Digital PDFs

EC-N0180-19

January 1992

48 pages

Original

94.7kB

Document:	Digital's DECsystem Family Performance Summary
Order Number:	EC-N0180-19
Revision:
Pages:	48
Original Filename:	https://web-docs.gsi.de/~kraemer/COLLECTION/DEC/decsystem_family_perf_summary_jan_92.pdf

OCR Text

Digital’s
DECsystem Family
Performance
Summary

January 1992
Order Number: EC-N0180-19

Digital Equipment Corporation
Maynard MA
TM

________________________________________________________________________________________________
First Printing, January 1992
The information in this document is subject to change without notice and should not be construed as a commitment by Digital Equipment Corporation.
Digital Equipment Corporation assumes no responsibility for any errors that may appear in this document.
Any software described in this document is furnished under a license and may be used or copied only in
accordance with the terms of such license. No responsibility is assumed for the use or reliability of software
or equipment that is not supplied by Digital Equipment Corporation or its affiliated companies.
Restricted Rights: Use, duplication, or disclosure by the U.S. Government is subject to restrictions as set
forth in subparagraph (c)(1)(ii) of the Rights in Technical Data and Computer Software clause at DFARS
252.227 7013.
Copyright ©1992 Digital Equipment Corporation
All right reserved.
Printed in U.S.A.
The following are trademarks of Digital Equipment Corporation: DEC, DECsystem, DECwrite , the DIGITAL Logo, and ULTRIX.
The following are third-party trademarks:
AIM Suite III Multiuser Benchmark is a trademark of AIM Technology.
AIX, IBM, and RISC System/6000 are trademarks of International Business Machines Corporation.
Ethernet is a trademark of Xerox Corporation.
HP and PA-RISC are trademarks of Hewlett-Packard Company.
NFS, SPARCserver, SunOS, and Sun are trademarks of Sun Microsystems, Incorporated.
Prestoserve is a registered trademark of Legato Systems, Inc.
SPEC, SPECratio, and SPECmark are trademarks of the Standard Performance Evaluation Corporation.
TPC Benchmark and TPC Benchmark B are trademarks of the Transaction Processing Performance Council.
UNIX is a registered trademark of UNIX System Laboratories, Inc.
This document was prepared using DECwrite, Version 1.1.

Contents
Part I Executive Summary ................................................................................................... 5
1

Introduction ......................................................................................................................... 6

Methodology ........................................................................................................................ 7

Summary of Relative Performance and System Attributes ................................................. 7

Part II Benchmark Results ............................................................................................... 11
AIM Performance Rating ..................................................................................................... 13
3.1

AIM Performance Ratings Results ................................................................................. 13

AIM Suite III Multiuser Benchmark ................................................................................. 14
3.2

AIM III Suite Results ....................................................................................................... 14

AIM Milestone Benchmark ................................................................................................. 16
3.3

AIM Milestone Results .................................................................................................... 16

Nhfsstone NFS File Server Benchmark ........................................................................ 18
3.4

Nhfsstone Results .......................................................................................................... 18

ULTRIX NFS Workload ......................................................................................................... 19
3.5

ULTRIX NFS Workload Results ..................................................................................... 19

SPEC SDM Release 1.0 Benchmark ............................................................................... 21
3.6

SPEC SDM Release 1.0 Results .................................................................................... 21

3.7

SPEC SDM V1.0 Release Test Results and Prestoserve .............................................. 22

SPEC Benchmark Suite ....................................................................................................... 24
3.8

SPEC Results ................................................................................................................. 24

TPC Benchmark B ................................................................................................................. 28
3.9

TPC-B Results ................................................................................................................ 28

Dhrystone Integer Benchmark ......................................................................................... 30
3.10

Dhrystone Results .......................................................................................................... 30

Linpack Benchmark .............................................................................................................. 32
iii

3.11

Linpack Results .............................................................................................................. 32

Whetstone Benchmark ........................................................................................................ 34
3.12

Whetstone Results ......................................................................................................... 34

DR Labs CPU 2 Benchmark ............................................................................................... 36
3.13

DR Labs CPU2 Results .................................................................................................. 36

Khornerstone Benchmarks ................................................................................................ 37
3.14

Khornerstone Results ..................................................................................................... 37

A Test Configurations ......................................................................................................... A-1
B References .......................................................................................................................... B-1
Figures
Figure 3-1:
Figure 3-2:
Figure 3-3:
Figure 3-4:
Figure 3-5:
Figure 3-6:
Figure 3-7:
Figure 3-8:
Figure 3-9:
Figure 3-10:
Figure 3-11:
Figure 3-12:
Figure 3-13:
Figure 3-14:
Figure 3-15:

AIM Performance Rating Results ......................................................................... 13
AIM Milestone MPH Performance Ratings ........................................................... 17
Average Server Response Time Using nhfsstone ............................................... 18
SPECmark Ratings .............................................................................................. 24
SPECint Ratings .................................................................................................. 25
SPECfp Ratings .................................................................................................. 25
TPC-B tpsB Benchmark Results .......................................................................... 29
TPC-B Price-per-TPS ($/tpsB) Benchmark Results ............................................. 29
Dhrystone MIPS Ratings ...................................................................................... 30
Dhrystones/second Ratings ................................................................................ 31
Linpack Single Precision MFLOPS Ratings ........................................................ 32
Linpack Double Precision MFLOPS Ratings ........................................................ 33
Whetstone Single Precision KWIPS Ratings ....................................................... 34
Whetstone Double Precision KWIPS Ratings ...................................................... 35
Khornerstone Results ........................................................................................... 37

Tables
Table 3-1:
Table 3-2:
Table 3-3:
Table 3-4:
Table 3-5:
Table 3-6:
Table 3-7:
Table 3-8:
Table 3-9:

Digital DECsystems’ and Competitive Systems’ Benchmark Results I .................. 8
Digital DECsystems’ and Competitive Systems’ Benchmark Results II ................. 9
Key to Graphs ...................................................................................................... 12
AIM Suite III Maximum User Load ....................................................................... 15
ULTRIX NFS Workload Recommended Optimum Number of Clients ................. 20
SPEC SDM Release 1.0 Benchmark Results ...................................................... 22
DECsystems’ SPECratios (SPEC Reference Time/Elapsed Time) ..................... 26
Competitors’ SPECratios (SPEC Reference Time/Elapsed Time) ....................... 27
DR Labs CPU2 MVUPs Ratings .......................................................................... 36

Part I
Executive Summary

Introduction
The DECsystem Family Performance Summary presents the results of system characterization tests performed on Digital’s newest UNIX-based RISC server systems:
•

DECsystem 5000 Model 25

•

DECsystem 5000 Model 133

The DECsystem 5000 Model 25 is a desktop, TURBOchannel-based system that
provides RISC power for PC database environments. The DECsystem 5000 Model
133 is Digital’s newest low cost-of-entry server system. It has a 33MHz CPU. Like
all members of the DECsystem family, the DECsystem 5000 Model 25 and Model
133 deliver optimum performance through balanced CPU, memory, and disk I/O.
Also included in this report are the performance results of the other current members
of the DECsystem product set: DECsystem 5100, DECsystem 5000 Model 200, and
DECsystem 5500.
The DECsystem 5100, DECsystem 5000 Models 133, and 200, and the DECsystem
5500 were specifically designed to meet such server requirements and activities as
NFS file serving, database serving, and multiuser serving. All DECsystems provide
excellent overall performance, particularly commercial performance, when compared to the competition.
AIM Performance Rating, AIM III Multiuser, AIM Milestone, nhfsstone, ULTRIX
NFS Workload, SPEC SDM, and TPC-B benchmarks measure complete system performance, including processor, I/O, and memory subsystems. SPEC, MIPS, and
MFLOPS are single-stream, compute-intensive benchmarks and basically measure
just one component of system performance: the processor. DECsystems were designed to provide excellent balanced system performance, and the DECsystem family of computers has superior system price performance ratings.
The DECsystem Family Performance Summary is a technical reference document
for Digital sales support personnel, customers, and other individuals who need to understand the performance characteristics of the DECsystem family. This document
provides information about the server and system-level performance of the
DECsystem family running standard industry benchmarks and workloads.

Digital’s DECsystem Family Performance Summary

The benchmarks and workloads results appear in the following order:
•

AIM Performance Ratings

•

AIM Suite III

•

AIM Milestone

•

Nhfsstone NFS File Server Benchmark

•

ULTRIX NFS Workload

•

SPEC SDM Release 1.0 Benchmark

•

SPEC Benchmarks

•

TPC Benchmark B

•

Dhrystone

•

Linpack

•

Whetstone

•

DR Labs CPU2

•

Khornerstones

The metrics for recording system performance vary according to the workload and
focus on useful work done by the system.
The tests discussed in this document were run in October through December, 1991.
Subsequent sections briefly describe the workloads and graphically depict the results. Test configurations and references appear in the document’s two appendices.

Methodology
Wherever possible, we executed the same benchmark code on all systems. However, due to equipment availability of competitive systems and time constraints,
some information is based on the most current, previously published data for those
particular systems and has not been independently verified. See Appendix A, titled
Test Configurations, for hardware configuration, operating system and compiler versions, and source of each benchmark result.

Summary of Relative Performance and System Attributes
Table 3-1 and Table 3-2 show relative application performance positioning only. It
is important to consider other factors when selecting a system to solve a specific
problem. Other factors include operating environment, expandability (such as expanded memory and storage), and price. Using this data, the most appropriate system can be selected based on system attributes, as well as performance characterization.

Digital’s DECsystem Family Performance Summary

Table 3-1: Digital DECsystems’ and Competitive Systems’ Benchmark Results I

Server

SPECmark SPECint

SPECfp

Dhrystone
MIPS

Dhrystone
per second

Linpack
Single
MFLOPS

Linpack
Double
MFLOPS

Whetstone
Single
KWIPS

Whetstone
Double
KWIPS

DR Labs
CPU2
Khorner
MVUPs stones

DS 5000/25¹

19.1

15.7

21.7

26.70

46927

6.60

2.80

25860

20882

27.71

32822

DS 5100¹

18.9

15.6

21.4

21.60

38060

5.38

3.01

20547

16700

22.74

29511

DS 5000/133¹

25.5

20.9

29.1

34.42

60475

8.79

5.93

33292

26724

37.10

39606

DS 5000/200¹

23.5

19.5

26.7

27.27

47920

6.81

3.73

25679

20899

28.61

27604

DS 5500¹

27.3

22.7

30.9

32.51

57117

8.23

4.25

30864

25189

33.87

82650

IBM RS/6000 320H²

41.2

20.0

66.8

37.10

65185

11.70

6.91

n/a³

n/a

IBM RS/6000 520²

32.6

15.8

52.9

29.50

51832

9.20

8.50

n/a

27.86

n/a

IBM RS/6000 530H²

57.4

26.6

95.9

37.10

65185

20.00

n/a

IBM RS/6000 550²

72.2

33.9

119.7

56.00

98392

25.20

23.00

n/a

IBM RS/6000 930²

43.4

20.1

72.4

37.10

65185

15.20

13.70

25000

27778

35.75

n/a

IBM RS/6000 950²

72.2

33.8

119.7

56.00

98392

25.20

23.00

30303

33333

42.80

n/a

SPARCserver ELC²

20.3

18.0

22.0

23.07

40540

3.61

2.20

23148

14663

17.29

23909

SPARCserver IPX²

24.4

21.7

26.5

26.68

46875

4.34

2.65

27778

19120

20.80

29111

SPARCserver 2²

25.0

21.7

27.4

28.50

50075

6.10

4.20

19920

14641

21.16

27142

SPARCserver 600MP²

26.8

22.5

30.2

n/a

Refer to Appendix A for test configurations
Table current as of January 1992

¹ SPEC results recorded using Kuck and Associates’ pre-compilers which will be available in Spring 1992
² SPEC results recorded using Kuck and Associates’ pre-compilers
³ n/a = not available

Table 3-2: Digital DECsystems’ and Competitive Systems’ Benchmark Results II
SPEC SDM
SDET
SPEC SDM
SDET Peak
Throughput
Scripts/Hour

SPEC SDM
Concurrent
Workloads
at Peak
Throughput

SPEC SDM
KENBUS1
KENBUS1
Peak
Throughput
Scripts/Hour

AIM
Performance
Rating

AIM
Max.
User
Load

AIM
Milestone
MPH

DS 5000/25

15.3

136

33.4

n/a

474.9

DS 5100

17.5

153

37.4

28.2

2345

131.2

965.4

120

DS 5000/133

16.6

146

40.9

n/a

795.8

108

DS 5000/200

14.0

131

53.2

n/a

133.7

1055.4

144

DS 5500

25.9

229

58.4

40.6

3944

195.0

1261.1

160

IBM RS/6000 320H

20.7

193

30.6

41.4

2600

138.9

n/a

IBM RS/6000 520

16.9

156

24.7

n/a

IBM RS/6000 530H

27.9

244

34.2

n/a

183.4

n/a

IBM RS/6000 550

34.5

307

48.8

69.2

3005

234.6

n/a

IBM RS/6000 930

20.8

195

31.7

n/a

IBM RS/6000 950

34.6

310

52.0

74.2

4521

n/a

HP 9000/817S

30.1

226

n/a

64.8

1940

n/a

SPARCserver ELC

11.3

25.7

n/a

SPARCserver IPX

14.4

117

30.4

n/a

SPARCserver 2

15.4

142

31.2

39.7

2600

140.4

935.9

136

SPARCserver 630MP

n/a

121.4

2081

n/a

SPARCserver 670MP

n/a

121.4

2222

n/a

Server

TPC-B TPC-B
tpsB
$/tpsB

Concurrent
Workloads
at Peak
Throughput

Part II
Benchmark Results

Results of individual benchmarks can be changed dramatically by the choice of operating system version, compiler version, level of optimization used, memory size,
configuration, cache size, process scheduling, buffer management, and so on. Because some systems have more than one compiler available from the vendor, using
different compilers can have a significant impact on benchmark performance. RISC
architectures are very dependent on compiler technology to keep pipeline operating
at peak performance.
When making comparisons between systems on the same benchmark, it is standard
practice to normalize all results to that of a certain system. In this report, we have in
some cases, normalized the results to the DECsystem 5000 Model 25. This system
is then defined to represent 1.0 on the scale, and all other systems fall higher or
lower than that system.
The following table explains the abbreviations used in the graphs and tables contained in this section.

Table 3-3: Key to Graphs

Abbreviation

Full Product Description

DS 5000/25
DS 5100
DS 5000/133
DS 5000/200
DS 5500
IBM RS/6000 320H
IBM RS/6000 520
IBM RS/6000 530H
IBM RS/6000 930
IBM RS/6000 950
HP 9000/817S
SPARCserver ELC
SPARCserver IPX
SPARCserver 2
SPARCserver 600MP
SPARCserver 630MP
SPARCserver 670MP

Digital DECsystem 5000 Model 25
Digital DECsystem 5100
Digital DECsystem 5000 Model 133
Digital DECsystem 5000 Model 200
Digital DECsystem 5500
IBM Risc System/6000 POWERserver 320H
IBM Risc System/6000 POWERserver 520
IBM Risc System/6000 POWERserver 530H
IBM Risc System 6000 POWERserver 930
IBM Risc System 6000 POWERserver 950
Hewlett-Packard 9000 Series 817S
Sun SPARCserver ELC
Sun SPARCserver IPX
Sun SPARCserver 2
Sun SPARCserver 600MP (single processor)
Sun SPARCserver 630MP (multiprocessor)
Sun SPARCserver 670MP (multiprocessor)

Digital’s DECsystem Family Performance Summary

AIM Performance Rating
According to AIM Technology, Santa Clara, CA., "The AIM Performance Rating reflects the peak performance of the system measured in AIM Multiuser
Performance Units. A VAX 11/780 typically rates 1 AIM. Performance Ratings of a wide range of UNIX systems can be compared using available AIM
Performance Reports."
3.1

AIM Performance Ratings Results
The AIM Performance Ratings appear in the following chart.
Note: AIM Performance Ratings were not available for the SPARCserver
630MP and SPARCserver 670MP.

Figure 3-1:

AIM Performance Rating Results

DS 5000/25

15.3

DS 5100

17.5

DS 5000/133

16.6

DS 5000/200

14.0

DS 5500 (SCSI)

25.9

DS 5500 (DSSI)

26.1
20.7

IBM RS/6000 320H
IBM RS/6000 520

16.9

IBM RS/6000 530H

27.9

IBM RS 6000/550

34.5
20.8

IBM RS/6000 930
IBM RS/6000 950

34.6

HP 9000/817S

30.1

SPARCserver ELC

11.3

SPARCserver IPX

14.4

SPARCserver 2

15.4
0.0

10.0

20.0

30.0

40.0

50.0

60.0

70.0

AIM Performance Ratings

Digital’s DECsystem Family Performance Summary

AIM Suite III Multiuser Benchmark
The AIM Suite III, a product of AIM Technology of Santa Clara, CA., is a synthetic, multiuser benchmark for UNIX systems. It measures multiuser system
performance of multiple systems under a variety of application mixes and resource loads, and assesses the multiuser performance of a single system under
varying use conditions. Thirty-three function tests are used to evaluate system
performances under user-defined application mixes.
AIM Suite III contains no application level software. Each simulated user runs
a combination of subsystem tests. The load that all simulated users put on the
system is said to be characteristic of a UNIX timesharing environment.
AIM Technology uses "User Load Rating", the number of emulated users running when the throughput metric reaches 1.0 jobs/minute/user.
3.2

AIM III Suite Results
The AIM Multiuser Benchmark does not reflect NFS or X-Windows performance. DECsystem’s outstanding X-Windows and NFS performance give
DECsystems an advantage over comparable, competitive systems in networked
and client/server multi-tasking environments.
Note: AIM Suite III results were not available for the Sun SPARCserver
630MP and SPARCserver 670MP.

Digital’s DECsystem Family Performance Summary

Table 3-4: AIM Suite III Maximum User Load

System

Disk
Controller

Prestoserve
Enabled

User
Rating

Relative to
DECsystem
5000 Model 25

DECsystem 5000 Model 25
DECsystem 5100
DECsystem 5000 Model 133
DECsystem 5000 Model 200
DECsystem 5500
DECsystem 5500
IBM RS/6000 320H
IBM RS/6000 520
IBM RS/6000 530H
IBM RS/6000 550
IBM RS/6000 930
IBM RS/6000 950
HP 9000/817S
SPARCserver ELC
SPARCserver IPX
SPARCserver 2

SCSI
SCSI
SCSI
SCSI
DSSI
SCSI
SCSI
SCSI
SCSI
SCSI
SCSI
SCSI
SCSI
SCSI
SCSI
SCSI

N/A
No
N/A
N/A¹
Yes
Yes
N/A
N/A
N/A
N/A
N/A
N/A
N/A
N/A
N/A
N/A

136
153
146
131
235
229
193
156
244
307
195
310
226
69
117
142

1.00
1.13
1.07
.96
1.73
1.68
1.42
1.15
1.79
2.26
1.43
2.28
1.66
.51
.86
1.04

¹ N/A = Not Applicable

Digital’s DECsystem Family Performance Summary

AIM Milestone Benchmark
The AIM Milestone benchmark evaluates the performance of a system by simulating a pre-defined collection of user types. The seven user types are:
•

Administrative Assistant

•

Spreadsheet User

•

Database User

•

Manager

•

Scientist

•

Software Engineer

•

Technical Writer

Equal weight is assigned to each of the user types. Because each user performs
different tasks repetitively, excessive CPU instruction and data caching caused
by locality of reference is avoided. Milestone was designed to functionally represent real user loads.
The results of running AIM Milestone Benchmark are rolled up into one number called MPH (Milestone user loads the system under test completes Per
Hour). The MPH rating is intended to provide a relative measure of the speed
of one vendor platform to another from the user’s perspective.
3.3

AIM Milestone Results
The AIM Milestone MPH ratings for the DECsystem family of systems and
competitive servers appear on the next page.
Note: AIM MPH ratings for the HP 9000/817S, the SPARCserver 630MP, and
the SPARCserver 670MP were unavailable.

Digital’s DECsystem Family Performance Summary

Figure 3-2:

AIM Milestone MPH Performance Ratings

DS 5000/25

33.4

DS 5100

37.4

DS 5000/133

40.9

DS 5000/200

53.2

DS 5500 (DSSI)

54.3

DS 5500 (SCSI)

58.4

IBM RS/6000 320H

30.6

IBM RS/6000 520

24.7

IBM RS/6000 530H

34.2

IBM RS 6000/550

48.8

IBM RS/6000 930

31.7

IBM RS/6000 950

52.0

SPARCserver ELC

25.7

SPARCserver IPX

30.4

SPARCserver 2

31.2
0.0

10.0

20.0

30.0

Digital’s DECsystem Family Performance Summary

40.0

50.0

60.0

70.0

Nhfsstone NFS File Server Benchmark
The nhfsstone V1.22 workload, distributed by Legato Systems, Inc., facilitates
the analysis of NFS server performance from the NFS client’s perspective. It
emulates traffic patterns of typical networked file system environments and represents the kind of loads seen during very heavy usage.
Nhfsstone executes on the NFS client. Nhfsstone creates subprocesses which
generate file operations on the specified remotely-mounted file systems. The
nhfsstone parent process waits for the subprocesses to finish and then reports
the results.
The metric used in nhfsstone is average response time (in milliseconds) for NFS
transfers as the load on the server is increased.
3.4

Nhfsstone Results
Below are the results for the DECsystem 5100 and DECsystem 5500 running
the nhfsstone utility. Each used a DECsystem 5500 configured with 16 MB of
memory, 10% buffer cache, cache_bufcache=1, and 4 block I/O daemons
(biods), as the client system.
The graph shows average server response time (in milliseconds) using the default nhfsstone mix over a range of applied load rates (in NFS operations/second). For each sample point, load was applied for 120 seconds.

Figure 3-3:

Average Server Response Time Using nhfsstone

Response
Time
(Msec.)

DS 5100/32MB/2RZ57
DS 5500/32MB/2RZ57

35
30
25
20
15
10
5
0

100
150
200
NFS Operations per Second

250

300

Digital’s DECsystem Family Performance Summary

ULTRIX NFS Workload
Digital’s ULTRIX NFS Workload emulates a UNIX software engineering environment in which one software engineer is assigned to one client workstation.
Each workstation executes a different sequence of ULTRIX commands, with
each command sequence being probabilistically generated from a single pool of
the 30 ULTRIX commands that are most frequently used in a software engineering environment. The overall performance of the DECsystem configured
as the NFS file server in this ULTRIX NFS environment is measured. The
server is then characterized in terms of
•

Client service time re-testing throughput trends as the load increases on the
server

•

Server resource consumption trends as the load increases on the server

•

and resource consumption trends as the load increases on the server

The metric used in this report for the ULTRIX NFS Workload is optimum number of supported clients.
3.5

ULTRIX NFS Workload Results
Table 3-5 shows the recommended optimum number of clients for the
DECsystem 5100, DECsystem 5000 Model 200, and DECsystem 5500 when
configured as an ULTRIX NFS server. Each number is based on the throughput/response ratio when applying the ULTRIX NFS Workload to progressive
numbers of diskless client workstations. A list of the benefits gained by enabling Prestoserve appears after the table.
Note: Different workloads and applications will most likely yield different results.

Digital’s DECsystem Family Performance Summary

Table 3-5: ULTRIX NFS Workload Recommended Optimum Number of
Clients

Configuration

Memory

DECsystem 5100¹
48 MB
DECsystem 5100²
48 MB
DECsystem 5000/200 24 MB
DECsystem 5500³
128 MB

Disks

Operating
System &
Version

Buffer Cache
Percent of Total
Physical Memory

4 RZ57s
4 RZ57s
6 RZ56s
7 RZ57s

ULTRIX T4.1 4.8 MB or 10%
ULTRIX T4.1 4.8 MB or 10%
ULTRIX T4.0 2.4 MB or 10%
ULTRIX V4.1 12.8 MB or 10%

Optimum
Number of Clients
Supported

36
36
36
>70

¹ Prestoserve enabled
² Prestoserve disabled
³ Prestoserve enabled

Prestoserve-related findings from the DECsystem 5100 ULTRIX NFS Workload
characterization study were

•

As a NFS server, the DECsystem 5100 with Prestoserve enabled, improved response time of the I/O intensive elements of the workload. While enhancing the
client performance, the Prestoserve option does not negatively impact the server
CPU utilization.

•

The response/throughput peak value for DECsystem 5100 with Prestoserve enabled was greater than that for the DECsystem 5100 with Prestoserve disabled.
The server performance for the 36 diskless clients on the DECsystem 5100 with
Prestoserve enabled was better than the server performance for the 36 diskless
clients on the same system with Prestoserve disabled.

•

The DECsystem 5100, with Prestoserve enabled, has a high potential for network throughput.

Digital’s DECsystem Family Performance Summary

SPEC SDM Release 1.0 Benchmark
In May 1991, Standard Performance Evaluation Corporation (SPEC) announced
the availability of a new benchmark suite called SPEC System Development
Multi-Tasking (SDM) Release 1.0. SDM measures overall system performance
using two workloads derived from two different application environments. It is
designed to measure the capacity of a system in a multi-tasking UNIX environment and consists of two multi-tasking system level benchmarks named SDET
and KENBUS1. The SDET benchmark uses UNIX commands that are found in
a C-based software development environment; KENBUS1 benchmark uses
commands and procedures that are found in a UNIX/C research and development environment. Both benchmarks are designed to exercise a variety of system resources.
SDM measures performance by gradually increasing the workload on the system. The system throughput increases until a constraint is reached in some system component (e.g., CPU, memory, I/O). SDM Release 1.0 Benchmark is an
accurate predictor for multi-tasking environments where multiple processes are
running simultaneously, and increased amounts of operation system services are
needed.
The throughput metric is defined as the total amount of work done in a given
time. The metric for both SDET and KENBUS1 is called scripts/hour. A
throughput curve is plotted as scripts/hour against number of scripts that generated the scripts/hour throughput. The highest scripts/hour measured is called
the peak value and is reported as the peak system throughput.
Note: The SDET and KENBUS1 benchmarks are unrelated and results should
not be compared.
3.6

SPEC SDM Release 1.0 Results
SPEC SDM Release 1.0 results, for the vendors who have submitted to SPEC,
are shown in Table 3-6. A list of performance gains gotten by enabling
Prestoserve on the DECsystem 5100 running SPEC SDM appear after the table.
Note: SPEC SDM SDET and KENBUS1 results were unavailable for the IBM
RS/6000 950; the HP 9000/817S; and the SPARCserver Models ELC, IPX,
630MP, and 670MP. SPEC SDM SDET results were not available for DS
5000/25 and DS 5000/133. SPEC SDM KENBUS1 results were unavailable
for the IBM RS/6000 Models 320H and 530H.

Digital’s DECsystem Family Performance Summary

Table 3-6: SPEC SDM Release 1.0 Benchmark Results

System

SDET Peak
Throughput
Scripts/Hour

DS 5000/25

SDET
Concurrent
Workloads
at Peak
Throughput

KENBUS1
Peak
Throughput
Scripts/Hour

KENBUS1
Concurrent
Workloads
at Peak
Throughput

n/a

474.9

131.2

965.4

120

DS 5000/133

n/a

795.8

108

DS 5000/200

133.7

1055.4

144

DS 5500²

195.0

1261.1

160

IBM RS/6000 320H 138.9

n/a

IBM RS/6000 530H 183.4

n/a

IBM RS/6000 550

234.6

n/a

SPARCserver 2

140.4

935.91

136

DS 5100

¹ Prestoserve enabled
² Prestoserve enabled

3.7

SPEC SDM V1.0 Release Test Results and Prestoserve
As stated earlier in this section, SPEC SDM Release 1.0 Benchmark measures
overall system performance using two different multi-tasking workloads simulating
two different software development computing environments. Basic software development activities apply to such user environments as software engineering, applications development done to support particular job functions, and to students
learning programming. The DECsystem 5100 SPEC SDM characterizations were
performed with and without the Prestoserve option enabled. Our analysis of the
data produced the following Prestoserve-related findings:
•

Examination of the results of the two SPEC SDM benchmarks performed on
the DECsystem 5100 showed that the performance improvement achieved
solely by enabling Prestoserve varied from 20% for KENBUS1 to 120% for
SDET.

•

Enabling Prestoserve on the DECsystem 5100 demonstrated that it increases
system performance in a software development environment.

Digital’s DECsystem Family Performance Summary

•

For environments that are disk file manipulation intensive and have CPU processing available to increase disk I/O processing, enabling Prestoserve is recommended. Prestoserve uses CPU processing to improve disk I/O performance.

•

In general, as the disk file manipulation load increases, the performance improvement realized by enabling Prestoserve will also increase.

Digital’s DECsystem Family Performance Summary

SPEC Benchmark Suite
This section presents the results of the Standard Performance Evaluation Corporation (SPEC) Benchmark Suite. SPEC is a nonprofit organization formed to
develop a standard suite of benchmark programs that characterize system performance. Digital is a member of SPEC and endorses its goals.
The release suite consists of ten codes/programs, four of which are written in C
and considered to be compute intensive. The geometric mean of these make up
the SPEC metric called SPECint and are classified as integer benchmarks. The
remaining six programs are FORTRAN based and floating point intensive.
They make up the SPECfp rating. The SPECmark is the geometric mean of the
ten programs’ elapsed times normalized to the VAX 11/780.
3.8

SPEC Results
The SPECmark recorded for the DECsystem 5000/25 was 19.1 and the
DECsystem 5000 Model 133’s rating was 25.5. Figure 3-4 compares these ratings with competitive systems.
Notes: DECsystems’ SPEC results were recorded using Kuck and Associates’
pre-compilers which will be available in Spring 1992. IBM, HP, and
SPARCserver SPEC results were recorded using Kuck and Associates’ precompilers.

Figure 3-4:

SPECmark Ratings

DS 5000/25

19.1

DS 5100

18.9

DS 5000/133

25.5
23.5

DS 5000/200
DS 5500

27.3

IBM RS/6000 320H

41.2

IBM RS/6000 520

32.6
57.4

IBM RS/6000 530H
IBM RS/6000 550

72.2

IBM RS/6000 930

43.4

IBM RS/6000 950

72.2

SPARCserver ELC

20.3

SPARCserver IPX

24.4

SPARCserver 2

25.0

SPARCserver 600MP

26.8
0.0

10.0

20.0

30.0

40.0

50.0

60.0

70.0

80.0

SPECmark

Digital’s DECsystem Family Performance Summary

Figure 3-5:

SPECint Ratings
15.7
15.6

DS 5000/25
DS 5100
DS 5000/133
DS 5000/200
DS 5500
IBM RS/6000 320H
IBM RS/6000 520
IBM RS/6000 530H
IBM RS/6000 550
IBM RS/6000 930
IBM RS/6000 950
SPARCserver ELC
SPARCserver IPX
SPARCserver 2
SPARCserver 600MP

0.0

Figure 3-6:

20.9
19.5
22.7
20.0
15.8
26.6
33.9
20.1
33.8
18.0
21.7
21.7
22.5
5.0

10.0

15.0

20.0

30.0

25.0

SPECint

35.0

40.0

SPECfp Ratings

DS 5000/25

21.7

DS 5100

21.4

DS 5000/133

29.1
26.7

DS 5000/200
DS 5500

30.9

IBM RS/6000 320H

66.8

IBM RS/6000 520

52.9
95.9

IBM RS/6000 530H

119.7

IBM RS/6000 550
IBM RS/6000 930

72.4

IBM RS/6000 950

119.7

SPARCserver ELC

22.0

SPARCserver IPX

26.5

SPARCserver 2

27.4

SPARCserver 600MP

30.2

0.0

20.0

40.0

60.0

80.0

100.0

120.0

140.0

SPECfp

Digital’s DECsystem Family Performance Summary

The SPECratio for a benchmark is the quotient derived from dividing the SPEC Reference Time by a particular machine’s corresponding run time. For Release 1.2, the
SPEC Reference Time is the time (in seconds) that it takes a VAX 11/780 machine
to run each particular benchmark in the suite.
Note: DECsystem SPECratios were recorded using Kuck and Associates’ precompilers which will be available in Spring 1992. Competitors’ SPECratios were
recorded using Kuck and Associates’ pre-compilers.

Table 3-7: DECsystems’ SPECratios (SPEC Reference Time/Elapsed
Time)

Benchmark
No. & Name

001.gcc
008.espresso
013.spice 2g6
015.doduc
020.nasa7
022.li
023.eqntott
030.matrix300
042.fpppp
047.tomcatv

Type

DS
5000/
25

DS
5100

DS
5000/
133

DS
5000/
200

DS
5500

Integer
Integer
Floating point
Floating point
Floating point
Integer
Integer
Floating point
Floating point
Floating point

12.2
17.4
11.4
18.7
19.9
17.5
16.5
57.4
22.1
19.5

14.9
14.7
11.5
16.5
21.8
18.1
15.1
53.9
20.1
21.2

16.8
22.7
14.8
25.2
25.4
22.4
22.3
82.7
28.4
27.1

18.1
18.3
14.3
20.8
27.7
23.1
18.7
66.5
25.1
26.4

21.0
22.0
17.0
24.3
30.7
25.7
22.5
78.4
29.3
29.9

Digital’s DECsystem Family Performance Summary

Table 3-8: Competitors’ SPECratios (SPEC Reference Time/Elapsed Time)
Benchmark
No. & Name

Type

IBM
6000
320H

001.gcc
Integer
16.9
008.espresso Integer
19.9
013.spice 2g6 Floating point 26.0
015.doduc
Floating point 28.8
020.nasa7
Floating point 72.6
022.li
Integer
20.0
023.eqntott
Integer
23.6
030.matrix300 Floating point 404.0
042.fpppp
Floating point 53.0
047.tomcatv Floating point 76.6

IBM IBM IBM
6000 6000 6000
520 530H 550

IBM IBM SPARC SPARC SPARC
6000 6000 server server server
930 950 ELC
IPX
2

13.3 22.1 28.9 17.0 28.6
15.8 26.9 33.8 20.2 33.9
20.5 37.5 47.2 28.8 47.2
22.9 39.0 49.4 29.6 49.6
57.5 126.8 144.4 88.1 144.4
15.8 26.9 33.8 20.2 33.8
18.7 31.4 39.9 23.4 40.0
320.9 572.8 729.8 435.1 729.8
41.7 66.6 86.8 53.3 86.6
60.8 109.9 119.7 83.0 138.0

Digital’s DECsystem Family Performance Summary

16.6
18.0
13.4
14.0
23.6
19.0
18.4
67.7
19.1
19.7

19.9
21.7
16.1
16.6
28.4
23.0
22.2
81.5
23.0
23.8

20.0
21.7
16.5
18.2
29.1
23.1
22.3
82.6
23.8
24.9

TPC Benchmark B
Formed in 1988, the Transaction Processing Performance Council (TPC) charter is to define standard benchmark specifications for commercial transaction
processing systems. To date, the TPC has defined two different types of TPC
benchmarks: TPC Benchmark A (TPC-A), established in 1989; and TPC
Benchmark B (TPC-B), established in August 1990.
TPC-B environments are characterized by:
•

Significant disk input/output

•

Moderate system and application execution time

•

Transaction integrity

This benchmark is not on-line transaction processing (OLTP) in that it does not
require any terminals, networking, or think time. It is similar in many respects
to a "batch" test focusing on the database throughput. TPC-B is most likely to
be performed by database vendors.
The metrics used in TPC-B are throughput as measured in transactions per second (tpsB), subject to response time constraint, and the associated price-perTPS ($/tpsB). The only benchmark results comparable to TPC-B are other
TPC-B results. In spite of similarities to TPC-A, TPC-B contains substantial
differences which make the TPC-B results not comparable to TPC-A. Also, as
benchmark results are dependent upon workload, specific application requirements, and system design and implementation, comparative system performance will vary as a result. TPC-B should not be used as a substitute for specific
and/or critical capacity customer application benchmarking.
The TPC Benchmark B Standard requires that test sponsors publish a full disclosure report of the implementation details along with the published test results.
3.9

TPC-B Results
TPC-B transactions per second (tpsB) and TPC-B price-per-TPS ($/tpsB) results appear in the following charts. All of the TPC-B tests used INFORMIXOnLine database software except for the SPARCserver 630MP and 670MP
which used Sysbase SQL Server. SPARCserver 2 TPC-B tests using Informix
or Sybase are shown.
The DECsystem 5100 and DECsystem 5500 TPC-B tpsB and $/tpsB results are
shown in the following charts.
Note: TPC-B results were unavailable for the DECsystem 5000 Models 25,
133, and 200; the IBM RS/6000 Models 520 and 530H; and the SPARCserver
Models ELC and IPX.

Digital’s DECsystem Family Performance Summary

Figure 3-7:

TPC-B tpsB Benchmark Results

tpsB
140.0
121.4

120.0

121.4

100.0
80.0

74.2
69.2

64.8

60.0
51.9

40.6

40.0

41.4

39.7

28.2

20.0
0.0
DECsystem 5100
IBM RS/6000 320H
IBM RS/6000 950
SPARCserver 2
SPARCserver 630MP
DECsystem 5500
IBM RS/6000 550
HP 9000 817S
SPARCserver 2
SPARCserver 670MP

RDBMS:

Informix

Figure 3-8:

Informix

Sybase

TPC-B Price-per-TPS ($/tpsB) Benchmark Results

US $/tpsB

5000
$4,521
$3,944

4000

$3,700

$3,005

3000

$2,600

$2,600
$2,345

$2,081

$1,940

2000

$2,222

1000

0
DECsystem 5100
IBM RS/6000 320H
IBM RS/6000 950
SPARCserver 2
SPARCserver 630MP
DECsystem 5500
IBM RS/6000 550
HP 9000 817S
SPARCserver 2
SPARCserver 670MP
RDBMS:

Informix

Digital’s DECsystem Family Performance Summary

Informix

Sybase

Dhrystone Integer Benchmark
The Dhrystone benchmark was introduced in 1984 as an ADA program by
Reinhold P. Weicker. It has since been translated into C and TURBO
PASCAL.
This synthetic benchmark1 measures processor and compiler efficiency. Its emphasis is on the type of data and operations encountered in a system rather than
numerical programming. Dhrystones is CPU-intensive.

Dhrystones are most commonly expressed in Integer MIPS (Millions of Instructions Per Second) where 1 MIP is the number Dhrystones per second that can be
performed by a VAX 11/780 (1757 Dhrystones/second).
3.10

Dhrystone Results
The following graph shows the MIPS ratings for the DECsystem family and
competitive servers.
Notes: Dhrystone results were unavailable for the HP 9000/817S and the
SPARCserver 600MP series.

Figure 3-9:

Dhrystone MIPS Ratings
26.70

DS 5000/25
21.60

DS 5100

34.42

DS 5000/133
27.27

DS 5000/200
DS 5500

32.51
37.10

IBM RS/6000 320H
29.50

IBM RS/6000 520
IBM RS/6000 530H

37.10
56.00

IBM RS/6000 550
IBM RS/6000 390

37.10
56.00

IBM RS/6000 950
23.07

SPARCserver ELC

26.68

SPARCserver IPX

28.50

SPARCserver 2
0.00

10.00

20.00

30.00

40.00

50.00

60.00

70.00

MIPS

1 Synthetic benchmarks are mostly loops which time different system calls and are not based on actual applications.

Digital’s DECsystem Family Performance Summary

Dhrystones per second are charted below.

Figure 3-10:

Dhrystones/second Ratings
46927

DS 5000/25
38060

DS 5100

60475

DS 5000/133
47920

DS 5000/200
DS 5500

57117
65185

IBM RS/6000 320H
51832

IBM RS/6000 520
IBM RS/6000 530H

65185
98392

IBM RS/6000 550
IBM RS/6000 930

65185
98392

IBM RS/6000 950
40540

SPARCserver ELC

SPARCserver IPX

46875
50075

SPARCserver 2

20000

40000

60000

80000

100000

120000

Dhrystones/second

Digital’s DECsystem Family Performance Summary

Linpack Benchmark
Developed at Argonne National Laboratories, Linpack is a FORTRAN benchmark that solves a 100x100 system of linear equations. The benchmark is used
to compare the performance of mathematical and scientific applications where
floating point computations are prevalent. When running, the benchmark gives
little weight to I/O.
The results are measured in millions of floating point operations per second
(MFLOPS). Both single and double precision operations are reported here.
3.11

Linpack Results
Single precision MFLOPS ratings are shown in the following graph.
Note: Single Precision MFLOPS ratings were unavailable for HP 9000 Model
817S and SPARCserver 600MP Series.

Figure 3-11:

Linpack Single Precision MFLOPS Ratings
6.60

DS 5000/25

5.38

DS 5100
DS 5000/133

8.79

6.81

DS 5000/200
DS 5500

8.23
11.70

IBM RS/6000 320H

9.20

IBM RS/6000 520
IBM RS/6000 530H

20.00
25.20

IBM RS/6000 550
IBM RS/6000 930

15.20
25.20

IBM RS/6000 950
3.61

SPARCserver ELC
SPARCserver IPX

4.34

6.10

SPARCserver 2
0.00

5.00

10.00

15.00

20.00

25.00

30.00

Single Precision MFLOPS

Digital’s DECsystem Family Performance Summary

The following chart contains the double precision MFLOPS results.
Note: Double precision MFLOPS for the IBM RS/6000 530H, HP 9000/817S, and
the SPARCserver 600MP Series were unavailable.

Figure 3-12:

Linpack Double Precision MFLOPS Ratings

DS 5000/25

2.80

DS 5100

3.01

DS 5000/133

5.93
3.73

DS 5000/200

4.25

DS 5500
IBM RS/6000 320H

6.91
8.50

IBM RS/6000 520

23.00

IBM RS/6000 550
13.70

IBM RS/6000 930

23.00

IBM RS/6000 950
SPARCserver ELC

2.20
2.65

SPARCserver IPX

4.20

SPARCserver 2
0.00

5.00

10.00

15.00

20.00

25.00

Double Precision MFLOPS

Digital’s DECsystem Family Performance Summary

Whetstone Benchmark
The Whetstone benchmark was developed at Great Britain’s National Physical
Laboratory in Whetstone, England in 1970. It is a synthetic benchmark designed to represent small engineering/scientific programs.
The Whetstone benchmark has been implemented in single precision and double precision FORTRAN programs, each arranged to defeat most compiler
optimizations. The results are measured in KWIPS (thousands of Whetstone Instructions Per Second).
3.12

Whetstone Results
Charted below are the Whetstone single and double precision benchmark tests
results.
Note: Single and Double Precision KWIPS were unavailable for the
IBM RS/6000 Models 320H, 520, 530H, and 550; the Hewlett-Packard 9000
Models 817S; and the SPARCserver 600MP Series.

Figure 3-13:

Whetstone Single Precision KWIPS Ratings
25860

DS 5000/25
20547

DS 5100

33292

DS 5000/133
25679

DS 5000/200
DS 5500

30864

IBM RS/6000 930

25000

IBM RS/6000 950

30303

23148

SPARCserver ELC

27778

SPARCserver IPX
19920

SPARCserver 2

10000

20000

30000

40000

50000

Single Precision KWIPS

Digital’s DECsystem Family Performance Summary

Whetstone double precision KWIPS ratings appear in the following chart.

Figure 3-14:

Whetstone Double Precision KWIPS Ratings
20882

DS 5000/25
16700

DS 5100

26724

DS 5000/133
20899

DS 5000/200

25189

DS 5500
IBM RS/6000 930

27778

IBM RS/6000 950

33333
14663

SPARCserver ELC

19120

SPARCserver IPX
14641

SPARCserver 2

5000

10000

15000

20000

25000

30000

35000

40000

Double Precision KWIPS

Digital’s DECsystem Family Performance Summary

DR Labs CPU 2 Benchmark
The DR Labs CPU2 benchmark from Digital Review Magazine is a floatingpoint intensive series of FORTRAN programs that include thirty-four separate
tests. The benchmark is most relevant in predicting the performance of engineering/scientific applications.
The geometric mean of the CPU times for the 34 kernels tested is compared.
Performance is expressed as a multiple of MicroVAX II Units of Performance
(MVUPs).
3.13

DR Labs CPU2 Results
The DR Labs CPU2 results appear in the following table. All values have also
been normalized to the DECsystem 5000 Model 25.
Note: MVUPs for the IBM RS/6000 320H, IBM RS/6000 530H, IBM RS/6000
550, HP 9000/817S, and SPARCserver 600MP Series were unavailable.

Table 3-9: DR Labs CPU2 MVUPs Ratings

System

DECsystem 5000 Model 25
DECsystem 5100
DECsystem 5000 Model 133
DECsystem 5000 Model 200
DECsystem 5500
IBM RS/6000 520
IBM RS/6000 930
IBM RS/6000 950
SPARCserver ELC
SPARCserver IPX
SPARCserver 2

MVUPs
Rating

27.71
22.74
37.10
28.61
33.87
27.86
35.75
42.80
17.29
20.80
21.16

Relative to
DECsystem
5000 Model 25

1.00
.82
1.34
1.03
1.22
1.01
1.29
1.54
.62
.75
.76

Digital’s DECsystem Family Performance Summary

Khornerstone Benchmarks
The Khornerstone and Khornerstone2 Benchmarks were developed by Workstation
Laboratories, Irving, TX. Consisting of 21 separate tests, the benchmarks rate the
overall performance of a system. The tests include both public domain (Whetstone,
Dhrystone, Sieve, etc.) and proprietary routines. The purpose of these tests is to
measure single-user loads on a system and to provide one number representing that
load condition.
The Khornerstone and Khornerstone2 benchmarks differ in that the Khornerstone2
test runs 10 times as many repetitions of each test.
3.14

Khornerstone Results
Khornerstone and Khornerstone 2 results are shown in the following graph.
Note: Khornerstone results were unavailable for the IBM RS/6000 Models 320H,
520, 530H, 550, 930, and 950. HP 9000/817S and SPARCserver 600MP Series
Khornerstone results were also unavailable.

Figure 3-15:

Khornerstone Results

DS 5000/25

32822
29511

DS 5100

39606

DS 5000/133
DS 5000/200

27604
82650

DS 5500
SPARCserver ELC

23909
29111

SPARCserver IPX

27142

SPARCserver 2

20000

40000

60000

80000

100000

Double Precision KWIPS

Digital’s DECsystem Family Performance Summary

A
Test Configurations

Digital’s DECsystem Family Performance Summary

A-1

Test Configurations

The benchmarks were run on systems with the following configurations:
Personal DECsystem 5000 Model 25 Server/Multiuser Computer:
CPU Chipset
CPU MHz
FPU Chipset
FPU MHz
Memory
Disk
Cache Size
Network Interface
Operating System
Compilers
File System
Tuning Parameters

R3000A
25 MHz
R3010A
25 MHz
16 MB
SCSI 1 426 MB RZ25, 1 665 MB RZ56
64KB data/64KB instruction
Ethernet
ULTRIX X4.2A-4 (Rev. 20)
DEC Fortran T3.1 (ft2), DEC C V1.0
Berkeley FFS
10% bufcache, cache_bufcache=1, delay_wbuffers=1

SPEC benchmarks were recorded using Kuck and Associates pre-compilers will be available in the Spring
1992.
SPEC KENBUS1 benchmark ran with 40 MB of memory; 3 426MB RZ25s; MIPS C V2.10; system state was
multi-user; system tuning parameters were MAXUSER=1000, MAXUPRC=1000, BUFCACHE=50,
cache_bufcache=1, delay_wbuffers=1; process tuning parameters none; and background load was standard
multi-user.
Khornerstone2 tests were run with 10% buffer cache.
All tests were performed by Digital.
DECsystem 5100 Server/Multiuser Computer:
CPU Chipset
CPU MHz
FPU Chipset
FPU MHz
Memory
Disk Controller
Disk
Cache Size
Network Interface
Operating System
Compilers
File System
Tuning Parameters
Background Load
System State

R3000
20
R3010
20
48 MB
SCSI
1.0 GB RZ57
64 KB data/64 KB instruction
Ethernet
ULTRIX T4.2 (Rev. 54)
DEC FORTRAN EFT5, DEC C
Berkeley FFS
Unlimited Stack Size, cache_bufcache=1, delay_wbuffers=1
None
single user

The SPEC benchmarks were recorded using Kuck and Associates pre-compilers which will be available in
Spring 1992.
Khornerstone 2 tests were run with 10% buffer cache.
Dhrystone V1.1 was compiled with -O3 optimization.
Linpack, Whetstones, Dhrystone and DR Labs CPU2 benchmarks were compiled with default -O optimization.
For nhfsstone test, system had 32 MB of memory, 2 1.0 Gbyte RZ57s, 16 nfs daemons, 25% buffer cache ,
and cache_bufcache=1.
The AIM benchmarks ran with 64 MB of memory, 1 SCSI controller with 3 RZ56s, ULTRIX V4.1, C V3.1, and
50% of memory was allocated as file buffer cache. All C source compilations were performed using level
-O3 optimization.
TPC-B results from TPC-B Full Disclosure report submitted to TPC council 12/3/91. Configuration consisted
of 32 MB of main memory, one embedded SCSI Controller supporting 2 RZ57 disk drives (1 Gbyte each) and
4 RZ25 disk drives (426 Mbytes each).

A-2

Digital’s DECsystem Family Performance Summary

Test Configurations

All tests were performed by Digital.
DECsystem 5000 Model 133 Server/Multiuser Computer:
CPU Chipset
CPU MHz
FPU Chipset
FPU MHz
Memory
Disk
Cache Size
Network Interface
Operating System
Compilers
File System
Tuning Parameters
Background Load
System State

R3000A
33
R3010A
33
64 MB
2 209 MB RZ24, 1 332 MB RZ55
128KB data/64KB instruction
Ethernet
ULTRIX X4.2A-5
DEC FORTRAN T3.1, DEC C V1.0
Berkeley FFS
10% bufcache, delay_wbuffers=1, cache_bufcache=1
none
single-user

SPEC benchmarks were recorded using Kuck and Associates pre-compilers which will be available in the
Spring 1992.
Khornerstone2 benchmarks were run with 10% buffer cache.
All AIM benchmarks ran with 48 MB of memory, 2 426 MB RZ25s, ULTRIX 4.2a operating system, and
MIPS C 2.1 compiler.
All tests were performed by Digital.
DECsystem 5000 Model 200 Server/Multiuser Computer:
CPU Chipset
CPU MHz
FPU Chipset
FPU MHz
Memory
Disk Controller
Disk
Cache Size
Network Interface
Operating System
Compilers
File System
Tuning Parameters
Background Load
System State

R3000
25
R3010
25
16 MB
SCSI
665 MB RZ56
64 KB data/64 KB instruction
Ethernet
ULTRIX T4.2 (Rev. 54)
DEC FORTRAN EFT5, DEC C
Berkeley FFS
Unlimited Stack Size, cache_bufcache=1, delay_wbuffers=1
None
single user

The AIM benchmarks used 64 MB of memory, 1 SCSI controller with 4 660 MB disks, ULTRIX V4.1-32, cc
compiler, and 20% of memory was allocated as buffer cache.
The SPEC benchmarks were recorded using Kuck and Associates pre-compilers which will be available in
Spring 1992.
Dhrystone V1.1 was compiled with -O3 optimization.
Linpack, Whetstones, Dhrystone and DR Labs CPU2 benchmarks were compiled with default -O optimization.
Khornerstone2 tests were run with 10% buffer cache.
All tests were performed by Digital.

Digital’s DECsystem Family Performance Summary

A-3

Test Configurations

DECsystem 5500 Server/Multiuser Computer:
CPU Chipset
CPU MHz
FPU Chipset
FPU MHz
Memory
Disk Controller
Disk
Cache Size
Network Interface
Operating System
Compilers
File System
Tuning Parameters
Background Load
System State

R3000A
30
R3010A
30
32 MB
SCSI/DSSI/SDA
1.2 GB RA90
64 KB data/64 KB instruction
Ethernet
ULTRIX T4.2 (Rev. 54)
DEC FORTRAN EFT5, DEC C
Berkeley FFS
Unlimited Stack Size, cache_bufcache=1, delay_wbuffers=1
None
single user

The AIM benchmarks used 128 MB of memory, SCSI controller with 3 665 MB RZ56s, ULTRIX V4.1 Rev
52, cc V3.1, and 50% of memory was allocated as buffer cache.
The AIM benchmarks used 128 MB of memory, DSSI disk controller with 3 381MB RF31s, ULTRIX V4.1
Rev 52, cc v3.1, and 50% of memory was allocated as buffer cache.
The SPEC benchmarks were recorded using Kuck and Associates pre-compilers which will be available in
Spring 1992.
Khornerstone tests were run with 10% buffer cache.
Dhrystone V1.1 was compiled with -O3 optimization.
For nhfsstone test, system had 32 MB of memory, 2 1.0 Gbyte RZ57s, 16 nfs daemons, 25% buffer cache ,
and cache_bufcache=1.
Linpack, Whetstones, and DR Labs CPU2 were compiled with default -O optimization.
TPC-B results from TPC-B Full Disclosure report submitted to TPC council 12/3/91. Configuration consisted
of 32 MB of main memory, two embedded controllers; a DSSI controller supporting seven 381 MB RF31 disk
drives and a SCSI controller supporting three 665 MB RZ56 disk drives and 2 1 Gbyte RZ57 disk drives.
All tests were performed by Digital.

Hewlett-Packard Apollo 9000 Series 817S System:
AIM results from Aim Technology. TPC-B results from TPC Council .
IBM Risc System/6000 POWERserver 320H Computer:
CPU
FPU
Memory
Disk Controller
Disk
Cache Size
Network Interface
Operating System
Compilers
File System

25 MHz IBM 2564
Integrated
16
DBA/SCSI
320 MB
64 KB data/8 KB instruction
Ethernet
AIX
C, FORTRAN
AIX

The SPEC results, recorded using Kuck and Associates’ pre-compilers, and SPEC SDM results from SPEC
Newsletter, Volume 3, Issue 3, September 1991.
Dhrystone, and Linpack ratings from IBM information dated March 1991.

A-4

Digital’s DECsystem Family Performance Summary

Test Configurations

The AIM benchmarks used 64MB of memory, IBM SCSI disk controller, 2 400 MB disks, AIX v3.1.5 operating system, x1c v3.1 compiler, and dynamic I/O buffers. CPU type was POWER 2532, ext. cpu cache was
32K data, 8K instruction.
IBM Risc System/6000 POWERserver 520 Computer:
CPU
Memory
Disk Controller
Disk
Cache Size
Network Interface
Operating System
Compilers
File System

20 MHZ
16 MB
SCSI
355 MB
32 KB data/8 KB instruction
Ethernet
AIX
C, FORTRAN
AIX

The SPEC results, recorded using Kuck and Associates’ pre-compilers, Dhrystone, and Linpack ratings from
IBM information dated March 1991.
The AIM benchmarks used 48 MB of memory, SCSI disk controller, 1 857 MB disk, AIX v3 was the operating system, compiler x1c, and dynamic I/O buffers. CPU type was IBM POWER 2032.
DR Labs CPU2 benchmark ran operating system AIX 3.0. Rating from Digital Review, October 15, 1990,
page 24.
IBM Risc System/6000 POWERserver 530H Computer:
CPU
FPU
Memory
Disk Controller
Disk
Cache Size
Operating System
Compilers
File System
Tuning Parameters
Background Load
System State

33 MHz POWER 3364
Integrated
32 MB
SCSI
1-400 MB
64 KB Data/8KB Instruction
AIX v3.1.5
AIX XL C/6000 Version 1.1.5, AIX XL FORTRAN Version 2.2
AIX
None
None
Multiuser (single user login)

SPEC and SPEC SDM SDET results from SPEC Newsletter, Volume 3, Issue 4, December 1991. SPEC
results recorded using Kuck and Associates’ pre-compilers. SPEC SDM SDET results used 2 SCSI 400 MB
disks (feature 2560) and 3 SCSI 320 MB disks (7204-320). System state was multiuser, MAXUPROC=2000.
Tuning parameters were default.
AIM benchmark results from AIM Technology.
IBM Risc System/6000 POWERserver 550 Computer:
CPU
FPU
Memory
Disk Controller
Disk
Cache Size
Network Interface
Operating System
Compilers
File System

41.67 MHz IBM 4164
Integrated
64 MB
SCSI
800 MB
64 KB data/8 KB instruction
Ethernet
AIX
C, FORTRAN
AIX

SPEC results , recorded using Kuck and Associates’ pre-compilers, from SPEC Newsletter, Volume 3, Issue
3, September 1991.
SPEC SDM results from SPEC Newsletter, Volume 3, Issue 2, Spring 1991.
The AIM benchmarks used 128 MB of memory, 2 IBM SCSI disk controllers, 2 400 MB disks, operating
system AIX v3, compiler x1c, and dynamic I/O buffers

Digital’s DECsystem Family Performance Summary

A-5

Test Configurations

TPC-B results from Sun Microsystems Computer Corp., Sun SPARCserver 600MP Performance Brief, September 1991, pages 2-3.
IBM Risc System/6000 POWERserver 930 Computer:
CPU
FPU
Memory
Disk
Cache Size
Operating System
Compiler
File System

25MHz POWER 2564
Floating Point
64 MB
SCSI 857 MB
64 KB data/8 KB instruction
AIX v3.1
xle v3.1
AIX

SPEC results were recorded using Kuck and Associates’ pre-compilers. Information from IBM dated 5/8/91.
AIM results from AIM Technology.
IBM Risc System/6000 POWERserver 950 Computer:
CPU
Memory
Disk Controller
Cache Size
Operating System
File System

41.6 MHz
64 MB
SCSI
64 KB data/8 KB instruction
AIX

AIM results from AIM Technology.
TPC-B results from Sun Microsystems Computer Corp., Sun SPARCserver 600MP Performance Brief, September 1991, pages 2-3.
Sun SPARCstation ELC Computer:
Processor Type & Frequency
Floating Point Unix & Frequency
Cache Memory Size & Speed
RAM Memory Size & Speed
Hardfile Brands & Model Numbers
Hardfile Quantity/Interface
Hardfile Size(s) (Unformatted/for.)
Operating System Name & Level
Fortran Supplier & Version
Fortran Compiler Switches Used
C supplier & Version
Compiler Switches Used
Graphics Libraries Used

SPARC (LSI) - 33MHz
SPARC (Fujitsu) - 33 MHz
64 Kilobytes
8 Megabytes / 80 ns (except khorner1 @ 16Mb)
CDC (Imprimis) 94191
1 / SCSI
760 / 680 MB
Sun OS 4.1.1
Sun Fortran 1.4
-O4
Sun C 1.1
-O4
X11

SPEC benchmark numbers from SPEC Newsletter, Volume 3, Issue 3, September 1991, page 24. Memory
configured 16MB, Disk subsystem 207 MB SCSI, other software KAP/SUN pre-processor. System state single user, no tuning parameters in use, and no background load.
Dhrystones, Whetstones, Khornerstone, Linpack, DR Labs CPU2, and X11perf benchmark numbers from
Workstation Laboratories, 9/1/1991, Volume 15, Chapter 21.
Sun SPARCstation IPX Workstation:
Processor Type & Frequency
Floating Point Unix & Frequency
Cache Memory Size & Speed
RAM Memory Size & Speed
Hardfile Brands & Model Numbers
Hardfile Quantity/Interface
Hardfile Size(s) (Unformatted/for.)
Operating System Name & Level
Fortran Supplier & Version
Fortran Compiler Switches Used

A-6

SPARC (LSI) - 40MHz
SPARC (Fujitsu) - 40MHz
64 Kilobytes
16 Megabytes / 80 ns
Maxtor 3.5" SCSI
1 SCSI
? / 207 Mb
Sun OS 4.1.1
Sun Fortran 1.4
-O4

Digital’s DECsystem Family Performance Summary

Test Configurations

C Supplier & Version
Compiler Switches Used
Graphics Libraries Used

Sun C 1.1
-O4
X11

SPEC benchmark numbers from SPEC Newsletter, Volume 3, Issue 3, September 1991, page 22. Other
software used was KAP/SUN pre-processor. Disk Subsystem was 424 MB SCSI. No tuning parameters in
use, no background load and system state was single user.
Dhrystones, Whetstones, Khornerstone, Linpack, DR Labs CPU2, and X11perf benchmark numbers from
Workstation Laboratories, 9/1/91, Volume 15, Chapter 22.
Sun SPARCserver 2 Computer:
CPU Processor
CPU MHz
FPU
FPU MHz
Memory
Disk Controller
Disk
Cache Size
Network Interface
Operating System
Compilers
Compiler Switches

SPARC
40 MHz
SPARC (TI)
40 MHz
16 MB
SCSI
207 MB
64 KB data/64 KB instruction
Ethernet
SunOS 4.1.1
Sun Fortran, Sun C
-O3

SPEC results from SPEC Newsletter, Volume 3, Issue 3, September 1991, page 24. Memory configuration
16MB, disk subsystem 207 MB SCSI, other software KUP/SUN pre-processor. System state single user, no
tuning parameters in use, and no background load.
SPEC SDM results from SPEC Newsletter, Volume 3, Issue 2, pages 37-38. Configuration was 64 MB of
memory, 2 SCSI controllers, 4 SCSI disks, and SBUS Prestoserve.,
DR Labs CPU2 rating from Digital Review, March 11, 1991, page 19. Configuration tested was running was
Sun SPARCstation 2GX , SunOS 4.1.1-1/91. (From same article, "The SPARCstation 2 has a server counterpart, the SPARCserver, that is virtually identical to the SPARCstation 2 without the graphics option." )
Linpack and Dhrystone benchmark ratings from SPARCstation 2 Performance Brief, Sun Microsystems, Inc.,
November 1990. Tested configuration was SPARCstation 2 running SunOS 4.1.1beta, Sun FORTRAN
1.4beta and Sun C1.1beta.
Whetstones results from Workstation Laboratories, Inc., Volume 12, page V12-23-Config.
Sun SPARCserver 600MP Series Server Systems (per processor):
CPU
40 MHz
FPU
40 MHz
Memory
64 MB
Disk Subsystem:
SCSI
Operating System/compilers:
Sun OS 4.1.2 Alpha 2.2, NC400 driver 1.7, Prestoserve driver
2.3
Other Hardware:
Sun SBus Prestoserve NFS Accelerator Board
2 or 4 Sun Network CoProcessor Boards
SPEC results for Sun SPARCserver 600MP (single processor) from Sun SPARCserver 600MP Performance
Brief, September 1991, pages 10-11.
TPC-B results for Sun SPARCserver 630MP and 670MP from Sun SPARCserver 600MP Performance Brief,
September 1991, pages 2-3.

Digital’s DECsystem Family Performance Summary

A-7

B
References

Digital’s DECsystem Family Performance Summary

B-1

References

Digital Equipment Corporation, Multiuser Benchmark Comparisons: AIM
Multiuser Benchmark (Suite III) DECsystem’s versus IBM RS/6000,
March, 1991.
Digital Review, Digital Review, Product Review/Workstations,
October 1, 1990.
Digital Review, Digital Review, Product Review/Workstations,
October 15, 1990.
Standard Performance Evaluation Corporation, SPEC Newsletter, Volume 3,
Issue 2, Spring 1991.
Standard Performance Evaluation Corporation, SPEC Newsletter, Volume 3,
Issue 1, Winter 1991.
Standard Performance Evaluation Corporation, SPEC Newsletter, Volume 3,
Issue 4, December 1991.
Sun Microsystems Computer Corp., SPARCstation 2 Performance Brief,
November 1990.
Sun Microsystems Computer Corp., Sun SPARCserver 600MP Performance
Brief, September 1991.
Workstation Laboratories, Workstation Laboratories, Volume 11, Chapter 12,
August 1, 1990.
Workstation Laboratories, Workstation Laboratories, Volume 12, Chapter 23,
February 1, 1991.

B-2

Digital’s DECsystem Family Performance Summary