IBM 9077 SP Switch Router:
Get Connected to the SP Switch
November 1998
SG24-5157-00
Take Note!
Before using this information and the product it supports, be sure to read the general information in
Appendix D, “Special Notices” on page 305.
First Edition (November 1998)
This edition applies to PSSP Version 2, Release 4 for use with AIX 4.3.1 and Ascend Embedded/OS
Version 1.4.6.4.
Comments may be addressed to:
IBM Corporation, International Technical Support Organization
Dept. HYJ Mail Station P099
522 Sou th Road
Poughkeepsie, New York 12601-5400
When you send information to IBM, you grant IBM a non-exclusive right to use or distribute the
information in any way it believes appropriate without incurring any obligation to you.
Note to U.S Government Users – D oc umen tation r elated to restricted r ights – Use, duplic ation or dis clos ure is
subject to restrictions set for th in GS A ADP S che dule Contr act with IBM C orp.
xiiIBM 9077 SP Switch Router: Get Connected to the S P S witch
Preface
The GRF is a high-performance switched IP Router which provides
high-speed data communication links between IBM RS/6000 SP and external
networks or hosts. It acts as a special-purpose SP node that routes IP traffic
between SP nodes on the SP Switch and the out side world. Connected
directly to the SP Switch, the router offers significantly improved SP I/O
performance. When packaged with an IBM SP system, the GRF router is
referred to as the SP Switch Router.
This redbook helps you i nstal l , tailor and configure the SP Swi tch Router, I BM
machine type 9077. The SP Switch Router is also known as the "Gigarouter"
or High Performance Gateway Node (HPGN).
The first part of the book gives an overview of the GRF architecture and how
the router was integrated into the SP. It emphasizes the advantages of
choosing a dedicated router node in some configurations, as opposed to
using standard nodes for the routing task. This part also describes some
routing fundamentals, particularly focusing on concepts like IP- and
switch-routing.
The second part presents sample configurations that were carefully chosen to
match frequently occurring customer situations. The basic configurations
shown are building blocks for more complex networking topologies that
include the SP Switch Router and may inspire more sophisticated
configurations. All configurations described were tested and provide some
comparable performance figures.
This publication is intended to give IBM customers, system engineers, and
marketing personnel a br oad understanding of this new arch itec ture and what
it is used for.
The Team That Wrote This Redbook
This redbook was produced by a team of specialists from around the world
working at the International Technical Support Organization, Poughkeepsie
Center.
Dr Hajo Kitzhöfer is an Advisory International Technical Support
Organization (ITSO) Specialist for RS/6000 SP at the Poughkeepsie Center.
He holds a Ph.D. degree in electrical engineering from the Ruhr-University of
Bochum (RUB). Before joining ITSO, he worked as an SP Specialist at the
RS/6000 and AIX Competence Center, IBM Germany. He has worked at IBM
for eight years. His areas of expertise include RS/6000 SP, SMP, and
Benchmarks. He now specializes in SP System Management, SP
Performance Tuning and SP hardware.
Dr Steffen Eisenblätter is an AIX Software Specialist in the RS/6000 SP
Software Support Center, Germany. He holds a Ph.D. degree in physics from
the University of Leipzig. He joined IBM in 1997 and has focused on RS/6000
SP products and TCP/IP.
Uwe Untermarzoner is an RS/6000 SP Technical Support Specialist with
IBM Germany. He joined IBM 1989. He has ten y ears of experience in AIX
and five years of experience with the SP, mostly in the commercial
environment. He joined IBM at 1989. His areas of expertise include AIX,
RS/6000 SP, SMP, PSSP, Networking, Performance Tuning and Systems
Management.
Thanks to the following people for their invaluable contributions to this
project:
Ronald Linton
IBM PPS Lab Poughkeepsie
Gene Novitsky
Ascend Communications, Inc.
Frank May
IBM Worldwide RS/6000 SP Product Marketing
Wes Kinard
IBM RS/6000 Networking Technologies
Marcelo R. Barrios
International Technical Support Organization, Poughkeepsie Center
xivIBM 9077 SP Switch Router: Get Con nected to the SP Switc h
Comments Welcom e
Your comments are important to us!
We want our redbooks to be as helpful as possible. Please send us your
comments about this or other redbooks in one of the following ways:
• Fax the evaluation form found in “ITSO Redbook Evaluation” on page 323
to the fax number shown on the form.
• Use the electronic evaluation form found on the Redbooks Web sites:
For Internet users
http://www.redbook s.ibm.com
For IBM Intranet usershttp://w3.itso.ibm .com
• Send us a note at the following address:
redbook@us.ibm. com
xv
xviIBM 9077 SP Switch Router: Get Con nected to the SP Switc h
2IBM 9077 SP Switch Router: Get Connected to the S P Swi tch
Chapter 1. Dependent Node
This chapter provides an overview of a
start by defining the dependent node and the rationale behind its design.
1.1 Dependent Node Architecture
The Dependent Node Architecture refers to a processor or node, possi bl y not
provided by IBM, for use with the RS/6000 SP.
Since a dependent node may not be a regular RS/6000 SP node, not all the
functions of a node can be performed on it, which is why it is called
"dependent". For example, it does not allow all the functions of the fault
service (Worm) daemon, as other RS/6000 SP nodes with access to the SP
Switch do.
The objective of this architecture i s to allow the other processors or hardware
to easily work together with the RS/6000 SP, extending the scope and
capabilities of the system.
The dependent node connects to the RS/6000 SP Switch (but not to the
earlier High Performance Switch, HiPS).
The SP Switch Router Adapter is the first product to exploit the Dependent
Node Architecture.
dependent node
in RS/6000 SP. We
1.2 Limitations of the De pendent N ode
The following are limitations associated with use of the dependent node:
• To use the dependent node in an RS/6000 SP requires the SP Extension
Node SNMP Manager to be installed in the Control Workstation. The SP
Extension Node SNMP Manager requires UDP port 162 in the Control
Workstation. Other SNMP managers, such as Netview, also require this
port. To allow the two SNMP managers to coexist, the SP Extension Node
SNMP Manager must use an alternative UDP port.Dependent nodes are
not allowed in Node Groups.
• Only the 8-port and 16-port SP Switch are supported. The 8-port and
16-port High Performance Switch (the old SP Switch) are not supported.
spmon command on the RS/6000 SP is not enhanced to support
dependent nodes. Dependent nodes can only be viewed with the
perspectives command.
• The fault service daemon runs on all switch nodes in the RS/6000 SP, but
not on the dependent node. Therefore, the dependent node does not have
the full functionality of a normal RS/6000 SP Switch node.
• The dependent node requires the SP Switch’s primary node to compute its
switch routes. Therefore, the primary node must have at least PSSP 2.3
installed, otherwise the dependent node cannot work with the RS/6000
SP.
• In the RS/6000 SP, SP Switch nodes occasionally send service packets
from one node to the next to keep track of status and links. Sometimes
these packets are sent indirectly through another switch node. As the
dependent node is not a standard RS/6000 SP Switch node, it cannot be
used to forward service packets to other nodes.
4IBM 9077 SP Switch Router: Get Connected to the S P Swi tch
Chapter 2. Router N ode
The first dependent node is actually a new SP Switch Router Adapter in a
router. This chapter offers more details about the implementation.
Section 2.1, “Overview” on page 5 gives you an overview of SP Switch
Router. This is probably the best to get an impression what the GRF is good
for. Also a functional- and a price-comparison between using an RS/6000 SP
node and the SP Switch Router is included.
More details about the underlaying Software and Hardware can be found in
Section 2.2, “GRF Softw are” on page 18 and Section 2. 3, “GRF Hardware” on
page 24.
Section 2.4, “PSSP Enhancements” on page 40 describes the enhancements
in the PSSP Software for the support of the dependent node.
Some planning considerations which should be considered can found in
Section 2.5, “Planning f or the GRF” on page 63 and Section 2. 6, “Planning for
the Dependent Node” on page 65.
2.1 Overview
The purpose of the SP Switch Router Adapter is to allow the GRF ("goes
really fast"), manufactured by Ascend, to forward SP Switch IP traff i c to other
networks. The GRF was known as the High Performance Gateway Node
(HPGN) during the development of the adapter. IBM remarkets models of the
GRF that connect to the SP Switch as the SP Switch Router model 04S
(9077-04S) and model 16S (9077-16S). These models are not available
directly from Ascend.
Note: In the remainder of this book, we refer to the SP Switch Router as the
GRF.
The distinguishing feature of the GRF, when c ompared with other routers, is
that it has an SP Switch Router Adapter and therefore can connect directly to
the SP Switch (see Figure 1 on page 6).
The RS/6000 SP software treats this adapter as an extension node. It is a
node because it takes up one port in the SP S witch and is assigned a node
number. It is described as an extension because it is not a standard RS/6000
SP node, but an adapter card that extends the scope of the RS/6000 SP.
Although the term
extension node
represents the node appearance of the
adapter, i t does not def i ne t he connec tion. An
for that purpose. Each extension node has an extension node adapter to
represent its connection to the SP Switch.
2.1.1 Motivation
A thin node, which has a single microchannel, is unable to deliver more than
about 30 MB/s to or from the SP Switch. Using a wide node, this number
increases to 65 MB/s but is still unable to provide full bandwidth to even one
HIPPI interface. It is also unable to feed 4 FDDI or 4 Ethernet 100BaseTx
cards at full bandwidth.
A 135 MHz wide node’s CPU becomes saturated at about 5000
packets/second. A 10 Mb/s Ethernet uses a maximum of 1500 byt es for a
6IBM 9077 SP Switch Router: Get Connected to the S P Swi tch
extension node adapter
is used
packet size. This would only enable a wide node to handle approximately 7.5
MB/s of IP traffic.
Since Ascend’s business depends on keeping pace with networking
technology, they already support the major interfaces today. The 9077 will be
able to take advantage of any new interfaces that are developed in the future
as well, with no further development time or money expended.
With some interfaces requiring up to 5 slots, even a wide node can run out of
available slots. This forces additional nodes to be added even if there are no
performance limitations in the current configuration.
Since there are no hot plug capabilities with an SP node, any failure means
downtime on all interfaces configured in that node, and at times a lengthy
maintenance procedure. Redundancy is not built into the SP node’s
architecture.
These fact s a re illu s trated i n Figu re 2:
SP Node9077
Bus
Route Table
Scalability
Throughput
Support
Shared
1 MCA per thin node
2 MCA per wide node
Centralized
Cache,Software Based
Cache hits <50% typical
Single port per card
Single CPU
Limited by shared bus
5000 pps
30 MB/s per thin node
65 MB/s per wide node
No support for:
HSSI
ATM OC12
Sonet
Multiple SP Switch Adapters
Non-blocking Crosspoint Switch
250 ns path setup
Independent lookups per card
Hardware based, <2.5 µs
150,000 route capacity per card
High per card port density
Per card
Processors
Route Tables
Lookup engines
Each card has dedicated bandwidth
Up to 130,000 pps
100 MB/s per card slot, full duplex
Support for:
multiple SP Switch interfaces
High-speed networks such as
HIPPI
Protocols
Figure 2. Functional Comparison
Router Node 7
2.1.2 Design Objectives
Because the dependent node is part of the RS/6000 SP, it had to be
packaged and assigned some roles consistent with other RS/6000 SP nodes.
Changes were made to the RS/6000 SP to incorporate management
requirements for the dependent node.
Ease of design and implementation were important objectives in the design.
These were accomplished by limiting the amount of switchcontrol prot ocol for
the dependent node.
SDR
New
dependent nodes. This was done to minimize the scope of the changes and
the exposure to side effects that dependent nodes may cause if they were
represented as standard nodes in the SDR.
(System Data Repository) classes were created to manage
2.1.3 What is a Router
One of the basic functions of the
between different networks. This is due its routing algorithm and its flexibility
to use almost any physical network below. A system that connects different
physical or logical net works and di rects traffic is termed a
older term
Again, IP routing is the passing of an IP packet from one device to another by
sending it on a physical or logical interface. routers interconnect networks so
that IP traffic can be routed between the systems in the networks, as shown
in Figure 3 on page 9.
IP gateway
Internet Protocol
is also used.
(IP) is its ability to connect
router
, although the
8IBM 9077 SP Switch Router: Get Connected to the S P Swi tch
Network 1
Network 3
Router
Network 2
Figure 3. Typical Router Configuration
Network n
Routers help to reduce the amount of processing required on local systems,
since they perform the computation of routes to remote systems. For
example, a system can communicate with a remote system by passing the
message (or packets) to the router. The router works out how to get to the
remote system and forwards the message appropriately.
Storing routes on the system takes up memory. But because a system does
not have to store routes t o systems not i n i t s own s ubnet, the rout e table uses
less storage space and thereby frees up memory for other work.
The use of routing reduces network traffic, because routers encourage
subnetting, which creates a smaller network of systems. By having smaller
networks, network traffic congestion is reduced and overall network
performance and traffic control are improved.
A network’s routing configuration does not always require a routing protocol.
In situations where the routing information does not change, for example,
when there is only one possible route, the system administrator usually builds
the routing table manually. Some networks have no access to any other
TCP/IP networks, and therefore do not require routing tables at all. The three
most common routing configurations are:
Router Node 9
• Minimal routing
A network completely isolated from all other TCP/IP networks requires
only minimal routing. A minimal routing table is usually built by
ifconfig
when the network interfaces ar e conf igured. If your network does not have
direct access to other TCP/IP networks, and if you are not using
subnetting, this may be the only routing table you require.
• Static routing
A network with a limited number of gateways to other TCP/IP networks
can be configured with static routing. When a network has only one
gateway, a static route is the best choice. A static routing table is
constructed manually by the system administrator using the
route
command. See Figure 4. Static routing tables do not adjust to network
changes, so they work best where routes do not change.
Source HostDestination Host
ApplicationApplication
Transport
Gateway
Transport
Destination
192.168.1.0
192.168.12.0
default192.168.12.1
Network AccessNetwork AccessNetwork Access
Figure 4. Table-Based Routing
192.168.12.3
192.168.12.2
192.168.12.2192.168.12.3192.168.1.2
192.168.12.0192.168.1.0
DestinationDestinationGatewayGatewayGateway
192.168.1.0
192.168.12.0
default192.168.12.1
192.168.1.5
192.168.12.3
192.168.1.5
192.168.1.0
default
192.168.1.2
192.168.1.5
• Dynam ic routing
A network with more than one possible route to the same destination
should use dynamic routing. A dynamic routing table is built from the
information exchanged by the routing protocols. The protocols are
designed to distribute information that dynamically adjusts routes to reflect
changing network conditions. Routing protocols handle complex routing
10IBM 9077 SP Switch Router: Get Connec ted to the SP S witch
situations more quickly and accurately than a system administrator can
do. Routing protocols are designed not only to switch to a backup route
when the primary route becomes inoperable; they are also designed to
decide which is the "best" route to a destination. On any network where
there are multiple paths to the same destination, a dynamic routing
protocol should be used.
2.1.4 Routing without the GRF
Before the GRF was available, there were only two ways to get IP traffic from
remote systems to reach the RS/6000 SP nodes:
1. By putting an addit ional IP adapter into every RS/6000 SP node.
2. By designating one or two nodes to act as a router (as shown in Figure 5).
Node
Router
Internet/Intranet
Node
SP Switch
Node
. . .
ATM
FDDI
Node
Ethernet
Figure 5. Routing without GRF
The first option was usually not chosen because it was too costly for the
following reasons:
• For systems with a large number of nodes having multiple IP adapters for
each RS/6000 SP node can be expensive.
• The number of I/O slots in the RS/6000 SP node is limited. In addition,
these slots are required to perform other tasks for the system, such as
connecting to disk or tape. Using these I/O slots to connect IP adapters
restricts the functions of the RS/6000 SP node.
Router Node 11
The second case has proven to be very expensive as well. The RS/6000 SP
node was not designed for routing. It is not a cost- effective way to route tr af fic
for the following reasons:
• It takes many CPU cycles to process routing. The CPU is not a dedicated
router and is very inefficient when used to route IP traffic (this processing
can result in usage of up to 90%).
• It takes a l ot of memory to store route tables. The memory on the RS/6000
SP node is typically more expensive than router memory.
The CPU on a node can only drive the system I/O bus at less than 80
megabytes per second, which is less than what a high-end router can do.
For these reasons, the performance of routers in handling IP traffic from
remote systems to the RS/6000 SP nodes was limited.
2.1.5 Routing with the GRF
The GRF is a dedicated, high-performance router (see Figure 6). Each SP
Switch Router adapter can route up to 30,000 packets per second and up to
100 MB/s into the SP Switch network in each direction simultaneously.
Node
Node
. . .
SP Switch
Node
GRF
Figure 6. Routing with GRF
12IBM 9077 SP Switch Router: Get Connec ted to the SP S witch
Internet/Intranet
ATM
FDDI
Ethernet
Loading...
+ 312 hidden pages
You need points to download manuals.
1 point = 1 manual.
You can buy points or you can get point for every manual you upload.