Hp COMPAQ PROLIANT 5000, COMPAQ PROSIGNIA 500, COMPAQ PROLIANT 4500, COMPAQ PROLIANT 4000, COMPAQ PROSIGNIA 720 Backup of Windows NT SAP R/3 Systems with ARCserve

WWHITE HITE PPAPERAPER

.

December 1996 Compaq Computer

Corporation

CONTENTS

Test EnvironmentTest Environment .................... 33

Test ConceptTest Concept ................................ 44

ARCserveARCserve

FunctionalityFunctionality.................................. 55

ARCserve Installation............ 5

ARCserve Manager............... 6

ARCserve Tape Label

Handling............................10

ARCserve Engine

Administrator ......................12

ARCserve RAIDARCserve RAID

EngineEngine ................................................1313

Tape Array Functionality.......13

Compression TestsCompression Tests ..............1515

Capacity of DLT Cartridges...15 Compression Factor of an

Initial R/3 System ................15

Compression Factor of a

Productive R/3 System.........16

Summary ............................16

Performance of thePerformance of the

Disk SubsystemDisk Subsystem ......................1717

ARCserve BackupARCserve Backup ..................2121

Backup on Windows NT File

System Level......................21

Backup Using the Image

Option ................................22

Online Backup Using ARCserve

R/3 Agent for Oracle ............24

Online Backup Using R/3

SAPDBA .............................28

Network BackupsNetwork Backups .................... 2929

Pull Agent...........................30

Push Agent .........................32

ARCserve RestoreARCserve Restore ................3232

Restore of Windows NT File

Backup...............................32

Restore of Image Backup......34

Restore of Online R/3

Database Agent Backup .......35

Restore of an SAPDBA

Backup...............................36

SummarySummary..........................................3838

AppendixAppendix .......................................... 3939

466A/1196

.

Backup of Windows NT SAP R/3 SystemsBackup of Windows NT SAP R/3 Systems

.

with ARCservewith ARCserve

.

EXECUTIVE SUMMARY

.

Cheyenne is the main partner of Compaq for backup solutions. According to the growing

.

interest on professional backup solutions Compaq did an evaluation of the functionality

.

and performance of Cheyenne ARCserve 6 for Windows NT, which is documented in this

.

White Paper.

.

Compaq tested the functionality of ARCserve 6 but the main focus of the tests was to get

.

performance data of backup and restore in various corner scenarios that will allow

.

predictions for real-life behavior. All the tests were done with SAP R/3 on an Oracle

.

system.

.

The maximum backup rate that was achieved for a backup on Windows NT file system

.

level was 39 GB/h. Using the ARCserve Image Option that bypasses the Windows NT file

.

system for even higher performance between 46 GB/h and 53 GB/h could be stored,

.

depending on the type of tape RAID array. Compaq also did online R/3 backups,

.

simulating different load conditions by using the SAP benchmark suite. The backup rate

.

was between 20 GB/h with high R/3 load and 43 GB/h with low R/3 load.

.

The restore rate was not primarily determined by the tape drive subsystem, but in the

.

high performance cases was restricted to the write performance of the disk subsystem.

.

The maximum restore rate was between 19 GB/h and 25 GB/h depending on the type of

.

the disk RAID array.

.

11

WWHITE HITE PPAPERAPER (cont.)

.

NOTICE

.

The information in this publication is subject to change without notice.

.

466A/1196

.

COMPAQ COMPUTER CORPORATION SHALL NOT BE LIABLE FOR TECHNICAL

.

OR EDITORIAL ERRORS OR OMISSIONS CONTAINED HEREIN, NOR FOR

.

INCIDENTAL OR CONSEQUENTIAL DAMAGES RESULTING FROM THE

.

FURNISHING, PERFORMANCE, OR USE OF THIS MATERIAL.

.

This publication does not constitute an endorsement of the product or products that were tested.

.

The configuration or configurations tested or described may or may not be the only available

.

solution. This test is not a determination of product quality or correctness, nor does it ensure

.

compliance with any federal, state or local requirements. Compaq does not warrant products other

.

than its own strictly as stated in Compaq product warranties.

.

Product names mentioned herein may be trademarks and/or registered trademarks of their

.

respective companies.

.

Compaq, Contura, Deskpro, Fastart, Compaq Insight Manager, LTE, PageMarq, Systempro,

.

Systempro/LT, ProLiant, TwinTray, ROMPaq, LicensePaq, QVision, SLT, ProLinea, SmartStart,

.

NetFlex, DirectPlus, QuickFind, RemotePaq, BackPaq, TechPaq, SpeedPaq, QuickBack, PaqFax,

.

Presario, SilentCool, CompaqCare (design), Aero, SmartStation, MiniStation, and PaqRap,

.

registered United States Patent and Trademark Office.

.

Netelligent, Armada, Cruiser, Concerto, QuickChoice, ProSignia, Systempro/XL, Net1, LTE

.

Elite, Vocalyst, PageMate, SoftPaq, FirstPaq, SolutionPaq, EasyPoint, EZ Help, MaxLight,

.

MultiLock, QuickBlank, QuickLock, UltraView, Innovate logo, Wonder Tools logo in

.

black/white and color, and Compaq PC Card Solution logo are trademarks and/or service marks

.

of Compaq Computer Corporation.

.

Other product names mentioned herein may be trademarks and/or registered trademarks of their

.

respective companies.

.

Microsoft, Windows, Windows NT, Windows NT Advanced Server, SQL Server for Windows NT

.

are trademarks and/or registered trademarks of Microsoft Corporation.

.

Backup of Windows NT SAP R/3 Systems with ARCserve

.

First Edition (December 1996)

.

466A/1196

.

22

WWHITE HITE PPAPERAPER (cont.)

.

TEST ENVIRONMENT

.

466A/1196

The following lists show what hardware and software was used in the test environment.

.

Hardware

.

The following hardware was used for evaluating the Cheyenne ARCserve product:

.

Computer Compaq ProLiant 5000

.

Processors 2 x 200 MHz Pentium Pro, L2 Cache 512 Kbytes

.

Memory 1 gigabyte

.

Disks C: 2 x 2.1 gigabyte RAID1

.

Controller 4 x Compaq 32 bit Fast Wide SCSI-2 Controller for Compaq 15/30

.

Tapes 8 x Compaq 15/30 DLT Tape Firmware 840B

.

Network 100 Mbit Ethernet

.

Software

.

The following software was used for evaluating the Cheyenne ARCserve product:

.

Operating system Windows NT3.51, Build 1057, SSD 1.18, Service Pack 4

.

Database Oracle 7.2.2.4.5

.

Database Data SAP IDES Client Version 3.0D

.

SAP R/3 R/3 3.0D

.

SAP R/3 backup tools SAPDBA 3.0C

.

Oracle Agent Cheyenne R/3 Agent for Oracle 2.0

.

Backint Agent Cheyenne R/3 Agent for Oracle 2.0

.

Backup product Cheyenne ARCserve 6.0

.

DLT driver Cheyenne ARCserve DLT driver

.

33

E: 2 x 2.1 gigabyte RAID1 F: 10 x 4.3 gigabyte RAID 5 G: 2 x 2.1 gigabyte

DLT Tape Drives 2 x Compaq SMART-2 Array Controller for Disks

WWHITE HITE PPAPERAPER (cont.)

.

TEST CONCEPT

.

466A/1196

The test scenarios discussed in this White Paper focused on the parallel operation of DLT tape

.

devices to improve backup performance. Compaq testing verified the functionality of the

.

integration solution and then investigated different off-line and online scenarios.

.

NOTE: Compaq does not recommend using DAT tape drives for production systems

.

due to their low performance and reliability. All tests for this White Paper were run only

.

with DLT tape drives.

.

Testing revealed that the particular configuration of a disk subsystem has much more influence on

.

the performance than initially expected. As you read this White Paper, then, you will find some

.

test scenarios using different disk subsystem configurations to show how the configurations

.

effected performance.

.

Test scenarios, although carefully chosen, create certain results which might not always be

.

achievable in all production sites. In addition, these tests are not meant to substitute or

.

complement any of the vendor’s quality tests of these software products.

.

To have representative data Compaq used the SAP IDES client. Together with the SAP R/3

.

benchmark client the database had a size of about 16 gigabytes. See the Appendix for detailed

.

information.

.

The tests in this document are divided into the following parts:

.

• Functionality of ARCserve

.

• Compression tests

.

• Performance of the disk subsystem

.

• Backup tests

.

• Restore tests

.

44

Testing the functionality of the ARCserve software.

The goal of these tests was to well understand hard- and software compression and to determine typical compression figures for R/3 on Oracle systems.

These tests determine the performance characteristics of the disk drive array.

Testing on Backup of files, Image backup, Online backup with ARCserve R/3 Agent for Oracle, Off-line and online backup with the SAPDBA via backint. See section “ARCserve Functionality” for detailed information about the different backup types.

Testing on Restore of files, Image restore, Restore with ARCserve R/3 Agent for Oracle, Restore with the SAPDBA via backint.

WWHITE HITE PPAPERAPER (cont.)

.

ARCSERVE FUNCTIONALITY

.

466A/1196

The following section briefly describe the functionality of ARCserve. See the Cheyenne ARCserve

.

documentation for more detailed information.

.

ARCserve Installation

.

ARCserve is installed by simply calling the setup program on the ARCserve CD. The basic

.

ARCserve component is installed first. The following options have to be purchased and installed

.

separately:

.

• SAP R/3 agent for backup and restore of an Oracle based R/3 system. This agent includes the

.

backint interface for an R/3 backup using the SAPDBA program. The SAP R/3 agent is

.

similar to the Oracle database agent but additionally contains some special settings for R/3.

.

• Image option for fastest backup and restore of complete partitions, bypassing the Windows NT

.

file system.

.

• RAID engine for operating several physical tape drives in a drive array, which is very similar

.

to the RAID operation of hard disks in a disk drive array.

.

After you have installed the core product plus all available options of ARCserve you get a

.

program group in your Windows NT Program Manager as shown in Figure 1.

.

If you are using the backint interface you additionally have to:

.

• edit the init<SID>.sap parameter file that contains the settings for the R/3 backup

.

55

and restore : backup_dev_type = util_file util_par_file = <name of the ARCserve parameter file, e.g. initJM2.utl>

Figure 1: ARCserve program group

WWHITE HITE PPAPERAPER (cont.)

.

• edit the ARCserve parameter file for backint (e.g. initJM2.utl) by defining the tapes that

.

466A/1196

.

66

should be used for the backup (see Appendix for an example file) Both files can be found in the %ORACLE_HOME%\database directory. The following section will give you some brief information of the functionality of ARCserve 6.0

for Windows NT. See the ARCserve documentation or a product information of ARCserve for further information.

ARCserve Manager

The ARCserve Manager is the main program that is used for nearly all operations. After the startup of the ARCserve Manager you get the Quick Access window as shown in

Figure 2 where you can decide which operation you want to start. You can chose between:

• Job Status

• Backup

• Restore

• Copy

• Media Pool Management

• Device Management

• Database

• Reports

Figure 2: Quick Access window

You can switch to any other options after you started at any time.

WWHITE HITE PPAPERAPER (cont.)

.

BackupBackup

.

If you choose the backup item, the Backup window is displayed as shown in Figure 3. At this

.

point, you can select which kind of backup you want to start or schedule.

466A/1196

.

There are three major groups of backup that are provided by ARCserve for Windows NT.

.

The first group is the local backup (My Computer) where you can select single files, complete

.

disks, the registry, or databases if you have installed any database agent. If you select a complete

.

disk and you have installed the add-on Image Option you can make an image backup of this disk.

.

ARCserve Image Option for Windows NT allows the backup/restore of local hard drives in the

.

image mode. Image Option bypasses the Windows NT file system. Image Option reads/writes data

.

at sector level and also uses asynchronous I/O mechanisms for highest performance.

.

The R/3 database agent allows you to backup the complete database in online mode including the

.

SAPARCH directory and control files.

.

ARCserve provides the option to backup the registry of a running Windows NT.

.

The second group is the network backup (Network) that allows you to backup files via an

.

Windows NT share (pull concept). There is also an option to backup the registry of remote

.

servers.

.

The third group is the agent backups (Cheyenne Agents) that allows you to backup data from

.

Macintosh, Netware, OS/2, UNIX, Windows 3.x, Windows 95, and Windows NT clients (push

.

77

Figure 3: Backup window

WWHITE HITE PPAPERAPER (cont.)

.

concept). For this kind of backup you have to install the agent software on the client and

.

configure the new connection on the server.

.

466A/1196

.

RestoreRestore

.

If you choose the restore item, the Restore window is displayed as shown in Figure 4. At this

.

point, you can choose which kind of restore you want to start.

.

Each restore job requires a source and a destination. The files selected as your source will always

.

come from a tape. The location selected as your destination will always be a hard drive.

.

ARCserve provides you with four methods for selecting the data you want to restore (the source)

.

which are:

.

• Restore by Tree

.

• Restore by Tape

.

• Restore by Query

.

• Restore by Tape Media

.

The first three methods use information logged in ARCserve’s database, while the fourth method

.

does not. In addition, if you have purchased separately the Image Backup Option, Restore by

.

Image will also be available. You also have several methods for selecting a destination to restore

.

the data to the machine.

.

88

Figure 4: Restore window

WWHITE HITE PPAPERAPER (cont.)

.

The Restore by Tree view allows you restore a specific directory or drive. This view displays a

.

machine tree of files and directories that were backed up with ARCserve as shown in Figure 5. It

.

is also possible to restore individual files from an Image Backup.

.

466A/1196

.

First of all the Restore by Tape allows you to select the tape, then the session, and finally, the

.

files and directories you want to restore. This source view is useful if you know the tape that holds

.

the data you want to restore, but you are not sure of the right session it is in.

.

99

Figure 5: Restore by Tree

WWHITE HITE PPAPERAPER (cont.)

.

466A/1196

.

The Restore by Query source is used to restore files when you already know the name of the file

.

or directory you want to restore, but you do not know the machine it was backed up from or the

.

tape(s) it was backed up to. This view also uses the ARCserve database. As long as the file and

.

directory information is logged in the database, you can effectively use this view to select files to

.

restore.

.

The Restore by Tape Media allows you to restore complete sessions on a tape to a tape device.

.

All files in this session are restored to the destination, unless filters are added to the job. This

.

source is either useful in case information from your ARCserve database is unavailable, or in case

.

you try to recover your network after a disaster.

.

Image mode restore ("Restore by Image" mode in ARCserve Restore Manager) overwrites the

.

target volume data.

.

ARCserve Tape Label Handling

.

The ARCserve tape label handling is done by the different tools. The first one is the Device

.

Manager.

.

If you have more than one tape device connected to your Windows NT machine, ARCserve lets

.

you separate them into two or more "groups". Establishing tape groups is a key component of

.

ARCserve's flexibility and efficiency.

.

The tape groups enable you to perform the following options:

.

• parallel streaming - you can have several operations occurring simultaneously, one at each

.

1010

tape group configured for your system.

• automated tape spanning - if, during a backup session, the tape becomes filled, ARCserve

automatically spans to another tape in another device within the same tape group. By default, ARCserve is installed with each tape device assigned to its own group. ARCserve

finds identical tape drives during the installation, placing them automatically in the same group. Later, you can use the Device Manager to regroup your devices. As you can have up to seven tape devices connected to each SCSI bus in your machine, you can have up to seven tape groups per SCSI board.

Figure 6: Restore by Tape

WWHITE HITE PPAPERAPER (cont.)

.

There is only one rule when assigning tape devices to the same group; the devices in the group

.

must be identical. That means they must be the same make and model.

.

The Device Manager is also used for formatting tapes. By formatting a tape you can specify a tape

466A/1196

.

name, which represents together with the tape ID (4 digits chosen by ARCserve) a kind of tape

.

label.

.

The second tool is the Media Pool Management utility.

.

A media pool is a collection of media managed as a unit. Tape rotation or grandfather-father-son

.

(GFS) methods of backup are based on media pools. Each media pool is assigned a name and the

.

media is organized according to serial numbers. The serial numbers assigned are permanent and

.

if you are using a device with a bar code reader, the bar code labels will be used for the serial

.

number of the media. The media pools are organized by the range of serial numbers of the media

.

they contain and are divided into two sets, the Save Set and the Scratch Set.

.

The set of media containing important data that can not be overwritten is called the Save Set.

.

Once the media has passed several criteria these will be recycled and re-used. The media from the

.

Save Set, that has met several criteria (minimum number of media to save and retention period)

.

and therefore can be re-used and overwritten, is placed in the Scratch Set.

.

Each time a media is written to, it moves from the Scratch Set to the Save Set. That media will

.

move back to the Scratch Set once the specified criteria have been met.

.

Media pools apply to every media, regardless of the selected backup type and method. All rotation

.

backup jobs will create their own media pool based on the name entered in the Media Pool Name

.

field of the Schedule dialog box for simple rotation, or the prefix entered in the Media Pool Name

.

Prefix field with GFS enabled. The GFS backup uses three media pools:

.

• Daily

.

• Weekly

.

• Monthly

.

These three pools are also based on the information entered in the Media Pool Name Prefix field.

.

For example, if you enter ACTG as the Media Pool Name Prefix for an GFS backup job, the Daily

.

media pool name would be ACTG_DLY.

.

For a simple (single media pool) rotation, you specify the complete name for the media pool.

.

1111

Figure 7: Format window

WWHITE HITE PPAPERAPER (cont.)

.

ARCserve Engine Administrator

.

In the ARCserve Windows NT group, you can find the ARCserve Engine Administrator. With the

466A/1196

.

Engine Administrator you are able to start and stop the different ARCserve engines (Job, Tape,

.

and Database Engine). In this applet, you can also see the status and some more data of these

.

engines. Normally you do not need this utility because you can start and stop the engines with the

.

ARCserve Manager and the Windows NT Service Manager.

.

1212

Figure 8: ARCserve Server Admin

WWHITE HITE PPAPERAPER (cont.)

.

ARCSERVE RAID ENGINE

.

466A/1196

The ARCserve RAID Engine is generating out of a group of physical tape drives one or more

.

logical tape drives applying one of the RAID algorithms that are well known from hard disk

.

arrays. Currently there is support for RAID0 (striping) and RAID5 (distributed data guarding).

.

Support for RAID1 (mirroring) is planned for the next release of the RAID option.

.

You can only group the same kind of tape devices to a logical tape drive. It is not possible to mix

.

different drives like DLT-tapes with DAT-tapes. But you can define a group of DLT-tapes and a

.

group of DAT-tapes. The current implementation of the RAID engine allows for up to 8 physical

.

tape drives belonging to one logical tape drive. For performance reasons, you should never

.

connect more than 2 DLT drives to 1 SCSI controller. So in the test environment, Compaq used

.

up to 4x “32-Bit Fast-Wide SCSI-2/P Controllers” for the 8 DLT drives test scenarios.

.

You can choose between different RAID levels that are the same as in a disk environment.

.

Tape Array Functionality

.

ARCserve provides the capability to combine multiple physical tape drives into one physical

.

RAID array. Data is distributed equally over all tapes of the RAID system. As this increases the

.

probability of failure (if one tape of the RAID system is damaged, the whole backup is lost),

.

additional mechanisms for data safety are included in the RAID technology. There are three

.

RAID levels:

.

• RAID 0

.

RAID 0 only distributes data over all tapes of the RAID system. It does not provide any

.

features for error correction but provides a higher data throughput than single devices. 100%

.

of the tape net capacity can be used.

.

• RAID 1

.

RAID 1 uses the mirror technique to reduce the probability of data loss in case of a damaged

.

tape. Data is written on two tapes in parallel. For a RAID1 tape array at least two tape devices

.

are needed. 50% of the net capacity can be used for the backup. RAID 1 has not been available

.

in the tested version of ARCserve.

.

1313

Figure 9: ARCserve RAID options

WWHITE HITE PPAPERAPER (cont.)

.

• RAID 5

.

RAID 5 generates checksums for the data to be backed up and distributes data as well as

.

466A/1196

.

checksums over all tapes of the RAID system. If one tape is damaged, data can be restored

.

from the remaining tapes. This mechanism needs less tapes than RAID 1. A RAID 5 array

.

consists of at least three tape drives. It needs one additional tape, i.e. net capacity plus one

.

tape.

.

The tested ARCserve version supported RAID0 and RAID5 tape arrays. They can be configured

.

with the ARCserve RAID Option Setup program as shown in Figure 10.

.

1414

Figure 10: ARCserve RAID Option Setup

WWHITE HITE PPAPERAPER (cont.)

.

COMPRESSION TESTS

.

466A/1196

The following sections describe the compression tests.

.

Capacity of DLT Cartridges

.

Compaq determined the capacity of a 15 gigabyte DLT cartridge. The capacity was calculated by

.

writing data in no-compression mode to the tape until the „end of tape“ error occurred. The

.

following data was found:

.

Compression Factor of an Initial R/3 System

.

In this test, an initial R/3 system was used as it is created right after the installation of the CD-

.

ROM. The only deviation to this was the import of the benchmark client 900 that had been used

.

for the online scenarios. This resulted in a total size of the database of 4.0 gigabytes. For software

.

compression, Compaq used the compress command that comes with the R/3 software.

.

As there was no direct way to determine the hardware compression factor, this factor was

.

calculated by writing hardware-compressed data so many times to the tape until the „end of tape“

.

error occurred.

.

Datafiles of an initial R/3

.

system with Client900

.

Of course the compression factors of individual tablespaces of the R/3 database are different and

.

depend on the fill rate and the type of data that is stored in a specific datafile. In order to get a

.

better feeling for this effect the compression factor for individual tablespaces was determined.

.

PSAPSOURCED 1 : 2.00 1 : 2.24

.

PSAPLOADD 1 : 2.80 1 : 2.97

.

PSAPBTABD 1 : 22.37 1 : 13.92

.

1515

• DLT tape 15 gigabyte

SW compression

factor

1 : 5.02 1 : 4.92 1 : 4.36

Figure 11: Compression factors

Tablespace SW compression factor HW compression factor

Figure 12: Compression factors of SAP R/3 tablespaces

HW compression

factor DLT

HW compression

factor DAT

DLT

WWHITE HITE PPAPERAPER (cont.)

.

Compression Factor of a Productive R/3 System

.

It is clear that productive R/3 systems can have a lower compression factor than our initial R/3

466A/1196

.

system because tablespaces are more filled and the nature of data might be different.

.

In addition, we had the opportunity to determine the compression factor of a productive system

.

with a 18 gigabyte database, using the command “brbackup -k only”. Depending on the installed

.

processing power this can easily become an overnight operation and unfortunately had to be

.

stopped in our case before completion.

.

However, Compaq can compare the results gained up to the present:

.

Tablespaces PSAPBTABD .. PSAPSOURCED of

.

18 GB database of a productive R/3 system

.

Tablespaces PSAPBTABD .. PSAPSOURCED of

.

our initial R/3 system with Client 900

.

As expected, the compression factor of the test system was slightly better (about 10% ) than the

.

compression factor of a database in a large productive system. By talking with some SAP basis

.

consultants, Compaq believes that typically a factor of 1:3 could be used as a conservative

.

estimate. This would mean that 1 DLT cartridge can hold 15 gigabyte * 3 = 45 gigabyte of data,

.

as they are typically found in a R/3 system.

.

Summary

.

It has been realized that the hardware compression factor is about the same as the software

.

compression factor, at least for small compression rates up to 1:3. This shows that SAP’s

.

assumption hardware compression equals software compression is valid and their mechanism can

.

be used in the native mode via cpio for our tape devices.

.

Nevertheless it is important to understand what compression rates can be expected. As shown a

.

rate of 1:3 should be a good conservative estimate for a typical R/3 system. This means that a 15

.

gigabyte DLT cartridge can hold 45 gigabyte of R/3 data. Should a customer decide to employ 2

.

DLT drives in parallel for performance reasons, they could do an unattended backup of a 90

.

gigabyte database. So in a lot of cases, a tape library might not be needed as quickly as it initially

.

appears.

.

In general, you want to do a backup using a hardware compression mechanism. This helps to

.

reduce the demand for CPU power that is especially needed in online scenarios. Therefore all

.

ARCserve testing has been done only with hardware compression.

.

1616

Figure 13: Compression factors of a compete R/3 oracle database

SW compression factor

1 : 3.63

1 : 4.12

WWHITE HITE PPAPERAPER (cont.)

.

PERFORMANCE OF THE DISK SUBSYSTEM

.

466A/1196

The maximum performance with which a backup or restore job can be done is among other

.

factors determined by the performance of the disk subsystem. In order to understand this influence

.

on the overall ARCserve performance, Compaq did some isolated disk performance tests with

.

different disk subsystem configurations. Compaq used the Completion port I/O test utility

.

provided with the Compaq Resource Kit for Windows NT. Basically, 2 configurations were tested.

.

In one case, Compaq just operated an increasing number of disk drives on 1 SCSI channel of the

.

SMART-2 controller. In the other case, Compaq distributed an increasing number of disk drives

.

equally over the 2 SCSI channels of the SMART-2 controller. As RAID5 is a very cost effective

.

fault tolerance mechanism, Compaq started out with RAID5 configurations. However as naturally

.

write accesses to a RAID5 volume always carry the burden of the additional read and write access

.

for the parity generation, Compaq also looked at a few RAID1 configurations.

.

The following diagram depicts more details of the 1 and 2 SCSI channel configuration with the

.

example of a 6 disks configuration.

.

1717

• 6 * 4,3 GByte disks

• SMART-2 controller

• 6 disks connected to

one SCSI channel

SCSI channel 1 SCSI channel 2

In the tests with two SCSI channels, Compaq distributed the disks equally to the two disk cabinets.

• 6 * 4,3 GByte disks

• SMART-2 Controller

• 3 disks at SCSI channel 1,

3 disks at SCSI channel 2

SCSI channel 2SCSI channel 1

Compaq did tests from three disks, which means the minimum for a RAID 5 configuration was up to ten disks.

Available disks Disks used for

RAID 5 system

Available disks

Disks used for RAID 5 system

WWHITE HITE PPAPERAPER (cont.)

.

466A/1196

.

1818

RAID 5 disk configuration Read performance in

GB/h

3 x 4.3 GB disks on 1 SCSI channel 3 x 4.3 GB disks on 2 SCSI channels

4 x 4.3 GB disks on 1 SCSI channel 4 x 4.3 GB disks on 2 SCSI channels

5 x 4.3 GB disks on 1 SCSI channel 5 x 4.3 GB disks on 2 SCSI channels

6 x 4.3 GB disks on 1 SCSI channel 6 x 4.3 GB disks on 2 SCSI channels

7 x 4.3 GB disks on 2 SCSI channels 81 18 8 x 4.3 GB disks on 2 SCSI channels 91 19 9 x 4.3 GB disks on 2 SCSI channels 86 19 10 x 4.3 GB disks on 2 SCSI channels 94 19

Figure 14: Disk subsystem performance of RAID 5 configurations

To get the maximum performance, Compaq also did some tests with RAID 0 and RAID 1 configurations. It was found out, that in order to obtain the best write performance, you have to set the Stripe Factor of the Compaq SMART Array controller to 32. For RAID 5 configurations the default number is already 32. For RAID 0 and RAID 1 configurations the default value is 256. To change the Stripe Factor of the Compaq SMART-2 Array controller,the Compaq utility CONFIG.EXE was used. Figure 15 shows the maximum numbers optimized for the write performance.

RAID0, RAID1 disk configuration

Stripe Factor 32

12 x 4.3 GB RAID 0 90 32 14 x 4.3 GB RAID 0 99 33 12 x 4.3 GB RAID 1 63 26 14 x 4.3 GB RAID 1 74 26 7 x 4.3 GB RAID 0 on 1 SCSI channel 50 27

Figure 15: Disk subsystem performance of RAID 0 and RAID 1 configurations

As you see in Figure 15, the best write performance that is important for the restore process is 33 GB/h with 14 disks in a RAID 0 configuration. For SAP systems, the fault tolerance is of high significance. The fastest fault tolerance configuration which can be implemented is a RAID 1 configuration with 14 disks. With the RAID 1 configuration, Compaq got a read rate of 73 GB/h and a write rate of 26 GB/h. The last two numbers show that the overhead in the write process of a RAID 1 configuration is not very high (26 GB/h vs. 27 GB/h).

Read performance in

45 51

46 73

48 72

49 83

GB/h

Write performance

in GB/h

12 12

15 15

16 17

17 17

Write performance

in GB/h

WWHITE HITE PPAPERAPER (cont.)

.

SCSI channel 1

SCSI channel 2

RAID5

Drive F

SMART-2/P Array Controller

2.1 GB disk

RAID1

Drive C

RAID1

Drive E

2.1 GB disk

RAID1

Drive G

2.1 GB disk2.1 GB disk2.1 GB disk2.1 GB disk2.1 GB disk2.1 GB disk

4.3 GB disk

.

There is also a feature to change the cache settings of the Compaq SMART2 Array controller. For

.

all these tests, the SMART2 Array controller cache was set to 50% read and 50% write. Changing

.

the SMART-2 cache configuration did not improve the disk performance in the particular test

466A/1196

.

environment, as Figure 16 shows:

.

Cache configuration

.

(10 x 4 GB disks on 2 SCSIs)

.

50% read, 50% write 934 19

.

0% read, 100% write 93 19

.

100% read, 0% write 94 13

.

If not stated differently, Compaq used the following disk subsystem configuration as a result of

.

these test results. If highest restore rates possible are required the only (expensive) solution then

.

would be to utilize a mirrored disk subsystem for the volume(s) carrying the database files.

.

1919

4.3 GB disk

All drives are formatted with NTFS, block size 4 Kbytes. Drive C: NT operating system and page file Drive D: CD-ROM Drive E: SAP R/3 and Oracle executables, SAPARCH directory Drive F: Database Data Drive G: Online Redo Logs

DLT configurationDLT configuration

DLT tapes : 8 x Compaq 15/30 DLT tape firmware 840B For performance reasons Compaq never connected more than two DLT tape drives to one SCSI

Controller.

Figure 16: Comparison of different cache configurations

Figure 17:Disk configuration

Read performance in

GB/h

Write performance

in GB/h

WWHITE HITE PPAPERAPER (cont.)

.

Primary

PCI Bus

Secondary

PCI Bus

Slot 8: Fast Wide SCSI-2/P Controller

Slot 7: Fast Wide SCSI -2/P Controller

Slot 6: Fast Wide SCSI-2/P Controller

Slot 5: SMART-2/P Array Controller

Slot 4: Fast Wide SCSI-2/P Controller

Slot 3: SMART-2/P Array Controller

Slot 2: 10/100 TX PCI UTP Controller

DLT

DLTDLTDLT

F:

C: E: G:

.

PCI configurationPCI configuration

.

The assignment of PCI controllers to PCI slots was chosen very carefully in order to get the

.

maximum performance out of the 2 PCI buses of the ProLiant 5000. There is an oversimplified

466A/1196

.

rule which says that controllers among which a lot of I/O is expected to occur should be grouped

.

on one PCI bus and I/O traffic across the 2 PCI busses should be avoided.

.

There is a White Paper titled “Configuring the Compaq ProLiant 5000 Server for Peak

.

Performance“ available on the Web that can provide you with more details.

.

2020

Figure 18: Controller configuration

WWHITE HITE PPAPERAPER (cont.)

.

ARCSERVE BACKUP

.

466A/1196

As already described in the ARCserve functionality section, you have different options to do a

.

backup. With ARCserve you are able to make off-line and online backups. The off-line backup

.

types are backup on Windows NT file system level and backup with the IMAGE option, which

.

means that you backup complete logical drives without using the Windows NT File system. For

.

all the following backups, Compaq used an SAP R/3 Oracle database with the SAP IDES system

.

(which is an SAP system that already contains data for training purposes) and the SAP standard

.

benchmark 900 client. The total database size was about 16 gigabytes. This database was located

.

on a RAID 5 disk array that had the following layout.

.

2121

• 10 * 4,3 GByte disks

• SMART-2 Controller

• 5 disks at SCSI channel 1,

5 disks at SCSI channel 2

SCSI channel 1

SCSI channel 2

Free

Disks used for RAID 5 system

Backup on Windows NT File System Level

In this backup mode, you get a window where you can select from single files up to all drives on your computer shown in the upper part of Figure 18.

For the backup on Windows NT file system level, Compaq tested the performance of several tape array configurations. It always did a full backup of all the database files of our R/3 system. To get an impression if there is any performance difference between only one tape connected to a SCSI controller, Compaq did some test with only using 1 tape per SCSI controller.

WWHITE HITE PPAPERAPER (cont.)

.

In the tests, the following numbers were received:

.

466A/1196

.

2222

Number of tapes /

Number of SCSI-

Controllers

1 Tape / 1 Controller 10 not possible

2 Tapes / 1 Controller 14 not possible 2 Tapes / 2 Controllers 19 not possible 3 Tapes / 2 Controllers 21 13 3 Tapes / 3 Controllers 28 19 4 Tapes / 2 Controllers 27 20 4 Tapes / 4 Controllers 37 27 6 Tapes / 3 Controllers 39 33 8 Tapes / 4 Controllers 39 38

Figure 19: Performance of backup on NT file system level

From these tests, Compaq derived two results.

First, the backup performance is depending on the number of tapes connected to one SCSI controller. As you can see there is a big difference if you connect only one or more tapes to a SCSI Controller. The RAID 0 backup with 4 tapes connected to 4 controllers is about 10 GB/h faster as the 4 tapes connected to 2 controllers. Never connect more than 2 DLT drives to 1 SCSI controller because of performance reasons.

Second, the performance of a RAID 5 tape configuration is worse than that of a RAID 0 configuration. The worse performance of the RAID 5 configuration is caused by the fact that the parity information needed for a restore with one tape missing has to be calculated and then written on the tape. That is why you only need 2 tapes / 2 controllers with RAID 0 to get the same performance as for 3 tapes / 3 controllers with RAID5.

Backup performance in

GB/h with RAID 0

Backup performance in

GB/h with RAID 5

Backup Using the Image Option

The ARCserve Image Option for Windows NT allows the backup and restore of local hard drives in the image mode. This mechanism bypasses the Windows NT file system. Image Option reads/writes data at a sector level and also uses asynchronous I/O mechanisms for highest performance.

WWHITE HITE PPAPERAPER (cont.)

.

The image option is set in Backup menu by choosing Source Context Menu -> Use Image

.

Option…:

.

466A/1196

.

2323

Figure 20: ARCserve Image Option

WWHITE HITE PPAPERAPER (cont.)

.

Compaq saw the following Image Option performance results:

.

466A/1196

.

2424

Number of tapes /

Number of SCSI-

Controllers

1 Tape / 1 Controller 9 not possible

2 Tapes / 1 Controller 14 not possible 2 Tapes / 2 Controllers 19 not possible 3 Tapes / 2 Controllers 21 14 3 Tapes / 3 Controllers 28 18 4 Tapes / 2 Controllers 26 20

4 Tape / 4 Controllers 37 27 6 Tapes / 3 Controllers 40 34 8 Tapes / 4 Controllers 53 46

Figure 21: Performance of Image Option backups

The results of the Image Option backup almost correspond to those of the regular backup via the NTFS file system. There is a slower backup rate in the RAID 5 configuration which is caused by additional parity data. There is no gain of performance compared to the regular backup up to the 6 tapes on 3 controllers configuration. Only the 8 tapes on 4 controllers configuration is higher than the number of the normal backup. There you can see the advantage of the Image option over an NTFS file system backup.

Backup performance in

GB/h with RAID 0

Backup performance in

GB/h with RAID 5

Online Backup Using ARCserve R/3 Agent for Oracle

ARCserve provides an R/3 database agent for Oracle and SQL Server. In this test scenario, Compaq determined the backup and restore performance of the ARCserve R/3 database agent for Oracle. Compaq only did some online backups, as the off-line backup performance is covered with the NTFS file system and image backup.

WWHITE HITE PPAPERAPER (cont.)

.

466A/1196

.

The purpose of the online backup tests via ARCserve database agent is to understand how the R/3

.

load, generated by active R/3 users, affects the performance of the backup, and how the

.

responsiveness of the R/3 system is reduced by the backup activity.

.

The active R/3 users were simulated by using the standard SAP benchmark environment in

.

version 3.0D. Out of several benchmark modules the Financial module (FI) was chosen to

.

generate the system activity. It might happen, that particular customer installations will observe a

.

heavier load caused by their active R/3 users and that a lower online backup performance is

.

experienced. As each R/3 system is different in terms of configuration and customization, it is

.

impossible to come up with general guidelines. The FI benchmark load has been chosen to

.

establish some point of reference. The absolute number of users is not really important. However,

.

the CPU utilization and the average response time, that reflects the load situation the system is in,

.

are of more significance.

.

The “average R/3 dialog response time“ is a value indicating how long the R/3 system needs on

.

average to handle a dialog transaction started by a user. Typically a value of up to 2 seconds ( for

.

FI 1 second) is accepted by most companies.

.

The backup was made by using the ARCserve R/3 database agent for the whole database

.

including all data and control files.

.

2525

Figure 22: ARCserve architecture

WWHITE HITE PPAPERAPER (cont.)

.

Please note that the performance numbers of the tests delivered by ARCserve are not totally

.

correct. ARCserve opens a new backup session for each tablespace that is in the Oracle database.

.

The time between two sessions in not added to the total backup time. In the test case, the time to

466A/1196

.

be added to the backup was about 5-6 minutes. If you include this time the results diminish.

.

For example, the measured throughput of 52 GB/h with 8 tapes decreases to about 39 GB/h

.

including the 5-6 minutes overhead. The 20 GB/h of the 3 tapes RAID 0 configuration decrease to

.

18 GB/h real backup performance. The time not included in the total backup time depends on the

.

number of sessions which is equal to the number of database tablespaces.

.

Online Test without FI benchmark usersOnline Test without FI benchmark users

.

Number of Tapes /

.

Number of SCSI-

.

2626

Controllers

3 Tapes / 2 Controllers 20 15 13 15 4 Tapes / 2 Controllers 27 20 20 19 4 Tapes / 4 Controllers 36 25 27 25 6 Tapes / 3 Controllers 39 33 33 26 8 Tapes / 4 Controllers 52 42 45 38

Online Test with 40 FI benchmark users = 30% CPU loadOnline Test with 40 FI benchmark users = 30% CPU load

Number of Tapes /

Number of SCSI-

Controllers

3 Tapes / 2 Controllers 19 49 13 48 4 Tapes / 2 Controllers 24 53 19 50 4 Tapes / 4 Controllers 31 60 22 57 6 Tapes / 3 Controllers 33 62 28 57 8 Tapes / 4 Controllers 43 65 35 63

Average data

[GB/h]

RAID 0

Figure 23: Online backup with no system load

Average data

[GB/h]

RAID 0

Figure 24: Online backup with 30% system load

Average CPU

load

[%]

Average CPU

load

[%]

Average data

[GB/h]

RAID 5

Average data

[GB/h]

RAID 5

Average CPU

load

[%]

Average CPU

load

[%]

WWHITE HITE PPAPERAPER (cont.)

.

Online Test with 62 FI benchmark users = 50% CPU loadOnline Test with 62 FI benchmark users = 50% CPU load

.

466A/1196

.

Number of Tapes /

.

Number of SCSI-

.

2727

Controllers

3 Tapes / 2 Controllers 16 65 12 65 4 Tapes / 2 Controllers 19 68 15 67 4 Tapes / 4 Controllers 24 72 18 72 6 Tapes / 3 Controllers 24 73 22 73 8 Tapes / 4 Controllers 32 76 24 76

Online Test with 85 FI benchmark users = 70% CPU loadOnline Test with 85 FI benchmark users = 70% CPU load

Number of Tapes /

Number of SCSI-

Controllers

3 Tapes / 2 Controllers 12 78 8 79 4 Tapes / 2 Controllers 14 82 11 82 4 Tapes / 4 Controllers 17 85 14 85 6 Tapes / 3 Controllers 19 86 16 87 8 Tapes / 4 Controllers 20 88 18 88

As you can see the backup performance slows down if the load of SAP R/3 increases. In case of no system load, you almost get the backup rates of the file backup on NTFS file system level shown in Figure 26. With 30 percent R/3 system load caused by 40 FI users the rates are lower but still acceptable. With 50 and 70 percent R/3 system load the performance gets worse and you almost have no gain by using more tapes.

For an online backup the influence of the running backup on the performance of the R/3 system is of importance. To see how the R/3 performance develops during a backup we ran some benchmarks while backing up the R/3 system to 4 and 8 tapes. Compaq used the SAP standard benchmark of the module FI:

Tape configuration 62 FI Benchmark users 85 FI Benchmark users

No backup 171 ms 178 ms 4 Tapes RAID 0 174 ms 428 ms 8 Tapes RAID 0 183 ms 451 ms

Average data

[GB/h]

RAID 0

Figure 25: Online backup with 50% system load

Average data

[GB/h]

RAID 0

Figure 26: Online backup with 70% system load

Figure 27: R/3 response times during the backup

Average CPU

load

[%]

Average CPU

load

[%]

Average data

[GB/h]

RAID 5

Average data

[GB/h]

RAID 5

Average CPU

load

[%]

Average CPU

load

[%]

WWHITE HITE PPAPERAPER (cont.)

.

All these numbers came from our test scenario. There is no guarantee that the same results are

.

obtained in a customer system. The customer transactions may be completely different from those

.

in the FI benchmark and therefore the system load can be different. Because of this, the R/3

.

466A/1196

.

response time during backup may be different, too.

.

Online Backup Using the R/3 SAPDBA

.

Online and off-line R/3 backups are also possible using the SAPDBA program that comes with

.

the R/3 and Oracle software.

.

The SAPDBA program is preconfigured to use the native cpio program that is delivered with the

.

R/3 system for writing to and reading from tape. As this program is rather slow especially when

.

using multiple tapes in parallel, SAP offers the possibility to connect an external backup program

.

using the so-called backint interface.

.

SAPDBA uses the R/3 brbackup and brarchive programs to backup the Oracle datafiles and the

.

archive logs. The connection to ARCserve is implemented in the backint interface that is

.

delivered by Cheyenne. backint submits a regular file system job into the ARCserve job queue and

.

ARCserve just does a normal backup. Using this mechanism the R/3 Agent for Oracle is not

.

involved.

.

SAP certification of the Cheyenne backinit interface is pending.

.

2828

Figure 28: Backup using the R/3 SAPDBA

WWHITE HITE PPAPERAPER (cont.)

.

466A/1196

.

Compaq only did some online backups, as the off-line backup performance is covered with the

.

Windows NT file system and image backup.

.

The performance of the SAPDBA backups is similar to the performance of the NTFS file system

.

backup. There is only a small overhead for the SAPDBA administrational tasks (e.g. starting and

.

stopping of the database if necessary, backing up internal protocol files) and the backint call. This

.

overhead is depending on different factors as the database size or the kind of backup (online or

.

off-line).

.

As the off-line backup performance is covered with the Windows NT file system and image

.

backup and the online backup performance is covered with the online ARCserve Database Agent

.

performance tests, Compaq only tested the backint functionality that worked without errors.

.

In contrast to the R/3 Agent backup where each tablespace is stored in a separate session, the

.

backup using SAPDBA and backint is done in one session.

.

NETWORK BACKUPS

.

There are two methods of performing network backups.

.

• Pull method

.

Any machine in the Microsoft network can be backed up by simply choosing the shared

.

2929

drives in the backup window. Backup is initiated by the ARCserve server that pulls the data from the client using a share.

brbackup brarchive

Figure 29: Backint interface

SAPDBA

BACKINT

ARCserve

WWHITE HITE PPAPERAPER (cont.)

.

466A/1196

.

• Push method

.

Backup of a remote Windows NT machine is also possible by using the ARCserve Windows

.

3030

NT agent that utilizes the Push Agent technology to accomplish, backup, restore, and compare jobs. Here backup is initiated by the ARCserve Windows NT agent and data are pushed to the ARCserve server holding the backup devices.

Pull Agent

The pull method is a backup over Microsoft network shares. You can select all drives or simple files which are shared by Windows NT or another LAN Manager Server like Windows 95 or Windows for Workgroups 3.11. It is also possible to backup a registry over the network with this method.

For pull method backups you don’t have to install any additional software on the client. The ARCserve user only needs sufficient privileges to access the shares. For our network benchmarks, Compaq used a 100 Mbit Ethernet network with two ProLiants over a 100BASE T repeater with no other machines connected to the LAN.

Figure 30: ARCserve network backup

WWHITE HITE PPAPERAPER (cont.)

.

466A/1196

.

The source machine is a ProLiant 4500 with a RAID 5 disk array containing the data. The backup

.

host is the ProLiant 5000 which has already been used for the local backup tests.

.

Compaq optimized the network throughput by changing the TCP Window size parameter in the

.

registry from the default value 8760 to 17520. The performance gain for our test environment was

.

about 15%. Be careful with tuning that parameter because it depends on the kind of network and

.

other hardware you use.

.

Compaq saw the following network backup results:

.

As you can see from Figure 32, the maximum backup rate of about 17 GB/h was reached with 3

.

tapes. Adding more tape devices didn't improve the network backup performance.

.

To guarantee that the bottleneck was not the disk subsystem of the source machine (Compaq

.

ProLiant 4500), Compaq measured its read performance. The read rate was about 73 GB/h, which

.

is quite enough for the tests. As the backup rate with one or two tapes nearly corresponds to the

.

local backup rates and the read performance of the source disk subsystem is large enough, the

.

backup rate should increase with a higher number of tapes.

.

As it did not, the network had to be the bottleneck. To prove this we did some network

.

performance tests. Compaq used a small program that generated point to point transfer on a

.

TCP/IP connection by transferring about 250 megabytes located in RAM. The average throughput

.

was about 24 GB/h. As expected this value is higher than the ARCserve network throughput

.

because with the test program Compaq only had data in memory that were directly put on the

.

network. There was no disk overhead.

.

3131

Number of Tapes /

Number of SCSI-Controllers

1 Tape / 1 Controller 10

2 Tapes / 1 Controller 13 3 Tapes / 2 Controllers 17 4 Tapes / 4 Controllers 17 8 Tapes / 4 Controllers 17

Figure 32: Performance of network backups

100 TX

Figure 31: Network layout

Average data

[GB/h]

WWHITE HITE PPAPERAPER (cont.)

.

So if the disk and tape hardware is fast enough the network is the bottleneck in backing up data

.

over the wire. Until the maximum network throughput is reached, you get the same performance

.

for network and for local backups.

.

466A/1196

.

Push Agent

.

The second kind of network backup tested was the backup through an ARCserve agent. ARCserve

.

delivers the already mentioned agents for Windows NT, Windows 95, Windows for Workgroups,

.

UNIX, Macintosh, and some other platforms.

.

In the ARCserve evaluation, Compaq only tested the agent for Windows NT. Compaq used the

.

same test configuration as described in the Pull Agent section of this white paper. Compaq saw a

.

backup performance of 21 GB/h and a restore performance of 16 GB/h. Here again the network is

.

the bottleneck.

.

In a 10 Mbit network the throughput will be worse than the backup rate we measured because the

.

network is slower.

.

ARCSERVE RESTORE

.

ARCserve provides you with four methods of selecting the data you want to restore:

.

• Restore by Tree

.

• Restore by Tape

.

• Restore by Query

.

• Restore by Image

.

Using the Restore by Tree, Tape, and Query the selected items are restored file by file. Only the

.

Restore by Image uses another method which is bypassing the Windows NT file system.

.

In the restore tests, Compaq found out that there are three parameters which influence the restore

.

performance. The first is the number of tapes used for the backup and restore. The restore rate

.

will not be higher as the backup rate with the same tape configuration. The second parameter is

.

the write performance of the disk configuration you are using. The third one is the behavior and

.

the overhead of the ARCserve software.

.

In the following section of this white paper, you see different tables which show the influence of

.

these parameters on different hardware configurations.

.

Restore of an Windows NT File Backup

.

In the test environment, Compaq used a RAID 5 disk array containing 10x4.3 GB disks. As you

.

can see from Figure 33, the write performance of this disk configuration is 19.3 GB/h. The

.

maximum restore rate is expected to be less than 19.3 GB/h as there is also some ARCserve

.

overhead.

.

3232

RAID 0 configuration of

3 tapes on 2 controllers

Backup 20 Restore 19

Figure 33: Restore performance of RAID 0

Performance in

GB/h

WWHITE HITE PPAPERAPER (cont.)

.

Compaq used this small configuration (regarding the number of tapes) to get an impression of the

.

difference between backup and restore performance. The numbers of Figure 33 show that you can

.

restore with nearly the backup speed if the performance of the disk subsystem is larger than the

.

466A/1196

.

backup rate.

.

Compaq also did some tests with a missing tape in a RAID 5 tape configuration to determine if a

.

damaged tape has any influence on the read performance of the RAID tape array.

.

As you can see there is almost no difference between the restore from three tapes configured as a

.

RAID 5 array and the restore with one damaged or missing tape if the 2 remaining tapes are

.

connected to one SCSI controller. So obviously calculating data from the RAID checksums is

.

faster than reading data from tape, i.e. the calculation doesn't lead to additional overhead.

.

If the remaining tapes are connected to 2 controllers the restore is even faster. You can verify this

.

behavior by comparing the restore results to the backup results of Figure 35. (Remember that in

.

the 3 tape configuration backup and restore performance are nearly the same.). The restore

.

performance of 3 tapes connected to 2 controllers corresponds to the backup performance of a

.

RAID5 array consisting of 3 tapes on 2 controllers. The restore performance of only 2 tapes on 2

.

controllers corresponds to the backup performance of a RAID 0 array consisting of 2 tapes on 2

.

controllers.

.

RAID configuration of

.

The numbers of Figure 35 show that the maximum restore rate in the test configuration is 19

.

GB/h. Here the write performance of the RAID 5 disk array is the bottleneck.

.

To get higher restore rates tests with a RAID 0 and a RAID 1 disk configuration were set up:

.

3333

RAID 5 configuration of 3 tapes on

2 controllers

Backup on 3 Tapes 14 Restore from 3 Tapes 12 Restore from 2 Tapes on 2 Controllers 19 Restore from 2 Tapes on 1 Controller 12

Figure 34: Restore performance of RAID 5 with a damaged tape

Restore performance with

8 tapes

RAID0 19 not possible RAID5 19 19

Figure 35: Restore performance of 8 tapes

8 tapes in GB/h

Performance in

GB/h

Restore performance with

7 tapes in GB/h

WWHITE HITE PPAPERAPER (cont.)

.

466A/1196

.

In Figure 36, you can see that using RAID 0 and RAID 1 disk subsystem configurations the

.

maximum restore rate is limited to 24.6 GB/h. With a RAID 0 disk array of 14 disks, the write

.

performance of the disk subsystem which is 33 GB/h is no longer the bottleneck.

.

The problem in achieving higher restore rates is that ARCserve allocates as much memory as the

.

size of the file to be restored, in maximum 512 megabytes. If the file is larger than 512

.

megabytes, the restore performance gets worse due to memory allocation problems. As the test

.

configuration contains tablespaces that are bigger than 1 gigabytye, Compaq got these

.

performance problems.

.

If only small files are restored you get restore rates which are nearly the same as the backup rates.

.

Normally only the disk write rate is the bottleneck.

.

According to Cheyenne the problem is solved with Windows NT 4.0. Cheyenne is also developing

.

a fix for Windows NT 3.51.

.

3434

Disk configuration Restore Rate

[GB/h]

10 Disks RAID 5 19 19 10 Disks RAID 0 25 31 14 Disks RAID 5 19 20 14 Disks RAID 1 25 26 14 Disks RAID 0 25 33

Figure 36: Restore performance using different disk subsystem configurations

Figure 37: Available memory during restore of PSAPSTABD (2 files of 1.2 GB)

Write Performance

Disk Subsystem

[GB/h]

WWHITE HITE PPAPERAPER (cont.)

.

Restore of an Image Backup

.

The only restore type different from the restore of an Windows NT file backup is the image

.

restore. In order to perform an image restore the target of the restore has to be free from access of

466A/1196

.

other programs or services like SAPSID_00. After the restore on a partition which is bigger than

.

the backup partition, you can only use the size of the backed up one.

.

As you can see in Figure 38, the disk subsystem is the bottleneck using the image restore. The

.

image backup is not allocating memory like the normal restore process. With image restore you

.

nearly get the number as in the backup.

.

Note that the restore rate for 14 disks is larger than the measured write performance of the disk

.

subsystem. This is because the two values were generated by different programs (ARCserve and

.

the Completion Port I/O Test Utility).

.

There is the possibility to restore partial files from an image backup. The performance is the same

.

as the restore of a file backup.

.

Restore of an Online R/3 Database Agent Backup

.

The restore of an online backup has the same performance as the Windows NT file restore (see

.

section, Restore of an Windows NT File Backup).

.

But there is one big difference: the ARCserve R/3 agent is writing every tablespace in a separate

.

session during backup and therefore reading every tablespace in a separate session during restore.

.

After each session the restore process is rewinding the tape. In a complete restore, it takes quite a

.

while to reach the end of the tape. In this case, about one hour had to be added to the net restore

.

time. The additional rewind time depends on the number of tablespaces.

.

After you have restored the datafiles and the off-line redologs the user has to do some actions to

.

get the database in a consistent state.

.

• Copy the controlfile from the database agent directory to the original locations

.

• Startup database in mount status

.

• Recover database until cancel using backup controlfile

.

• Alter database open resetlogs

.

3535

Disk configuration Restore

Rate

[GB/h]

14 Disks RAID 0 / 8 Tapes 35 33 10 Disks RAID 5 / 4 Tapes 19 20 10 Disks RAID 5 / 2 Tapes 12 20

Figure 38: Restore of an Image Backup

Write Performance

Disk Subsystem

[GB/h]

WWHITE HITE PPAPERAPER (cont.)

.

466A/1196

.

See the Oracle documentation for more detailed information.

.

Restore of an SAPDBA Backup

.

Restore of the R/3 system is also possible using the SAPDBA program that comes with the R/3

.

and Oracle software.

.

SAPDBA calls the R/3 brrestore program. The connection to ARCserve is implemented in the

.

backint interface as described before in the SAPDBA backup section of this white paper.

.

3636

Figure 39: Recovery of the Oracle database

Figure 40: Restore using the R/3 SAPDBA

WWHITE HITE PPAPERAPER (cont.)

.

SAPDBA

brrestore

BACKINT

ARCserve

.

466A/1196

.

The restore performance is similar to the restore performance of Windows NT files. There is a

.

small constant overhead for SAPDBA administration tasks and the backint call. A maximum

.

restore performance of 19 B/h was possible using 8 tapes.

.

To evaluate the backint functionality, Compaq tested the restore of single tablespaces as well as

.

the restore of a complete R/3 database.

.

3737

Figure 41: Backint interface

WWHITE HITE PPAPERAPER (cont.)

.

SUMMARY

.

466A/1196

ARCserve provides different methods of backing up and restoring Windows NT and SAP R/3

.

data. Compaq tested the functionality of :

.

• backup and restore on Windows NT file system level

.

• backup and restore with the Image Option, bypassing the Windows NT file system

.

• integration into R/3 using the R/3 Agent for Oracle

.

• integration into R/3 using the backint interface together with the SAPDBA

.

All the different backup and restore methods worked fine. All the different backup and restore

.

methods worked fine. With the next version of the ARCserve Database Agent, it will even be

.

possible to backup the R/3 database in one session that will further improve the restore

.

performance. The next version of the ARCserve Database Agent will be available in the first

.

quarter of 1997. In the case of restore, you loose a lot time by changing between the sessions.

.

In general with the right tape configuration ARCserve delivers excellent backup and restore rates.

.

Compaq achieved the following backup rates:

.

• 39GB/h for a backup on Windows NT file system level

.

• between 46GB/h and 53GB/h (depending on the type of tape RAID array) using the

.

3838

ARCserve Image Option

• between 20GB/h and 43 GB/h for an online R/3 backup (depending on the R/3 load) The restore rate was only limited by the write performance of the disk subsystem. The maximum

restore rate was between 19GB/h and 25GB/h depending on the type of the disk RAID array. The restore of very large files (>1GB) where some memory allocation problems occurred should be improved.

As the backup and restore performance is heavily depending on the hardware configuration you have to configure the PCI and the disk subsystem very carefully to avoid the hardware of becoming the bottleneck in backup and restore. Therefore this White Paper contains some recommendations for the disk subsystem, DLT and PCI configuration.

In total ARCserve is a really good backup solution because it is providing a lot of features like the RAID and the Image Option. The product can be used for small backups with only one DLT up to complete online backups of big databases with at 8 DLT devices configured to a RAID array.

WWHITE HITE PPAPERAPER (cont.)

.

APPENDIX

.

466A/1196

.

List of All Tablespaces

.

***************************************************************

.

09.10.96 12:27 --- List of all tablespaces / fragmentation:

.

The values of total space, allocated space and largest free space area are in

.

KBytes.

.

TABLESPACE Total Allocated %-Alloc. Files Free areas Largest

.

---------------------------------------------------------------

.

PSAPBTABD 3063808 1692136 55 2 1 1371656

.

PSAPBTABI 2197504 1199760 55 2 2 997640

.

PSAPCLUD 245760 238240 97 1 1 7512

.

PSAPCLUI 33792 28424 84 1 1 5360

.

PSAPDDICD 61440 41024 67 1 1 20408

.

PSAPDDICI 33792 14088 42 1 1 19696

.

PSAPDOCUD 102400 67616 66 1 1 34776

.

PSAPDOCUI 20480 11480 56 1 1 8992

.

PSAPEL30DD 839992 657400 78 2 2 111880

.

PSAPEL30DI 40960 25056 61 1 1 15896

.

PSAPES30DD 2055376 1654048 80 2 2 399008

.

PSAPES30DI 616352 491624 80 2 2 118192

.

PSAPLOADD 70656 4504 6 1 1 66144

.

PSAPLOADI 41984 6080 14 1 1 35896

.

PSAPPOOLD 857296 552008 64 2 2 197832

.

PSAPPOOLI 841728 371456 44 2 2 420496

.

PSAPPROTD 292864 254536 87 1 1 38320

.

PSAPPROTI 29696 26336 89 1 1 3352

.

PSAPROLL 204800 75920 37 1 22 42552

.

PSAPSOURCED 67584 19280 29 1 1 48296

.

PSAPSOURCEI 14336 2808 20 1 1 11520

.

PSAPSTABD 2401896 1398848 58 2 2 900664

.

PSAPSTABI 1344512 1046552 78 2 7 297568

.

PSAPTEMP 167936 0 0 1 1 167928

.

PSAPUSER1D 5120 2128 42 1 1 2984

.

PSAPUSER1I 5120 32 1 1 1 5080

.

SYSTEM 153600 72896 47 1 2 75496

.

Total number of files in database: 40 including 4 redo log files

.

TABLESPACE File-Id Filename Size in K

.

------------------------------------------------------------------------

.

PSAPBTABD 4 f:\oracle\jm2\sapdata1\btabd_1\btabd.data1 1531904

.

28 f:\oracle\JM2\sapdata1\btabd_2\btabd.data2 1531904

.

PSAPBTABI 7 f:\oracle\jm2\sapdata2\btabi_1\btabi.data1 1098752

.

29 f:\oracle\JM2\sapdata2\btabi_2\btabi.data2 1098752

.

PSAPCLUD 9 f:\oracle\jm2\sapdata2\clud_1\clud.data1 245760

.

PSAPCLUI 12 f:\oracle\jm2\sapdata3\clui_1\clui.data1 33792

.

PSAPDDICD 5 f:\oracle\jm2\sapdata1\ddicd_1\ddicd.data1 61440

.

PSAPDDICI 24 f:\oracle\jm2\sapdata5\ddici_1\ddici.data1 33792

.

PSAPDOCUD 22 f:\oracle\jm2\sapdata5\docud_1\docud.data1 102400

.

PSAPDOCUI 13 f:\oracle\jm2\sapdata3\docui_1\docui.data1 20480

.

PSAPEL30DD 19 f:\oracle\jm2\sapdata4\el30dd_1\el30dd.data1 646144

.

32 f:\oracle\JM2\sapdata4\el30dd_2\el30dd.data2 193848

.

PSAPEL30DI 25 f:\oracle\jm2\sapdata5\el30di_1\el30di.data1 40960

.

PSAPES30DD 27 f:\oracle\JM2\sapdata6\es30dd_2\es30dd.data2 474320

.

36 f:\oracle\JM2\sapdata6\es30dd_1\es30dd.data1 1581056

.

PSAPES30DI 26 f:\oracle\jm2\sapdata5\es30di_1\es30di.data1 474112

.

33 f:\oracle\JM2\sapdata5\es30di_2\es30di.data2 142240

.

PSAPLOADD 17 f:\oracle\jm2\sapdata4\loadd_1\loadd.data1 70656

.

PSAPLOADI 11 f:\oracle\jm2\sapdata3\loadi_1\loadi.data1 41984

.

PSAPPOOLD 23 f:\oracle\jm2\sapdata5\poold_1\poold.data1 659456

.

34 f:\oracle\JM2\sapdata5\poold_2\poold.data2 197840

.

PSAPPOOLI 6 f:\oracle\jm2\sapdata1\pooli_1\pooli.data1 420864

.

30 f:\oracle\JM2\sapdata1\pooli_2\pooli.data2 420864

.

3939

WWHITE HITE PPAPERAPER (cont.)

.

PSAPPROTD 16 f:\oracle\jm2\sapdata4\protd_1\protd.data1 292864

.

PSAPPROTI 15 f:\oracle\jm2\sapdata3\proti_1\proti.data1 29696

.

PSAPROLL 3 f:\oracle\JM2\sapdata1\roll_1\roll.data1 204800

.

PSAPSOURCED 10 f:\oracle\jm2\sapdata2\sourced_1\sourced.data1 67584

.

PSAPSOURCEI 18 f:\oracle\jm2\sapdata4\sourcei_1\sourcei.data1 14336

466A/1196

.

PSAPSTABD 14 f:\oracle\jm2\sapdata3\stabd_1\stabd.data1 1501184

.

31 f:\oracle\JM2\sapdata3\stabd_2\stabd.data2 900712

.

PSAPSTABI 21 f:\oracle\jm2\sapdata5\stabi_1\stabi.data1 1034240

.

35 f:\oracle\JM2\sapdata5\stabi_2\stabi.data2 310272

.

PSAPTEMP 2 f:\oracle\JM2\sapdata1\temp_1\temp.data1 167936

.

PSAPUSER1D 20 f:\oracle\jm2\sapdata4\user1d_1\user1d.data1 5120

.

PSAPUSER1I 8 f:\oracle\jm2\sapdata2\user1i_1\user1i.data1 5120

.

SYSTEM 1 f:\oracle\JM2\sapdata1\system_1\system.data1 153600

.

ARCserve Parameter File for backint

.

########################################################################

.

# This is the parameter file of BACKINT for ARCserve containing parameters that #

.

# determine the backup procedure. #

.

# BACKINT reads this file and submits a job to the ARCserve queue for Execution.#

.

# Instructions for filling out each line start with a pound sign (#) and precede#

.

# the actual information line which the user needs to fill in. #

.

########################################################################

.

# Tape Name that ARCserve should use for the backup (eg. tape0, tape1,.., or '*')

.

DESTTAPE=*

.

# Group Name that ARCserve should use for the backup (eg. GROUP0, GROUP1,.., or '*')

.

DESTGROUP=*

.

4040

Hp COMPAQ PROLIANT 5000, COMPAQ PROSIGNIA 500, COMPAQ PROLIANT 4500, COMPAQ PROLIANT 4000, COMPAQ PROSIGNIA 720 Backup of Windows NT SAP R/3 Systems with ARCserve

Specifications and Main Features

Frequently Asked Questions

User Manual