Compaq 312134-B21 - Intel Pentium II Xeon Processor... User manual

7(&+12/2*<%5,()

ECG075.0997
6HSWHPEHU 
&RPSDT &RPSXWHU
&RUSRUDWLRQ
CONTENTS
,QWURGXFWLRQ 
7HVW  7DSH 'ULYHV LQ 5HODWLRQ WR
6&6, &DUGV

:LQGRZV 17  
7HVW  7DSH 'ULYHV LQ 5HODWLRQ WR
6&6, &DUGV

1RYHOO ,QWUDQHW:DUH
 *%KU 
7HVW   *%KU 3HUIRUPDQFH

1RYHOO ,QWUDQHW:DUH  
7HVW  /RFDO %DFNXS RI
:LQGRZV 17 WR 9DULRXV 1XPEHUV RI
'/7 'ULYHV  
7HVW  8VH RI &RQFXUUHQW -REV WR
,PSURYH 3HUIRUPDQFH

:LQGRZV
17

 *%KU 
7HVW  &RPSUHVVLELOLW\ DQG 
'/7 WR  '/7 3HUIRUPDQFH 
7HVW  5$,7 9HUVXV 5$,7

:LQGRZV 17 
7HVW  1HWZRUN 3HUIRUPDQFH

6LPSOH %DFNXS RI D 5HPRWH &OLHQW

:LQGRZV 17 
7HVW  1HWZRUN 3HUIRUPDQFH

:LQGRZV 17 6HUYHU 3XOOLQJ IURP DQ
,QWUDQHW:DUH 5HPRWH &OLHQW 
7HVW  )LEUH &KDQQHO$UELWUDWHG
/RRS )&$/ /RRS%DVHG %DFNXS
RI  *%

:LQGRZV 17  
7HVW  )HHG 6SHHG
5HODWLRQVKLSV 
&RQFOXVLRQ 
$SSHQGL[  &RPSDT  *% '/7
'ULYH 1RWHV 
$SSHQGL[  &RPSDT '/7 7DSH $UUD\
,, 1RWHV 
$SSHQGL[  6WDQGDUG +LJK
3HUIRUPDQFH+LJK6SHHG %DFNXS
7HVW 6\VWHPV 
&RPSDT'/77DSH$UUD\,,
+LJK3HUIRUPDQFH%DFNXSRI(QWHUSULVH
6HUYHUV
EXECUTIVE SUMMARY
Since its introduction in February, 1996, the Compaq DLT Tape Array (which uses Compaq
15/30 GB DLT drives) has been an industry leader in backup performance capabilities. The
DLT Tape Array II, which uses 35/70 GB DLT drives, sets new standards in high-speed backup
performance.
Early in 1997, Compaq performed a series of tests to examine the performance of the Compaq
35/70 GB DLT drive and the DLT Tape Array II. The objective of these tests was to find the
highest-performance configurations in a variety of backup scenarios on Compaq servers
running Windows NT and Novell IntranetWare network operating systems.
The Compaq DLT Tape Array II provides high-capacity, high-speed tape storage for use in
local and remote backup strategies. It uses the Compaq 35/70 GB DLT drive, which is
engineered for superior performance. This brief addresses the technical aspects of high-
performance backups that can reach and exceed 165 gigabytes per hour (GB/hr) with off-the-
shelf Compaq hardware.
0
20
40
60
80
100
120
140
160
180
Network Backup Local Backup:
Novell IntranetWare Local Backup: Windows NT
GB/hr
One 100TX Ethernet
connection. Four
concurrent backups
from four clients to
four tape drives
(Test 9).
One backup to
eight tape drives
(Test 2).
Eight concurrent
backups to eight
tape drives
(Test 5).
Figure 1: Maximum results in each backup scenario tested.
Please direct comments regarding this communication to the ECG Technology Communications Group at this Internet
address: [e-mail]
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

7
(&+12/2*<
%
5,() (cont.) DLT Tape Array II

ECG075.0997
NOTICE
The information in this publication is subject to change without notice and is provided “AS IS”
WITHOUT WARRANTY OF ANY KIND. THE ENTIRE RISK ARISING OUT OF THE
USE OF THIS INFORMATION REMAINS WITH RECIPIENT. IN NO EVENT SHALL
COMPAQ BE LIABLE FOR ANY DIRECT, CONSEQUENTIAL, INCIDENTAL, SPECIAL,
PUNITIVE OR OTHER DAMAGES WHATSOEVER (INCLUDING WITHOUT
LIMITATION, DAMAGES FOR LOSS OF BUSINESS PROFITS, BUSINESS
INTERRUPTION OR LOSS OF BUSINESS INFORMATION), EVEN IF COMPAQ HAS
BEEN ADVISED OF THE POSSIBILITY OF SUCH DAMAGES.
The limited warranties for Compaq products are exclusively set forth in the documentation
accompanying such products. Nothing herein should be construed as constituting a further or
additional warranty.
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 and ProLiant are registered with the United States Patent and Trademark Office.
Netelligent is a trademark and/or service mark of Compaq Computer Corporation.
Microsoft, Windows, and Windows NT are trademarks and/or registered trademarks of
Microsoft Corporation.
NetWare and Novell are registered trademarks and IntranetWare, is a trademark of Novell, Inc.
Pentium is a registered trademark of Intel Corporation.
DLT is a trademark applied for by Quantum.
Other product names mentioned herein may be trademarks and/or registered trademarks of their
respective companies.
©1997 Compaq Computer Corporation. All rights reserved. Printed in the U.S.A.
Compaq DLT Tape Array II:
High-Performance Backup of Enterprise Servers
Second Edition (September 1997)
ECG075.0997
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

7
(&+12/2*<
%
5,() (cont.) DLT Tape Array II

ECG075.0997
INTRODUCTION: THE NEED FOR SPEED
As companies migrate to standardized server solutions, there are significant changes in
backup/recovery paradigms. The five-days-a-week, nine-hours-a-day operation has expanded to
seven days a week and 24 hours a day. Backups must be performed concurrently (live) or
during very small scheduled outage windows. In either case, the backup must be performed in a
minimum amount of time: Speed is essential.
With the size of data sets and databases approaching and exceeding the terabyte mark,
customers require improved performance for large backups. The chart below illustrates typical
peak server backup speeds for the last five years. Today, Compaq passes 165 gigabytes per
hour (GB/hr) with off-the-shelf technology.
0
10
20
30
40
50
60
70
80
90
100
110
120
130
140
150
160
170
1993 1994 1995 1996 1997
GB/hr
Figure 2: Backup speed improvement in the last five years.
The 1997 figure represents results achieved by Compaq.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

7
(&+12/2*<
%
5,() (cont.) DLT Tape Array II

ECG075.0997
RESULTS: 165 GB/HR BACKUPS
Testing revealed three important factors in a high-speed backup solution:
•The base hardware capabilities of the system, centering on the PCI bus architecture.
•The speed at which the disk drive storage system can supply data to the tape system
(referred to as feed speed).
•The tape technology selected to perform the backup.
In the tests, the system that provided optimum performance consisted of a Compaq ProLiant
6000 with four or eight drives in a DLT Tape Array II, which was fed data from a pair of
Compaq SMART-2/P Array Controllers and associated SCSI disk drives. This configuration
worked very well for Novell IntranetWare and Microsoft Windows NT network operating
systems.
Test results show the DLT Tape Array II can exceed backup speeds of 150 GB/hr in multiple
operating system environments.
256 MB
Memory
4 Pentium
Pro
Processors
Secondary PCI Bus
Primary PCI Bus
7
8
9
10
11
3
4
5
6
Slots
T
T
T
S
T
S
S
S
T T
T T
S S
S S
SCSI
SCSI
SCSI
SCSI
SMART-
2
SMART-
2
SMART-
2
SMART-
2
Networ
k
T = 35/70 Gb DLT Drive
S = Storage Unit
Cards Devices
Figure 3: High-speed backup off-the-shelf-solution: ProLiant 6000 test system.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

7
(&+12/2*<
%
5,() (cont.) DLT Tape Array II

ECG075.0997
HIGH-SPEED BACKUP RESEARCH
Single and multi-processor systems were used to identify hardware factors that affect backup
performance, including processors, memory, SCSI, and PCI channel bottlenecks. Some tests
were run in a network environment; however, the focus was on local backups, because network
bottlenecks significantly reduce feed speed. Future testing will address this problem and
explore other ways to improve network-based backups.
Listed below are test machine configuration details, along with factors that were taken into
consideration in developing the highest performance configurations.
Processor Quantity
The average backup uses from 6 to 25 percent of a single 200 MHz Pentium Pro (256K cache
version) processor’s power. The lower number is typical when using a single disk or tape drive;
usage percentage grows substantially when striping data across a redundant array of
independent tape drives (RAIT) in a RAIT-5 configuration. All testing was performed on
Compaq ProLiant 5000 and ProLiant 6000 systems so that up to four 200 MHz Pentium Pro
processors could be used. An assumption was made that most large systems would have at least
two processors, and the majority of tests were performed with that number.
Memory
A system with 256 MB of RAM (regardless of the number of processors) was the standard for
all tests, to eliminate the remote possibility of complications due to inadequate RAM.
PCI Bus
Although some testing was done with EISA bus-based systems, the higher throughput of PCI
(up to three times faster) clinched the decision to use it for high-speed backup testing. The 32-
bit x 33 MHz PCI bus peaked at 454 GB/hr in load tests with six Compaq SMART-2/P Array
Controllers. To maximize bandwidth, tests were also performed to determine optimum card
placement in the dual PCI bus system of the ProLiant 5000 and ProLiant 6000. The dual peer-
to-peer PCI buses connect to a common 1,919-GB/hr, or 540megabytes per second (MB/s)
Gunning Transceiver Logic (GTL+) processor bus (the host bus), which permits a path to the
processor that is twice the width of buses in many servers on the market today (Figure 4).
Slots
Pentium Pro
Pentium Pro
Processors
Processors
Host Bus
Host Bus
Memory
Memory
Secondary PCI Bus
Secondary PCI Bus
Host-
Host-
to-PCI
to-PCI
PCI-to-EISA
PCI-to-EISA
EISA Bus
EISA Bus
33 MB/s
133 MB/s 133 MB/s
Primary PCI Bus
Primary PCI Bus
540 MB/s
Bridge Bridge
Bridge
Slots
Slots
Figure 4: ProLiant 5000 peer-to-peer bus architecture. The ProLiant 5000 has seven PCI slots
and three EISA slots across the two PCI buses (eight total; two are shared).
The ProLiant 6000 has eleven slots: nine PCI and two EISA.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

7
(&+12/2*<
%
5,() (cont.) DLT Tape Array II

ECG075.0997
The dual peer-to-peer buses provide throughput that is superior to that of single PCI-buses
typical of smaller servers. Many dual buses are stacked buses. They use a PCI-to-PCI-bridge,
which feeds one PCI bus to the other and then goes through a host-to-PCI-bridge to get to the
host bus (Figure 5). Although this permits the use of additional cards by providing more slots, it
does not increase performance. In contrast, the additional PCI card space in the ProLiant 5000
and ProLiant 6000 permits use of multiple SMART-2/P Array Controllers, multiple SCSI cards
for tape drives, and multiple network interface cards. All of these devices run at peak PCI
speeds, coexisting with minimal bottlenecks in performance. Although the PCI bus is rated at
133 MB/second, that is a burst data transfer speed. Actual measurements are typically from 60
MB/second to a maximum sustained peak of about 100 MB/second.
PCI-to-PCI
Bridge
Processors Memory
Host Bus
Host-to-PCI
Bridge
PCI Bus
133 MB/s
PCI Bus 133 MB/s
Slots
Slots
Figure 5: PCI-to-PCI bridge architecture.
Feed Speed
The disk drive subsystem is a critical component in high-speed backups. There is a direct
relationship between the number of disk drives in the array and the speed at which the array can
provide data to the processor (more drives allow greater speed). There is also a direct
relationship between the number of RAID controllers and performance (more is better). For
example, a server using a single SCSI card to access hard disks is limited to 10, 20 or 40 MB/s,
depending on the type of SCSI cards and drives in use. On the other hand, an array controller
can often deliver the maximum limit of performance that the PCI bus can deliver: 133 MB/s.
The Compaq SMART-2/P Array Controller is the best choice. It has two channels, and a
minimum of four drives per channel should be used for high performance. If only six drives are
available, the best performance will be achieved by assigning three drives to each channel.
Adding a fourth drive to each channel improves performance significantly. Adding more than
four drives per channel makes only a minor improvement in throughput, although it does add
capacity.
Tests were performed with an assortment of narrow and wide drives ranging in size from 2 to 9
gigabytes (GB). Superior performance was seen on the Compaq Fast-Wide 2 GB, 4 GB and 9
GB drives. See Test 12 for more information.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

7
(&+12/2*<
%
5,() (cont.) DLT Tape Array II

ECG075.0997
Data Set
To simulate the types of file structures and data mixtures customers typically want to back up,
mixes of data types included large files, large numbers of files, and large numbers of directory
structures.
The size of a data set can affect backup speed. With a data sample of less than 1 GB,
performance is sometimes affected by the tape drive size, as well as the tape drive cache and
disk cache. For example, a 32-MB file might appear to have been backed up in less than 1
second, simply because it went to the 8-MB cache in each of the four tape drives of the array.
With data sets from 1 GB to 1 terabyte, performance was found to remain constant. The
minimum test data set size for this project was 2 GB.
Compression
Hardware data compression is built into the 35/70 GB DLT drive. The default setting is to
compress data. Compression has a direct impact on backup performance. Tests were performed
with 2:1 and 4:1 compressible data sets, to simulate the most common levels of compressibility
in actual working environments. Highly random sets of data may be less compressible, while
large empty databases may be compressible to much higher levels. Determine how
compressible the data sets are by testing them, usually by observing the speed of backup that
the backup program reports. (The backup programs used for these tests [Cheyenne ARCserve
and JetServe] report the compressibility of the data set being backed up. This number is
reported to the backup application by the drive.) See Test 6 for more information.
Backup Types
Two types of backup techniques are generally available: image and by file.
In image backups, the backup program looks at the disk drive or file system as a single unit and
takes a snapshot of it, backing it up in one session. Image is usually the fastest method of
backup and restore for an entire file system; however, some image backups can only be restored
as an entire unit, meaning the entire backup must be restored to obtain a single file.
By file backups are made on a file-by-file basis, retaining the information needed to retrieve a
single file from the backup tape and restore it to the correct location on the system. The by file
method usually takes much longer than the image method, because the full path/directory
information is saved with each file. Nevertheless, it permits random access to the files for faster
retrieval of individual files.
Both methods were included in Compaq’s testing scenarios. It should be noted that some image
backup programs (notably Cheyenne ARCserve for NT) have the ability do a by file restore
from an image backup, but using that functionality usually extends the restore operation time
significantly.
Test Sites
All tests except test 10 were performed on systems located in the Compaq Secondary Storage
Systems Engineering lab. Test 10 was performed on the Max-Storage system in the Compaq
Maximum Configuration Lab. All network loads were performed on lab networks.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

7
(&+12/2*<
%
5,() (cont.) DLT Tape Array II

ECG075.0997
TEST 1T
APE DRIVES IN RELATION TO SCSI CARDS 
WINDOWS NT
2SHUDWLQJ 6\VWHP
Windows NT 4.0
%DFNXS 6RIWZDUH
Cheyenne ARCserve 6.0 for Windows NT
%DFNXS 0HWKRG
Image and by file
'DWD 6HW
2 GB
&RPSUHVVLRQ
4:1
7DSH )RUPDW
RAIT-0
+DUGZDUH
ProLiant 5000, two Pentium Pro 200 processors, 256K
cache, 256 MB RAM
80-GB array (two SMART-2/P Array Controllers
with twenty 4-GB drives)
35/70 GB DLT drive(s)
Wide-Ultra SCSI-3 cards
(All Compaq off-the-shelf products)
3XUSRVH
To observe the performance of various numbers of tape drives in relation to various numbers of
SCSI cards under Windows NT 4.0.
5HVXOWV
Of the configurations tested, two SCSI cards driving two 35/70 GB DLT drives offer the fastest
speed. Although a typical configuration (two SCSI cards driving four 35/70 GB DLT drives)
tested two percent slower, it doubles the capacity of the backup, and is therefore recommended
to avoid tape changes during the backup of large data sets. Tape change delays might easily
offset any performance advantage of using the smaller two-drive configuration.
There is no benefit to using a separate SCSI card for each drive when using the Wide-Ultra
SCSI Controller (part number 272514-001). The bandwidth of this card can easily drive two
35/70 GB DLT drives. On a system with only Fast-Wide SCSI Controllers (part number
199533-001), splitting the four drives among four cards will help somewhat. It is also evident
that image backups are considerably faster in a multi-drive configuration.
The ProLiant 5000 uses the slower Fast/Wide technology on its single embedded SCSI
controller. It is recommended that this be used only for slower devices, not for multiple tape
drives. The ProLiant 6000 has two embedded Wide-Ultra SCSI Controllers, of which the B-
port is available for use with externally mounted tape devices. The A port can also be used;
however, an optional cable is required.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

7
(&+12/2*<
%
5,() (cont.) DLT Tape Array II

ECG075.0997
0
10
20
30
40
50
60
1 to 1 1 to 2 2 to 2 2 to 4 4 to 4
Image Backup Image Restore By File Backup By File Restore
GB/hr
Figure 6: Performance results from tests comparing various proportions of
53c875 SCSI cards to DLT drives (shown at the bases of the columns).
1XPEHU RI
6&6, &DUGV
1XPEHU RI 
*% '/7 'ULYHV
%DFNXS
*%KU
5HVWRUH
*%KU
Image Backup/Restore
1 1 34.38 21.89
1 2 51.91 31.09
2 2 56.54 56.54
2 4 51.91 31.15
4 4 49.73 30.21
Backup/Restore by File
1 1 34.95 31.79
1 2 45.51 47.89
2 2 44.96 59.00
2 4 44.96 57.62
4 4 43.13 58.53
Table 1: Test data for tape drives versus SCSI card test

Windows NT.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

7
(&+12/2*<
%
5,() (cont.) DLT Tape Array II

ECG075.0997
TEST 2T
APE DRIVES IN RELATION TO SCSI CARDS 
NOVELL INTRANETWARE 151 GB/HR BACKUP
2SHUDWLQJ 6\VWHP
Novell IntranetWare 4.11
%DFNXS 6RIWZDUH
Cheyenne JETserve 4.0 Plus
%DFNXS 0HWKRG
Image
'DWD 6HW
8 GB
&RPSUHVVLRQ
4:1
7DSH )RUPDW
RAIT-0
+DUGZDUH
ProLiant 6000, three Pentium Pro 200 processors, 512 K
cache, 384 MB RAM
224-GB array (four SMART-2/P Array Controllers
with fifty-six 4-GB drives)
35/70 GB DLT drive(s)
Wide-Ultra SCSI-3 cards
(All Compaq off-the-shelf products)
3XUSRVH
To observe the performance of various numbers of tape drives in relation to various numbers of
SCSI cards under Novell IntranetWare 4.11 on a very high-performance backup server
configuration.
5HVXOWV
In a typical DLT Tape Array II solution, two SCSI cards driving four 35/70 GB DLT drives
offer a reasonable backup speed (121 GB/hr). A configuration of four SCSI cards and eight
35/70 GB DLT drives provides very high speed (151.58 GB/hr) and large capacity (more than
560 GB on a single set of mounted tapes). At 2:1 compression, storage capacity reaches 1.12
terabytes.
Restore speeds varied, but were generally about 60 GB/hr.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

7
(&+12/2*<
%
5,() (cont.) DLT Tape Array II

ECG075.0997
0
20
40
60
80
100
120
140
160
1 to 1 1 to 2 2 to 3 2 to 4 3 to 5 3 to 6 4 to 7 4 to 8
Backup Restore
GB/hr
Figure 7: Performance of IntranetWare backups and restores by numbers of SCSI cards in relation to
numbers of tape drives (shown at the bases of the columns).
1XPEHU RI
6&6, &DUGV
1XPEHU RI
'/7 'ULYHV
%DFNXS
*%KU
5HVWRUH
*%KU
1 1 31.61 31.32
1 2 61.94 53.38
2 3 92.61 79.12
2 4 121.78 58.42
3 5 126.59 59.69
3 6 144.36 60.57
4 7 146.57 61.28
4 8 151.58 56.52
Table 2: Test data for numbers of tape drives in relation to numbers of SCSI cards
under Novell IntranetWare.
The relationship between backup and restore speeds is inconsistent for a variety of reasons.
Restore speed varies with the compression of the files. Compression and decompression times
vary with file content and have different effects on CPU use.
Another influence on restore speed is the rate at which the target can receive and write data.
The use of large, multi-megabyte caches on disk drive arrays occasionally allows restore speeds
to exceed backup speeds.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

7
(&+12/2*<
%
5,() (cont.) DLT Tape Array II

ECG075.0997
TEST 3 118 GB/HR PERFORMANCE TEST NOVELL
INTRANETWARE
2SHUDWLQJ 6\VWHP
Novell IntranetWare 4.11 Build 2884
%DFNXS 6RIWZDUH
Cheyenne JETserve 4.0 Plus
%DFNXS 0HWKRG
Image
'DWD 6HW
2 GB
&RPSUHVVLRQ
2:1, 4:1
$JHQWV
None
7DSH )RUPDW
RAIT-0
+DUGZDUH
ProLiant 5000, two Pentium Pro 200 processors, 256K
cache, 256 MB RAM
80-GB array (two SMART-2/P Array Controllers
with twenty 4-GB drives)
35/70 GB DLT drive(s)
Wide-Ultra SCSI-3 cards
(All Compaq off-the-shelf products)
3XUSRVH
The initial purpose of this test was to find the fastest backup solution using Cheyenne JETserve
and to examine performance variables. Long after this set of tests was performed, it was
discovered that the addition of two SMART-2/P Array controllers and disk drives dramatically
improved performance. Test 2 demonstrates this peak performance system.
5HVXOWV
Of the configurations tested, a typical DLT Tape Array II solution with two SCSI cards driving
four 35/70 GB DLT tape drives offers the fastest backup speed (118 GB/hr). The four
card/eight drive configuration provides high speed (89 GB/hr) and very large capacity (more
than 560 GB at 2:1 compression and more than 1 terabyte at 4:1) on a single set of mounted
tapes. Numerous tests showed the two card/four drive setup to be the highest performer. The
2:1 versus 4:1 test results prompted other tests to measure the compressibility factor (see Test 6,
page 19).
This set of tests illustrates the levels of performance that might be seen in this configuration. To
achieve the highest speeds possible, a very fast source hard disk drive storage system is
required. Other factors, such as CPU speed and bus speeds, also influence backup speed.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

7
(&+12/2*<
%
5,() (cont.) DLT Tape Array II

ECG075.0997
0
20
40
60
80
100
120
1 to 1 1 to 2 2 to 2 2 to 3 2 to 4
2 to 4*
2 to 4**
3 to 3 3 to 6 4 to 4 4 to 8
GB/hr
Backup Restore
Figure 8: High-performance backup and restore under Novell IntranetWare with different proportions of
SCSI cards to tape drives (shown at the bases of the columns).
*Hardware compression turned off.
**This test used 4:1 compression; all others, 2:1 compression.
&RPSUHVVLRQ
5DWLR 6&6, &DUGV 7DSH 'ULYHV %DFNXS *%KU 5HVWRUH *%KU
2:1 1 1 29.03 29.49
2:1 1 2 55.81 39.62
2:1 2 2 59.70 52.17
2:1 2 3 81.63 59.80
2:1 2 4 104.35 79.22
2:1 (but drive compression was
turned off) 2 4 66.67 79.12
4:1 2 4 118.98 79.55
2:1 3 3 79.65 69.10
2:1 3 6 90.45 51.43
2:1 4 4 94.99 72.00
2:1 4 8 89.55 57.05
Table 3: Test data for performance test of backups under Novell IntranetWare.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

7
(&+12/2*<
%
5,() (cont.) DLT Tape Array II

ECG075.0997
TEST 4L
OCAL BACKUP OF WINDOWS NT TO VARIOUS
NUMBERS OF DLT DRIVES
2SHUDWLQJ 6\VWHP
Windows NT 4.0
%DFNXS 6RIWZDUH
Cheyenne ARCserve 6.0 for Windows NT
%DFNXS 0HWKRG
Image
'DWD 6HW
2 GB
&RPSUHVVLRQ
2:1
7DSH )RUPDW
RAIT-0
+DUGZDUH
ProLiant 5000, two Pentium Pro 200 processors, 256K
cache, 256 MB RAM
40-GB array (one SMART-2/P Array Controller
with ten 4-GB drives)
35/70 GB DLT drive(s)
Wide-Ultra SCSI-3 cards
(All Compaq off-the-shelf products)
3XUSRVH
To determine the highest performance configuration for an image backup to DLT drive or
RAIT-0 group of drives (RAIT-0 is not significant when using one drive). The test consisted of
running a single job to back up a 2 GB file set to various numbers of DLT drives (from one to
eight).
5HVXOWV
Results show a limit of about 40 GB/hr while running a single job. Other tests with faster feed
sources show an upper limit of 57 GB/hr for a single job writing to eight tape drives. This limit
is based on the limit to the number of threads that a backup program can run under Windows
NT. Additional tests will be run to show that if multiple jobs are run simultaneously, there is
still available bandwidth from the disk drives, the CPU, the operating system, and the tape
drives to permit a much higher aggregate speed.
In this configuration, the restore speed exceeded the backup speed in some cases, apparently
because the SMART-2/P Array Controllers cache much of the data, making the restore data
flow more consistently to the target.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

7
(&+12/2*<
%
5,() (cont.) DLT Tape Array II

ECG075.0997
0.0
5.0
10.0
15.0
20.0
25.0
30.0
35.0
40.0
45.0
1 to 1 1 to 2 2 to 2 2 to 4 4 to 4 4 to 8
GB/hr
Restore Backup
Figure 9: Local backup and restore performance under Windows NT
with various proportions of SCSI cards and 35/70 GB DLT drives (shown at bases of columns).
&DUGV 'ULYHV %DFNXS *%KU 5HVWRUH *%KU
1 1 30.1 14.2
1 2 41.1 36.3
2 2 25.7 33.6
2 4 37.0 38.4
4 4 34.8 44.3
4 8 38.5 38.7
Table 4: Test data for local image backups to various numbers of DLT drives under Windows NT.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

7
(&+12/2*<
%
5,() (cont.) DLT Tape Array II

ECG075.0997
The following test results are included to demonstrate the performance increase achieved when
using even larger systems with faster feed speeds. These tests differ only in that a ProLiant
6000 was used, with three Pentium Pro processors and two additional SMART-2/P Controllers
with 56 drives.
0
20
40
60
80
100
GB/hr
2:1 Compression 4:1 Compression
1 to 1 1 to 2 2 to 2 2 to 4 4 to 4 3 to 6 4 to 6 4 to 8
Figure 10: Backup speed results on the ProLiant 6000 test under Windows NT,
with various proportions of SCSI cards and 35/70 GB DLT drives (shown at bases of columns).
-REV &DUGV 'ULYHV  'DWD 6HW
%DFNXS *%KU 
 'DWD 6HW
%DFNXS *%KU
1 1 1 29.88 32.17
1 1 2 52.70 55.17
1 2 2 58.92 67.38
1 2 4 76.54 86.50
1 4 4 78.93 84.50
1 3 6 84.96 86.23
1 4 6 84.51 85.02
1 4 8 83.56 82.48
Table 5: Test data for local image backups to various numbers of DLT drives
under Windows NT using the ProLiant 6000.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

7
(&+12/2*<
%
5,() (cont.) DLT Tape Array II

ECG075.0997
TEST 5U
SE OF CONCURRENT JOBS TO IMPROVE
PERFORMANCE WINDOWS NT 165 GB/HR
2SHUDWLQJ 6\VWHP
Windows NT 4.0
%DFNXS 6RIWZDUH
Cheyenne ARCserve 6.0 for Windows NT
%DFNXS 0HWKRG
Image multiple jobs
'DWD 6HW
2 GB
&RPSUHVVLRQ
2:1, 4:1
7DSH )RUPDW
RAIT-0
+DUGZDUH
ProLiant 6000, three Pentium Pro 200 processors, 512 K
cache, 384 MB RAM
224 GB array (four SMART-2/P Array Controllers
with fifty-six 4-GB drives)
35/70 GB DLT drive(s)
Wide-Ultra SCSI-3 cards
(All Compaq off-the-shelf products)
3XUSRVH
To determine whether throughput can be improved by performing concurrent backups on
multiple drives.
7HVW 6HWXS
To show what effect multiple tape drives would have, a large variety of situations were tested.
In all cases, the Compaq Array Configuration Utility was used to configure the Compaq disk
drive arrays. To join multiple disk array groups between adapter groupings, Windows NT Disk
Administrator was used to group these as one striped drive.
5HVXOWV
These tests demonstrated performance limits during image backups under Windows NT. The
bottleneck has to do with the way an image is processed. To get around the bottleneck,
concurrent backups of multiple drives were run. This technique works very well. It helps obtain
higher overall backup speeds, but its use depends on the drive/data structure. It is of no
advantage in an environment with a single large database, but it is of great value in many
applications server or file server backups.
The following graph demonstrates one, two, four and eight jobs running under Windows NT
4.0, backing up in each case to a set of eight tape drives. The graph shows the high performance
capability under Windows NT; however, the raw data below the chart shows that performance
with a four-drive array is still very impressive.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

7
(&+12/2*<
%
5,() (cont.) DLT Tape Array II

ECG075.0997
0
20
40
60
80
100
120
140
160
180
1 Job 2 Jobs 4 Jobs 8 Jobs
GB/hr
2:1 4:1
Data Compression
Figure 11: Concurrent backup performance running Cheyenne ARCserve 6.0
(image option) under Windows NT.
Jobs SCSI
Cards DLTs RAIT Level Smart Cards per Job Drives per Job 2:1 Backup (GB/Hr) 4:1 Backup (GB/Hr)
1 4 8 0 4 56 83.56 82.48
2 1 2 No RAIT 2 28 54.02 57.71
2 2 2 No RAIT 2 28 60.05 68.79
2 2 4 0 2 28 97.03 107.11
2 4 4 0 2 28 118.36 136.52
2 4 8 0 2 28 135.64 137.15
4 2 4 No RAIT 1 14 107.71 112.73
44 4NoRAIT 1 14 121.88 135.04
4 4 8 0 1 14 154.27 147.14
8 4 8 No RAIT 1 channel 7 165.26 164.18
Table 6: Test data for concurrent backup tests under Windows NT.
•Jobs: Number of Cheyenne backup jobs started concurrently under Windows NT.
•SCSI Cards: Number of Wide-Ultra SCSI-3 cards used to connect to the 35/70 GB DLT drives.
•DLT: Number of tape drives used.
•RAIT Level: When a job used more than one drive, the drives were configured as RAIT 0.
•SMART Cards – Number of SMART-2/P Array Controllers used per job (to help determine
loading factors).
•Drives per Job: Number of hard disk drives per job (to give perspective on feed source to each
tape job running).
•Note: In all testing, care was taken to balance the loading of the two PCI buses.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

7
(&+12/2*<
%
5,() (cont.) DLT Tape Array II

ECG075.0997
TEST 6C
OMPRESSIBILITY AND 35/70 GB DLT TO 15/30 GB
DLT PERFORMANCE
2SHUDWLQJ 6\VWHP
Windows NT 4.0
%DFNXS 6RIWZDUH
Cheyenne ARCserve 6.0 for Windows NT
%DFNXS 0HWKRG
Image
'DWD 6HW
2 GB
&RPSUHVVLRQ
1:1, 2:1, 4:1, 8:1
7DSH )RUPDW
RAIT-0
+DUGZDUH
ProLiant 5000, two Pentium Pro 200 processors, 256K
cache, 256 MB RAM
96-GB array (two SMART-2/P Array Controllers
with twenty-four 4-GB drives)
35/70 GB DLT drive(s)
Wide-Ultra SCSI-3 cards
(All Compaq off-the-shelf products)
3XUSRVH
To compare the performance of the new 35/70 GB DLT drive with that of the 15/30 GB DLT
drive. Though the new drive is faster and holds more, there are cases where the 15/30 GB DLT
drive may still be a better match for the backup application.
5HVXOW
The 35/70 GB DLT drive, with its larger cache and higher throughput rating, is three times
faster than the 15/30 GB DLT drive. In all compression categories, the 35/70 GB DLT drive is
the top in its class in performance. However, in some backup scenarios (feeding the enterprise
backup server with multiple 10MB Ethernet connections) the 15/30 GB DLT drive speed still
exceeds the feed speed of the network link. In these cases, the 15/30 GB DLT drive is an
excellent choice, and the use of the Compaq DLT Library is a very attractive and logical option.
The decision between the DLT Library and the DLT Tape Array II is based on backup
requirements and the environment. The DLT Tape Array II is fast, but a single (4 tape drive)
unit is only capable of storing 280 GB native (560 GB compressed) of tape online. Any backup
exceeding those amounts risks significant delays while waiting for an operator to change tapes.
The DLT Library has a 15-tape capacity. Eight single-drive DLT Library units (with 15/30 GB
DLT drives) would deliver a capacity of 1,800 GB native (3,600 GB compressed).
The DLT Library can have two drives per unit. The speed and capacity of this system makes it
an ideal solution for network backups. An example of a typical setup might be a central backup
server with four DLT Library units (each with two tape drives). This will support eight
simultaneous backup jobs from eight network-connected remote servers or systems. Using
groups of tapes, this system (which has been tested by Compaq with Cheyenne software) can
perform a weekly full backup of the remote systems, as well as a daily incremental backup.
This is a totally automated backup procedure, with no human intervention required. Of course,
sets of tapes can be rotated for archiving.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

7
(&+12/2*<
%
5,() (cont.) DLT Tape Array II

ECG075.0997
Data compression is an important factor in backup speeds, and in most cases, it is not
something the customer can control. In investigations into real customer data set composition, it
was found that most servers contain data with a compressibility between 2:1 and 4:1.
The backup speed for non-compressible data highlights the advantage of the data compression
hardware built into the drive: The 2:1 data backed up nearly twice as fast as data backed up
with the tape drive compression feature turned off. If data set compressibility is 2:1 or greater,
customers will achieve significantly faster backup speeds using the 35/70 GB DLT drive, with
its built-in compression. As the data set becomes more compressible, the advantage gained is
proportionally less: Some of the tests documented in this brief were run with 2:1 data
compression; others were run with 4:1 compression. The performance difference between these
compression levels is only about 14 percent.
0
5
10
15
20
25
30
35
40
1:1 2:1 4:1 8:1
GB/hr
15/30 15/30 Restore 35/70 35/70 Restore
Figure 12: Data compression effects at various levels of compression (indicated at bases of columns).
&RPSUHVVLRQ
5DWLR

%DFNXS
*%KU

5HVWRUH
*%KU

%DFNXS
*%KU

5HVWRUH
*%KU
1:1 3.90 3.93 15.35 15.39
2:1 7.51 6.97 29.50 26.46
4:1 9.68 9.35 33.32 23.07
8:1 9.99 9.80 36.13 28.64
Table 7: Test data for compressibility/performance tests.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.