enable use of non-IB specific Fast Fabric tools (such as cmdall) against the given hosts.
3.1
Design the Fabric
Prior to beginning the installation and setup of the fabric, its important to carefully design and plan the installation. Part of the design plan must include identification of which servers will be the administration nodes for the cluster and hence where Fast Fabric will be installed.
For large clusters, cable, power, and cooling plans are very important and must be carefully considered. These plans drive the ultimate layout of equipment in the racks. A typical configuration involves leaf switches and servers in the same racks, with core switches in centrally located racks. This minimizes both cable lengths and complexity. It is also recommended to place the IB switches at the bottom of a rack. This allows inter-rack cables to be cleanly routed below the floor (some sites use cable routing above the racks in which case placing the IB switches near the top of the rack is recommended).
NOTE: The overall physical design has many complex aspects, such as power,
cooling and rack layout which are beyond the scope of this document.
D000006-000 Rev A 3-1
Page 16
3 – Getting Started Set Up the Fabric
3.2
Set Up the Fabric
1. (All) The first step in any installation is to physically install the hardware:
Servers
Core and leaf InfiniBand switches, such as the SilverStorm 9024 and 9000
Multi-Protocol Fabric Directors (9020, 9040, 9080, 9120 and 9240).
Virtual I/O systems, such as the EVIC and FVIC cards for the SilverStorm
9000 Multi-Protocol Fabric Directors Series.
NOTE: When installing externally managed switches (such as the SilverStorm
2. (All) Within each server a host channel adapter (HCA), such as the QuickSilver HCA 7000 or 9000 must be installed. Refer to the QuickSilver Fabric Access Quick Start Guide for instructions.
Q
9024-FC switch), take note of the Node GUID. This is typically on a label on the case of the switch. The Node GUID will be needed later to configure and manage the switch(es).
3. (All) Prior to installing software, the hardware configuration should be reviewed to ensure everything was installed according to plan. Later during the installation Fast Fabric tools may also be used to help verify the installation.
4. (Linux) Install the desired Linux OS version (with the same kernel distribution) on all hosts. Generally the IB Management node(s) (i.e., the host which will run Fast Fabric) should have a full install and must include the Tcl, Expect and TCLx packages. If Redhat Enterprise Server 3 or later is being installed, only the Tcl and Expect packages are required. For MPI clusters install the C and Fortran compilers along with their associated tools on the IB Management node(s).
NOTE: All hosts must have a command-line prompt ending in "# " or "$ ". Make
certain there is a space after either "#" or "$". Such a prompt must be used for the root user as well as any other user codes the user intends Fast Fabric to make use of.
NOTE: To simplify the use of Fast Fabric to setup ssh security, it is recommended
to install all servers with the same root password. If desired after ssh has been setup using Fast Fabric, the user may change the root passwords.
NOTE: Consult the QuickSilver Fabric Access Linux Host Release Notes for a
list of supported OS versions.
5. (Linux) Enable remote login as root to each host: In order for Fast Fabric to manage the hosts, the IB Management Node must be able to securely login as root to each host. This can be accomplished using either ssh or rsh. SSH is recommended due to its higher level of security. If
3-2 D000006-000 Rev A
Page 17
Q
3 – Getting Started
ssh is used, no additional manual steps are require at this stage (typically Linux OS installation will enable ssh) Alternatively, if its desired to use rsh during fabric installation and/or operation, the following steps must be performed on each node such that the IB Management Node can login using rsh as user root.
a. Each node must be configured such that the IB management node can rsh
into it. The IB management node must also be able to rsh into itself. Typically this requires that a .rhosts file be created in /root such as:
<mgmthost name> root
<mgmthost name.domain name> root
localhost root
<mgmthost IP address>
where mgmthost is the network name of the IB Management Node and domain is the network domain name of the master. The .rhosts file must have permissions of 640. Also, rsh should be enabled on each node.
Enable rsh by editing the /etc/xinetd.d/rsh file and setting:
disable=no
This can also be accomplished using:
chkconfig rsh on
Also enable rexec and rlogin using the above steps.
b. Execute mv /etc/securetty /etc/securetty.bak
6. (All) TCP/IP Host Name resolution: Fast Fabric and TCP/IP will need to resolve hostnames to Management Network and/or IPoIB IP addresses. If the management network is not IPoIB, each host will need both a management network name and an IPoIB network name. In which case, a recommended convention is to use the actual hostname as the management network name and <HOSTNAME>-ib as the IPoIB network name (where <HOSTNAME> is the management network name of the given host) Typically name resolution is accomplished by configuring a DNS server on the management network with both management network and IPoIB addresses for each host (and QLogic internally managed IB chassis). Alternately a /etc/hosts file may be created on the IB Management node. Fast Fabric can then propagate this /etc/hosts file to all the other hosts.
If using the /etc/hosts approach:
On the master node, add all the Ethernet and IPoIB addresses into the
/etc/hosts file. For the IPoIB convention, use <HOSTNAME>-ib. The localhost line should not be edited.
D000006-000 Rev A 3-3
Page 18
3 – Getting Started Using Fast Fabric
Q
The /etc/hosts file should not have any node-specific data (the following section will step through the task of copying this file to all the nodes).
If using DNS:
Consult the documentation for the DNS server being used. Make sure to edit the /etc/resolv.conf configuration on the IB Management Node to use the proper DNS server. Consult the Linux OS documentation for more information on configuring /etc/resolv.conf. This file is typically configured during OS installation.
If /etc/resolv.conf must be manually configured for each host, Fast Fabric can aid in copying this to all the hosts. In which case, the /etc/resolv.conf file created on the IB Management Node must not have any node-specific data and must be appropriate for use on all hosts. A later section will step through the task of copying this file to all the nodes.
7. (All) NTP setup - it is recommended to configure an NTP server for the cluster and have all the hosts and Internally-Managed chassis synchronize their clocks with the NTP server. Consult the Linux OS documentation for information on how to configure NTP servers and clients.
8. (All) On the IB Management node, install the Fabric Access Software using the procedure documented in the Fabric Access Software Users Guide. The IB Management Node must have at least Fast Fabric, the IB Stack and IPoIB installed and configured. For MPI clusters running the QuickSilver Host stack, the IB Management Node should also include the MPI Runtime and MPI Development packages, and if the user desires to rebuild MPI itself, the IB Development package and MPI Source packages will also be required.
After completing the install, reboot the IB Management node.
NOTE: When managing a cluster where compute nodes are not running the
3.3
Using Fast Fabric
The initial installation and verification process is best performed using the Fast Fabric TUI menu system. The main menu can be invoked using the iba_config command. The main menu is as follows:
QuickSilver host stack or where the IPoIB settings on the compute nodes are incompatible with the IB Management node (for example when a 4K MTU is used on the compute nodes), it is recommended not to run IPoIB on the IB management nodes.
3-4 D000006-000 Rev A
Page 19
Q
3 – Getting Started
SilverStorm Technologies Inc. InfiniBand 4.1.1.0.15 Software
1) Show Installed Software
2) Reconfigure IP over IB
3) Reconfigure Driver Autostart
4) Update HCA Firmware
5) Generate Supporting Information for Problem Report
6) Host Setup via Fast Fabric
7) Host Admin via Fast Fabric
8) Chassis Admin via Fast Fabric
9) Externally Managed Switch Admin via Fast Fabric a) Uninstall Software
X) Exit
In the above menu, items 6-9 represent the Fast Fabric menus. The operation of this menu is the same as the INSTALL and iba_config functions documented in the QuickSilver Fabric Access Users Guide. Pressing a key 1-9 or a will invoke the given submenu. Pressing X will exit the menu system.
Selection of a Fast Fabric menu (6-9) will present a submenu similar to the following:
SilverStorm Technologies Inc. IB Host Setup Menu (4.1.1.0.15) Fast Fabric Host List: /etc/sysconfig/iba/hosts
0) Edit Config and Select/Edit Hosts Files [Perform]
1) Verify Hosts via Ethernet ping [Perform]
2) Verify rsh/rcp Configured [ Skip ]
3) Setup Password-less ssh/scp [Perform]
4) Copy /etc/hosts to all hosts [ Skip ]
5) Show uname -a for all hosts [Perform]
6) Install/Upgrade InfiniServ Software [Perform]
7) Configure IPoIB IP Address [Perform]
8) Build MPI Test Apps and Copy to Hosts [Perform]
9) Reboot Hosts [Perform] a) Refresh ssh Known Hosts [Perform] b) Rebuild MPI Library and Tools [ Skip ] c) Run a command on all hosts [ Skip ] d) Copy a file to all hosts [ Skip ] e) View ibtest result files [ Skip ]
P) Perform the selected actions N) Select None
X) Return to Previous Menu (or ESC)
The submenus typically present operations in the typical order they would be used during an installation. Pressing the keys corresponding to menu items (0-e in the example above) will toggle the Skip/Perform selection for the given item. As shown in the example above, more than 1 item may be selected. Once the desired set of
D000006-000 Rev A 3-5
Page 20
3 – Getting Started Installing and Verifying Firmware on the SilverStorm IB Chassis
items have been selected, press P. To unselect all items, press N. Pressing X or ESC will exit this menu and return to the Main Menu.
If more than 1 item is selected, the items will be performed in the order shown in the menu. This is the typical order desired during fabric setup. If it's desired to perform items in a different order, select a single item and press P to perform it by itself. Then repeat. An opportunity will be presented after each item to abort:
Hit any key to continue (or ESC to abort)...
If ESC is pressed, the sequence of operations will be aborted and return to the previous menu. Any other key will result in the next selected menu item being performed. This prompt is also shown after the last selected item completes, hence permitting an opportunity to review the results before the screen is cleared to display the menu.
At the top of each Fast Fabric menu, the file listing the components to operate on is shown. For example:
Fast Fabric Host List: /etc/sysconfig/iba/hosts
On each Fast Fabric menu, item 0 will permit a different file to be selected and will permit the editing of the file (using the editor selected via the EDITOR environment variable). In addition it will also permit review and editing of the fastfabric.conf file. The fastfabric.conf file guides the overall configuration of Fast Fabric and describes cluster specific attributes of how Fast Fabric will operate. It is discussed in greater detail in appendix B.
Q
During the execution of each menu selection, the actual Fast Fabric command line tool being used will be shown. This can be used as an educational aid to learn the tools.
3.4
Installing and Verifying Firmware on the SilverStorm IB Chassis
If the fabric contains SilverStorm 9000 series internally-managed IB switches, Fast Fabric may be used to aid the installation and configuration of the switches.
Prior to using Fast Fabric the following minimal steps need to be performed:
1. (Switch) Connect each SilverStorm chassis to the management network via its Ethernet management port. Chassis with redundant management should have both Ethernet management ports connected.
2. (Switch) Assign each SilverStorm chassis a unique IP address and appropriately configure the chassis Ethernet management port network settings.
3. (Switch) Select a unique name which will be used for each SilverStorm Chassis. This name should be configured in DNS or /etc/hosts as the TCP/IP name for the chassis Ethernet management port. In addition this should be configured as the IB Node Description for the chassis via the chassis GUI or CLI.
3-6 D000006-000 Rev A
Page 21
Q
3 – Getting Started
a. When Virtual I/O controllers (VIC) are installed in a chassis, each VIC should
also be assigned a unique name.
4. (Switch) Configure the administrator password on each SilverStorm Chassis
NOTE: Newer versions of SilverStorm chassis firmware permit SSH keys to be
configured within the chassis for secure password-less login. In this case it is recommended to configure SSH keys in the chassis at this point such that the IB Management Node can login as admin without a password.
NOTE: When using versions of SilverStorm chassis firmware that do not support
SSH keys, to simplify the use of Fast Fabric it is recommended to install all chassis with the same admin password.
5. (Switch) Mount or copy the relevant chassis firmware CD(s) or files onto the Fast Fabric management node. During the steps below the *.pkg files on the CD will be used to upgrade the firmware on each chassis.
NOTE: When copying files its best to place all files at a given firmware level into
a single directory whose name indicates the firmware revision number
Once the above steps have been completed, additional setup of the Chassis may be performed using Fast Fabric.
1. (Switch) Select the "Chassis Admin via Fast Fabric" option from the main menu.
2. (Switch) Select the items shown as "Perform" in the menu below and press the P key to perform them:
SilverStorm Technologies Inc. IB Chassis Admin Menu (4.1.1.0.15) Fast Fabric Chassis List: /etc/sysconfig/iba/chassis
0) Edit Config and Select/Edit Chassis Files [Perform]
1) Verify Chassis via Ethernet ping [Perform]
2) Update Chassis Firmware [Perform]
3) Show Status of Chassis IB Ports [ Skip ]
4) Reboot Chassis [ Skip ]
5) Generate all Chassis Problem Report Info [ Skip ]
6) Run a command on all chassis [ Skip ]
7) View ibtest result files [ Skip ]
P) Perform the selected actions N) Select None
X) Return to Previous Menu (or ESC)
3. (Switch) "Edit Config and Select/Edit Chassis Files" will permit the chassis and fastfabric.conf files to be edited. When placed in the editor for fastfabric.conf, review all the settings. Especially review the FF_CHASSIS_LOGIN_METHOD
D000006-000 Rev A 3-7
Page 22
3 – Getting Started Installing and Verifying Firmware on the SilverStorm IB Chassis
and FF_CHASSIS_ADMIN_PASSWORD. Consult appendix B for more information about fastfabric.conf.
NOTE: Fast Fabric will provide the opportunity to enter the chassis password
interactively when needed. Hence it's not necessary to place it within fastfabric.conf. If it is desired to instead keep the QLogic Chassis admin password in fastfabric.conf, its recommended to change the fastfabric.conf permissions to be 0x600 (eg. root only access).
NOTE: Newer versions of chassis firmware permit ssh keys to be configured
within the chassis for secure password-less login. In which case there is no need to configure a FF_CHASSIS_ADMIN_PASSWORD and FF_CHASSIS_LOGIN_METHOD can be ssh. Consult the SilverStorm 9000 Users Guide for more information.
When placed in the editor for chassis, create the file with a list of the chassis names (the TCP/IP Ethernet management port names assigned above) or IP addresses (Use of names is recommended). One entry per line. Such as:
Chassis1
Chassis2
Q
NOTE: Do not list externally managed switches, such as the SilverStorm 9024FC
switches in this file. Those will be covered in the next section.
For further details about the file format refer to the section “Selection of Chassis”
on page 5-4
4. (Switch) "Verify Chassis via Ethernet ping" will ping each selected chassis over the management network. If all chassis were found, continue to the next step. If some chassis were not found, abort out of the menu and review the following for those chassis which were not found:
Is chassis powered on and booted
Is chassis connected to management network
Is chassis IP address and network settings consistent with DNS or /etc/hosts
Is Management node connected to the management network
Are Management node IP address and network settings correct
Is the management network itself up (switches, routers, etc)
Is correct set of chassis listed in the chassis file (the previous step may be
repeated to review and edit the file as needed)
5. (Switch) "Update Chassis Firmware" will permit the chassis firmware version to be verified and updated as needed.
3-8 D000006-000 Rev A
Page 23
Q
3 – Getting Started
NOTE: The chassis must be running firmware version 4.0.0.4.3 or later to perform
this function. If the chassis is not up to this level, it will need to be manually updated via the chassis GUI. See the SilverStorm 9000 Users Guide for more information.
NOTE: Consult the relevant chassis firmware release notes to ensure any
prerequisites for the upgrade to the new firmware level have been met prior to performing the upgrade via Fast Fabric.
When prompted:
Multiple Firmware files and/or Directories may be space separated
Shell wildcards may be used
For Directories all .pkg files in the directory tree will be used
Enter Files/Directories to use (or none):
specify the directory where the relevant firmware files have been stored. This can be the mount point of the CD or the directory to which the files were copied in a previous step.
Since the fabric is not yet operational, it's recommended to answer "y" to:
Would you like to run the firmware now? [n]:
Fast Fabric will ensure that all chassis are running the firmware level provided and install and/or reboot each chassis as needed
If any chassis fails to be updated, use the "View ibtest result files" option to review the result files from the update. Refer to the section “Interpreting the ibtest log files” on page 5-68for more details.
6. (Switch) If there are any other operations which need to be performed on all chassis, they may be performed using the "Run a command on all chassis" option.Each time this is executed a single chassis CLI command may be specified to be executed against all selected chassis. Using such commands, additional setup or verification of the chassis may be performed.
3.5
Installing and Configuring the Subnet Manager
(All) At this point the subnet manager (SM) for the fabric must be installed or
enabled. Consult the QuickSilver Fabric Manager and Fabric Viewer Users Guide for information on how to install, enable and configure the SM.
When using the QuickSilver host-based SM, a typical installation will place Fast Fabric and the host SM on the same IB Management Node. If desired, it is also valid to place Fast Fabric on its own independent management node, perhaps along with other 3rd party management applications (such as MPI job schedulers, etc).
D000006-000 Rev A 3-9
Page 24
3 – Getting Started Installing and Verifying Firmware on the IB Switches
The steps which follow will require that an SM be operational within the fabric.
3.6
Installing and Verifying Firmware on the IB Switches
If the fabric contains SilverStorm 9024FC series externally managed switches, Fast Fabric may be used to aid the installation and configuration of the switches.
Prior to using Fast Fabric the following minimal steps need to be performed:
1. (Switch) Select a unique name which will be used for each Switch. This name will be configured as the IB Node Description for the switch in the steps below.
NOTE: Externally managed switches do not have an Ethernet port and hence
will not have a TCP/IP name.
2. (Switch) Mount or copy the relevant switch firmware CD(s) or files onto the Fast Fabric management node. During the steps below the *.emfw files on the CD will be used to upgrade the firmware on each switch.
NOTE: When copying files its best to place all files at a given firmware level into
a single directory whose name indicates the firmware revision number
Q
Once the above steps have been completed, additional setup of the switches may be performed using Fast Fabric.
3. (Switch) Select the "Externally Managed Switch Admin via Fast Fabric" option from the main menu.
4. (Switch) Select the items shown as "Perform" in the menu below and press the P key to perform them:
SilverStorm Technologies Inc. IB Switch Admin Menu (4.1.1.0.15) Fast Fabric Externally Managed Switch List: /etc/sysconfig/iba/ibnodes
0) Edit Config and Select/Edit Switch Files [Perform]
1) Verify Switch via Firmware dump [ Skip ]
2) Update Switch Firmware [Perform]
3) Reboot Switch [ Skip ]
4) View ibtest result files [ Skip ]
P) Perform the selected actions N) Select None
X) Return to Previous Menu (or ESC)
5. (Switch) "Edit Config and Select/Edit Switch Files" will permit the ibnodes and fastfabric.conf files to be edited. When placed in the editor for fastfabric.conf,
3-10 D000006-000 Rev A
Page 25
Q
3 – Getting Started
review all the settings. Refer to appendix B for more information about fastfabric.conf.
When placed in the editor for ibnodes, create the file with a list of the switch node guids and desired switch names, one entry per line. Such as:
0x00066a00d9000138,edge1
0x00066a00d9000139,edge2
NOTE: Do not list Internally managed chassis, such as the SilverStorm 9000
chassis in this file. Those were covered in the previous section.
NOTE: If the IB path from the IB Management node to other switch nodes is
through a 9024FC which is to be updated, the ibnodes file should omit that 9024FC switch at this time. Otherwise the reboot for the 9024FC in the path could disrupt the updates of other switches.
For further details about the file format, refer to section “Selection of Switches” on page 5-7
If needed, a SA query such as the following can be used to get a list of all switches, however this will include both internally and externally managed switches and hence the output must be edited to leave only the SilverStorm externally-managed switches:
saquery -t sw -o nodeguid
6. (Switch) "Update Switch Firmware" will permit the switch firmware version to be updated and the switch node name set.
NOTE: Consult the relevant switch firmware release notes to ensure any
prerequisites for the upgrade to the new firmware level have been met prior to performing the upgrade via Fast Fabric.
When prompted:
Multiple Firmware files and/or Directories may be space separated
Shell wildcards may be used
For Directories all .emfw files in the directory tree will be used
Enter Files/Directories to use (or none):
specify the directory where the relevant firmware files have been stored. This can be the mount point of the CD or the directory to which the files were copied in a previous step.
Since the fabric is not yet operational, it's recommended to answer "y" to:
Would you like to run the firmware now? [n]:
Fast Fabric will update the firmware on all switches and set the node names as per the ibnodes file created in a previous step. Each switch will then be rebooted.
D000006-000 Rev A 3-11
Page 26
3 – Getting Started Installing InfiniBand on the Remaining Servers
If any switch fails to be updated, use the "View ibtest result files" option to review the result files from the update. Refer to the section “Interpreting the ibtest log files” on page 5-68 for more details.
If some switches were not found review the following for those switches which were not found:
Is switch powered on
Is switch connected to IB network
Is Management node connected to the IB network
Is the SM running on the IB network
Is correct set of switches listed in the ibnodes file (the previous step may be
repeated to review and edit the file as needed)
7. (Switch) If any 9024FC switches were skipped above in step 5 and 6, these steps should be repeated for those switches. In this case it is recommended to create a separate file with a name other than ibnodes. An alternate name may be specified at the prompt:
Q
Select Switch File to Use/Edit
[/etc/sysconfig/iba/ibnodes]:
3.7
Installing InfiniBand on the Remaining Servers
Fast Fabric may now be used to install and configure the remaining hosts and verify overall operation of the fabric.
1. (Linux) Select the "Host Setup via Fast Fabric" option from the main menu.
2. Select the items shown as "Perform" in the menu below and press the P key to perform them:
3-12 D000006-000 Rev A
Page 27
Q
3 – Getting Started
SilverStorm Technologies Inc. IB Host Setup Menu (4.1.1.0.15) Fast Fabric Host List: /etc/sysconfig/iba/hosts
0) Edit Config and Select/Edit Hosts Files [Perform]
1) Verify Hosts via Ethernet ping [Perform]
2) Verify rsh/rcp Configured [ Skip ]
3) Setup Password-less ssh/scp [Perform]
4) Copy /etc/hosts to all hosts [ Skip ]
5) Show uname -a for all hosts [Perform]
6) Install/Upgrade InfiniServ Software [Perform]
7) Configure IPoIB IP Address [Perform]
8) Build MPI Test Apps and Copy to Hosts [Perform]
9) Reboot Hosts [Perform] a) Refresh ssh Known Hosts [ Skip ] b) Rebuild MPI Library and Tools [ Skip ] c) Run a command on all hosts [ Skip ] d) Copy a file to all hosts [ Skip ] e) View ibtest result files [ Skip ]
P) Perform the selected actions N) Select None
X) Return to Previous Menu (or ESC)
NOTE: If passwordless root login via rsh is to be used during fabric setup and
operation, also select "Verify rsh/rcp Configured". However it is instead recommended that ssh be used, in which case this step can be skipped.
NOTE: If /etc/hosts will be used for name resolution (as opposed to using DNS),
also select "Copy /etc/hosts to all hosts"
3. (All) "Edit Config and Select/Edit Hosts Files" will permit the hosts and fastfabric.conf files to be edited. When placed in the editor for fastfabric.conf, review all the settings. Especially review the FF_IPOIB_SUFFIX, .ff_host_basename_to_ipoib, ff_host_basename, FF_IPOIB_NETMASK, FF_PRODUCT, FF_PACKAGES, FF_INSTALL_OPTIONS, FF_UPGRADE_OPTIONS and FF_ALL_ANALYSIS. Consult appendix B for more information about fastfabric.conf.
NOTE: During setup of passwordless ssh, Fast Fabric will provide the opportunity
to enter the host root password interactively when needed. Therefore, it is recommended not to place it within fastfabric.conf. If it is desired to instead keep the root password for the hosts in fastfabric.conf, its recommended to change the fastfabric.conf permissions to be 0x600 (eg. root only access).
D000006-000 Rev A 3-13
Page 28
3 – Getting Started Installing InfiniBand on the Remaining Servers
When placed in the editor for hosts, create the file with a list of the hosts names (the TCP/IP management network names) except the IB Management node from which Fast Fabric is presently being run, one entry per line. Such as:
host1
host2
NOTE: Do not list the IB Management Node itself (i.e., the node where Fast
Fabric is currently running).
If multiple IB Management Nodes are to be used, they may be listed at this time and Fast Fabric can aid in their initial installation and verification.
For further details about the file format, refer to the section “Selection of Hosts” on page 5-3.
4. (All) "Verify Hosts via Ethernet ping" will ping each selected host over the management network. If all hosts were found, continue to the next step. If some hosts were not found, abort out of the menu and review the following for those hosts which were not found:
Q
Is host powered on and booted
Is host connected to management network
Is host management network IP address and network settings consistent with
DNS or /etc/hosts
Is Management node connected to the management network
Are Management node IP address and network settings correct
Is the management network itself up (switches, routers, etc)
Is correct set of hosts listed in the hosts file (the previous step may be repeated
to review and edit the file as needed)
5. (Linux) "Verify rsh/rcp Configured" will confirm that passwordless rsh/rcp is properly configured such that the IB Management Node can access all the other hosts.
NOTE: It is recommended that ssh be used instead in which case this step may
be skipped
6. (Linux) "Setup Password-less ssh/scp" will setup secure password-less ssh such that the IB Management Node can securely login to all the other hosts as root via the management network without requiring a password.
Password-less ssh is required by Fast Fabric, MPI test applications and most versions of MPI (including QuickSilver MPI).
7. (Linux) "Copy /etc/hosts to all hosts" will copy the /etc/hosts file on this host to all the other selected hosts.
3-14 D000006-000 Rev A
Page 29
Q
3 – Getting Started
NOTE: If DNS is being used, this step should be skipped.
NOTE: Typically, /etc/resolv.conf is setup as part of OS installation for each host.
However, if /etc/resolv.conf was not setup on all the hosts during OS installation, the Fast Fabric "Copy a file to all hosts" operation could be done at this time to copy /etc/resolv.conf from the IB Management Node to all the other nodes.
8. (Linux) "Show uname -a for all hosts" will show the OS version on all the hosts. Review the results carefully to verify all the hosts have the expected OS version. In typical clusters all hosts will be running the same OS and kernel version.
If any hosts are identified with an incorrect OS version, the OS on those hosts should be corrected at this time and operation of this sequence should be aborted when prompted. As necessary all the preceding setup steps should then be repeated for those hosts (there is no harm in repeating them for all the hosts).
9. (Host) "Install/Upgrade InfiniServ Software" will install the IB software on all the hosts. By default it will look in the current directory for the $FF_PRODUCT.<VERSION>.tgz file. If it is not found in the current directory, it will prompt for input of a directory name where this file can be found.
When prompted, select to do an initial installation as follows:
Would you like to do an upgrade install? [y]: n
Would you like to do an initial install/load? [n]: y
NOTE: An initial installation will uninstall any existing InfiniServ software on the
selected hosts. An upgrade install is not appropriate at this step.
If any hosts fail to be installed, use the "View ibtest result files" option to review the result files from the update. For more details, see “Interpreting the ibtest log files” on page 5-68.
10. (Host) "Configure IPoIB IP Address" will create the ifcfg-ib1 files on each host. The file will be created with a statically assigned IP address. The IPoIB IP address for each host will be determined by the resolver (Linux host command). If not found via the resolver, /etc/hosts on the given host will be checked.
11. (Host) "Build MPI Test Apps and Copy to Hosts" will build the MPI sample benchmarks on the IB Management Node and copy the resulting object files to all the hosts. This is in preparation for execution of MPI performance tests and benchmarks in a later step.
12. (Linux) "Reboot Hosts" will reboot all the selected hosts and ensure they go down and come back up (as verified via ping over the management network). When the hosts come back up, they will be running the IB software installed.
13. (Linux) If there are any other setup operations which need to be performed on all hosts, they may be performed using the "Run a command on all hosts" option.
D000006-000 Rev A 3-15
Page 30
3 – Getting Started Verifying InfiniBand on the Remaining Servers
Each time this is executed a Linux shell command (or sequence of commands separated by semicolons) may be specified to be executed against all selected hosts.
NOTE: It is recommended at this time to run the "date" command to verify the
the date and time is consistent on all hosts. If needed "Copy a file to all hosts" may be used to copy the appropriate files to all hosts to enable and configure NTP.
3.8
Verifying InfiniBand on the Remaining Servers
Upon completion of the preceeding sections, the hosts are all booted, installed and operational. The subsequent steps will verify the operation of the hosts and fabric.
1. (All) Select the "Host Admin via Fast Fabric" option from the main menu.
2. Select the items shown as "Perform" in the menu below and press the P key to perform them:
Q
SilverStorm Technologies Inc. IB Host Admin Menu (4.1.1.0.15) Fast Fabric Host List: /etc/sysconfig/iba/allhosts
0) Edit Config and Select/Edit Hosts Files [Perform]
1) Verify Hosts via Ethernet ping [Perform]
2) Summary of Fabric Components [Perform]
3) Show Status of Host IB Ports [ Skip ]
4) Verify Hosts see each other [Perform]
5) Verify Hosts ping via IPoIB [Perform]
6) Refresh ssh Known Hosts [Perform]
7) Check MPI Performance [Perform]
8) Generate all Hosts Problem Report Info [ Skip ]
9) Run a command on all hosts [ Skip ] a) View ibtest result files [ Skip ]
P) Perform the selected actions N) Select None
X) Return to Previous Menu (or ESC)
3. (All) "Edit Config and Select/Edit Hosts Files" will permit the hosts and fastfabric.conf files to be edited. When placed in the editor for fastfabric.conf, review all the settings. Especially review the FF_IPOIB_SUFFIX, ff_host_basename_to_ipoib, and ff_host_basename. Consult appendix B for more information about fastfabric.conf.
When placed in the editor for allhosts, create the file with the IB Management node's hosts names (the TCP/IP management network names) (shown as
3-16 D000006-000 Rev A
Page 31
Q
3 – Getting Started
mgmthost below for example) and include the hosts file previously created, one entry per line. Such as:
mgmthost
include /etc/sysconfig/iba/hosts
For further details about the file format refer to section “Selection of Hosts” on page 5-3.
4. (All) "Verify Hosts via Ethernet ping" will ping each selected host over the management network. If all hosts were found, continue to the next step. If some hosts were not found, abort out of the menu and review the following for those hosts which were not found:
Is the host powered on and booted
Is the host connected to management network
Is the host management network IP address and network settings consistent
with DNS or /etc/hosts
Is the Management node connected to the management network
Are the Management node IP address and network settings correct
Is the management network itself up (switches, routers, etc)
Is the correct set of hosts listed in the hosts file (the previous step may be
repeated to review and edit the file as needed)
5. (All) "Summary of Fabric Components" will provide a brief summary of the counts of components in the fabric including how many switch chips, hosts, and links are in the fabric. It will also indicate if any 1x links were found (which could indicate a poorly seated or bad cable). Review the results against the expected configuration of the cluster.
NOTE: The link count includes some internal links within the switch boxes. This
means that the count displayed will be greater than the actual number of cables.
If components are missing or 1x links are found, they should be corrected. Subsequent steps will aid in locating any 1x links.
6. (Host) If desired "Show Status of Host IB Ports" will allow the state and symbol error counts of all ports to be manually reviewed.
Instead it is recommended to run:
iba_report -i 10 -o errors -o slowlinks
on the IB Management node. This will check all the ports in the fabric for any links which have high error rates or are running at a lower speed than expected. Any such identified links should be diagnosed and corrected.
D000006-000 Rev A 3-17
Page 32
3 – Getting Started Complete Installation of additional IB Management Nodes
7. (Host) "Verify Hosts see each other" will verify that each host can see all the others via queries to the Subnet Administrator and the SA replica on each host has been fully populated.
8. (Host) "Verify Hosts ping via IPoIB" will verify that IPoIB is properly configured and running on all the hosts. This is accomplished via the IB management node pinging each host via IPoIB.
9. (Linux) "Refresh ssh Known Hosts" will refesh the ssh knowhosts file on the IB management node to include the IPoIB hostnames of all the hosts
10. (Host) "Check MPI Performance" will do a quick check of PCI and MPI performance.
This displays the MPI latency and bandwidth between pairs of hosts (1-2, 3-4, 5-6, etc). The results are also written to the test.res file which may be viewed via the "View ibtest result files". Refer to the section “Interpreting the ibtest log files” on page 5-68 for more details.
The numbers reported should be checked against the practical PCI speeds in the Performance Impact table below. If any pairs are not in the expected performance range, carefully examine the two hosts involved to verify that the PCI slot used, BIOS settings and any motherboard jumpers related to devices on PCI buses or slot speeds. Also verify HCA and riser cards are properly seated.
Q
Table 3-1. Performance Impact
PCI Speed Theoretical Max Practical Bandwidth
133 1024MB/sec 800-900 MB/sec
100 770MB/sec 600-680 MB/sec
66 512MB/sec 400-450 MB/sec
3.9
Complete Installation of additional IB Management Nodes
If the fabric is to have more than one IB Management Node, the setup of the additional management nodes may be completed now. The previous steps will have performed basic software installation, setup and verification on those nodes. Now the management software itself must be installed and configured.
NOTE: The steps below assume a symmetrical configuration where all IB
management nodes have the same connectivity and capabilities. In assymetrical configurations where the IB management nodes are not all connected to the same set of management networks and IB subnets, the files copied to each management node may need to be slightly different. For example configuration files for fabric_analysis may indicate different port numbers, or host files used for FastFabric and MPI may need to list different hosts. For multiple subnet configurations, refer to “Multi-Subnet Fabrics” on page D-1.
3-18 D000006-000 Rev A
Page 33
Q
3 – Getting Started
Repeat the following steps on each additional IB Management Node:
1. (All) Install the additional Fabric Access Software components using the procedure documented in the Fabric Access Software Users Guide. The IB Management Node must have at least Fast Fabric, the IB Stack and IPoIB installed and configured. For MPI clusters the IB Management Node should also include the MPI Runtime and MPI Development packages, and if the user desires to rebuild MPI itself, the IB Development package and MPI Source packages will also be required.
NOTE: Do not uninstall or replace existing configuration files which were previous
created, especially IPoIB related configuration files.
2. (All) Copy the Fast Fabric configuration files from the initial IB Management Node. At least the following files should be copied:
/etc/sysconfig/fastfabric.conf
/etc/sysconfig/iba/hosts
/etc/sysconfig/iba/allhosts
/etc/sysconfig/iba/ibnodes
/etc/sysconfig/iba/chassis
After copying the files, edit the hosts and allhosts files such that the file on each IB Management Node omits itself from the hosts files (but lists all other IB Management Nodes) and specifies itself in the allhosts file.
See appendix B for a complete list of Fast Fabric configuration files.
3. (Linux) Perform "Setup Password-less ssh/scp" in the "Host Setup via Fast Fabric" menu and "Refresh ssh Known Hosts" in the "Host Admin via Fast Fabric" menu.
3.10
Configure and Initialize Health Check Tools
For more information on the health check tools, see the detailed discussion in “Health Check and Baselining Tools” on page 5-69. The Health check tools may be run on one or more IB management nodes within the cluster. This procedure should be followed on each IB management node from which the health check tools will be used.
1. (All) Edit fastfabric,conf and review the following parameters: FF_ANALYSIS_DIR, FF_ALL_ANALYSIS, FF_FABRIC_HEALTH, FF_CHASSIS_CMDS,_FF_CHASSIS_HEALTH, and FF_ESM_CMDS. FF_ALL_ANALYSIS should be updated to reflect the type of SM (esm or hostsm).
2. (All) If using Embedded SM(s) in QLogic IB Chassis, create /etc/sysconfig/iba/esm_chassis listing the chassis which are running SMs.
D000006-000 Rev A 3-19
Page 34
3 – Getting Started Running HPL
Q
Create the file with a list of the chassis names (the TCP/IP Ethernet management port names assigned above) or IP addresses (Use of names is recommended). One entry per line. Such as:
Chassis1
Chassis2
For further details about the file format refer to the section “Selection of Chassis” on page 5-4.
3. (All) Perform a health check using: all_analysis -e. If any errors are encountered resolve the errors and rerun all_analysis -e until a clean run occurs.
4. (All) Create a cluster configuration baseline using: all_analysis -b
5. (All) If desired, schedule regular runs of all_analysis via cron or other mechanisms. Consult the Linux OS documentation for more information on cron. Also consult the section “Health Check and Baselining Tools” on page 5-69 for more information about all_analysis and its automated use.
3.11
Running HPL
As part of the installation process, a set of common MPI benchmarks have been installed. One of the more popular measures of overall performance is HPL. This is the application used to rate systems on the Top 500 list. The steps allow some initial runs of HPL to be made and provide some initial baseline numbers. The defaults provided should perform within 10%-20% of optimal HPL results for the cluster. Tuning for that additional 10%-20% is beyond the scope of this document.
1. (Host) To run HPL, first select a configuration file appropriate to your cluster.
2. (Host) Now create the file /opt/iba/src/mpi_apps/mpi_hosts listing the host
3. (Host) Now run HPL:
It is best to start with a small configuration to verify HPL has been properly compiled:
a. cd /opt/iba/src/mpi_apps
b. /config_hpl 2t
will configure a two process test run of HPL.
names of all the hosts. Depending of your selection of VIADEV_PATH_METHOD in /opt/iba/src/mpi_apps/mpi.param.hpl the user can specify Ethernet or IPoIB host names. The default config will allow either.
./run_hpl 2
Since this is a very small problem size the performance of the run will be much lower than the potential of the machine. So do not worry about performance, just whether or not the run was successful.
3-20 D000006-000 Rev A
Page 35
Q
3 – Getting Started
At this point the user is ready to move onto full scale HPL runs. Assorted sample HPL.dat files are provided in /opt/iba/src/mpi_apps/hpl-config. These files are a good starting point for most clusters and should get within 10-20% of the optimal performance for the cluster. The problem sizes used assume a cluster with 1GB of physical memory per processor (e.g., for a 2 processor node, 2GB of node memory is assumed). For each cluster size, 4 files are provided:
t - a very small test run (5000 problem size)
s - a small problem size on the low end of optimal problem sizes
m - a medium problem size
l - a large problem size
These can be selected using config_hpl. The following command displays the preconfigured problem sizes available:
./config_hpl
For example, to do a small run for a 256 processor cluster (i.e., 128 nodes of dual CPU systems):
1. Type ./config_hpl 256s and press Enter.
2. Type ./run_hpl 256 and press Enter.
During these runs the user should use top on a node to monitor memory and CPU usage. The xhpl should use 98-99% of the CPU. If any other processes are taking more than 1-2%, review the host configuration and stop these extra processes if possible. HPL is very sensitive to swapping. If a lot of swapping is seen, and xhpl is dropping below 97% for long durations, this may indicate a problem size that is too large for the memory and OS configuration.
At this point the user can continue to tune HPL to refine performance. Parameters in /opt/iba/src/mpi_apps/mpi.param.hpl and in HPL.dat can all affect HPL performance. In addition the selection of compiler and BLAS Math library may also significantly affect performance. The new HPL.dat files may be placed in /opt/iba/src/mpi_apps/hpl-config and use config_hpl to select them and copy them to all nodes in the run. Alternately scpall may be used to copy the file to all nodes. Refer to the section “Basic Setup and Administration Tools” on page 5-11 for more information on scpall.
3.12
Upgrading IB software
If an existing cluster which has been installed and verified needs to have Infiniband software upgraded, the following steps may be followed.
1. (All) On each IB Management Node, perform an upgrade installation of the Fabric Access Software using the procedure documented in the Fabric Access Software Users Guide. Each IB Management Node must have at least Fast Fabric, the IB Stack and IPoIB installed and configured. For MPI clusters the
D000006-000 Rev A 3-21
Page 36
3 – Getting Started Upgrading IB software
NOTE: Ensure that existing configuration is appropriately upgraded, especially
2. (All) Select the "Host Setup via Fast Fabric" option from the main menu.
3. Select the items shown as "Perform" in the menu below and press the P key
SilverStorm Technologies Inc. IB Host Setup Menu (4.1.1.0.15) Fast Fabric Host List: /etc/sysconfig/iba/hosts
0) Edit Config and Select/Edit Hosts Files [Perform]
1) Verify Hosts via Ethernet ping [ Skip ]
2) Verify rsh/rcp Configured [ Skip ]
3) Setup Password-less ssh/scp [ Skip ]
4) Copy /etc/hosts to all hosts [ Skip ]
5) Show uname -a for all hosts [ Skip ]
6) Install/Upgrade InfiniServ Software [Perform]
7) Configure IPoIB IP Address [ Skip ]
8) Build MPI Test Apps and Copy to Hosts [ Skip ]
9) Reboot Hosts [Perform] a) Refresh ssh Known Hosts [ Skip ] b) Rebuild MPI Library and Tools [ Skip ] c) Run a command on all hosts [ Skip ] d) Copy a file to all hosts [ Skip ] e) View ibtest result files [ Skip ]
Q
IB Management Nodes should also include the MPI Runtime and MPI Development packages, and if the user desires to rebuild MPI itself, the IB Development package and MPI Source packages will also be required.
After completing the install, reboot each of the IB Management Nodes to ensure they are running the new IB software.
Fast Fabric and IPoIB related configuration files. Consult the Fabric Access Software Users Guide and release notes for further information.
to perform them:
P) Perform the selected actions N) Select None
X) Return to Previous Menu (or ESC)
4. (All) "Edit Config and Select/Edit Hosts Files" will permit the hosts and fastfabric.conf files to be edited. When placed in the editor for fastfabric.conf, review all the settings. Especially review the FF_PRODUCT, FF_PACKAGES, and FF_UPGRADE_OPTIONS. See appendix B for more information about fastfabric.conf.
Select a hosts list file which lists all the hosts except the IB Management nodes. If necessary create a new file at this time, potentially based on the existing /etc/sysconfig/iba/hosts file.
3-22 D000006-000 Rev A
Page 37
Q
3 – Getting Started
NOTE: Do not list any of IB Management Nodes (eg. The nodes which have
fast fabric installed)
NOTE: The file may list the Management Network or IPoIB hostnames for the
selected hosts
5. (Host) "Install/Upgrade InfiniServ Software" will upgrade the IB software on all the selected hosts. By default it will look in the current directory for the $FF_PRODUCT.<VERSION>.tgz file. If it is not found in the current directory, it will prompt for input of a directory name where this file can be found.
When prompted, select to do an upgrade installation as follows:
Would you like to do an upgrade install? [y]: y
NOTE: An upgrade installation will update any existing InfiniServ software on the
selected hosts. An upgrade install is only valid for hosts which already have a previous version of InfiniServ software installed.
If any hosts fail to be updated, use the "View ibtest result files" option to review the result files from the update. See the section “Interpreting the ibtest log files” on page 5-68 for more details.
6. (Linux) If there are any other setup operations which need to be performed on all hosts, they may be performed using the "Run a command on all hosts" option. Each time this is executed a Linux shell command (or sequence of commands separated by semicolons) may be specified to be executed against all selected hosts.
NOTE: NOTE: Check the relevant release notes for the new InfiniServ release
being installed for any such additional required steps.
7. (Linux) "Reboot Hosts" will reboot all the selected hosts and ensure they go down and come back up (as verified via ping over the management network). When the hosts come back up, they will be running the IB software installed.
8. Repeat the verification steps for the fabric as discussed in the section “Verifying InfiniBand on the Remaining Servers” on page 3-16.
.
D000006-000 Rev A 3-23
Page 38
3 – Getting Started Upgrading IB software
Q
3-24 D000006-000 Rev A
Page 39
Section 4
Fast Fabric TUI Menu
Fast Fabric is easiest to use from the textual user interface (TUI) menu system. The menu system provides a way to perform all common tasks and presents common options. Additional less common options are available directly via the Command Line Tools documented in the next section.
In the sections that follow, the menu system will be discussed. The majority of menu items directly invoke various Fast Fabric command tools. As such the section on each menu item will indicate what command tool it invokes and a summary of the operation performed. For further details about the given command tool, consult the relevant section within “Basic Setup and Administration Tools” on page 5-11.
Some of the menu items are only applicable when Linux is being used. They will be marked with (Linux). Similarly some of the menu items are only applicable when QuickSilver Linux IB software is being used on the hosts. Those will be marked with (Host). All menu items which are applicable only when SilverStorm IB Switches or Chassis are being used will be marked with (Switch). All remaining menu items are generally applicable to all environments and will be marked with (All).
NOTE: Some of the Linux menu items may be applicable to other Unix-like
operating systems if it is desired to enable the use of non-IB specific Fast Fabric tools (such as cmdall) against the given hosts.
The main menu can be invoked using the iba_config command. The main menu is as follows:
SilverStorm Technologies Inc. InfiniBand 4.1.1.0.15 Software
1) Show Installed Software
2) Reconfigure IP over IB
3) Reconfigure Driver Autostart
4) Update HCA Firmware
5) Generate Supporting Information for Problem Report
6) Host Setup via Fast Fabric
7) Host Admin via Fast Fabric
8) Chassis Admin via Fast Fabric
9) Externally Managed Switch Admin via Fast Fabric a) Uninstall Software
X) Exit
In the above menu, items 6-9 represent the Fast Fabric menus. The operation of this menu is the same as the INSTALL and iba_config functions documented in the
D000006-000 Rev A 4-1
Page 40
4 – Fast Fabric TUI Menu
QuickSilver Fabric Access Software Users Guide. Selecting items 1-9 will display the given submenu. Pressing X will exit the menu system.
Selection of a Fast Fabric menu (6-9) will present a submenu such as below:
SilverStorm Technologies Inc. IB Host Setup Menu (4.1.1.0.15) Fast Fabric Host List: /etc/sysconfig/iba/hosts
0) Edit Config and Select/Edit Hosts Files [Perform]
1) Verify Hosts via Ethernet ping [Perform]
2) Verify rsh/rcp Configured [ Skip ]
3) Setup Password-less ssh/scp [Perform]
4) Copy /etc/hosts to all hosts [ Skip ]
5) Show uname -a for all hosts [Perform]
6) Install/Upgrade InfiniServ Software [Perform]
7) Configure IPoIB IP Address [Perform]
8) Build MPI Test Apps and Copy to Hosts [Perform]
9) Reboot Hosts [Perform] a) Refresh ssh Known Hosts [Perform] b) Rebuild MPI Library and Tools [ Skip ] c) Run a command on all hosts [ Skip ] d) Copy a file to all hosts [ Skip ] e) View ibtest result files [ Skip ]
Q
P) Perform the selected actions N) Select None
X) Return to Previous Menu (or ESC)
The submenus typically present operations in the typical order they would be used during an installation. Pressing the keys corresponding to menu items (0-9, a-e in the example above) will toggle the Skip/Perform selection for the given item. As shown in the example above, more than 1 item may be selected. Once the desired set of items have been selected, press P. To unselect all items, press N. Pressing X or ESC will exit this menu and return to the Main Menu.
If more than 1 item is selected, the items will be performed in the order shown in the menu. This is the typical order desired during fabric setup. If it's desired to perform items in a different order, select a single item and press P to perform it by itself. Then repeat. An opportunity will be presented after each item to abort:
Hit any key to continue (or ESC to abort)...
If ESC is pressed, the sequence of operations will be aborted and return to the previous menu. Any other key will result in the next selected menu item being performed. This prompt is also shown after the last selected item completes, hence permitting an opportunity to review the results before the screen is cleared to display the menu.
4-2 D000006-000 Rev A
Page 41
Q
At the top of each Fast Fabric menu, the file listing the components to operate on is shown. For example:
Fast Fabric Host List: /etc/sysconfig/iba/hosts
On each Fast Fabric menu, item 0 will permit a different file to be selected and will permit the editing of the file (using the editor selected via the EDITOR environment variable). In addition it will also permit review and editing of the fastfabric.conf file. The fastfabric.conf file guides the overall configuration of Fast Fabric and describes cluster specific attributes of how Fast Fabric will operate. It is discussed in greater detail in appendix B.
During the execution of each menu selection, the actual Fast Fabric command line tool being used will be shown. This can be used as an educational aid to learn the tools.
4.1
Host Setup via Fast Fabric
This menu is focused on initial host setup and installation of IB software on all the hosts.
4 – Fast Fabric TUI Menu
D000006-000 Rev A 4-3
Page 42
4 – Fast Fabric TUI Menu Host Setup via Fast Fabric
0) Edit Config and Select/Edit Hosts Files [ Skip ]
1) Verify Hosts via Ethernet ping [ Skip ]
2) Verify rsh/rcp Configured [ Skip ]
3) Setup Password-less ssh/scp [ Skip ]
4) Copy /etc/hosts to all hosts [ Skip ]
5) Show uname -a for all hosts [ Skip ]
6) Install/Upgrade InfiniServ Software [ Skip ]
7) Configure IPoIB IP Address [ Skip ]
8) Build MPI Test Apps and Copy to Hosts [ Skip ]
9) Reboot Hosts [ Skip ] a) Refresh ssh Known Hosts [ Skip ] b) Rebuild MPI Library and Tools [ Skip ] c) Run a command on all hosts [ Skip ] d) Copy a file to all hosts [ Skip ] e) View ibtest result files [ Skip ]
P) Perform the selected actions N) Select None
X) Return to Previous Menu (or ESC)
Q
4.1.1
Edit Configuration and Select/Edit Hosts Files
(All) This will permit the hosts and fastfabric.conf files to be edited. The hosts file
selected and created via this menu should not list the Fast Fabric host itself. After editing the two files, an opportunity is given to edit them again or continue forward.
Selected Host File: /etc/sysconfig/iba/hosts
Do you want to edit/review/change the files? [y]:
The default will repeat the editing process, answer "n" to proceed to continue forward.
Refer to the section “Selection of Hosts” on page 5-3 for more details about the format of the hosts file.
4.1.2
Verify Hosts via Ethernet ping
(All) This will run the pingall command. All the hosts listed will be pinged via the
Management Network.
4.1.3
Verify RSH/RCP Configured
(Linux) This will run the check_rsh command. This will confirm that passwordless
rsh/rcp is properly configured such that the IB Management Node can access all the other hosts.
4-4 D000006-000 Rev A
Page 43
Q
NOTE: It is recommended that SSH be used in place of the check_rsh
command.
4.1.4
Setup Password-less SSH/SCP
(Linux) This will run the setup_ssh -i "" command. This will setup secure
password-less SSH such that the IB Management Node can securely login to all the other hosts as root via the management network without requiring a password.
Password-less SSH is required by Fast Fabric, MPI test applications and most versions of MPI (including QuickSilver MPI).
4.1.5
Copy /etc/hosts to all hosts
(Linux) This will run the scpall /etc/hosts /etc/hosts command to copy the /etc/hosts
file on this host to all the other selected hosts. This is not necessary when using a DNS server to resolve hostnames for the cluster.
4 – Fast Fabric TUI Menu
4.1.6
Show uname -a for all hosts
(Linux) This run the cmdall "uname -a" command to show the OS version on
all the hosts. Review the results carefully to verify all the hosts have the expected OS version. In typical clusters all hosts will be running the same OS and kernel version.
4.1.7
Install/Upgrade QuickSilver Software
(Host) This will run the ibtest load or ibtest update command to install the
IB software on all the hosts. By default it will look in the current directory for the $FF_PRODUCT.<VERSION>.tgz file. If it is not found in the current directory, it will prompt for input of a directory name where this file can be found.
Prompts will guide the user through options:
upgrade - updates all servers with new release. Only components previously
installed are upgraded. Will fail for any hosts which have no InfiniServ IB software installed
initial install/load - uninstalls any existing InfiniServ IB software and installs the
given release based on fastfabric.conf installation options specified.
After the install is completed, the hosts will still need to be rebooted to bring up the new IB drivers. This can be performed using the "Reboot Hosts" menu item discussed below.
D000006-000 Rev A 4-5
Page 44
4 – Fast Fabric TUI Menu Host Setup via Fast Fabric
If any hosts fail to be updated, use the View ibtest result files option to review the result files from the update. For more details, see “Interpreting the ibtest log files” on page 5-68.
4.1.8
Configure IPoIB IP Address
(Host) This will run the ibtest configipoib command to create the ifcfg-ib1 files
on each host. The file will be created with a statically assigned IP address. The IPoIB IP address for each host will be determined by the resolver (Linux host command). If not found via the resolver, /etc/hosts on the given host will be checked.
4.1.9
Build MPI Test Apps and Copy to Hosts
(Host) This will build the MPI sample benchmarks on the IB Management Node
and copy the resulting object files to all the hosts. This is in preparation for execution of MPI performance tests and benchmarks in a later step.
Q
4.1.10
Reboot Hosts
(Linux) This will run the ibtest reboot command to reboot all the selected hosts
and ensure they go down and come back up (as verified via ping over the management network). When the hosts come back up, they will be running the IB software installed.
4.1.11
Refresh SSH Known Hosts
(Linux) This will run the setup_ssh -C -i"" command to refresh the ssh known
hosts list on this server for the Management Network. This may be used to update security for this host if hosts are replaced, reinstalled, renamed, or repaired.
4.1.12
Rebuild MPI Library and Tools
(Host) This will rebuild the InfiniServ MPI Library itself and related tools (such as
mpirun). This will be performed via the dobuild tool supplied with the InfiniServ MPI Source. Consult the QuickSilver Fabric Access Software Users Guide for more information.
4-6 D000006-000 Rev A
Page 45
Q
4.1.13
Run a command on all hosts
(Linux) This will run the cmdall command. A Linux shell command (or sequence
of commands separated by semicolons) may be specified to be executed against all selected hosts.
4.1.14
Copy a file to all hosts
(Linux) This will run the scpall command. A file on the local host may be specified
to be copied to all selected hosts.
4.1.15
View ibtest result files
(All) This permits viewing of the test.log and test.res files that reflect the
results from ibtest runs (such as for installing QuickSilver software or rebooting all hosts per menu items above). The user is also given the option to remove these files after viewing them.
4 – Fast Fabric TUI Menu
If not removed, subsequent runs of ibtest from within the current directory will continue to append to these files.
D000006-000 Rev A 4-7
Page 46
4 – Fast Fabric TUI Menu Host Admin via Fast Fabric
4.2
Host Admin via Fast Fabric
This menu is focused on verifying hosts and the fabric as well as administration of all the hosts.
SilverStorm Technologies Inc. IB Host Admin Menu (4.1.1.0.15) Fast Fabric Host List: /etc/sysconfig/iba/allhosts
0) Edit Config and Select/Edit Hosts Files [ Skip ]
1) Verify Hosts via Ethernet ping [ Skip ]
2) Summary of Fabric Components [ Skip ]
3) Show Status of Host IB Ports [ Skip ]
4) Verify Hosts see each other [ Skip ]
5) Verify Hosts ping via IPoIB [ Skip ]
6) Refresh ssh Known Hosts [ Skip ]
7) Check MPI Performance [ Skip ]
8) Generate all Hosts Problem Report Info [ Skip ]
9) Run a command on all hosts [ Skip ] a) View ibtest result files [ Skip ]
Q
P) Perform the selected actions N) Select None
X) Return to Previous Menu (or ESC)
4.2.1
Edit Config and Select/Edit Hosts Files
All This will permit the allhosts and fastfabric.conf files to be edited. The allhosts file selected and created via this menu should list the Fast Fabric host itself. After editing the two files, an opportunity is given to edit them again or continue forward.
Selected Host File: /etc/sysconfig/iba/allhosts
Do you want to edit/review/change the files? [y]:
The default will repeat the editing process, answer n to roceed to continue forward.
Refer to the section “Selection of Hosts” on page 5-3 for more details about the format of the allhosts file.
4.2.2
Verify Hosts via Ethernet Ping
(All) This will run the pingall command. All the hosts listed will be pinged via the
Management Network.
4-8 D000006-000 Rev A
Page 47
Q
4.2.3
Summary of Fabric Components
(All) This will run the fabric_info command to provide a brief summary of the
counts of components in the fabric including how many switch chips, hosts, and links are in the fabric. It will also indicate if any 1x links were found (that could indicate a poorly seated or bad cable). Review the results against the expected configuration of the cluster.
NOTE: The link count includes some internal links within the switch boxes. This
means that the count displayed will be greater than the actual number of cables.
4.2.4
Show Status of Host IB Ports
(Host) This will run the showallports command to allow the state and symbol error
counts of all host ports to be manually reviewed.
(All) Instead it is recommended to run:
4 – Fast Fabric TUI Menu
iba_report -i 10 -o errors -o slowlinks
on the IB Management node. This will check all the ports in the fabric for any links which have high error rates or are running at a lower speed than expected. Any such identified links should be diagnosed and corrected.
4.2.5
Verify Hosts see each other
(Host) This will run the ibtest sacache command to verify that each host can
see all the others via queries to the Subnet Administrator and the SA replica on each host has been fully populated.
4.2.6
Verify Hosts ping via IPoIB
(Host) This will run the ibtest ipoibping command to verify that IPoIB is
properly configured and running on all the hosts. This is accomplished via the IB management node pinging each host via IPoIB.
4.2.7
Refresh SSH Known Hosts
(Linux) This will run the setup_ssh -C command to refresh the SSH known hosts
list on this server for the IPoIB and Management Networks. This may be used to update security for this host if hosts are replaced, reinstalled, renamed, or repaired.
D000006-000 Rev A 4-9
Page 48
4 – Fast Fabric TUI Menu Host Admin via Fast Fabric
4.2.8
Check MPI Performance
(Host) This will run the ibtest mpiperf command to do a quick check of PCI
and MPI performance.
This displays the MPI latency and bandwidth between pairs of hosts (1-2, 3-4, 5-6, etc). The results are also written to the test.res file which may be viewed via the View ibtest result files. Refer to the section “Interpreting the ibtest log files” on page 5-68 for more details.
The numbers reported should be checked against the practical PCI speeds in the Performance Impact section. If any pairs are not in the expected performance range, carefully examine the two hosts involved to verify that the PCI slot used, BIOS settings and any motherboard jumpers related to devices on PCI buses or slot speeds. Also verify HCA and riser cards are properly seated.
4.2.9
Generate all Hosts Problem Report Info
Q
(Host) This will run the captureall command to collect configuration and status information from all hosts and generate a single *.tgz file which can be sent to the Support Representative.
4.2.10
Run a command on all hosts
(Linux) This will run the cmdall command. A Linux shell command (or sequence
of commands separated by semicolons) may be specified to be executed against all selected hosts.
4.2.11
View ibtest result files
(All) This permits viewing of the test.log and test.res files which reflect the
results from ibtest runs (such as those for installing QuickSilver software or rebooting all hosts per menu items above). The user is also given the option to remove these files after viewing them.
If not removed, subsequent runs of ibtest from within the current directory will continue to append to these files.
4-10 D000006-000 Rev A
Page 49
Q
4.3
QLogic IB Chassis Admin via Fast Fabric
This menu is focused on administration of QLogic 9000 series internally managed IB chassis.
SilverStorm Technologies Inc. IB Chassis Admin Menu (4.1.1.0.15) Fast Fabric Chassis List: /etc/sysconfig/iba/chassis
0) Edit Config and Select/Edit Chassis Files [ Skip ]
1) Verify Chassis via Ethernet ping [ Skip ]
2) Update Chassis Firmware [ Skip ]
3) Show Status of Chassis IB Ports [ Skip ]
4) Reboot Chassis [ Skip ]
5) Generate all Chassis Problem Report Info [ Skip ]
6) Run a command on all chassis [ Skip ]
7) View ibtest result files [ Skip ]
P) Perform the selected actions N) Select None
4 – Fast Fabric TUI Menu
X) Return to Previous Menu (or ESC)
4.3.1
Edit the Configuration and Select/Edit Chassis Files
(Switch) This will permit the chassis and fastfabric.conf files to be edited.
The chassis file selected and created via this menu should not list the Fast Fabric host itself. After editing the two files, an opportunity is given to edit them again or continue forward.
Selected Chassis File: /etc/sysconfig/iba/chassis
Do you want to edit/review/change the files? [y]:
The default will repeat the editing process, answer n to proceed to continue forward.
Refer to section “Selection of Chassis” on page 5-4 for more details about the format of the chassis file.
4.3.2
Verify Chassis via Ethernet Ping
(Switch) This will run the pingall -C command to ping each selected chassis
over the management network.
D000006-000 Rev A 4-11
Page 50
4 – Fast Fabric TUI Menu QLogic IB Chassis Admin via Fast Fabric
4.3.3
Update Chassis Firmware
(Switch) This will run the ibtest -C update command to permit the chassis
firmware version to be verified and updated as needed.
NOTE: The chassis must be running firmware version 4.0.0.4.3 or later to perform
this function. If the chassis is not up to this level, it will need to be manually updated via the chassis GUI. See the SilverStorm 9000 Users Guide for more information.
NOTE: Consult the relevant chassis firmware release notes to ensure any
prerequisites for the upgrade to the new firmware level have been met prior to performing the upgrade via Fast Fabric.
Prompts will guide the user through options:
push - push firmware to each chassis but do not change selected nor running
firmware
select - push firmware to each chassis and select it for use on next reboot
Q
run - push firmwarew to each chassis, select it for use and if its not the presently
running firmware, reboot the chassis
Additional options prompted for:
parallel vs serial update
selection of firmware files or directory containing .pkg files
prompting for chassis password (default is to have password in fastfabric.conf)
If any chassis fails to be updated, use the View ibtest result files option to review the result files from the update. Refer to the section “Interpreting the ibtest log files” on page 5-68 for more details.
4.3.4
Show Status of Chassis IB Ports
(Switch) This will run the showallports -C command to allow the state and symbol
error counts of all chassis ports to be manually reviewed.
(All) Instead it is recommended to run:
iba_report -i 10 -o errors -o slowlinks
on the IB Management node. This will check all the ports in the fabric for any links which have high error rates or are running at a lower speed than expected. Any such identified links should be diagnosed and corrected.
4-12 D000006-000 Rev A
Page 51
Q
4.3.5
Reboot Chassis
(Switch) This will run the ibtest -C reboot command to reboot all the selected
chassis and ensure they go down and come back up (as verified via ping over the management network).
4.3.6
Generate all Chassis Problem Report Information
(Switch) This will run the captureall -C command to collect configuration and
status information from all chassis and generate a single *.tgz file that can be sent to the Support Representative.
4.3.7
Run a command on all chassis
(Switch) This will run the cmdall -C command. A Chassis CLI command may
be specified to be executed against all selected chassis.
4 – Fast Fabric TUI Menu
4.3.8
View ibtest results files
(All) This permits viewing of the test.log and test.res files which reflect the
results from ibtest runs (such as for updating Chassis Firmware or rebooting all chassis per menu items above). The user is also given the option to remove these files after viewing them.
If not removed, subsequent runs of ibtest from within the current directory will continue to append to these files.
D000006-000 Rev A 4-13
Page 52
4 – Fast Fabric TUI Menu SilverStorm Externally Managed IB Switch Administration via Fast Fabric
4.4
Q
SilverStorm Externally Managed IB Switch Administration via Fast Fabric
This menu is focused on administration of SilverStorm 9024FC externally managed switches.
SilverStorm Technologies Inc. IB Switch Admin Menu (4.1.1.0.15) Fast Fabric Externally Managed Switch List: /etc/sysconfig/iba/ibnodes
0) Edit Config and Select/Edit Switch Files [ Skip ]
1) Verify Switch via Firmware dump [ Skip ]
2) Update Switch Firmware [ Skip ]
3) Reboot Switch [ Skip ]
4) View ibtest result files [ Skip ]
P) Perform the selected actions N) Select None
X) Return to Previous Menu (or ESC)
4.4.1
3.4.1Edit Config and Select/Edit Chassis Files
(Switch) This will permit the ibnodes and fastfabric.conf files to be edited.
The ibnodes file selected and created via this menu should not list the Fast Fabric host itself. After editing the two files, an opportunity is given to edit them again or continue forward.
Selected Chassis File: /etc/sysconfig/iba/chassis
Do you want to edit/review/change the files? [y]:
The default will repeat the editing process, answer n to proceed to continue forward.
Refer to the section “Selection of Switches” on page 5-7 for more details about the format of the ibnodes file.
4.4.2
Verify Switch via Firmware Dump
(Switch) Use of this option is not recommended.
4.4.3
Update Switch Firmware
(Switch) This will run the ibtest -n upgrade command to permit the switch
firmware version to be updated and the switch node name to be set.
4-14 D000006-000 Rev A
Page 53
Q
NOTE: Consult the relevant switch firmware release notes to ensure any
Prompts will guide the user through options:
select - push firmware to each switch and select it for use on next reboot
run - push firmware to each switch, select it for use and reboot switches
Additional options prompted for:
parallel vs serial update
select of firmware files or directory containing .emfw files
If any switches fail to be updated, use the View ibtest result files option to review the result files from the update. Refer to the section “Interpreting the ibtest log files” on page 5-68 for more details.
4.4.4
Reboot Switch
4 – Fast Fabric TUI Menu
prerequisites for the upgrade to the new firmware level have been met prior to performing the upgrade via Fast Fabric.
(Switch) This will run the ibtest -n reboot command to reboot all the selected switches.
4.4.5
View ibtest result files
(All) This permits viewing of the test.log and test.res files that reflect the
results from ibtest runs (such as those for updating Switch Firmware or rebooting all switches per menu items above). The user is also given the option to remove these files after viewing them.
If not removed, subsequent runs of ibtest from within the current directory will continue to append to these files.
D000006-000 Rev A 4-15
Page 54
4 – Fast Fabric TUI Menu SilverStorm Externally Managed IB Switch Administration via Fast Fabric
Q
4-16 D000006-000 Rev A
Page 55
Detailed Descriptions of Command LineTools
Some of the commands are only applicable when Linux is being used. They will be marked with (Linux). Similarly some of the commands are only applicable when QuickSlver Linux IB software is being used on the hosts. Those will be marked with (Host). All commands which are applicable only when SilverStorm IB Switches or IB Chassis are being used will be marked with (Switch). All remaining commands are generally applicable to all environments and will be marked with (All).
NOTE: Some of the Linux commands may be applicable to other Unix-like
operating systems if it is desired to enable use of non-IB specific Fast Fabric tools (such as cmdall) against the given hosts.
The Fast Fabric tools are installed in directories which are part of the standard Linux root PATH. Most of the tools are installed in /sbin.
5.1
Common Tool Options
There are some common options to the assorted command line tools. These options are applicable to most of the tools:
Section 5
5.1.1
-?
5.1.2
-p
Will display Usage information for any of the commands (as will any invalid option)
Runs the operation/command in parallel. This means the operation is performed simultaneously on batches of 20 hosts. As such this option allows the overall time of an operation to be much lower. However, a side effect is that any output from the command will be bursty and intermingled. Therefore this option should be used for commands where there is no output or the output is of limited interest. For some commands (such as scpall), this will perform the operation in a quiet mode to limit output. If the user wants to change the number of parallel operations export TEST_MAX_PARALLEL=# where # is the new number (such as 30).
For more advanced operations (such as ibtest), parallel operation is the default mode.
Parallel operation can also be disabled by setting FF_MAX_PARALLEL to 1.
D000006-000 Rev A 5-1
Page 56
5 – Detailed Descriptions of Command LineTools Common Tool Options
5.1.3
-S
Prompt for password for admin on chassis. By default Fast Fabric operations against SilverStorm chassis (such as cmdall, captureall, showallports, and ibtest) obtain the chassis admin password from the FF_CHASSIS_ADMIN_PASSWORD environment variable which may be directly exported or part of fastfabric.conf. Alternatively the -S option may be used on these commands in which case the chassis admin password will be prompted for interactively. The password is prompted for once and the same password is then used to login to each chassis during the operation.
NOTE: Newer versions of SilverStorm chassis firmware permit ssh keys to be
configured within the chassis for secure password-less login. In which case there is no need to configure a FF_CHASSIS_ADMIN_PASSWORD and FF_CHASSIS_LOGIN_METHOD can be ssh. Consult the SilverStorm 9000 Users Guide for more information.
5.1.4
-C
Q
5.1.5
-n or -I
Specifies that the given operation should be performed against chassis. By default Fast Fabric operations are performed against hosts. However, selected Fast Fabric commands (such as cmdall, pingall, captureall, and ibtest) can also operate against SilverStorm internally managed IB chassis. When -C is specified, the operation will be performed against chassis instead of hosts (and the selection of chassis options discussed below will be used).
Specifies that the given operation should be performed against externally-managed switches (such as the SilverStorm 9024FC model IB switch). By default Fast Fabric operations are performed against hosts. However, selected Fast Fabric commands (such as ibtest) can also operate against externally-managed switches. When specified, the operation will be performed against switches instead of hosts (and the selection of switches options discussed below will be used).
NOTE: Some commands use -n while others use -I. In a future release this
will be made consistent among all commands.
5-2 D000006-000 Rev A
Page 57
Q
5.1.6
Selection of Hosts
For operations that are performed against a set of hosts, there are multiple ways to specify the hosts on which to operate:
1. Small sets of hosts can be easily specified on the command line via the -h option discussed below.
2. When multiple commands are performed against the same small set of hosts, the environment variable HOSTS can be used to specify a space separated lists of hosts.
3. For groups of hosts that will be used often, a file may be created listing the hosts. The defauilt file is /etc/sysconfig/iba/hosts that should list all hosts in the cluster except the host running Fast Fabric itself. Such a file may then be specified via the -f command line option or the HOSTS_FILE environment variable.
Within the tools the options are considered in the following order, the first item listed below that is specified is used for the given command.
5 – Detailed Descriptions of Command LineTools
1. -h option
2. HOSTS environment variable
3. -f option
4. HOSTS_FILE environment variable
5. /etc/sysconfig/iba/hosts file
For example if the -h option is used and the HOSTS_FILE environment variable is also exported, the command will operate only on hosts specified via the -h option.
5.1.6.1
Host List Files
The -f option or the HOSTS_FILE environment variable may be used to provide the name of a file containing the list of hosts on which to operate. The default is /etc/sysconfig/iba/hosts. In some fabrics it may be useful to create multiple files in /etc/sysconfig/iba representing different subsets of the fabric from which the user may operate. For example:
/etc/sysconfig/iba/hosts-mpi: list of MPI hosts
/etc/sysconfig/iba/hosts-fs: list of file server hosts
/etc/sysconfig/iba/hosts: list of all hosts except for the Fast Fabric
node
/etc/sysconfig/iba/allhosts: list of all hosts including the Fast Fabric node
D000006-000 Rev A 5-3
Page 58
5 – Detailed Descriptions of Command LineTools Common Tool Options
If a relative path is specified for the -f option or HOSTS_FILE, the current directory will be checked first, followed by /etc/sysconfig/iba/
5.1.6.1.1
Host List File Format
Below is a sample host list file:
# this is a comment
192.168.0.4# host identified by IP address
n001 # host identified by resolvable TCP/IP name
include /etc/sysconfig/iba/hosts-mpi # included file
Each line of the host list file may specify a single host, a comment or another host list file to include.
Hosts may be specified by IP address or a resolvable TCP/IP host name. Typically, host names are used for readability. Also, some Fast Fabric tools will translate the supplied host names to IPoIB hostnames, in which case names are generally easier to translate than numeric IP addresses. Typically management network hostnames are specified. However, if desired, IPoIB hostnames or IP addresses may be used. This can accelerate large file transfers and other operations.
Q
Files to be included may be specified via an include directive followed by a file name. File names specified should generally be absolute pathnames. If relative pathnames are used, they will be searched for within the current directory then /etc/sysconfig/iba.
Comments may be placed on any line. By using a # to precede the comment. On lines with hosts or include directives, the # must be white space separated from any preceding hostname, IP address or included file name.
5.1.6.2
Explicit host names
When hosts are explicitly specified via the -h option or the HOSTS environment variable, a space separated list of host names (or IP addresses) may be supplied. For example: -h 'host1 host2 host3'.
5.1.7
Selection of Chassis
For operations which are performed against a set of chassis, there are multiple ways to specify the chassis on which to operate:
5-4 D000006-000 Rev A
Page 59
Q
5 – Detailed Descriptions of Command LineTools
1. Small sets of chassis can .be easily specified on the command line via the -H option discussed below
2. When multiple commands will be performed against the same small set of chassis, the environment variable CHASSIS can be used to specify a space separated lists of chassis.
3. For groups of chassis which will be used often, a file may be created listing the chassis. The defauilt file is /etc/sysconfig/iba/chassis which should list all chassis in the cluster. Such a file may then be specified via the -F command line option or the CHASSIS_FILE environment variable.
Within the tools the options are considered in the following order, the first item listed below that is specified is used for the given command.
1. -H option
2. CHASSIS environment variable
3. -F option
4. CHASSIS_FILE environment variable
5. /etc/sysconfig/iba/chassis file
For example if the -H option is used and the CHASSIS_FILE environment variable is also exported, the command will operate only on chassis specified via the -H option.
5.1.7.1
Chassis List Files
The -F option or the CHASSIS_FILE environment variable may be used to provide the name of a file containing the list of SilverStorm IB chassis to operate on. The default is /etc/sysconfig/iba/chassis. In some fabrics it may be useful to create multiple files in /etc/sysconfig/iba representing different subsets of the fabric the user may operate from. For example:
/etc/sysconfig/iba/chassis-core: list of core switching chassis
/etc/sysconfig/iba/chassis-edge: list of edge switching chassis
/etc/sysconfig/iba/esm_chassis: list of chassis running an SM
/etc/sysconfig/iba/chassis: list of all chassis
If a relative path is specified for the -F option or CHASSIS_FILE, the current directory will be checked first, followed by /etc/sysconfig/iba/.
D000006-000 Rev A 5-5
Page 60
5 – Detailed Descriptions of Command LineTools Common Tool Options
5.1.7.1.1
Chassis List File Format
Below is a sample chassis file:
# this is a comment
192.168.0.5# chassis IP address
edge1 # chassis resolvable TCP/IP name
include /etc/sysconfig/iba/corechassis # included file
Each line of the chassis list file may specify a single chassis, a comment or another chassis that list file to include.
A chassis may be specified by chassis management network IP address or a resolvable TCP/IP name. Typically names are used for readability.
Files to be included may be specified via an include directive followed by a file name. File names specified should generally be absolute path names. If relative path names are used, they will be searched for within the current directory then /etc/sysconfig/iba.
Q
Comments may be placed on any line. By using a # to precede the comment. On lines with chassis or include directives, the # must be white space separated from any preceding name, IP address or included file name.
5.1.7.2
Explicit Chassis names
When chassis are explicitly specified via the -H option or the CHASSIS environment variable, a space separated list of names (or IP addresses) may be supplied. For example: -H 'chassis1 chassis2 chassis3'.
5.1.7.3
Selection of slots within a chassis
Normally, operations are performed against the management card in the chassis. For operations such as cmdall, the command is executed against the management interface for the given chassis. For more sophisticated operations such as firmware update, a directory with firmware for each chassis card type can be supplied and all cards in the chassis will be updated with the appropriate firmware from that directory.
However, in some cases it may be desirable to perform operations against a specific subset of cards within the chassis. In this case the chassis IP address or name within a chassis list or a chassis file can be augmented with a list of slot numbers on which to operate. This is done in the form:
chassis:slot1,slot2,…
5-6 D000006-000 Rev A
Page 61
Q
5 – Detailed Descriptions of Command LineTools
For example:
i9k229:0
i9k229:0,1,5
192.168.0.5:0,1,5
NOTE: There must be no spaces within the chassis name and/or slot list.
This format is used by cmdall and chassis firmware update. This format may be used anyplace a chassis name or IP address is valid, such as the -H option, the CHASSIS environment variable or chassis list files. The slot number specified is ignored on some operations (such as pingall). Only slots containing management cards, EVICs and FVICs may be specified with this format. For all 9000 series chassis, slot 0 is always an alias for the presently active management card for the chassis. For the remainder of slot usages in the chassis, the chassisQuery command can be executed against a given chassis to identify which slots have management, EVIC or FVIC cards.
NOTE: For any operation, care should be taken that a given chassis is listed only
once with all relevant slots as part of that single specification. This is important so that parallel operations do not cause conflicting concurrent operations against a given chassis.
5.1.8
Selection of Switches
For operations that are performed against a set of fixed configuration externally-managed switches, there are multiple ways to specify the switch on which to operate:
1. Small sets of switches can be easily specified on the command line via the -N option discussed below.
2. When multiple commands will be performed against the same small set of switches, the environment variable IBNODES can be used to specify a space separate lists of switches
3. For groups of switches which will be used often, a file may be created listing the switches. The defauilt file is /etc/sysconfig/iba/ibnodes that should list all switches in the cluster. Such a file may then be specified via the -L command line option or the IBNODES_FILE environment variable.
Within the tools the options are considered in the following order, the first item listed below which is specified is used for the given command.
1. -N option
2. IBNODES environment variable
3. -L option
D000006-000 Rev A 5-7
Page 62
5 – Detailed Descriptions of Command LineTools Common Tool Options
4. IBNODES_FILE environment variable
5. /etc/sysconfig/iba/ibnodes file
For example if the -N option is used and the IBNODES_FILE environment variable is also exported, the command will operate only on switches specified via the -N option.
5.1.8.1
Switch List Files
The -L option or the IBNODES_FILE environment variable may be used to provide the name of a file containing the list of SilverStorm IB switches on which to operate. The default is /etc/sysconfig/iba/ibnodes. In some fabrics it may be useful to create multiple files in /etc/sysconfig/iba representing different subsets of the fabric from which the user may operate.
If a relative path is specified for the -L option or CHASSIS_FILE, the current directory will be checked first, followed by /etc/sysconfig/iba/.
5.1.8.1.1
Switch List File Format
Q
Below is a sample switch list file:
# this is a comment
0x00066a00d9000138,i9k138 # Node GUID with desired Name
0x00066a00d9000139,i9k139 # Node GUID with desired Name
include /etc/sysconfig/iba/moreswitches # included file
Each line of the switch list file may specify a single switch, a comment or another switch list file to include.
Switches can be specified by node GUID optionally followed by a comma and the IB Node Description (i.e., the name) to be assigned to the switch. The GUID will be used to select the switch and on firmware update operations, the node description will be written to the switch such that other Fast fabric tools (such as saquery and iba_report) can provide a more easily readable name for the switch.
Files to be included may be specified via an include directive followed by a file name. File names specified should generally be absolute path names. If relative path names are used, they will be searched for within the current directory then /etc/sysconfig/iba.
Comments may be placed on any line. By using a # to precede the comment. On lines with chassis or include directives, the # must be white space separated from any preceding GUID, name or included filename.
5-8 D000006-000 Rev A
Page 63
Q
It is recommended that a unique node description be specified for each switch. This name should follow typical naming rules and use the characters a-z, A-Z, 0-9, and underscore. No spaces are allowed in the node description. Additionally, names should not start with a digit.
For 9024FC switches, the node GUID can be found on a label on the bottom of the switch. Alternately, the node GUIDs for switches in the fabric can be found use a command such as:
saquery -t sw -o nodeguid
NOTE: The above command will report all switch node GUIDs, including those
of internally-managed chassis such as the 9120 model. GUIDs for internally-managed chassis cannot be specified for use in -N, IBNODES,
-L, or IBNODES_FILE specified lists.
5.1.8.2
Explicit Switch names
When switches are explicitly specified via the -N option or the IBNODES environment variable, a space separated list of GUIDs (optionally with name) may be supplied. For example:
5 – Detailed Descriptions of Command LineTools
-N '0x00066a00d9000138,i9k138 0x00066a00d9000139,i9k139'
5.1.9
Selection of local Ports (subnets)
Some of the fabric health commands (fabric_analysis, all_analysis) permits a specific set of local HCA ports to be used for fabric analysis. The default is to use the first active port. However, for IB management nodes connected to more than 1 IB subnet, it is necessary to specify the local HCA and port such that the desired subnet will be analyzed. When the non-default behavior is desired, there are multiple ways to specify the local ports to use:
1. Small sets of ports can be easily specified on the command line via the -p option discussed below.
2. When multiple commands will be performed against the same small set of ports, the environment variable PORTS can be used to specify a space separated lists of ports.
3. For groups of ports that will be used often, a file may be created listing the ports. The defauilt file is /etc/sysconfig/iba/ports that should list all local ports connected to unique subnets. Such a file may then be specified via the -t command line option or the PORTS_FILE environment variable.
Within the tools the options that are considered in the following order, the first item listed below that is specified is used for the given command.
D000006-000 Rev A 5-9
Page 64
5 – Detailed Descriptions of Command LineTools Common Tool Options
1. -p option
2. PORTS environment variable
3. -t option
4. PORTS_FILE environment variable
5. /etc/sysconfig/iba/ports file
6. default of the first active port on system (0 :0 port specification)
For example, if the -p option is used and the PORTS_FILE environment variable is also exported, the command will operate only on ports specified via the -p option.
5.1.9.1
Port List Files
The -t option or the PORTS_FILE environment variable may be used to provide the name of a file containing the list of local HCA ports to use. The default is /etc/sysconfig/iba/ports. In some fabrics it may be useful to create multiple files in /etc/sysconfig/iba representing different subsets of the ports from which the user may operate. For example:
Q
/etc/sysconfig/iba/ports-primary: ports for which this node is primary
/etc/sysconfig/iba/ports-plain1: port(s) for plain1 subnet
/etc/sysconfig/iba/ports: list of all unique subnet ports
If a relative path is specified for the -t option or PORTS_FILE, the current directory will be checked first, followed by /etc/sysconfig/iba/.
5.1.9.1.1
Port List File Format
Below is a sample port list file:
# this is a comment
1:1 # first port on 1st HCA
1:2 # second port on 1st HCA
2:1 # first port on 2nd HCA
3:0 # first active port on 3rd HCA
include /etc/sysconfig/iba/ports-plain2# included file
Each line of the port list file may specify a single port, a comment or include another port list file.
5-10 D000006-000 Rev A
Page 65
Q
5.1.9.2
Explicit ports
5 – Detailed Descriptions of Command LineTools
Ports are specified as hca:port. No spaces are permitted. The first HCA is 1 and the first Port is 1. The value 0 for HCA or Port has special meaning. The allowed formats are:
0:0 = 1st active port in system
0:y = port y within system
x:0 = 1st active port on HCA x
x:y = HCA x, port y
Files to be included may be specified via an include directive followed by a file name. File names specified should generally be absolute pathnames. If relative pathnames are used, they will be searched for within the current directory then /etc/sysconfig/iba.
Comments may be placed on any line. By using a # to precede the comment. On lines with a port or include directive, the # must be white space separated from any preceding port or included filename.
When ports are explicitly specified via the -p option or the PORTS environment variable, a space separated list of ports may be supplied.. For example: -p '1:1 1:2 2:1'.
5.2
Basic Setup and Administration Tools
5.2.1
pingall
(All): Pings a group of hosts or chassis to verify that they are powered on and
accessible via TCP/IP ping
Usage:
pingall [-Cp] [-f hostfile] [-F chassisfile] [-h 'hosts'] [-H 'chassis']
-C - performs a ping against a chassis. The default is hosts
-p - ping all hosts/chassis in parallel
-f hostfile - file with hosts in cluster, default is /etc/sysconfig/iba/hosts
-F chassisfile - file with chassis in cluster default is /etc/sysconfig/iba/chassis
-h hosts - list of hosts to ping
-H chassis - list of chassis to ping
D000006-000 Rev A 5-11
Page 66
5 – Detailed Descriptions of Command LineTools Basic Setup and Administration Tools
Example:
pingall
pingall -h 'arwen elrond'
HOSTS='arwen elrond' pingall
pingall -C
pingall -C -H 'chassis1 chassis2'
CHASSIS='chassis1 chassis2' pingall -C
Environment Variables:
The following environment variables are also used by this command:
HOSTS, HOSTS_FILE - see discussion on selection of hosts above
CHASSIS, CHASSIS_FILE - see discussion on selection of chassis above
FF_MAX_PARALLEL - when -p option is used maximum number of parallel
operations to perform at once.
Q
5.2.2
check_rsh
(Linux) Verifies that rsh is set up to allow passwordless file copies (RCP) and
commands (rsh) to be run from this host to all the other hosts (and to itself via localhost) as a specific user (default is root). Additionally, this command can be used to verify rsh is setup to allow MPI to use rsh for job startup.
NOTE: For security reasons, configuration and use of rsh/rcp/rlogin is no
longer recommended. Instead ssh is recommended. SSH may be used by MPI as well as setup_ssh.
Usage:
check_rsh [-i ipoib_suffix] [-f hostfile] [-h 'hosts'] [-u user]
-i 'ipoib_suffix '- suffix to apply to host names to create IPoIB host names. The default is '-ib'. Use -i '' to indicate no suffix.
-h hosts - list of hosts to setup
-f hostfile - file with hosts in cluster, default is /etc/sysconfig/iba/hosts
-u user - user on remote system to verify this user can rsh to. The default
is current user code.
Example:
check_rsh
check_rsh -h 'arwen elrond'
HOSTS='arwen elrond' check_rsh
5-12 D000006-000 Rev A
Page 67
Q
5.2.3
setup_ssh
5 – Detailed Descriptions of Command LineTools
Environment Variables
The following environment variables are also used by this command:
HOSTS, HOSTS_FILE - see discussion on selection of hosts above
(Linux): creates ssh keys and configures them on all hosts so the system can ssh and scp into all other hosts without a password prompt. Typically, during cluster setup this tool is used to enable the root user on the IB Management node to login to the other hosts via password-less ssh. However, if desired, this tool can also aid the setup of password-less ssh login for other user codes as well.
Usage:
setup_ssh [-C] [-s] [-i ipoib_suffix] [-f hostfile] [-h 'hosts'] [-u user] [-S]
-C - only perform connect (to enter in local hosts knownhosts). When run in this mode, -S and -s options are ignored).
-s - use ssh/scp to transfer files, default is rsh/rcp.
-i ipoib_suffix - suffix to apply to host names to create IPoIB host names. The default is '-ib'.
-h hosts - list of hosts to setup.
-f hostfile - file with hosts in cluster, default is /etc/sysconfig/iba/hosts.
-u user - user on remote system to allow this user to ssh to, default is
current user code.
-S - securely prompt for password for user on remote system.
Example:
setup_ssh -s -S -I""
setup_ssh -C
setup_ssh -h 'arwen elrond' -C
HOSTS='arwen elrond' setup_ssh -C
Environment Variables
The following environment variables are also used by this command:
HOSTS, HOSTS_FILE - see discussion on selection of hosts above.
FF_IPOIB_SUFFIX - suffix to append to hostname to create IPoIB
hostname. Used in absence of -i.
D000006-000 Rev A 5-13
Page 68
5 – Detailed Descriptions of Command LineTools Basic Setup and Administration Tools
Fast Fabric provides additional flexibility in the translation between IPoIB and management network hostnames. Refer to appendix C for more information.
Setup_ssh provides an easy way to create ssh keys and distribute them to the hosts in the cluster. Many of the Fast Fabric tools (as well as many versions of MPI) require ssh be set up for password-less operation. Therefore, setup_ssh is an important setup step.
This tool also sets up ssh to the local host and the local hosts IPoIB name. This capability is required by selected Fast Fabric commands and may be used by some applications (such as MPI).
Setup_ssh has two modes of operation. The mode is selected by the presence or absence of the -C option. Typically, setup_ssh will first be run without the
-C option, then it may later be run with the -C option.
Initial key exchange
When run without the -C option, setup_ssh will perform the initial key exchange and enable password-less ssh and scp. The key exchange can be accomplished using ssh and scp (in a password prompting manner) via the -s option or using password-less rsh and rcp (omitting the -s option).
Q
The preferred way to use setup_ssh for initial key exchange is with the -s and -S options. This requires all hosts have been configured with the same password for the specified "user" (typically root). In this mode the password will be prompted for once and then ssh and scp are used in conjunction with that password to complete the setup for the hosts. Use in this manner also avoids the need to setup rsh/rcp/rlogin (which can be a security risk).
If -s is used without the -S option, the user will be prompted by ssh and scp for each host as they are setup. There will be multiple prompts per host. For a handful of hosts this is manageable, however for a significant number of hosts this can become cumbersome. Therefore, the -S option is recommended in this case.
If the -s option is not specified, rsh and rcp will be used to perform the ssh key exchange. This requires password-less rcp and rlogin be enabled on each host (check_rsh can perform verification).
Setup_ssh will configure password-less ssh/scp for both the management network and IPoIB. Typically, the management network will be used for Fast Fabric while IPoIB will be used for MPI and other applications. If IPoIB is not yet running (for example, during initial cluster installation IB software will not yet be installed on all the hosts), the -i option can be specified with an empty string:
setup_ssh -i ''
This will cause the last part of the setup of ssh for IPoIB to be skipped.
Refreshing local systems known hosts
5-14 D000006-000 Rev A
Page 69
Q
5.2.4
cmdall
5 – Detailed Descriptions of Command LineTools
If hosts have IP addresses added (for example by installing IB software and enabling IPoIB), IP addresses changes, MAC addresses changed or other aspects have changed (such as server OS reinstallation), the local hosts ssh known_hosts file can be refreshed by running setup_ssh with the -C option. This option will not transfer the keys, but rather will connect to each host (management network and IPoIB) in order to refresh the ssh keys. Existing entries for the specified hosts are replaced within the local known_hosts file. When run in this mode the -S and -s options are ignored. This mode assumes ssh has previously been setup for the hosts, as such no files are transferred to the specified hosts and no passwords should be required.
Typically after completing the installation and booting of IB software, setup_ssh will need to be rerun with the -C option to update the knownhosts file
(Linux and Switch): Executes a command on all hosts or SilverStorm IB chassis. This is very powerful and can be used for everything from configuring servers or chassis, verifying that they are running, starting and stopping host processes, etc.
Usage:
cmdall [-Cpq] [-f hostfile] [-F chassisfile] [-h 'hosts'] [-H 'chassis'] [-u user] [-S] 'cmd'
-C - perform command against chassis, default is hosts
-p - run command in parallel on all hosts
-q - quiet mode, do not show command to execute
-f hostfile - file with hosts in cluster, default is /etc/sysconfig/iba/hosts
-F chassisfile - file with chassis in cluster default is /etc/sysconfig/iba/chassis
-h hosts - list of hosts on which to execute the command
-H chassis - list of chassis on which to execute the command
-u user - the user to perform the command as. For hosts, the default is current user code. For chassis, the default is admin (this argument is ignored)
-S - securely prompt for password for admin on chassis
D000006-000 Rev A 5-15
Page 70
5 – Detailed Descriptions of Command LineTools Basic Setup and Administration Tools
Host Examples:
cmdall date
cmdall 'uname -a'
cmdall -h 'elrond arwen' date
HOSTS='elrond arwen' cmdall date
Chassis Examples:
cmdall -C 'ismPortStats'
cmdall -C -H 'chassis1 chassis2' ismPortStats
CHASSIS='chassis1 chassis2' cmdall ismPortStats
Environment Variables
The following environment variables are also used by this command:
HOSTS, HOSTS_FILE - see discussion on selection of hosts above
CHASSIS, CHASSIS_FILE - see discussion on selection of chassis above
FF_MAX_PARALLEL - when -p option is used maximum number of parallel
operations to perform at once.
Q
FF_CHASSIS_LOGIN_METHOD - how to login to chassis. Can be ssh or telnet
FF_CHASSIS_ADMIN_PASSWORD - password for admin on all chassis. Used in absence of -S option.
NOTE: All commands performed with cmdall must be non-interactive in nature.
cmdall will wait for the command to complete before proceeding. For
example, when running host commands such as rm, the -i option (interactively prompt before removal) should not be used (Note that this option is sometimes part of a standard bash alias list). Similarly, when running chassis commands such as fwUpdateChassis, the -reboot option should not be used (this option causes an immediate reboot therefore, the command never returns). Similarly, the chassis command reboot should not be executed via cmdall. Instead use the ibtest -C reboot Fast Fabric command to reboot one or more chassis For further information about individual chassis CLI commands consult the SilverStorm 9000 CLI Reference Guide. For further information about Linux OS commands, consult the Linux man pages and any other documentation supplied with the OS by the OS supplier.
When performing cmdall against hosts, internally ssh is used. The command cmdall requires that password-less ssh be setup between the host running Fast Fabric and the hosts cmdall is operating against. The setup_ssh Fast Fabric tool can aid in setting up password-less ssh.
When performing cmdall against a set of chassis, all chassis must be configured with the same admin password.
5-16 D000006-000 Rev A
Page 71
Q
5.2.5
captureall
5 – Detailed Descriptions of Command LineTools
For operations against chassis use of the -S option is recommended. This avoids the need to keep the password in configuration files.
(Switch and Host): Captures supporting information for a problem report from all hosts or SilverStorm IB chassis and uploads to this system
Usage:
captureall [-Cp] [-f hostfile] [-F chassisfile] [-h 'hosts'] [-H 'chassis'] [-d upload_dir] [-S] [file]
-C - perform capture against chassis, default is hosts
-p - perform capture in parallel
[for a host capture this only affects the upload phase]
-f hostfile - file with hosts in cluster, default is /etc/sysconfig/iba/hosts
-F chassisfile - file with chassis in cluster, default is /etc/sysconfig/iba/chassis
-h hosts -a list of hosts to perform a capture of
-H chassis - a list of chassis to perform a capture of
-d upload_dir - directory to upload to, default is uploads. If not specified,
the environment variable UPLOADS_DIR will be used. If that is not exported, the default (./uploads) will be used.
-S - securely prompt for password for administrator on a chassis
file - name for capture file [.tgz will be appended]
When a host captureall is performed, iba_capture will be run to create the specified capture file within ~root on each host (with the .tgz suffix added). The files will be uploaded and unpacked into a matching directory name within
upload_dir/hostname/ on the local system. The default file name is hostcapture.
When a chassis capture all is performed, the chassis capture CLI command will be run on each chassis and its output will be saved to upload_dir/chassisname/file on the local system. The default file name is chassiscapture.
For both host and chassis capture, the uploaded captures will be combined into a tgz file with the file name specified and the suffix .all.tgz added
Host Capture Examples:
captureall
D000006-000 Rev A 5-17
Page 72
5 – Detailed Descriptions of Command LineTools Basic Setup and Administration Tools
The above example creates a hostcapture directory in
./uploads/<HOSTNAME>/ for each host in /etc/sysconfig/iba/hosts then creates hostcapture.all.tgz.
captureall mycapture
The above example creates a mycapture directory in ./uploads/<HOSTNAME>/ for each host in /etc/sysconfig/iba/hosts then creates mycapture.all.tgz.
captureall -h 'arwen elrond' 030127capture
Chassis Capture Examples:
captureall -C
The above example creates a chassiscapture file in ./uploads/<CHASSISNAME>/ for each chassis in /etc/sysconfig/iba/chassis then creates chassiscapture.all.tgz.
captureall -C mycapture
Q
The above example creates a mycapture.tgz file in ./uploads/<CHASSISNAME>/ for each chassis in /etc/sysconfig/iba/chassis then creates mycapture.all.tgz.
captureall -C -H 'chassis1 chassis2' 030127capture
Environment Variables
The following environment variables are also used by this command:
HOSTS, HOSTS_FILE - see discussion on selection of hosts above.
CHASSIS, CHASSIS_FILE - see discussion on selection of chassis above.
UPLOADS_DIR - directory to upload to, used in absence of -d.
FF_MAX_PARALLEL -maximum number of parallel operations to perform at
once.
FF_CHASSIS_LOGIN_METHOD - how to login to chassis. Can be SSH or telnet.
FF_CHASSIS_ADMIN_PASSWORD - password for administrator on all chassis. Used in absence of -S option.
When performing captureall against hosts, internally SSH is used. The command captureall requires that password-less SSH be setup between the host running Fast Fabric and the hosts captureall is operating against. The setup_ssh Fast Fabric tool can aid in setting up password-less SSH.
When performing captureall against a set of chassis, all chassis must be configured with the same administrator password.
5-18 D000006-000 Rev A
Page 73
Q
For operations against chassis use of the -S option is recommended. This avoids the need to keep the password in configuration files.
NOTE: The resulting host capture files can require significant amounts of space
on the Fast Fabric host. Actual size will vary, but sizes can be multiple megabytes per host. As such it is recommended to ensure adequate space is available on the Fast Fabric system. In many cases it may not be necessary to run captureall against all hosts or chassis, but rather a representative subset may be sufficient. Consult with your support representative for further information.
5.3
File Management Tools
The following tools aid in copying files to and from large groups of nodes in the fabric.
Internally, these tools make use of SCP and require that password-less SSH/SCP be setup between the host running Fast Fabric and the hosts files that are being transferred to and from. The setup_ssh Fast Fabric tool can aid in setting up password-less SSH/SCP.
5 – Detailed Descriptions of Command LineTools
5.3.1
scpall
(Linux): The scpall tool permits efficient copying of files or directories from the
current system to multiple hosts in the fabric. When copying large directory trees, performance can be improved by using the -t option. This will tar and compress the tree, then transfer the resulting compressed tarball to each node (and untar it on each node).
This can provide a powerful facility for copying data files, operating system files or even applications to all the hosts (or a subset of hosts) within the fabric.
Usage:
scpall [-p] [-r] [-f hostfile] [-h 'hosts'] [-u user] source_file ... dest_file
scpall -t [-p] [-f hostfile] [-h 'hosts'] [-u user] [source_dir [dest_dir]]
Options are:
-r - recursive copy of directories
-p - perform copy in parallel
-t - optimized recursive copy of directories using tar
-h hosts - list of hosts to copy to
D000006-000 Rev A 5-19
Page 74
5 – Detailed Descriptions of Command LineTools File Management Tools
-f hostfile - file with hosts in cluster, default is /etc/sysconfig/iba/hosts.
-u user - user to perform copy to, default is current user code
source_file: the name of files to copy from this system, relative to the
current directory. Multiple files may be listed.
source_dir: the name of directory to copy from this system, relative to the current directory.
dest_file or dest_dir: is the name of the file or directory on the destination system to copy to. It is relative to the home directory of the specified user code (an absolute path name may be specified if desired).
When performing directory copies using the -t option, the destination directory is optional. If not specified it defaults to the present directory name. If both the source and destination directory names are omitted, they both default to the current directory name.
Example:
Q
# copy a single file
scpall MPI-PMB /root/MPI-PMB
# efficiently copy an entire directory tree
scpall -p -t /opt/iba/src/mpi_apps /opt/iba/src/mpi_apps
# copy a group of files
scpall a b c /root/tools/
# copy to an explitly specified set of hosts
scpall -h 'arwen elrond' a b c /root/tools
HOSTS='arwen elrond' scpall a b c /root/tools
NOTE: The tool scpall can only copy from this system to a group of systems
in cluster. The user@ style syntax cannot be used in the arguments to scpall.
Environment Variables
The following environment variables are also used by this command:
HOSTS, HOSTS_FILE - see discussion on selection of hosts above
FF_MAX_PARALLEL - when -p option is used maximum number of parallel
operations to perform at once.
To copy from hosts in the cluster to this host, use uploadall.
5-20 D000006-000 Rev A
Page 75
Q
5.3.2
uploadall
5 – Detailed Descriptions of Command LineTools
(Linux): Copies one or more files from a group of hosts to this system. Since the file name will be the same on each host, a separate directory on this system is created for each host and the file is copied to it. This is a convenient way to upload log files or configuration files for review. It can also be used in conjunction with downloadall to upload a host specific configuration file, edit it for each host and download the new version to all the hosts.
Usage:
uploadall [-rp] [-f hostfile] [-d upload_dir] [-h 'hosts'] [-u user] source_file ... dest_file
Options:
-p - perform copy in parallel on all hosts
-r - recursive upload of directories
-f <HOST FILE> - file with hosts in cluster, default is /etc/sysconfig/iba/hosts
-h hosts - list of hosts to upload from
-u user - user to perform copy to, default is current user code
-d upload_dir - directory to upload to, default is uploads. If not specified
the environment variable UPLOADS_DIR will be used, if that is not exported the default (./uploads) will be used.
source_file - the name of files to copy to this system, relative to the current directory. Multiple files may be listed.
dest_file - is the name of the file or directory on this system to copy to. It is relative to upload_dir/<HOSTNAME>.
A local directory within upload_dir/ will be created for each host being uploaded from. Each uploaded file will be copied to upload_dir/<HOSTNAME>/dest_file. If more than one source file is specified, dest_file will be treated as a directory name and the directories
upload_dir/<HOSTNAME>/dest_file/ will be created for each host and the source_files will be uploaded to those directories.
D000006-000 Rev A 5-21
Page 76
5 – Detailed Descriptions of Command LineTools File Management Tools
Example:
# upload two files from 2 hosts
uploadall -h 'arwen elrond' capture.tgz /etc/init.d/ipoib.cfg .
# upload two files from all hosts
uploadall capture.tgz /etc/init.d/ipoib.cfg .
# upload network config files from all hosts
uploadall -r -p /etc/sysconfig/network-scripts network-scripts
# upload two files to a specific subdirectory of upload_dir
uploadall capture.tgz /etc/init.d/ipoib.cfg pre-install
The above example copies capture.tgz and /etc/init.d/ipoip.cfg to ./uploads/<HOSTNAME>/preinstall/ where there a <HOSTNAME> directory is created for each host in /etc/sysconfig/iba/hosts.
NOTE: The uploadall tool can only copy from a group of systems in a cluster
to this system. The user@ style syntax cannot be used in the arguments to uploadall.
Q
5.3.3
downloadall
To copy files from this host to hosts in the cluster use scpall or downloadall.
Environment Variables
The following environment variables are also used by this command:
HOSTS, HOSTS_FILE - see discussion on selection of hosts above.
FF_MAX_PARALLEL - when -p option is used maximum number of parallel
operations to perform at once.
UPLOADS_DIR - the directory to upload to, used in absence of -d.
(Linux): Copies one of more files to a group of hosts from a system. Since the file contents to copy may be different for each host, a separate directory on this system is used for the source files for each host. This can also be used in conjunction with uploadall to upload a host-specific configuration file, edit it for each host and download the new version to all the hosts.
Usage:
downloadall [-rp] [-f hostfile] [-d download_dir] [-h 'hosts'] [-u user] source_file ... dest_file
-p - perform copy in parallel on all hosts
-r - recursive download of directories
5-22 D000006-000 Rev A
Page 77
Q
5 – Detailed Descriptions of Command LineTools
-f hostfile - file with hosts in cluster. The default is /etc/sysconfig/iba/hosts.
-h hosts - the list of hosts to download files to
-u user - the user to perform the copy. The default is current user code
-d download_dir - the directory to download files to. The default is ./downloads. If not specified, the environment variable DOWNLOADS_DIR will be used. If that is not exported the default (./downloads) will be used.
source_file - the name of files to copy from the system. Multiple files may be listed. The option source_file is relative to download_dir/<HOSTNAME>.
A local directory within download_dir/ must exist for each host being downloaded to Each downloaded file will be copied from
download_dir/<HOSTNAME>/source_file.
dest_file - is the name of the file or directory on the destination hosts to
copy to.
If more than one source file is specified, dest_file will be treated as a directory name. The given directory must already exist on the destination hosts (the copy will fail for hosts where the directory does not exist).
Example:
# copy two files to 2 hosts
downloadall -h 'arwen elrond' ics_srp.cfg ics_inic.cfg /etc/sysconfig
# copy two files to all hosts
downloadall ics_srp.cfg ics_inic.cfg /etc/sysconfig
NOTE: The tool downloadall can only copy from this system to a group of
hosts in the cluster. The user@ style syntax cannot be used in the arguments to downloadall.
To copy files from hosts in the cluster to this host use uploadall.
Environment Variables
The following environment variables are also used by this command:
HOSTS, HOSTS_FILE - see discussion on selection of hosts above.
FF_MAX_PARALLEL - when -p option is used maximum number of parallel
operations to perform at once.
DOWNLOADS_DIR - directory to download from, used in absence of -d.
D000006-000 Rev A 5-23
Page 78
5 – Detailed Descriptions of Command LineTools File Management Tools
5.3.4
Simplified Editing of Node-Specific Files
(Linux): The combination of uploadall and downloadall provide a powerful
yet simple to use mechanism for reviewing and/or editing node-specific files without the need to login to each node.
This is best explained with an example.
Assume the file /etc./sysconfig/network-scripts/ifcfg-ib1 needs to be reviewed and possibly edited for each host. This file would typically contain the IP configuration information for IPoIB and may contain a unique IP address per host.
To upload the file from all the hosts:
uploadall /etc/sysconfig/network-scripts/ifcfg-ib1 ifcfg-ib1
Now edit the uploaded files with an editor, such as vi:
vi uploads/*/ifcfg-ib1
If via the editor the file was changed for some or all of the hosts, it can then be downloaded to all the hosts:
Q
downloadall -d uploads ifcfg-ib1 /etc/sysconfig/network-scripts/ifcfg-ib1
Alternatively, if there was no need to download the file to all hosts, a subset of hosts can be specified using the -h option or by creating an alternate host list file:
downloadall -d uploads -h 'host1 host32' ifcfg-ib1 /etc/sysconfig/network-scripts/ifcfg-ib1
NOTE: When downloading to a subset of hosts, make sure that only the hosts
uploaded from are specified.
5.3.5
Simplified Setup of Node-Generic Files
(Linux): In contrast scpall can provide a powerful yet simple to use mechanism
for transferring files to all nodes that are generic (i.e., not node-specific).
For example, if all nodes in the cluster will use the same DNS server and TCP/IP name resolution, they may be quickly set as follows:
Create an appropriate local file with the desired information. For example:
vi resolv.conf
Now copy the file to all hosts:
scpall resolv.conf /etc/resolv.conf
5-24 D000006-000 Rev A
Page 79
Q
5.4
Fabric Analysis Tools
5.4.1
Fabric_info
Fabric_info provides a brief summary of the components in the fabric. Fabric_info uses the first active IB port on the given local host to perform its
analysis.
Example output:
Fabric_info
Fabric_info has no options and uses no environment variables.
Example output:
# fabric_info
Fabric Information:
SM: i9k229 Guid: 0x00066a00d8000229 State: Master
SM: i9k3ff Guid: 0x00066a00d90003ff State: Standby
Number of CAs: 17
5 – Detailed Descriptions of Command LineTools
Number of CA Ports: 22
Number of Switch Chips: 6
Number of Links: 29
Number of 1x Ports: 2
The output is as follows:
SM - each subnet manger (SM) running in the fabric is listed along with its node name, port GUID and present SM state (Master, Standby, etc).
Number of CA - number of unique channel adapters (CA) in the fabric. A CA with two-connected ports is counted as a single CA.
NOTE: Channel adapters include both HCAs in servers as well has TCAs within
IO Modules, IB Native Storage, etc.
Number of CA ports - number of connected CA ports in the fabric.
Number of Switch chips - number of unique switches in the fabric.
NOTE: A large IB switch may be composed of many unique switch chips.
Number of Links - number of IB links in the fabric. Note that a large IB switch may have internal links.
D000006-000 Rev A 5-25
Page 80
5 – Detailed Descriptions of Command LineTools Fabric Analysis Tools
Number of 1x Ports - number of ports in the fabric running at 1x speed. Typically such ports represent a bad cable connection, a bad cable, too long a cable or perhaps faulty hardware on one side of the link.
Fabric_info can be very useful as a quick assessment of the fabric state. Fabric_info can be run against a known good fabric to identify its components
and then later run to see if anything has changed about the fabric configuration or state. When used in this manner it can be used to quickly identify if CAs are down, links are missing, SMs are missing, etc.
For more extensive fabric analysis, see iba_report.
5.4.2
showallports
(Switch and Host): Displays basic port state and statistics for all host nodes,
chassis or externally managed switches.
NOTE: iba_report is a newer and more powerful Fast Fabric command. For
general fabric analysis, use iba_report with options such as -o errors and/or -o slowlinks to perform a more efficient analysis of link speeds and errors.
Q
Usage:
showallports [-C|-I] [-f hostfile] [-F chassisfile] [-L ibnodefile] [-h 'hosts'] [-H 'chassis'] [-N 'ibnodes'] [-M 'host'] [-S]
-C - perform operation against chassis; the default is hosts
-I - perform operation against IB nodes; the default is hosts
-f hostfile - file with hosts in cluster; the default is /etc/sysconfig/iba/hosts
-F chassisfile - file with chassis in cluster; the default is /etc/sysconfig/iba/chassis
-L ibnodefile - a file with ib nodes in the cluster; the default is /etc/sysconfig/iba/ibnodes
-h hosts - a list of hosts to show port information
-H chassis - a list of chassis to show port information
-N ibnodes - a list of IB nodes to show port information
-M host - The management host. This is the remote host from which to run IB node queries; the default is localhost
-S - securely prompt for password for administrator on chassis
Environment:
5-26 D000006-000 Rev A
Page 81
Q
5 – Detailed Descriptions of Command LineTools
HOSTS - a list of hosts, used if the -h option is not supplied
CHASSIS - a list of chassis, used if the -C is used and the -h option is not
supplied
HOSTS_FILE - a file containing the list of hosts, used in absence of -f and -h
CHASSIS_FILE - a file containing the list of chassis, used in absence of -F and -H
Example:
showallports
showallports -h 'elrond arwen'
HOSTS-'elrond arwen' showallports
showallports -C
showallports -H 'chassis1 chassis2'
CHASSIS-'chassis1 chasssi2' showallports
showallports -I
showallports -I -N '0x00066a0005000105 0x00066a0005000110'
IBNODES='0x00066a0005000105 0x00066a0005000110' showallports
-I
Environment Variables
The following environment variables are also used by this command:
HOSTS, HOSTS_FILE - see discussion on selection of hosts above
CHASSIS, CHASSIS_FILE - see discussion on selection of chassis above
IBNODES, IBNODES_FILE - see discussion on selection of switches above
MGMT_HOST - host to use to perform IB node queries, used in absence of -M
FF_MAX_PARALLEL - when -p option is used maximum number of parallel
operations to perform at once.
FF_CHASSIS_LOGIN_METHOD - how to login to chassis. Can be SSH or Telnet
FF_CHASSIS_ADMIN_PASSWORD - password for the administrator on all chassis. Used in absence of -S option.
When performing showallports against hosts, internally SSH is used. showallports requires that password-less SSH be setup between the host running Fast Fabric and the hosts showallports is operating against. The setup_ssh Fast Fabric tool can aid in setting up password-less SSH.
When performing showallports against a set of chassis, all chassis must be configured with the same administrator password.
D000006-000 Rev A 5-27
Page 82
5 – Detailed Descriptions of Command LineTools Fabric Analysis Tools
For operations against chassis use of the -S option is recommended. This avoids the need to keep the password in configuration files.
When performing showallports against externally-managed switches it requires an IB-enabled management node with Fast Fabric installed. Typically this will be the Fast Fabric node from which showallports is being run. However, if desired an alternate node may be specified by the -M option or MGMT_HOST environment variable.
5.4.3
iba_report
(All): iba_report provides powerful fabric analysis and reporting capabilities.
It must be run on a host connected to the IB fabric with Fast Fabric installed.
iba_report obtains all its data in an IBTA-compliant manner. Therefore, it will interoperate with both SilverStorm and 3rd party IB components, provided those components are IBTA compliant and implement the IBTA optional features required by iba_report.
iba_report requires that the subnet manager implement all the IBTA SA queries defined in the standard (such as SM Info records, Link Records, Trace Routes, Port Records, Node Records, etc). As such, it is recommended that the QuickSilver Fabric Manager version 4.0 or later be used. iba_report requires all end nodes to implement the PMA PortCounters (IBTA mandatory counters). Also any end nodes which report support of a IBTA device management agent must implement the IOU Info, IOC Profile and Service Entry queries as outlined in the IBTA 1.1 standard.
Q
iba_report takes advantage of these interfaces to obtain extensive information about the fabric from the subnet manager and the end nodes. Using this information, iba_report is able to cross reference it and produce analysis greatly beyond what any single subnet manager request could provide. As such, it exceeds the capabilities previously available in tools such as saquery and
fabric_info.
iba_report internally cross references all this information so its output can be
in user-friendly form. Reports will include both GUIDs, LIDs and names for components. Obviously, these reports will be easiest to read if the end user has taken the time to provide unique names for all the components in the fabric (node names and IOC names). All SilverStorm components support this capability. For hosts, the node names automatically are assigned based on the network host name of the server. For switches and line cards the names can be assigned via the element managers for each component.
Each run of iba_report obtains up to date information from the fabric. At the start of the run iba_report will take a few seconds to obtain all the fabric data, then it will output it to stdout. The reports are sorted by GUIDs and other permanent information such that they can be rerun in the future and produce
5-28 D000006-000 Rev A
Page 83
Q
5 – Detailed Descriptions of Command LineTools
output in the same order even if components have been rebooted. This is useful for comparison using simple tools like diff. iba_report permits multiple reports to be requested for a single run (i.e., 1 of each report type).
By default iba_report uses the first active port on the local system. However, if the IB management node is connected to more than one fabric (e.g., a subnet), the HCA and port may be specified to select the fabric to analyze.
Usage:
iba_report [-v][-q] [-h hca] [-p port] [-o report] [-d detail] [-P|-H] [-N] [-x] [-s] [-i seconds] [-C] [-a] [-c file] [-L] [-F point] [-S point] [-D point] [-Q]
-v/--verbose - verbose output
-q/--quiet - disable progress reports
-h/--hca hca - HCA to send via, default is 1st HCA
-p/--port port - port to send via, default is 1st active port
-o/--output report - report type for output
-d/--detail level - level of detail 0-n for output, default is 2
-P/--persist - only include data persistent across reboots
-H/--hard - only include permanent hardware data
-N/--noname - omit node and IOC names
-x/--xml - output in XML
-s/--stats - get performance statistics for all ports
-i/--interval seconds - obtain performance statistics over interval
seconds, clears all statistics, waits interval seconds, then generates report. Implies -s
-C/--clear - clear performance stats for all ports. Only stats with error thresholds are cleared. A clear occurs after generating the report.
-a/--clearall - clear all performance stats for all ports
-c/--config file - error thresholds configuration file. The default is /etc/sysconfig/iba/iba_mon.conf
-L/--limit - For port error counters check (-o errors) and port counters
clear (-C or -i) with -F limit operation to exact specified focus. Normally the neighbor of each selected port would also be checked/cleared does not affect other reports
-F/--focus point - focus area for report used for all reports except route to limit scope of report
-S/--src point - source for trace route, default is local port
D000006-000 Rev A 5-29
Page 84
5 – Detailed Descriptions of Command LineTools Fabric Analysis Tools
-D/--dest point - destination for trace route
-Q/--quietfocus - do not include focus description in report
Report Types:
comps - summary of all systems and SMs in fabric
brcomps - brief summary of all systems and SMs in fabric
nodes - summary of all node types and SMs in fabric
brnodes - brief summary of all node types and SMs in fabric
ious - summary of all IO units in the fabric
links - summary of all links
extlinks - summary of links external to systems
slowlinks - summary of links running slower than expected
slowconfiglinks- summary of links configured to run slower than
supported
Q
includes slowlinks
slowconnlinks- summary of links connected with mismatched speed
potential
includes slowconfiglinks
misconfiglinks - summary of links configured to run slower than supported
misconnlinks - summary of links connected with mismatched speed potential
errors - summary of links whose errors exceed counts in the configuration file
otherports - summary of ports not connected to the fabric
all - comp, nodes, ious, links, extlinks, slowconnlinks, and errors reports
route - trace route between -S and -D points
none - no report, useful if just want to clear statistics
Point Syntax:
gid:value - value is numeric port gid of form: subnet:guid
lid:value - value is numeric lid
portguid:value - value is numeric port GUID
nodeguid:value - value is numeric node GUID
5-30 D000006-000 Rev A
Page 85
Q
5 – Detailed Descriptions of Command LineTools
nodeguid:value1:port:value2 - value1 is numeric node GUID, value2 is port #
iocguid:value - value is numeric IOC GUID
iocguid:value1:port:value2 - value1 is numeric IOC GUID, value2 is
port #
systemguid:value - value is numeric system image GUID
systemguid:value1:port:value2 - value1 is numeric system image
GUID
value2 is port #
ioc:value - value is IOC Profile ID String (IOC Name)
ioc:value1:port:value2 - value1 is IOC Profile ID String (IOC Name)
value2 is port #
iocpat:value - value is global pattern for IOC Profile ID String (IOC Name)
iocpat:value1:port:value2 - value1 is global pattern for IOC Profile ID
String
(IOC Name), value2 is port #
ioctype:value - value is IOC type (VNIC or SRP)
ioctype:value1:port:value2 - value1 is IOC type (VNIC or SRP)
value2 is port #
node:value - value is node description (node name)
node:value1:port:value2 - value1 is node description (node name)
value2 is port #
nodepat:value - value is glob pattern for node description (node name)
nodepat:value1:port:value2 - value1 is glob pattern for node
description
(node name), value2 is port #
nodetype:value - value is node type (SW, CA or RT)
nodetype:value1:port:value2 - value1 is node type (SW, CA or RT)
value2 is port #
sm - master subnet manager
route:point1:point2 - all ports along the routes between the 2 given
points
D000006-000 Rev A 5-31
Page 86
5 – Detailed Descriptions of Command LineTools Fabric Analysis Tools
Examples:
iba_report can generate hundreds of different reports. Following is a list of some commonly generated reports:
Analyze a fabric for bad cables:
iba_report -o slowlinks -o errors
Analyze a fabric for bad cables or misconfigured ports:
iba_report -o slowconfiglinks -o errors
Analyze a fabric for bad cables or misconfigured ports or misconnected ports:
iba_report -o slowconnnlinks -o errors
Reverse lookup a lid:
iba_report -o brnodes -F lid:5
Reverse lookup a nodeguid:
iba_report -o brnodes -F nodeguid: 0x00066a0098000380
Reverse lookup a portguid:
Q
iba_report -o brnodes -F portguid: 0x00066a00a0000380
Find all the connections to a server:
iba_report -o links -F node:duster
Find all the connections to a switch chip:
iba_report -o links -F 'node:i9k156'
Find all the connections to a multi-node system:
iba_report -o links -F systemguid:0x00066a0098000380
Report on all the components in a multi-node system:
iba_report -o comp -F node:goblin
Identify the routes between 2 servers:
iba_report -o route -S node:duster -D node:goblin
Identify the route between a server and a specific lid:
iba_report -o route -S node:duster -D lid:5
Identify the route between a server and the master SM:
iba_report -o route -S node:duster -D sm
Analyze the route between 2 nodes for bad cables or misconfigured ports or misconnected ports:
iba_report -o slowconnnlinks -o errors -F route:node:cuda:node:duster
5-32 D000006-000 Rev A
Page 87
Q
5 – Detailed Descriptions of Command LineTools
Identify the routes between this server and another server:
iba_report -o route -D node:goblin
Analyze a single switch for any high error counts:
iba_report -o errors -F 'node:i9k156'
Identify the routes between a server and an IOC:
iba_report -o route -S node:duster -D 'ioc:Chassis 0x00066A005000010C, Slot 2, IOC 2'
Clear all the port counters in the fabric:
iba_report -C -o none
Clear all the port counters on a multi-HCA server:
iba_report -C -F node:goblin -o none
Check all port counters, clear them, then recheck:
iba_report -o errors -C; sleep 10; iba_report -o errors
Check all port counters, clear them, then recheck:
iba_report -o errors -C; sleep 10; iba_report -o errors
Clear all port counters, wait 10 seconds, then check
Iba_report -i 10 -o errors
Check all port counters on a server:
iba_report -o errors -F node:goblin
Check all port counters on a specific port on a server:
iba_report -o errors -F node:goblin:port:2
Get all the detailed information for a server including port counters:
iba_report -o nodes -F node:goblin -d 5 -s
Get all the detailed information for an IOU including port counters:
iba_report -o nodes -F 'ioc:Chassis 0x00066A005000010C, Slot 2, IOC 2' -d 5 -s
5.4.3.1
Basics of Using iba_report
iba_report can be run with no options at all. In this mode it provides a brief list of the nodes in the fabric (the brnodes report). The report organizes nodes as CAs, Switches and Routers. It also includes a summary of all the SMs in the fabric.
D000006-000 Rev A 5-33
Page 88
5 – Detailed Descriptions of Command LineTools Fabric Analysis Tools
Here is a sample of iba_report for a small fabric:
[root@duster root]# iba_report
Node Type Brief Summary
14 Connected CAs in Fabric:
NodeGUID Type Name
Port LID PortGUID Width Speed
0x0002c9020020e0d4 CA coyote1
1 0x000d 0x0002c9020020e0d5 4x 2.5Gb
0x00066a00580001e0 CA VEx in Chassis 0x00066a005000010c, Slot 2
2 0x0014 0x00066a02580001e0 4x 2.5Gb
0x00066a0098000001 CA julio
1 0x000c 0x00066a00a0000001 4x 2.5Gb
0x00066a00980001b8 CA orc
1 0x000b 0x00066a00a00001b8 4x 2.5Gb
0x00066a0098000380 CA goblin
1 0x000a 0x00066a00a0000380 4x 2.5Gb
Q
0x00066a0098000384 CA cuda
1 0x0005 0x00066a00a0000384 1x 2.5Gb
2 0x0006 0x00066a01a0000384 4x 2.5Gb
0x00066a00980003a6 CA erik
1 0x0015 0x00066a00a00003a6 4x 2.5Gb
2 0x0016 0x00066a01a00003a6 4x 2.5Gb
0x00066a00980006a2 CA goblin
1 0x000f 0x00066a00a00006a2 4x 2.5Gb
0x00066a0098000849 CA rockaway
2 0x000e 0x00066a01a0000849 4x 2.5Gb
0x00066a0098002813 CA brady
1 0x0002 0x00066a00a0002813 4x 2.5Gb
2 0x0003 0x00066a01a0002813 4x 2.5Gb
0x00066a0098002854 CA brady
1 0x0004 0x00066a00a0002854 4x 2.5Gb
2 0x0008 0x00066a01a0002854 4x 2.5Gb
0x00066a0098003f81 CA ibm345
1 0x0007 0x00066a00a0003f81 4x 2.5Gb
0x00066a009800447b CA duster
5-34 D000006-000 Rev A
Page 89
Q
5 – Detailed Descriptions of Command LineTools
1 0x0011 0x00066a00a000447b 4x 2.5Gb
2 0x0012 0x00066a01a000447b 4x 2.5Gb
0x00066a0098004a73 CA erik
1 0x0009 0x00066a00a0004a73 4x 2.5Gb
3 Connected Switches in Fabric:
NodeGUID Type Name
Port LID PortGUID Width Speed
0x00066a00280002cd SW InfiniCon Systems InfiniFabric (Sw A Dev A)
0 0x0013 0x00066a00280002cd Noop Noop
3 4x 2.5Gb
5 4x 2.5Gb
0x00066a00d8000123 SW InfiniCon Systems InfinIO9024
0 0x0001 0x00066a00d8000123 4x 2.5Gb
1 4x 2.5Gb
2 1x 2.5Gb
3 4x 2.5Gb
4 4x 2.5Gb
5 4x 2.5Gb
6 4x 2.5Gb
7 4x 2.5Gb
8 4x 2.5Gb
9 4x 2.5Gb
10 4x 2.5Gb
11 4x 2.5Gb
12 4x 2.5Gb
14 4x 2.5Gb
15 4x 2.5Gb
16 4x 2.5Gb
17 4x 2.5Gb
18 4x 2.5Gb
19 4x 2.5Gb
20 4x 2.5Gb
0x00066a10280002cd SW InfiniCon Systems InfiniFabric (Sw A Dev
D000006-000 Rev A 5-35
Page 90
5 – Detailed Descriptions of Command LineTools Fabric Analysis Tools
B)
0 0x0010 0x00066a10280002cd Noop Noop
2 4x 2.5Gb
4 4x 2.5Gb
1 Connected SMs in Fabric:
State GUID Name
Master 0x00066a00d8000123 InfiniCon Systems InfinIO9024
Each iba_report allows for various levels of detail. Increasing detail is shown as further indentation of the additional information. The -d option to iba_report controls the detail level. The default is 2. Values from 0-n are
Q
5-36 D000006-000 Rev A
Page 91
Q
5 – Detailed Descriptions of Command LineTools
permitted. The maximum detail per report varies, but most have less than 5 detail levels.
For example, the above report when run at detail level 0 outputs:
[root@duster root]# iba_report -d 0
Node Type Brief Summary
14 Connected CAs in Fabric:
3 Connected Switches in Fabric:
1 Connected SMs in Fabric:
You will notice this is a nice summary of fabric components and is very similar to fabric_info.
At the next level of detail you get a report with a little more detail:
[root@duster root]# iba_report -d 1
Node Type Brief Summary
14 Connected CAs in Fabric:
NodeGUID Type Name
0x0002c9020020e0d4 CA coyote1
0x00066a00580001e0 CA VEx in Chassis 0x00066a005000010c, Slot 2
0x00066a0098000001 CA julio
0x00066a00980001b8 CA orc
0x00066a0098000380 CA goblin
0x00066a0098000384 CA cuda
0x00066a00980003a6 CA erik
0x00066a00980006a2 CA goblin
0x00066a0098000849 CA rockaway
0x00066a0098002813 CA brady
0x00066a0098002854 CA brady
0x00066a0098003f81 CA ibm345
0x00066a009800447b CA duster
0x00066a0098004a73 CA erik
3 Connected Switches in Fabric:
NodeGUID Type Name
0x00066a00280002cd SW InfiniCon Systems InfiniFabric (Sw A Dev A)
D000006-000 Rev A 5-37
Page 92
5 – Detailed Descriptions of Command LineTools Fabric Analysis Tools
0x00066a00d8000123 SW InfiniCon Systems InfinIO9024
0x00066a10280002cd SW InfiniCon Systems InfiniFabric (Sw A Dev B)
1 Connected SMs in Fabric:
State GUID Name
Master 0x00066a00d8000123 InfiniCon Systems InfinIO9024
The above examples were all performed with a single report, the brnodes (Brief Nodes) report. However this is just one of the many topology reports which iba_report can generate the others include:
nodes - a more verbose form of brnode which can provide much greater
levels of detail to drill down into all the details of every node, even down to all the port state, IOUs/IOCs/Services, Port counters.
comps and brcomps are very similar to brnodes and nodes, except the
reports are organized around systems. The grouping into systems is based on system image guids for each node. This report will help to present more complex systems (such as servers with multiple HCAs or large switches composed of multiple IB Switch chips).
Q
NOTE: All SilverStorm switches implement a system image GUID and will
therefore be properly grouped. However, some third-party devices do not implement the system image GUID and may report a value of 0. In such a case iba_report will treat each component as an independent system.
links - This report presents all the links in the fabric. The output is very
concise and helps to identify the connectivity between nodes in the fabric.
extlinks - All the external links in the fabric (eg. those between different
systems).
ious - This is somewhat similar to the nodes reports, however the focus is
around IOUs/IOCs and IO Services in the fabric. This report can be used to identify various IO devices in the fabric and their capabilities (such as the SilverStorm EVIC and FVIC Virtual IO Controllers or IBTA compliant direct-attach IB storage).
otherports - All the ports which are not connected to this fabric. This report
will identify additional ports on CAs or Switches which are not connected to this fabric. For switches these represent unused ports. For CAs these may be ports connected to other fabrics or unused ports.
The above reports are all summaries of the present state of the fabric. These reports can be very helpful to analyze the configuration of the fabric and or verify it was installed consistent with the desired design and configuration.
5-38 D000006-000 Rev A
Page 93
Q
5 – Detailed Descriptions of Command LineTools
However, iba_report does not stop there. Additionally, iba_report has reports that will help to analyze the operational characteristics of the fabric and help to identify bottlenecks and faulty components in the fabric.
To assist in this area, iba_report also supports the following reports:
slowlinks - identifies links which are running slower than expected. This
helps to pinpoint bad cables or components in the fabric, such as a 4x cable that is poorly-connected and therefore only runs at 1x link width. The analysis includes both link speed and width.
slowconfiglinks - this extends on the slowlinks report to also report links
which have been configured (most likely by software) to run at a width or speed below their potential. Such as DDR capable links which have been forced to run at SDR rates.
slowconnlinks - this further extends on the slowconfiglinks report to also
report links which are cabled such that one of the ends of the link will never run to its potential. Such as a DDR capable HCA connected to an SDR switch.
misconfiglinks - this is similar to slowconfiglinks in that it reports
links which have been configured to run below their potential. However it does not include links which are running slower than expected.
misconnlinks - this is similar to slowconnlinks in that it reports links
which have been connected between ports of different speed potential. However it does not include links which are running slower than expected, nor links which have been configured to run slower than their potential.
errors - this performs a single point in time analysis of the PMA port counters
for every node and port in the fabric. All the counters are compared against configured thresholds (defaults are those in the iba_mon.conf file). Any link whose counters exceed these thresholds are listed (and depending on the detail level the exact counter and threshold will be reported). This is a powerful way to identify marginal links in the fabric such as bad or loose cables or damaged components.
route - This permits the user to identify two end points in the fabric (by node
name, node GUID, port name, port GUID, system image GUID, LID, port GID, IOC GIUD or IOC name) and obtain a list of all the links and components used when these two end points communicate. If there are multiple paths between the end points (such as a CA with 2 connected ports or a system with 2 CAs), the route for every available path (based on presently configured routing tables) will be reported.
The above set of reports can therefore be very powerful ways to obtain point in time status and problem analysis for the fabric.
D000006-000 Rev A 5-39
Page 94
5 – Detailed Descriptions of Command LineTools Fabric Analysis Tools
5.4.3.2
Topology Verification
iba_report provides a flexible way to identify changes to the fabric or the appropriate reassembly of the fabric after a move (for example after staging and testing the fabric in a remote location before final installation at a customer site).
In this mode of operation, all the above reports are available, however the types of information output can be filtered. For example, using the -P option, information which would not persist across a fabric reboot (such as LIDs and error counters) will be omitted from the report (and marked out with xxx). Such a report can be saved for later comparison to a future report. Since iba_report produces simple text reports, standard tools such as sdiff (i.e., side by side diff) can be used for easy comparison and analysis of what changed.
Given the wealth of reports available, the user can select the information they want to save. For ease of use an all report is available which includes all the reports of general interest.
If software configuration changes are anticipated (such as adjusting the timeouts the SM configures in the fabric), the iba_report -H option can be used. This will further limit the report to only include hardware information. This is a superset of -P and omits more information.
Q
A related but independent option is -N. This will omit all the node and IOC names from the report. If changes are anticipated in this area, this option can be used so future diffs will not report changes in names.
5.4.3.3
Focused Reports
One of the more powerful features of iba_report is the ability to focus a report on a subset of the fabric. Using the -F option the user can specify a node name, node name pattern, node guid, node type, port guid, IOC name, IOC name pattern, ioc guid, ioc type, system image guid, port gid, port rate, lid or SM. The subsequent report will indicate the total components in the fabric but will only report on those which relate to the focus area. For example in a nodes report, if a port is specified for focus, only the node containing that port will be reported on. In a links report, only the link using that port will be reported.
Notice that a focus level that is different from the orientation of the report may be chosen. For example if a node name is specified as the focus for the links report, a report of all the links to that node will be provided. This could include multiple switch ports or CA ports.
By carefully using this feature of report focus, reverse lookups can be done. For example, doing a brnodes report with a focus on a LID will reverse lookup the LID and indicate what node it is for.
5-40 D000006-000 Rev A
Page 95
Q
When focusing a report, it can sometimes be helpful to also use a detail level of 0 or 1. In this case the report will show only a count of number of matches (for detail 0) and just the highest level of the entity which matches (for detail 1).
5.4.3.3.1
Advanced Focus
The node name, node name pattern, node guid, node type, IOC name, IOC name pattern, IOC GUID, IOC type and system image GUID also allow for a port number specifier. This permits the focus to be limited to the given port number. If the selection resolves to multiple switches or CAs (for example a system composed of multiple nodes), all ports on the present fabric matching the given port number will be selected.
An even more advanced form of focus is to focus on the route between any two points. This will focus on all the ports involved in that route and can be an excellent way to focus in quickly on a performance or error situation which is being reported between 2 specific points in the fabric (Such as a StatusTimeoutRetry that MPI may be reporting between 2 processes in its run).
5 – Detailed Descriptions of Command LineTools
Focus can use glob style patterns. This permits a wildcarded focus by node name or IOC name. If a naming convention is used for fabric components, this can provide a powerful way to focus reports on nodes. For example, if the host names are prefixed with an indication of their purpose, searches can be performed based on the purpose of the node. For example if the following naming convention is used: l### = login node ###, n### = compute node ###, s### = storage node ###, etc. Node purposes can be focused by using patterns such as 'l*', 'n*' or 's*'.
NOTE: A glob style pattern is a shell style wildcard pattern as used by bash and
many other tools. When using such patterns they should be single quoted so that the shell will not try to expand them to match local file names.
Typically a focused report will include a summary at its start of the items focused on. When the focus has a large scope, this list can be quite long. In this case the -Q option can be used to omit this section from the report.
D000006-000 Rev A 5-41
Page 96
5 – Detailed Descriptions of Command LineTools Fabric Analysis Tools
5.4.3.3.2
Focus Examples:
Below are some examples of using the focus options:
iba_report -o nodes -F portguid:0x00066a00a000447b
iba_report -o nodes -F nodeguid:0x00066a009800447b:port:1
iba_report -o nodes -F nodeguid:0x00066a009800447b
iba_report -o nodes -F node:duster
iba_report -o nodes -F node:duster:port:1
iba_report -o nodes -F 'nodepat:d*'
iba_report -o nodes -F 'nodepat:d*:port:1'
iba_report -o nodes -F nodetype:CA
iba_report -o nodes -F nodetype:CA:port:1
iba_report -o nodes -F lid:1
iba_report -o nodes -F gid:0xfe80000000000000:0x00066a00a000447b
iba_report -o nodes -F systemguid:0x00066a009800447b
iba_report -o nodes -F systemguid:0x00066a009800447b:port:1
iba_report -o nodes -F iocguid:0x00066a01300001e0
iba_report -o nodes -F iocguid:0x00066a01300001e0:port:2
iba_report -o nodes -F 'ioc:Chassis 0x00066A005000010C, Slot 2, IOC 1'
iba_report -o nodes -F 'ioc:Chassis 0x00066A005000010C, Slot 2, IOC 1:port:2'
iba_report -o nodes -F 'iocpat:*Slot 2*'
iba_report -o nodes -F 'iocpat:*Slot 2*:port:2'
iba_report -o nodes -F ioctype:VNIC
iba_report -o nodes -F ioctype:VNIC:port:2
iba_report -o nodes -F sm
iba_report -o nodes -F route:node:duster:node:cuda
iba_report -o nodes -F route:node:duster:port:1:node:cuda:port:2
Q
5.4.3.4
Scriptable output
iba_report permits custom scripting. As previously mentioned, options like
-H, -P and -N can aid the generation of reports that can be diff'ed.
In addition the -x option permits output reports to be generated in XML format. The XML hierarchy is similar to the textual reports. Use of XML permits other XML tools (such as PERL XML extensions) to easily parse iba_report output such that scripts can be created to further search and refine report output formats.
This allows iba_report to be integrated into custom scripts. It can also be used to generate customer-specific new report formats, cross reference iba_report with other site-specific information, etc.
5-42 D000006-000 Rev A
Page 97
5 – Detailed Descriptions of Command LineTools
Q
5.4.3.4.1
Using iba_report to monitor for fabric changes
iba_report can easily be used in other scripts. For example the following simple script could be run as a cron job to identify if the fabric has changed as compared to the initial design:
#!/bin/bash
# specify some filenames to use
expected_config=/usr/local/report.master # master copy of config previously created
config=/tmp/report$$ # where we will generate new report
diffs=/tmp/report.diff$$ # where we will generate diffs
iba_report -o all -d 5 -P > $config 2>/dev/null
if ! diff $config $expected_config > $diffs 2>/dev/null
then
# notify admin, for example mail the new report to the admin
cat $diffs $expected_config $config |
mail -s "fabric change detected" admin@somewhere
fi
rm -f $config $diffs
D000006-000 Rev A 5-43
Page 98
5 – Detailed Descriptions of Command LineTools Fabric Analysis Tools
5.4.3.5
Q
Sample Output
5.4.3.5.1
Analysis of all ports in fabric for errors, inconsistent connections, bad cables
[root@duster root]# iba_report -o errors -o slowconnlinks
Links running slower than faster port Summary
Links running slower than expected:
20 of 20 Links Checked, 0 Errors found
Links configured to run slower than supported:
Rate NodeGUID Port Type Name
Enabled Supported
2.5g 0x00066a0098000384 1 CA cuda
1x 2.5Gb 1-4x 2.5Gb
<-> 0x00066a00d8000123 2 SW InfiniCon Systems InfinIO9024
1-4x 2.5Gb 1-4x 2.5Gb
20 of 20 Links Checked, 1 Errors found
Links connected with mismatched speed potential:
20 of 20 Links Checked, 0 Errors found
-------------------------------------------------------------
Links with errors > threshold Summary
Configured Error Thresholds:
SymbolErrorCounter 100
LinkErrorRecoveryCounter 3
LinkDownedCounter 3
PortRcvErrors 100
PortRcvRemotePhysicalErrors 100
PortXmitDiscards 100
PortXmitConstraintErrors 10
PortRcvConstraintErrors 10
LocalLinkIntegrityErrors 3
5-44 D000006-000 Rev A
Page 99
Q
5 – Detailed Descriptions of Command LineTools
ExcessiveBufferOverrunErrors 3
VL15Dropped 100
Rate NodeGUID Port Type Name
10g 0x00066a0098000001 1 CA julio
<-> 0x00066a00d8000123 8 SW InfiniCon Systems InfinIO9024
LinkDownedCounter: 5 Exceeds Threshold: 3
10g 0x00066a00980001b8 1 CA orc
<-> 0x00066a00d8000123 10 SW InfiniCon Systems InfinIO9024
LinkDownedCounter: 5 Exceeds Threshold: 3
10g 0x00066a0098000380 1 CA goblin
SymbolErrorCounter: 65535 Exceeds Threshold: 100
LinkErrorRecoveryCounter: 255 Exceeds Threshold: 3
PortRcvErrors: 65535 Exceeds Threshold: 100
<-> 0x00066a00d8000123 15 SW InfiniCon Systems InfinIO9024
SymbolErrorCounter: 41079 Exceeds Threshold: 100
LinkErrorRecoveryCounter: 188 Exceeds Threshold: 3
10g 0x00066a0098003f81 1 CA ibm345
<-> 0x00066a00d8000123 12 SW InfiniCon Systems InfinIO9024
SymbolErrorCounter: 9533 Exceeds Threshold: 100
LinkErrorRecoveryCounter: 46 Exceeds Threshold: 3
PortRcvErrors: 617 Exceeds Threshold: 100
20 of 20 Links Checked, 4 Errors found
D000006-000 Rev A 5-45
Page 100
5 – Detailed Descriptions of Command LineTools Fabric Analysis Tools
5.4.3.5.2
Identification of the route between 2 nodes in the fabric
[root@duster root]# ./iba_report -o route -S node:orc -D node:julio
Routes Summary Between:
Node: 0x00066a00980001b8 CA orc
and Node: 0x00066a0098000001 CA julio
Routes between ports:
0x00066a00980001b8 1 CA orc
and 0x00066a0098000001 1 CA julio
1 Paths
SGID: 0xfe80000000000000:00066a00a00001b8
DGID: 0xfe80000000000000:00066a00a0000001
Q
SLID: 0x000b DLID: 0x000c Reversible: Y PKey: 0xffff
Raw: N FlowLabel: 0x00000 HopLimit: 0x00 TClass: 0x00
SL: 0 Mtu: 2048 Rate: 10g PktLifeTime: 67 ms Pref: 0
Rate NodeGUID Port Type Name
10g 0x00066a00980001b8 1 CA orc
-> 0x00066a00d8000123 10 SW InfiniCon Systems InfinIO9024
10g 0x00066a00d8000123 8 SW InfiniCon Systems InfinIO9024
-> 0x00066a0098000001 1 CA julio
2 Links Traversed
5-46 D000006-000 Rev A
Loading...