Hp PHOTOSMART R827 User Manual [sk]

SANworks by Compaq

Application Notes – HSG80 ACS Version 8.6-4P Data Replication Manager Design Guide

Part Number: AA–RQ78A-TE

First Edition (February 2002)

Product Version: ACS Version 8.6-4P

This document provides design assistance for Compaq customers, as well as for field and reseller presales technical support staff, for supportable SANworks design guide provides help by discussing the different design tradeoffs and by providing a checklist of required solution hardware and software components, based on the results of the design process.

For the latest version of these Application Notes and other DRM documentation, visit the Compaq storage website at: http://www.compaq.com/products/sanworks/drm/index.html.

Data Replication Manager by Compaq solutions. This

trademarks of Compaq Information Technologies Group, L.P. in the U.S. and/or other countries. Microsoft, Windows, and Windows NT are trademarks of Microsoft Corporation in the U.S. and/or other

countries. UNIX is a trademark of The Open Group in the U.S. and/or other countries. All other product names mentioned herein may be trademarks of their respective companies. Confidential computer software. Valid license from Compaq required for possession, use or copying. Consistent

with FAR 12.211 and 12.212, Commercial Computer Software, Computer Software Documentation, and Technical Data for Commercial Items are licensed to the U.S. Government under vendor's standard commercial license.

Compaq shall not be liable for technical or editorial errors or omissions contained herein. The information is provided “as is” without warranty of any kind and is subject to change without notice. The warranties for Compaq products are set forth in the express limited warranty statements accompanying such products. Nothing herein should be construed as constituting an additional warranty.

Compaq service tool software, including associated documentation, is the property of and contains confidential technology of Compaq Computer Corporation or its affiliates. Service customer is hereby licensed to use the software only for activities directly relating to the delivery of, and only during the term of, the applicable services delivered by Compaq or its authorized service provider. Customer may not modify or reverse engineer, remove, or transfer the software or make the software or any resultant diagnosis or system management data available to other parties without Compaq’s or its authorized service provider’s consent. Upon termination of the services, customer will, at Compaq’s or its service provider’s option, destroy or return the software and associated documentation in its possession.

Printed in the U.S.A.

Application Notes – HSG80 ACS Version 8.6-4P Data Replication Manager Design Guide First Edition (February 2002) Part Number: AA–RQ78A-TE

2 Application Notes – HSG80 ACS Version 8.6-4P Data Replication Manager Design Guide

Application Notes Contents

These Application Notes cover the following major topics:

• Other Documentation, page 4

• Introduction, page 4

• DRM Overview, page 4 — Supported Hardware and Software, page 5

• Design Tradeoffs, page 6 — Business Requirements, page 6 — High Availability, page 7 — Threat Radius, page 7 — Disaster Tolerance, page 9 — Physics of Distance, page 9 — Synchronous versus Asynchronous Replication, page 13 — Bidirectional Solution, page 14

Application Notes Contents

— Determining Write-to-Disk Rate, page 15

• DRM Storage Design Considerations, page 16 — Disaster-Tolerant LUNs, page 16 — Non-Disaster-Tolerant LUNs, page 16 — Array Limits, page 16 — Fabric Port Count Results, page 17 — Failover Requirements Planning, page 17

• DRM Configurations, page 18 — Basic DRM-over-Fiber, page 18 — Long Distance Solutions, page 21 — Special-Purpose DRM Solutions, page 26

• Checklist for Supportable Solution Configuration Confirmation, page 30 — Basic DRM or DRM-over-WDM Configurations, page 30 — DRM Fibre Channel over ATM Configurations, page 33 — DRM Fiber Channel over IP Configurations, page 35

• Glossary, page 38

Application Notes – HSG80 ACS Version 8.6-4P Data Replication Manager Design Guide 3

Introduction

DRM Overview

DRM is a controller-based data replication software solution for disaster tolerance and data movement. DRM works with the StorageWorks™ by Compaq Fibre Channel MA8000/EMA12000/EMA16000 and RA8000/ESA12000 storage systems.

Data Replication Manager by

DRM allows all data to be mirrored between storage elements in two different storage arrays that can be in separate geographical locations. Each write I/O is sent to both storage locations, and reads occur only at the local storage. Properly configured, DRM can be a complete disaster-tolerant storage solution that guarantees data integrity in case of a storage subsystem or site failure.

4 Application Notes – HSG80 ACS Version 8.6-4P Data Replication Manager Design Guide

DRM Overview

DRM copies data online and in real time to remote locations through a local or extended storage area network (SAN). For more about the features and benefits of DRM, refer to the Compaq Features and Benefits of ACS 8.6P Data Replication Manager white paper.

A basic DRM configuration is shown in Figure 1.

Server

Figure 1: SANworks Data Replication Manager basic configuration

Supported Hardware and Software

Table 1 details the hardware and operating systems supported by DRM as of the time this document was published. Refer to the DRM website for a current list of supported hardware and operating system versions:

http://www.compaq.com/products/sanworks/drm/index.html

Switch

Initiator

Switch

Target

Table 1: DRM-Supported Host Servers/Operating Systems

Host Server Supported Operating System

Compaq AlphaServer™ Compaq Proliant™ Server Hewlett-Packard HP9000—L, N, V Class IBM—RS6000 Sun—Ultra SPARC

OpenVMS™ , Tr u6 4 ™ UNIX Novell NetWare, Windows 2000, Windows NT HP-UX AIX Solaris

Table 2 lists the supported operating systems by version and the SCSI command level that each operating system supports. Additional information on sharing of Compaq storage subsystems is available in the Compaq StorageWorks Heterogeneous Open SAN Design Reference Guide. Other supported operating systems and specific versions may be available from the Compaq Custom Systems Group.

Application Notes – HSG80 ACS Version 8.6-4P Data Replication Manager Design Guide 5

Design Tradeoffs

Table 2: DRM-Supported Operating System Versions

Operating System Supported Version

Compaq OpenVMS V 7.2-2, 7.3

• VMS Clusters appropriate to OS version

Compaq Tru64 UNIX V5.1, 5.1a

• TruClusters appropriate to OS version

Hewlett-Packard HP-UX

IBM AIX V4.3.3

Microsoft Windows 2000 for Server, Advanced Server, Datacenter Server

Microsoft Windows NT Server

Novell NetWare V5.1, 6.0

SUN Solaris 2.6 (32-bit mode only)

V11.0

• HP MC/Service Guard Clusters VA.11.12

• Cluster Object Manager VA.01.01

• Secure Path for HP-UX V3.0

• HACMP Clusters V4.4.0

• Secure Path for AIX V2.0A

• Service Pack 2

• MSCS V1.1

• Secure Path for Windows V3.1A with SP1

V4.0

• Service Pack 6a with hotfix

• MSCS V1.1

• Secure Path for Windows V3.1A with SP1

• Secure Path for Novell V3.0B

7, 8 (32- and 64-bit mode)

• Secure Path for SUN V2.1D, 3.0

• VERITAS Clusters V1.3

• SUN Clusters V2.2

SCSI-2

Support

No Yes

Ye s Ye s

Yes Yes (V7, 8)

SCSI-3

Support

No (V2.6)

DRM supports the ability of compatible operating systems to share the same controller. To be compatible, all the operating systems must support the same level of SCSI command and control.

Design Tradeoffs

The following sections describe tradeoffs that need to be considered when you design a DRM solution.

Business Requirements

The first step in designing a data replication solution is to understand what is driving the need for it. In some cases, there is a business need for high availability, and disaster tolerance is seen as a side benefit. In other cases, there is a business requirement for disaster-tolerant data, with high availability of the applications being viewed as a side benefit. Neither configuration, however, satisfies both requirements without additional hardware or software. DRM provides disaster-tolerant data, while clustering technology (such as VMSClusters, TruClusters, Service Guard Clusters, HACMP, MSCS, VERITAS Clusters, and SUN Clusters) supports high availability of applications. Together, these solutions provide highly available applications with disaster-tolerant data.

6 Application Notes – HSG80 ACS Version 8.6-4P Data Replication Manager Design Guide

High Availability

DRM provides highly available and reliable access to data, but because it is a storage-centric solution, it does not provide highly available applications. To achieve highly available applications, standby servers are needed to provide the processing platforms used by the applications if the primary system fails.

Customers can deploy a cluster of servers at the primary site, with either a single backup server or cluster of servers at the remote site. The resulting configuration provides highly available and disaster-tolerant applications and data. In addition, DRM can function in both directions between sites to optimize the use of the equipment.

Table 3 shows the importance of high availability. Note that even at an availability of 99%, an application could be down for nearly 4 days per year.

Table 3: Real-world Availability: Outage Minutes per Year

Design Tradeoffs

Availability

Outage Minutes

per Year

Table 4 illustrates expected user outage minutes per year for several server types.

Table 4: User Outage Minutes and Relative Availability

User Outage

Minutes per

User Year

Relative

Availability

Threat Radius

The threat radius is the distance from the center of a threat to the outside perimeter of that threat. For example, half the diameter of a hurricane or typhoon is the threat radius of that storm. Another example might be where a chemical factory is at the center of the threat radius circle. The threat radius is defined by the downwind distance of danger if there is a leak of hazardous chemicals.

90% 95% 99% 99.9% 99.99% 99.999% 100%

50,000 25,000 5,000 500 50 5 0

Windows NT

Server

8,200 3,150 1,150 180

46 times less

reliable than

Himalaya

Windows NT

Server & MSCS

18 times less

reliable than

Himalaya

Server & MSCS &

NonStop Software

6 times less reliable than

Himalaya

NonStop Himalaya

OS & NonStop

Software

The three general classes of threat in a DRM solution are local, within the metropolitan area, and regional. These three classes make it easier to characterize the solution in terms of intersite link options, performance, and system reliability. These three types of threats are defined as follows:

Local—Any threat that is less than a few kilometers in radius (less than 25 square kilometers), such as a tornado or a plant fire, is a local threat. Local replication has the least impact on performance as compared to the other options.

Metropolitan—Any threat that is larger than a local threat, and which might extend from 25 square kilometers to 5,000 square kilometers, such as a large chemical incident, a moderate earthquake, or a severe storm, is considered a metropolitan threat. Replication outside of metropolitan threats is similar in performance cost to running older disk drives—it is slower but acceptable.

Application Notes – HSG80 ACS Version 8.6-4P Data Replication Manager Design Guide 7

Design Tradeoffs

Regional—Any disaster that affects a radius of hundreds of kilometers to tens of thousands of kilometers, such as a large flood or hurricane, is considered a regional disaster. By sheer size, the regional disaster requires the largest separation distance when planning disaster-tolerant solutions. Depending on the distances, data replication to the outside of a regional disaster threat radius will impact system performance. However, separation distances of more than 1,000 kilometers are rarely needed and only increase the cost of the link, rather than provide additional disaster tolerance.

In review, local disasters usually do not exceed a threat radius of a few kilometers. A metropolitan disaster covers threats with a radius greater than a few kilometers to tens of kilometers. The typical regional disaster covers hundreds of kilometers.

Examples of each disaster include:

• Building fire: local.

• Tornado: local, but possibly metropolitan if it stays on the ground for a long time.

• Hurricane/typhoon: metropolitan to regional, depending on size and intensity.

• Floods: local to regional along the flood plain.

• Earthquake: usually local to metropolitan, depending on severity.

• Environmental: local.

• Power loss: local to regional, depending on which component fails—a local distribution point or a

major portion of the regional power grid.

In considering the threat radius, planners must decide what the threats are to the primary system, and whether those threats also apply to the backup system. For example, it would be unwise to place both sites in the same flood plain because one flood could destroy both. Similarly, if severe storms tend to travel in a certain direction, then a good strategy would be to place the two sites perpendicular to the expected route of travel and as far apart as possible to prevent one storm from affecting both sites.

Figure 2 is an example of the relative relationship between the three classes of threats based on the radius of those threats.

Local (few km)

Regional (100s km)

Metropolitan (10s km)

Figure 2: Threat radius

8 Application Notes – HSG80 ACS Version 8.6-4P Data Replication Manager Design Guide

Disaster Tolerance

Disaster tolerance is the ability of a system to withstand a defined disaster, such as being outside the threat radius for that potential disaster. DRM enables applications to automatically and continuously build two copies of application data at geographically separated sites.

The larger the potential disaster (threat radius), the farther apart the primary and backup sites need to be. To determine an adequate distance, first determine what kinds of disasters are probable in the local area and understand the protection distances needed to separate the two sites. Consider any local regulatory requirements that might increase or limit the separation distance. For example, some counties in the United States require both sites to remain within the same 100 to 400 sq. km county. This restriction has limited the maximum separation distance to less than 30 km in an area prone to earthquakes. Such earthquakes have impacted buildings several hundred kilometers from the earthquake's epicenter.

On the east coast of the United States, and the southern and eastern coasts of the Asian subcontinent, hurricanes or typhoons can cover an area with a radius of 200 km or more. Other natural disasters include regional forest fires, localized tornadoes, and widespread flooding. Non-natural disasters include building fires or chemical contamination, either of which can limit access to computer facilities.

In order to deal with the types and scope of potential disasters, storage system planners must consider this tradeoff—with increasing distance comes lower performance.

Design Tradeoffs

Physics of Distance

A data replication product can move data at extreme distances, but due to propagation delays inherent in data transfers, it may take a long time for each replication I/O to be completed. The distance between the two sites, not the width of the pipe, is the limiting factor in replication performance. This section provides a brief overview of the impact of distance on the sustainable I/O rate.

i t

Figure 3: Impact of distance on I/O rate

In Figure 3 there are two I/O pipes: one narrow arrow representing a slow communications channel, and a wider arrow representing a high-speed intersite link. Both I/O packets (shown by the parallelograms) contain the same amount of data. They are moving from Site A to Site B (left to right), and the leading edge of both will arrive at Site B at the same time. The difference in pipe size (bandwidth of the link) allows the faster pipe to complete delivery of the packet before the slower pipe. The difference is that the wider pipe provides shorter times to load and unload the packet. At long distances, the communications bandwidth becomes a less important factor in determining the I/O rate.

i t

Application Notes – HSG80 ACS Version 8.6-4P Data Replication Manager Design Guide 9

Design Tradeoffs

The time required to load or unload a packet can be approximated using the bandwidth of the pipe and the size of the packet. For example, given a 16-KB packet and a T1 transmission speed of 1.54 Mbps, it takes approximately 0.1 seconds to load the packet onto the link. Using a Fibre Channel link running at 1,000 Mbps and the same 16-KB data packet, the time is reduced to 0.00016 seconds.

Table 5 provides the data for calculating how long it takes to complete a single I/O across a zero distance intersite link, based on the link bandwidth and the size of the data packet. The columns are the transmission technology, performance or bandwidth of common intersite links, the slope (m) of the line that approximates the time it takes to load the data onto a given pipe, and the Y-intercept (b) representing the replication and conversion overhead.

Enter the size of the data packet (X) in kilobytes and calculate the zero distance I/O completion times (Y) as follows:

Y (in milliseconds) = mX+b Note that intercept (b) represents the minimum amount of time it takes to load one 512-byte data

packet onto the intersite link. In addition, the slope (m) of the line represents the amount of additional time it takes to insert ever larger data packets onto the intersite link.

Table 5: Link Performance Table

Fibre Channel 1000 0.056 0.84 1-GbE (Gigabit Ethernet) IP 1000 0.056 0.84 OC3-ATM 155 0.134 1.8 100-Mbit IP 100 0.134 1.8 100-Mbit IP with very high compression 100 0.056 0.84 T3-ATM 44 0.32 2.8 10-Mbit IP 10 1.02 3.0 10-Mbit IP with very high compression 10 0.134 1.8

Intersite Link Technology

Intersite Link Bandwidth

(Mbps) Slope (m) Intercept (b)

NOTE: m = a number proportional to the amount of time it takes to send larger packets

versus smaller packets. b = the replication overhead—the amount of time it takes to replicate one 512-byte block.

The second factor in determining how long it takes to complete a single I/O over distance is the time it takes for the leading edge of the data to travel from one end of the link to the other. This time can be estimated using a speed of 200,000,000 m/s for light in standard fiber-optic cable, which equates to 5 microseconds per kilometer. For an intersite separation distance of 10,000 km, the time interval would be 50,000 microseconds or 0.05 seconds, and is 25 times the average rotational latency of a 15,000 rpm disk drive (which is 0.002 seconds).

DRM implementation of the FC-defined SCSI protocol requires four trips through the intersite link to complete a single write. First, the SCSI initiator must ask the SCSI target if it is ready to accept data. Second, the SCSI target indicates it is ready. Then, the SCSI initiator sends the data to the target. The final trip occurs when the SCSI target replies that it has received the data. The true time it takes to move data from the initiator controller to the target controller is four times the single-trip distance, plus the time it takes to load the data for a given pipe size. These four trips, consisting of three small control packets and one large data packet, equate to 20 microseconds of latency per kilometer of intersite link distance to complete the SCSI write. Based on the distance between the two sites, this latency is added to the previously calculated time to complete a zero distance I/O.

10 Application Notes – HSG80 ACS Version 8.6-4P Data Replication Manager Design Guide

Two methods are used to calculate the intersite latency based on the cable distance between the two sites. One is based on the driving distance, the other on the network latency. If an intersite network exists, use the more accurate network latency as the metric. If an intersite network does not exist, then use the driving distance approach.

Driving Distance

To estimate the cable distance between two sites, measure the distance by driving a car between the two sites. For point-to-point networks, multiply the driving distance by 1.5. For routed networks, multiply the driving distance by 2.25. For example, if two sites are located 150 km apart, the estimated cable distance for a point-to-point network would be 225 km.

Network Latency

If an intersite network connection exists, ask the network engineers for an estimate of the one-way or round-trip intersite latency. In the preceding example, the actual point-to-point network was measured to give one-way latency of approximately 5 milliseconds or 250 km, instead of the 225 km originally estimated by using driving distance.

Single Stream I/O Example

DRM invokes a process called normalization whenever a remote copy set (RCS) is created. The remote copy set function allows data to be copied (mirrored) from the initiator site to a target site. The result is mirrored data at two disparate sites. During the normalization process, data is copied from the initiator side of the logical unit to the target side of the RCS. Each write copies 128 blocks of 512 bytes (or 64 KB of data) from the initiator to the target. To prevent overwhelming the system, only one write per RCS is allowed to be outstanding. This is defined as a single synchronous I/O stream.

Design Tradeoffs

For the purpose of this example, assume that the two sites are located 100 km apart and that the RCS consists of two 9-GB disk drives, one at each site.

Table 6 provides the calculation results for loading 64 kilobytes of data onto the different links.

Table 6: Calculation of Time to Load Data onto an Intersite Link

Technology Slope Time to Load 64 KB of Data (ms)

Direct FC or GbE 0.056 3.58 FC/ATM OC-3 or 100 Mbps 0.134 8.58 FC/ATM T3 0.32 20.48 FC/IP 10-Mbit with no compression 1.02 65.28

Given the intersite distance of 100 km, the transfer latency is 100 x 20 microseconds/km = 2 milliseconds, and is approximately the same for all interconnects.

Table 7 provides the calculation for adding the two results together, plus the replication overhead.

Table 7: Calculation Time to Complete a Single I/O Over Distance

+ Intersite

Technology Time to Load (ms)

Direct FC or GbE 3.58 2 0.84 6.42 FC/ATM OC-3 or

100 Mbps FC/ATM T3 20.48 2 2.8 25.28 FC/IP 10 Mbps 65.28 2 3.0 70.28

8.58 2 1.8 12.38

Latency (ms) + Overhead (ms)

= Time to

Complete I/O (ms)

Application Notes – HSG80 ACS Version 8.6-4P Data Replication Manager Design Guide 11

Design Tradeoffs

Inverting the time-to-complete number (milliseconds per I/O) produces an approximate number of single-stream synchronous writes that can be completed every second for the various bandwidths of pipes for the example distance. The results are shown in Table 8 as approximate times needed to normalize a 9-GB logical unit number (LUN).

Table 8: Relationship of I/O Per Second to Throughput and Time to Normalize a 9-GB LUN

Approximate I/O

Technology

Direct FC or GbE 155.7 35.87 GB per hour 0.25 hour FC/ATM OC-3 or 100 Mbps 80.8 18.62 GB per hour 0.48 hour FC/ATM T3 39.6 9.11 GB per hour 0.98 hour FC/IP 10 Mbps 14.2 3.28 GB per hour 2.75 hours

You can also use the Compaq SANworks Data Replication Manager Inter-site Link Performance Analyzer Calculation Tool to perform these same calculations.

Multiple I/O Streams for a Single Application

After determining the worst case for the time duration to perform the replication of a single write, you must examine the impact of multiple I/O streams from a single application. Multiple writes can be replicated at the same time, as shown in Figure 4. The wider the pipe, the more writes it can hold, up to a maximum called the bandwidth-latency product. Multiply the net performance of the communications pipe (total bits per second, less any overhead) times the intersite latency in seconds. This number, when expressed in bytes (use 10 bits per byte) and divided by the average message size (in bytes), determines the average number of messages that can be in the communications pipe at one time.

In Figure 4, the parallelograms represent two I/O streams. Each I/O stream consists of data packets of equal size (same number of bytes). The narrow pipe is able to hold only three packets, but the wider pipe is able to hold nine packets.

per Second

Throughput (Based

on 64KB per I/O)

Approximate Time Required

to Normalize a 9-GB LUN

i t

Figure 4: The impact of multiple I/O streaming for single applications

Two other constraints will further limit the maximum number of I/O streams in the pipe. The first limit is the number of Fibre Channel buffer-to-buffer credits allowed between any Fibre Channel devices. Using Brocade Fibre Channel switches running firmware V2.1.9m, this number is currently defaulted at 16, for 16 open message exchanges. This limit is seen on any long distance direct Fibre Channel connection with very long distance Gigabit Interface Converters (GBICs) or wavelength division multiplexing (WDM) solution. It is not usually seen in DRM-over-ATM or DRM-over-IP because the credit is returned by the ATM or IP gateway to the sending switch.

i t

12 Application Notes – HSG80 ACS Version 8.6-4P Data Replication Manager Design Guide

+ 27 hidden pages