Hp COMPAQ PROLIANT 6000 High Performance Backup of Enterprise Servers

[February 1998

Compaq Computer
Corporation

CONTENTS

Introduction....................3

Test bed Description....... 4

Hardware......................... 4

Software.......................... 4

Performance

Demonstrations .............. 5

Case 1 – Local Flat File

Backup – Non RAIT .......... 5

Case 2 – Local Flat File

Backup – RAIT ................. 6

Case 3 – Local Relational

Database Backup ........... 11

Case 4 – Local Flat File

Restore – Non RAIT........ 12

Case 5 – Local Flat File

Restore – RAIT .............. 13

Case 6 – Local Relational

Database Restore........... 15

RAIT Tape Management .. 15
Data Migration Option

with Tape library Option .. 16
Disaster Recovery with

Tape Library Option ........ 16

DLT LIBRARY
PERFORMANCE BRIEF

High Performance Backup of Enterprise
Servers

Since its introduction in June1997, the Compaq DLT 15 Cartridge Library Model 1530
(Compaq 1530 Library) has been an industry leader in high performance backup
capabilities. The Compaq DLT 15 Cartridge Library Model 3570 (Compaq 3570 Library)
was announced in November 1997. This Library uses 35/70-GB DLT drives and sets new
standards in high-speed backup performance, capacity, and ease of use.

Late in 1997, Compaq performed a series of tests to examine the performance of the
Compaq 3570 Library. The objective of these tests was to find the highest-performance
configurations while performing realistic backup tasks on Compaq servers running
Windows NT 4.0 network operating system.

The Compaq 3570 Library provides high capacity, high-speed tape storage for use in
local and remote backup strategies. This brief addresses the technical aspects of high­performance backups that can reach and exceed 1651 gigabytes per hour (GB/hr) with
standard Compaq hardware.

Conclusion................... 17

APPENDIX 1.................. 18

Raw Performance Data.... 18

APPENDIX 2.................. 20

Standard Test System..... 20

APPENDIX 3.................. 21

Operating the Compaq

3570 Library .................. 21

APPENDIX 4.................. 22

DLT Media

Considerations ............... 22

APPENDIX 5.................. 25

Table of Figures ............. 25

ECG046/0298 1

Figure 1-- Compaq 3570 Library

1

Four Compaq 3570 Libraries with 8 35/70 DLT Drives running 8 concurrent jobs in a local (non-LAN) backup with data

compression of 4:1 can achieve this performance level. See the data in the appendix 1.

DLT LIBRARY PERFORMANCE BRIEF (cont.)

NOTICE

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

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, ProLiant, ProLinea, NetFlex, registered United States Patent and Trademark Office.
TriFlex Architecture and UltraView, CompaqCare and the Innovate logo, are trademarks and/or

service marks of Compaq Computer Corporation.
DLT is a trademark applied for by Quantum Corporation
Other product names mentioned herein may be trademarks and/or registered trademarks of their

respective companies.
©1998 Compaq Computer Corporation. Printed in the U.S.A.
Microsoft, Windows, Windows NT, Windows NT Advanced Server, SQL Server for Windows NT

are trademarks and/or registered trademarks of Microsoft Corporation.
ARCserve is a registered trademark of Computer Associates.
Lotus Notes is a registered trademark of IBM.

HIGH PERFORMANCE BACKUP OF ENTERPRISE SERVERS

High Performance Backup of Enterprise Servers

First Edition (January 1998)

ECG046/0298

ECG046/0298 2

Raid

RAID technology was
developed to improve
performance and
reliability of inexpensive
disk drives. Hence the
acronym Redundant

Array of Inexpensive
Drives. These systems

employ multiple drives to
improve the performance
and reliability of less
expensive drives.

When applied to tape
drives, system
performance and data
protection are most
important. For tape
systems, the acronym is

Redundant Array of
Independent Tape—

RAIT. Whatever the
acronym, the purpose is
the same, improved
performance and data
protection. Just remember
– RAIT is RAID
technology applied to
tape.

There are five techniques,
each offering unique
features to meet differing
speed and data protection
needs. These techniques
are referred to as levels.

RAID 0 offers greatest
speed. RAID 1 offers the
greatest data protection.
RAID 2, 3, 4 and 5 offer
high speed, with data
protection. The speed is
not as high as RAID 0,
and data protection is not
as great as RAID 1. Each
level offers features best
suited to a particular type
of hardware configuration
or operational

ECG046/0298 3

DLT LIBRARY PERFORMANCE BRIEF (cont.)

INTRODUCTION

The primary objective of this paper is to provide performance data for one or more Compaq Model
3570 DLT Libraries using RAID and non-RAID software technology. As more companies migrate
to server solutions to meet company data storage and retrieval requirements, system unavailability
during scheduled backup windows is a significant issue.

Operation has expanded to seven days a week and 24 hours a day. Concurrent (hot) backups with
the database in use, or small scheduled windows for backup are common practice. In either case,
shorter times to accomplish the task result in greater customer satisfaction.

Fault tolerance has become even more important with the large storage media prevalent today.
Statistically, as the volume of data increases, the number of faults increases. RAID 5 fault
tolerance technology allows a Library backup to complete successfully, even if a single tape drive
fails or one tape is missing or damaged. The data on the failed or missing tape is recoverable,
without requiring another backup.

Until the introduction of the Compaq Model 1530 DLT and Compaq Model 3570 DLT Libraries,
customers requiring tape RAID functionality needed to implement a Compaq DLT Tape Array. A
major disadvantage of this solution is that unattended data capacity is limited to a single tape
cartridge.

The Compaq Tape Array does not have the automation to remove and replace tapes. The Compaq
DLT Library products provide the automated functionality and capacity to eliminate the tedious
but necessary swapping of tapes. This automation permits backup scheduling during off-peak
hours, without an operator present to change tapes.

Libraries do not inherently have the ability to utilize RAIT. The backup software that supports the
library provides this functionality. Software products that support the Compaq Model 1530 and
Model 3570 DLT Libraries include Cheyenne ARCserve and Seagate Backup Exec. Of these, only
Cheyenne ARCserve 6.5 presently supports RAIT for the Compaq DLT Libraries.

Compaq, along with Cheyenne Software, a division of Computer Associates, can now provide the
automation and capacity of the DLT Library and the fault tolerance and performance of the DLT
Tape Array. Combining these features provides a reliable method of high-speed, automated
backups over a period of days or weeks without user intervention.

With the creation of the concept of a Virtual Library, Cheyenne ARCserve 6.5 allows management
of up to sixteen DLT Libraries and 32 DLT Drives as a single entity. A backup administrator can
easily manage multiple libraries from a single control point. The libraries may be physically
attached to any server on the LAN. The tape drives and cartridge slots in the physical libraries are
grouped together to form a single logical group, or pool. Multiple RAIDs of Libraries form pools
that can be viewed in ARCserve as a larger Virtual Library. The size of the pool is dynamically
adjustable by changing the defined RAID size to meet changing backup requirements. Pools of
Libraries can accommodate very large multi-terabyte databases, yet maintain the flexibility to
backup even the smallest remote office with the appropriately sized pool.

For ARCserve 6.5 running on one server, a maximum of sixteen Compaq 3570 Libraries (thirty­two 35/70 DLT Drives) per server can run 32 concurrent local backup jobs. Appendix 1 presents
the hardware environments used and details of performance measurements made for this white
paper.

The Compaq 1530 Library is the best choice for installations requiring a price/performance
tradeoff matching current LAN speeds and capacities2. Where maximum storage capacity and
performance are required, the Compaq 3570 Library is able to meet the demands. Regardless of the
needs of the installation, Compaq offers a state-of-the-art answer that meets current requirements
and will continue to do so in the future.

2

Current Ethernet 100Mbit/second LAN speeds allow a up to 35GB/Hr backup transfer rates with DB Agent and
compression at the source. An additional white paper, planned for later in 1998, will detail these limitations and techniques
to improve backups over the LAN.

DLT LIBRARY PERFORMANCE BRIEF (cont.)

.

.

TEST BED DESCRIPTION

.

.

.

.

.

.

.

Hardware

.

ECG046/0298 4

.

.

.

.

.

The Compaq ProLiant 6000 server was equipped with a Compaq

.

.

.

Netelligent 10/100Mbit/s TX PCI Ethernet Controller. Compaq

.

.

.

ProLiant 6000 features relevant to this test include four (4)

.

.

.

200-MHz Pentium processors, 33-MHz PCI bus, and 100-MHz

.

.

.

synchronous memory.

.

.

.

.

Additional test bed details are discussed in Appendix 2.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

Hard Drive Array

.

.

.

.

The server was equipped with four SMART/2 SCSI ‘RAID’ Hard Disk controllers feeding two

.

.

.

channels of five Compaq 10K 9.1-GB Wide Ultra SCSI-3 hard drives. This provides a total of 364-

.

.

.

GB per server. Older or non-Compaq hard drives may not be fast enough to keep pace with other

.

.

.

backup components, and may be a significant performance bottleneck.

.

.

.

.

.

The DLT Libraries were connected to SCSI cards with the Dual Channel Ultra Wide SCSI-3

.

.

.

controllers. The use of a Compaq 32-bit Fast-Wide (SCSI-2) controller or the Ultra–wide

.

.

.

integrated SCSI-3 controller for the DLT Library attachment is also possible. Use of other

.

.

.

controllers may result in lower performance, and is not recommended. For additional details of

.

.

.

system and SCSI channel-device attachment see Appendix 2, Standard Test System.

.

.

.

.

.

.

Memory

.

.

.

.

The server is populated with 320-MB main memory for these performance demonstrations. This is

.

.

.

sufficient to prevent performance limitations due to insufficient memory.

.

.

.

.

.

.

.

Software

.

.

.

.

The operating system and backup software chosen for these tests is commercially available. Since

.

.

.

Seagate Backup Exec does not support RAID, only ARCServe was used in testing. The software

.

.

.

on each server was configured for optimum performance in accordance with the manufacturers

.

.

.

recommendations.

.

.

.

.

.

.

Operating System

.

.

.

.

.

The server had Windows NT Server, version 4.0, build 1381 installed. Service pack 3 was also

.

.

.

installed.

.

.

.

.

.

Backup Software

.

.

.

.

.

Cheyenne ARCserve version 6.5, Enterprise edition, build 600, with the Tape Library Option,

.

.

.

RAIT Option, and Image Option was installed for these tests. The Tape Library Option specifically

.

.

.

supports the Compaq 3570 and 1530 Libraries.

.

.

.

.

.

.

.

.

.

Figure 2 -- Compaq ProLiant 6000

DLT LIBRARY PERFORMANCE BRIEF (cont.)

GB/Hr

PERFORMANCE DEMONSTRATIONS

Case 1 – Local Flat File Backup – Non RAIT

Case 1 demonstrates local backup performance. Figure 3 illustrates this network, consisting of a
single Compaq ProLiant 6000 server with four Compaq 3570 Libraries attached.

Figure 3 -- Flat File Network

Image Backup

This test demonstrates image backup performance from a single Compaq ProLiant 6000 server to
one, two, and four Compaq 3570 Libraries. Cheyenne ARCserve 6.5 image backup copies all used
blocks from the storage media to the backup media.

The total volume of data was 364-GB. The data for this test was compressible at a 4:1 ratio3. The
results of this test, presented in Figure 4, demonstrate the performance improvements obtainable
from additional Compaq 3570 Libraries. Each additional library, which houses two 35/70 DLT
drives, allows the addition of two jobs.

180
160
140
120
100

80
60
40
20

0

2 4 8

Jobs

ECG046/0298 5

Figure 4 -- Flat File Performance – Multiple Jobs

3

The compression ratio of data is a key factor in backup performance. For estimated compressibility of several types of data
files, refer to appendix 4.

DLT LIBRARY PERFORMANCE BRIEF (cont.)

Case 2 – Local Flat File Backup – RAIT

These tests demonstrate local backup tasks. A single Compaq ProLiant 6000 server has from one
to four Compaq Model 3570 DLT Libraries attached. Image backup tasks were tested. The files are
simple flat (non-relational) files. Figure 3 illustrates this network with four Compaq 3570 Libraries
attached to the Compaq ProLiant 6000.

ARCserve 6.5 will operate the Compaq Model 3570 DLT Libraries in RAIT 0, RAIT 1, or RAIT 5
configurations. RAIT 0 typically yields highest performance. RAIT 1 offers the greatest protection
for the data. RAIT 5 retains up to 80% of the speed performance, with data protection.

RAIT 0 stripes the data block across two or more tapes. These tapes may be mounted on one
library if only two drives are used, or the tapes may be in different libraries. This is the highest
performance configuration, but there is no data redundancy. If a fault occurs in any tape, the entire
data set may be lost. Figure 5 illustrates RAIT 0 data distribution across three tapes.

A potential performance limitation exists in the way applications software, such as ARCServe,
launches threads and the way the Windows NT processes those threads. Launching concurrent
tasks will start more threads, thus improving time slicing and backup efficiency.

ECG046/0298 6

Figure 5 -- RAIT 0 Data Distribution

DLT LIBRARY PERFORMANCE BRIEF (cont.)

ARCserve 6.5 RAIT 1 mirrors both 35/70 GB LT drives in one Compaq 3570 Library. Each 35/70
GB DLT drive receives identical data. If there is a failure in one library during either reading or
writing, the data on the other drive is intact. This configuration is more secure than RAIT 0, but
does not take full advantage of the capacity and performance benefits available from the Compaq
3570 Library. Figure 6 illustrates mirroring data to two tapes.

ECG046/0298 7

Figure 6 -- RAIT 1 Data Mirroring

It is also possible to create automatic off-site backups using a configuration with two Compaq
Model 3570 DLT Libraries physically in different locations while connected to the same network.
The server and one library are connected locally, with a second library attached to the network, but
physically located in a remote location. Since RAID is not currently supported over a network, one
job is run locally, and a second job backing up the same files is run to the remote library over the
network. This would limit backup performance to that of the LAN.

DLT LIBRARY PERFORMANCE BRIEF (cont.)

RAIT 5 retains up to 80% of the performance benefit of RAIT 0 while adding fault tolerance. Each
data block is striped across three or more tapes, a parity code is calculated for each block, and that
parity code is stored on another tape. The data and parity rotate among all tapes in the Virtual
Library. The Virtual Library concept permits mounting the tapes in a single backup set in different
libraries. Since each Compaq 3570 Library houses two 35/70 DLT Drives, this concept is essential
in order to operate the library in RAIT 5. Figure 7 illustrates RAIT 5 data and parity placement
across five tapes.

HOST SYSTEM

Parallel Data Queue

Parity

data
data
data
data

Parity

data
data

data

Parity

Figure 7-- RAIT 5 Writing Pattern

data

Parity

data
data

data

datadatadata data
data
data

Parity

data

datadata

data
data
data

Parity

data

ECG046/0298 8

DLT LIBRARY PERFORMANCE BRIEF (cont.)

GB/Hr

Image Backup – RAIT 0

For this test, a single Compaq ProLiant 6000 server had four Compaq Model 3570 DLT Libraries
attached. Tests were performed with one job running to eight 35/70 GB DLT drives, two jobs
running to four 35/70 GB DLT drives each, and four jobs running to two 35/70 GB DLT drives
each. The data was 364-GB, 4:1 compressible, as in the previous tests. Figure 8 shows the results
of these tests.

200
150

Gb/Hr

100

50

0

1 2 4

Jobs

Figure 8 -- Flat File RAIT 0 Performance – Image Backup

Image Backup – RAIT 1

Data Compression

For these tests, all
compression was
accomplished in the 35/70
GB DLT drives.

This test was performed with one Compaq ProLiant 6000 server and 1 dual drive Compaq 3570
Library attached. Both 35/70 GB DLT drives in the library were mirrored. Data volume was 364­GB. This test, shown in Figure 9, demonstrates the effect of data compressibility on performance.

70
60
50
40
30
20
10

0

1:1 2:1 4:1

Compression Ratio

Figure 9 -- Flat File RAIT 1 Performance – Image Backup

ECG046/0298 9

DLT LIBRARY PERFORMANCE BRIEF (cont.)

GB/Hr

Image Backup – RAIT 5

This test was performed with one Compaq ProLiant 6000 server with four Compaq Model 3570
DLT Libraries attached. Data volume was 364-GB. Though somewhat slower than the RAIT 0
performance presented in Figure 8, data security is much greater. The results of these tests are
shown in Figure 10. As demonstrated in Figure 9, compressibility of the data has a dramatic
influence on performance up to 2:1 compression ratio. Compressibility of the data beyond a 2:1
ratio yields diminishing performance improvements. Running two jobs with four drives per job
produced the data for Figure 10.

120
100

80
60
40
20

0

1:1 2:1 4:1

Compression Ratio

Figure 10 -- Flat File RAIT 5 Performance – Image Backup

ECG046/0298 10

DLT LIBRARY PERFORMANCE BRIEF (cont.)

Case 3 – Local Relational Database Backup

This case demonstrates a Lotus Notes database local backup task. One Compaq ProLiant 6000
server has two Compaq 3570 Libraries attached. Both 35/70 GB DLT drives in each library were
used. ARCserve 6.5 operated the Compaq 3570 Libraries in RAIT 0, RAIT 1, and RAIT 5
configuration during these tests. The database was backed up both while the database was in use
(hot) and with the database offline (cold). For each of the tests, one backup task was running.

During a hot backup, ARCServe 6.5 must be aware of the state of the individual records in the
database. If the record is being updated, the backup process must wait until the transaction is
complete before copying the data. NoteBench’s Mail and DB test simulated three hundred users on
the Lotus Notes database for the hot backup testing. Slightly higher performance is evident with
the database hot, but with no users. ARCserve uses an Open File Agent and part of its Lotus Notes
Agent to backup open or in use files.

The cold backup was with the server running, but with Lotus Notes shut down and the database
inaccessible to users. In this case, ARCServe treats the database as if it were a flat database file.
The Lotus Notes Agent and Open File Agent are not used.

Figure 11 illustrates comparative backup performance using RAIT 0, RAIT 1, and RAIT 5
configurations. The same 20-GB Lotus Notes database was used for each of the tests. Average data
compressibility was 1.7:1. In each test, one backup job used four DLT Drives in two Libraries.

Performance
improvements are possible
by running more than one
job with each job backing
up different databases.

45
40
35
30

GB/Hr

25
20
15

Hot (300 Users)
Hot (No Users)
Cold (Offline)

10

5
0

RAIT 0 RAIT 1 RAIT 5

Figure 11 – Comparative Performance – Relational Database Local Backup

ECG046/0298 11

DLT LIBRARY PERFORMANCE BRIEF (cont.)

GB/Hr

Case 4 – Local Flat File Restore – Non RAIT

Case 4 demonstrates local restore performance. A single Compaq ProLiant 6000 server has four
Compaq 3570 Libraries attached. These tests demonstrate simple flat file (non-relational) database)
file restore. Figure 3 illustrates this network with four Compaq 3570 Libraries attached to the
Compaq ProLiant 6000.

Image Restore

This test demonstrates image restore performance from four Compaq 3570 Libraries to one
Compaq ProLiant 6000 server. Cheyenne ARCserve 6.5 image restore copies data from the backup
media to the server. An image restore is usually the fastest method to restore an entire file system.
Cheyenne ARCserve 6.5 supports selective restore (single file selection and restore). This is slower
than when a selective restore archive is used for this purpose. Selective restore was not evaluated.

As was previously demonstrated during backup testing, increasing the number of drives and jobs
improves restore performance. This performance boost is due to greater utilization of the hardware
capabilities made possible by running multiple threads. Launching concurrent tasks that will use
these threads improves efficiency significantly. This improvement is apparent in Figure 12.

160
140
120
100

80
60
40
20

0

2 4 8

Jobs

Figure 12 -- Flat File NON-RAIT Performance – Image Restore

ECG046/0298 12

DLT LIBRARY PERFORMANCE BRIEF (cont.)

Case 5 – Local Flat File Restore – RAIT

This case demonstrates local restore performance. A single Compaq ProLiant 6000 server has four
Compaq 3570 Libraries attached. These tests demonstrate simple flat file (non-relational database)
file restore. Figure 3 illustrates the network hardware with four Compaq 3570 Libraries attached to
the Compaq ProLiant 6000. ARCServe 6.5 operated the Compaq 3570 Libraries in RAIT 0, RAIT
1, and RAIT 5 configuration for these tests.

Image Restore – RAIT 0

Four configurations were tested for this performance demonstration. All configurations used a
single Compaq ProLiant 6000 server. Tests were performed with one, two, and four Compaq
Model 3570 DLT Libraries attached. Both drives in each library were used. The data volume was
364-GB, as in the previous tests. Figure 13 shows the results of these tests.

140
120
100

GB/Hr

80
60
40
20

0

1 2 4

Jobs

Figure 13 -- Flat File RAIT 0 Performance – Image Restore

Image Restore – RAIT 1

This test was performed with one Compaq ProLiant 6000 server with a two drive Compaq 3570
Library attached and both 35/70 GB DLT drives mirrored. One job was launched. Data volume
was 364-GB. Compressibility of the data impacts the maximum achievable data rate.

60
50
40

GB/Hr

30
20
10

0

1:1 2:1 4:1

Compression Ratio

ECG046/0298 13

Figure 14 -- Flat File RAIT 1 Performance – Image Restore

DLT LIBRARY PERFORMANCE BRIEF (cont.)

Image Restore – RAIT 5

This test was performed with one Compaq ProLiant 6000 server with four Compaq Model 3570
DLT Libraries attached. Two jobs were running for each test. Data volume was 364-GB. Data
compressibility has a significant impact on data throughput, as indicated by Figure 15

75
70
65
60

GB/Hr

55
50
45
40

1:1 2:1 4:1

Compression Ratio

Figure 15 -- Flat File RAIT 5 Performance – Image Restore

A significant point to observe in the data presented above is that with 2:1 compressible data and a
single job running, 50- to 60-GB/Hr data rates are typical under Windows NT and ARCServe 6.5.
This is true regardless of the number of libraries or drives in the array. Other software
implementations and operating systems may yield different results.

ECG046/0298 14

DLT LIBRARY PERFORMANCE BRIEF (cont.)

Case 6 – Local Relational Database Restore

This case demonstrates a Lotus Notes restore task. In this scenario, the Compaq ProLiant 6000
server has two Compaq 3570 Libraries attached. The database is stored locally, e.g. attached to the
same server.

The data set for this test was a Lotus Notes database. The total volume of data was 20-GB of
approximately 1.7:1 compressible data. A single restore job was running. The results below are
normal: if a backup ran at 50- to 60-GB/Hr, restore normally runs at 40-GB/Hr. The database was
offline for all restore testing. The Lotus Notes server can not be running while restoring the
databases.

40
35
30
25

GB/Hr

20
15
10

5
0

RAIT 0 RAIT 1 RAIT 5

Figure 16 – Comparative Performance - Relational Database Restore

Figure 16 compares the results of RAIT 0, RAIT 1 and RAIT 5 testing. Backup performance is
slightly faster than restore performance.

RAIT Tape Management

The backup administrator can create a logical grouping of resources, a Virtual Library, larger than
is possible with a single Compaq 3570 Library. Physical resources are assigned to meet data
volume requirements exceeding a single library, or to allow implementation of RAIT functionality.
The Virtual Library concept requires advanced media management features to keep media
organized.

Media management features are included in ARCserve 6.5 through the RAIT Option and the
Library Option to support the Compaq 3570 Library. Media serial numbering and bar code support
assist in efficiently automating management of DLT media cassettes. If you use a
Grandfather/Father/Son backup rotation scheme, media automatically moves from scratch set to
rotation set, rotation set to save set and save set to scratch set based on user defined retention
periods.

ECG046/0298 15

DLT LIBRARY PERFORMANCE BRIEF (cont.)

Data Migration Option with Tape library Option

Data Migration or Hierarchical Storage Management (HSM) allows the backup administrator to
utilize available storage resources efficiently. Infrequently used data is migrated to less expensive
storage devices until users request it. ARCServe 6.5 allows the backup administrator to establish
policies to automatically migrate data to and from any NTFS formatted Windows NT volume on
tape drives and libraries, optical storage devices, and magnetic hard disks.

Disaster Recovery with Tape Library Option

Due to the size of the programs necessary to operate the changer, disaster recovery procedures are
more complex than with a single tape drive. ARCserve and other backup software support disaster
recovery that utilizes the tape changing ability of the library. The library can create a set of backup
tapes that can be used either on a single standalone drive or in a drive within the library for disaster
recovery. ARCServe 6.5 has the ability to create a set of disks for use with this backup set.

ECG046/0298 16

DLT LIBRARY PERFORMANCE BRIEF (cont.)

.

.

CONCLUSION

.

.

.

.

.

The Compaq 3570 Library sets new standards in high-speed backup performance where LAN

.

.

.

speed is not a limiting factor. Note that the 20+GB/Hr in performance backup rates measured

ECG046/0298 17

.

.

.

above are not sustainable by a single Ethernet 100Mbit/second LAN and the 15/30 Library, at 1/3

.

.

.

the cost, may be a better price performer.

.

.

.

.

For ARCserve, when using RAIT (0 or 5) and a single job, performance is limited to 60GB/Hr or

.

.

.

less regardless of the number of tape drives in the RAIT. When using multiple concurrent jobs

.

.

.

backup rates increase at approximately 30GB/Hr/tape up to a maximum of 165GB/Hr. These

.

.

.

numbers are consistent and equivalent whether the tape drives are used in a physical Compaq DLT

.

.

.

Tape Array or in multiple Compaq DLT Libraries.

.

.

.

.

.

Both Libraries, when used with ARCserve and other backup software applications, make operation

.

.

.

and control of multiple Libraries much easier to use. When combined with Cheyenne ARCseve

.

.

.

6.5, the capacity to make fast, reliable backups using similar configurations of high-performance

.

.

.

Compaq products is multiplied.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

DLT LIBRARY PERFORMANCE BRIEF (cont.)

APPENDIX 1

Raw Performance Data

FLAT FILE RESULTS

Jobs Compression Raid Level DLT Drives 35/70 Libraries Backup (GPH) Fig Restore (GPH) Fig

2 1:1 None 2 1 29.29 24.52
2 2:1 None 2 1 54.02 39.25
2 4:1 None 2 1 57.71 Figure 4 45.23 Figure 12
4 1:1 None 4 2 58.25 54.25
4 2:1 None 4 2 102.25 86.25
4 4:1 None 4 2 112.12 Figure 4 94.25 Figure 12
8 1:1 None 8 4 82.45 72.45
8 2:1 None 8 4 165.23 134.22
8 4:1 None 8 4 164.83 Figure 4 139.49 Figure 12
2 1:1 0 4 2 60.23 54.23
2 2:1 0 4 2 97.03 75.25
2 4:1 0 4 2 107.11 89.26
1 1:1 0 8 4 39.22 35.22
1 2:1 0 8 4 83.56 75.12
1 4:1 0 8 4 82.48 Figure 8 74.25 Figure 13
2 1:1 0 8 4 67.23 55.25
2 2:1 0 8 4 134.64 97.25
2 4:1 0 8 4 137.52 Figure 8 101.54 Figure 13
4 1:1 0 8 4 78.46 66.81
4 2:1 0 8 4 153.22 126.76

4 4:1 0 8 4 148.12 132.12 Figure 13
2 1:1 1 4 2 28.89 Figure 9 25.56 Figure 14
2 2:1 1 4 2 54.24 Figure 9 49.25 Figure 14
2 4:1 1 4 2 57.21 Figure 9 48.21 Figure 14
2 1:1 5 8 4 54.14 Figure 10 44.25 Figure 15
2 2:1 5 8 4 107.25 Figure 10 60.25 Figure 15
2 4:1 5 8 4 113.24 Figure 10 72.25 Figure 15

ECG046/0298 18

DLT LIBRARY PERFORMANCE BRIEF (cont.)

Lotus Notes 4.5.3a using ARCserve 6.5 with Raid and Lotus Notes Agent for ARCserve. Database
was 20-GB with an average of 1.7:1 compression. Backup data is presented in Figure 11. Figure 16
presents the results of the restore performance tests.

HOT (NO USERS) DATA

Raid Level DLT Drives 35/70 Lib Backup (GPH)

0 4 2 39.75
1 4 2 19.25
5 4 2 34.23

HOT (300 USERS)

Raid Level DLT Drives 35/70 Lib Backup (GPH)

0 4 2 28.12
1 4 2 14.45
5 4 2 26.11

COLD (OFFLINE)

Raid Level DLT Drives 35/70 Lib Backup (GPH) Restore (GPH)

0 4 2 41.12 37.45
1 4 2 21.45 22.11
5 4 2 35.23 19.44

ECG046/0298 19

DLT LIBRARY PERFORMANCE BRIEF (cont.)

APPENDIX 2

Standard Test System

System Compaq ProLiant 6000 with dual peer PCI bus architecture
Processor Quad Pentium Pro processors with 512K L2 cache
Memory 320-MB ECC DIMMs
Disk Array Configuration Four Compaq SMART-2 Array Controllers feeding Eight channels (40 drives total)
Disk Drives Compaq 9.1-GB Ultra Wide SCSI-3
SCSI Cards Compaq Wide-Ultra PCI SCSI Controller for the DLT Libraries
Network Cards Compaq Netelligent 10/100Mbit/s TX PCI Ethernet Controller
Operating System Windows NT 4.0, Build 1381 with Service Pack 3
Backup Software Cheyenne ARCserve 6.5 Build 600, w/Tape Library Opt. V3, Image and RAID 0pt. V2
Database Software Lotus Notes

The Compaq ProLiant 6000 features a dual peer-to-peer PCI bus connected to a common
1,919-GB/hr or 540-MB/s Gunning Transceiver Logic (GTL+) host processor bus. This feature
permits a path to the processor that is twice the width of buses in many servers on the market
today. Figure 17 illustrates the dual PCI bus architecture.

Slots

4 Pentium

Pro

Processors

256 MB
Memory

Secondary PCI Bus

7

8

SMART-

9

SMART-

10

Networ

11

Cards

SCSI

SCSI

Devices

TTT

T

S

2

SSS

2

k

ECG046/0298 20

3

4

Primary PCI Bus

5

6

Figure 17 -- ProLiant 6000 peer-to-peer bus architecture.

T T

SCSI

SCSI

T T

SMART-

SMART-

T = 35/70 Gb DLT Drive
S = Storage Unit

S S

2

S S

2

DLT LIBRARY PERFORMANCE BRIEF (cont.)

APPENDIX 3

Operating the Compaq 3570 Library

Cable Quality

The cables provided with the various Compaq option kits are of very high quality. If they are ever
damaged, replace them with identical Compaq parts. SCSI data transfers will operate poorly (or
not at all) when signal conditions are degraded due to low-quality cabling. Excessive cable length
also contributes to slower performance.

Data and Equipment Protection

Always screw the external cable heads into the SCSI cards and Compaq 3570 Library chassis to
prevent damage to the connectors and possible loss of data.

Transfer Speed

Use a Compaq 32-bit Ultra-Wide (SCSI-3) controller with the Compaq 3570 Library for best
performance. To identify the card, look at the main logic chip in the center of the card and verify
that it is a Symbios Logic 53c875 part. Cards with the NCR 53c825 part will not deliver maximum
library performance.

Cleaning

Clean the drive by loading a cleaning tape (PN 199704-001). It will clean the drive, then
automatically unload.

The cleaning tape is effective for 20 cleaning cycles. Do not exceed this. If a backup tape causes
the cleaning light to come on during a backup operation, copy the tape or recover it and remove it
from service.

Such a tape can contaminate the drive, which may then contaminate other tapes if the drive is not
cleaned between loads. Never try to clean a DLT drive with anything other than a cleaning tape.

Operating Considerations

Dropping or physically shocking DLT cassettes can damage the tape. Damaged tapes can damage
the drive. If the cartridge appears damaged, do not use it.

DLT cassettes should be kept in the magazine supplied with the Compaq 3570 Library. This
magazine protects the cassettes from handling damage or contamination.

ECG046/0298 21

DLT LIBRARY PERFORMANCE BRIEF (cont.)

APPENDIX 4

DLT Media Considerations

Performance

The 35/70 GB DLT drive can read half-inch DLT media labeled DLT Tape Type III, DLT Tape
Type III XT or DLT Tape IV. The DLT Tape Type IV delivers optimum performance and
maximum capacity when used with the 35/70 DLT drive. The DLT Tape IV medium holds a
minimum of 35 GB of uncompressed data.

The 70 in 35/70 GB DLT refers to the drive’s ability to compress data, usually at a factor of at least
2:1. Higher compression ratios are achievable, but to be conservative, the industry uses 2:1 as an
average figure. When backing up servers containing an assortment of general end-user file data, the
Compaq 3570 Library can exceed 100 GB per tape cartridge. Typical data compressibility of
uncompressed data is listed in the accompanying table.

TYPICAL DATA COMPRESSIBILITY

Data Type Compressibility

4

These figures are typical,
and were not tested by
Compaq. Your results may
vary, depending on the
actual data.

Bitmaps 6.9:1
CAD 3.8:1
Source 2.9:1
Spreadsheet 2.5:1
Desktop Publishing 2.5:1
Database 1.7:1

If the data is already compressed on the source drive, the opposite will occur. When compressed
data is passed through a compression process a second time, it typically occupies more space. This
is due to the overhead of the compression algorithm.

Durability

The DLT Tape IV has a minimum shelf life of 30 years if stored properly. Its rated usage is
500,000 passes; however, it is suggested that tapes be rotated on a more frequent schedule.

Media and Drive Compatibility

ECG046/0298 22

4

These compression ratios are meant to be approximations useful for estimating backup
performance and capacity. The numbers were originally published by Sony. Only the database
compression ratio was directly measured for this paper. This number was measured for the Lotus
Notes database used in testing for this whitepaper. The number of users, number of notes or
attachments and other variables can substantially affect this number. Databases other than Lotus
may compress at a different ratio.

DLT LIBRARY PERFORMANCE BRIEF (cont.)

The DLT Library uses either the 15/30 GB DLT Drive or the 35/70 GB DLT Drive specially
modified to work in the Library. Non-Library Ready drives cannot be used in the Library. The
15/30 and 35/70 Drives cannot be mixed in the Library. The 35/70 GB DLT Drive is ‘downward’
or ‘backward’ compatible. This means that the 35/70 GB DLT Drive can read and write any DLT
tape (data) cartridge type in the maximum density available for the cartridge type. The converse
statement, ‘upward’ or ‘forward’ compatible is not true. A 20/40 DLT Drive cannot read type IV
tape cartridges formatted and written on a GB DLT Drive. The 35/70 GB DLT Drive will
automatically select the proper density and recording pattern appropriate for the tape type and any
data recorded on it. The ‘recording pattern’ is essentially either 2 ‘head’ or 4 ‘head’ technology, the
35/70 GB DLT drive being the only 4 head (Symmetric Phase Recording SPR) recording pattern
technology at present.

ECG046/0298 23

DLT LIBRARY PERFORMANCE BRIEF (cont.)

Here are some examples to clarify this:

• if a type IV cartridge containing data written on a 20/40 GB DLT Drive is used in a 35/70 GB

DLT Drive, the 35/70 GB DLT Drive will read and write data in a manner readable by the
20/40 GB DLT Drive

• if a type IIIxt cartridge containing data written by a 15/30 DLT Drive is loaded into a GB DLT

Drive, it can be read and written in a density and recording pattern usable by the 15/30 GB
DLT Drive

• if a type IV cartridge containing no data is used in a 35/70 GB DLT Drive it will be formatted,

read and written using 70GB density and 4 head(SPR) recording pattern – thereafter if this
tape cartridge is inserted in a 20/40 GB DLT Drive it cannot be read

• If a type III cartridge is written by a 10/20 DLT Drive then used in a 15/30 DLT Drive it can

be read and written in 20 GB density – the type IIIxt cartridge is longer to hold more data but
is otherwise the same as the type III cartridge. The following table lists the compatibility of
DLT tape (data) cartridge types and DLT tape drives.

DLT Drive Type III Type IIIxt Type IV Notes

Read Write Read Write Read Write

10/20 (2000) Yes Yes No No No* No* *2 head, can read/write

but max 20GB(2:1),
can’t read/write if tape
formatted using 35/70

15/30 (2000xt) Yes Yes Yes Yes No* No*

20/40 (4000) Yes Yes Yes Yes Yes* Yes*

35/70 (7000) Yes Yes Yes Yes Yes Yes 4 head(SPR), max

*2 head, can read/write
but max 30GB(2:1),
can’t read/write if tape
formatted using 35/70

*2 head, max
40GB(2:1), cant
read/write if tape
formatted using 35/70

70GB(2:1), type IV
cartridges written in
SPR format cannot be
read by previous drives

Drive Duty Cycle

Most tape drives have a 15 to 25 percent duty cycle, which means the manufacturer expects them
to actually be writing tape some small percentage of time during their installed lives. The DLT
drive is a 100 percent duty cycle rated device. While it won’t be writing 100 percent of the time,
the drive is tougher and can be used more often, with greater confidence in its reliability than
drives with a lower percentage duty cycle.

ECG046/0298 24

DLT LIBRARY PERFORMANCE BRIEF (cont.)

APPENDIX 5

Table of Figures

Figure 1-- Compaq 3570 Library __________________________________________________ 1
Figure 2 -- Compaq ProLiant 6000_________________________________________________ 4
Figure 3 -- Flat File Network _____________________________________________________ 5
Figure 4 -- Flat File Performance – Multiple Jobs_____________________________________ 5
Figure 5 -- RAIT 0 Data Distribution _______________________________________________ 6
Figure 6 -- RAIT 1 Data Mirroring_________________________________________________ 7
Figure 7-- RAIT 5 Writing Pattern _________________________________________________ 8
Figure 8 -- Flat File RAIT 0 Performance – Image Backup ______________________________ 9
Figure 9 -- Flat File RAIT 1 Performance – Image Backup ______________________________ 9
Figure 10 -- Flat File RAIT 5 Performance – Image Backup ____________________________ 10
Figure 11 – Comparative Performance – Relational Database Local Backup_______________ 11
Figure 12 -- Flat File NON-RAIT Performance – Image Restore _________________________ 12
Figure 13 -- Flat File RAIT 0 Performance – Image Restore ____________________________ 13
Figure 14 -- Flat File RAIT 1 Performance – Image Restore ____________________________ 13
Figure 15 -- Flat File RAIT 5 Performance – Image Restore ____________________________ 14
Figure 16 – Comparative Performance - Relational Database Restore ____________________ 15
Figure 17 -- ProLiant 6000 peer-to-peer bus architecture. _____________________________ 20

ECG046/0298 25