COBOL Benchmark Timings / (11-Nov-2005)
| System |
O/S |
Cobol |
Prime |
File Build |
File Sort |
DB cpu
DB wall |
5 x DB cpu
5 x DB wall |
| Dell PE-500sc |
Linux |
Fujitsu NetCOBOL |
4.74 [x]
2.26 [y] |
1.26 |
7.83 |
10.06
29.00 |
12.83
2m21s |
| Dell PE-500sc |
Linux |
theKompany KOBOL |
7.03 [x]
3.56 [y] |
2.89 |
20.82 |
- |
- |
| Dell PE-2850 |
Linux |
Fujitsu NetCOBOL |
2.10 [x]
1.24 [y] |
0.46 |
2.61 |
- |
- |
| HP rp7410 |
HP-UX |
MicroFocus COBOL |
13.24 |
1.00 |
4.36 |
- |
- |
| HP rp2470 |
HP-UX |
Fujitsu NetCOBOL |
11.90 |
1.40 |
9.70 |
- |
- |
| HP e3K A400-110 |
MPE/iX |
HP COBOL |
317.60 [7]
174.13 [6] |
24.75 |
97.70 |
22.80
32.00 |
23.12
2m21s |
| HP e3K A400-150 |
MPE/iX |
HP COBOL |
82.04 [6] |
12.31 |
48.63 |
- |
- |
| HP e3K N4K-440 |
MPE/iX |
HP COBOL |
39.26 [7]
21.65 [6] |
3.39 |
13.80 |
- |
- |
| HP e3K N4K-220 |
MPE/iX |
HP COBOL |
77.51 [7]
42.59 [6] |
6.23 |
24.76 |
- |
- |
| HP e3K 979KS-300 |
MPE/iX |
HP COBOL |
109.56 [7]
57.92 [6] |
8.67 |
31.43 |
- |
- |
| HP e3K 969KS-220 |
MPE/iX |
HP COBOL |
129.21 [7]
72.72 [6] |
10.50 |
44.75 |
- |
- |
| HP e3K 987/150 |
MPE/iX |
HP COBOL |
169.21 [7]
95.65 [6] |
11.82 |
68.36 |
- |
- |
| HP e3K A500-200-140 |
MPE/iX |
HP COBOL |
219.88 [7]
120.43 [6] |
16.62 |
66.31 |
- |
- |
| HP e3K 928/RX |
MPE/iX |
HP COBOL |
367.12 [7]
196.40 [6] |
47.58 |
206.29 |
- |
- |
| HP e3K 918/LX |
MPE/iX |
HP COBOL |
503.22 [7]
269.55 [6] |
58.76 |
311.44 |
- |
- |
System Reference
Dell PE-500sc:
Dell PowerEdge 500SC Tower; Pentium III 1.0ghz;
1/2gb ram, 20gb IDE drive. This system is running SuSe 7.3 and was
installed December 2001.
Dell PE-2850:
Dell PowerEdge 2850 2-u Rack; 2 x Xeon 2.8ghz;
6gb ram; 2 x 73gb 15krpm scsi (hw raid-1); 4 x 36gb 15krpm scsi
(hw raid-5). This system is running SuSe Enterprise 9 sp2 (64 bit).
HP e3k N4K-440:
HP e3000 N4000; 1 440mhz cpu; 2.5gb memory; SCSI Jbod
HP e3k N4K-220:
HP e3000 N4000; 1 220mhz cpu; unknown memory config; SCSI Jbod
HP rp7410:
HP rp7410; 4 x 750mhz cpu; 8gb ram; dual Fibre-VA disk arrays
HP rp2470:
HP rp2470; 2 x 650mhz cpu; 2gb ram; SCSI Jbod
HP e3K 979KS-300:
HP e3000 979KS; 3 x 180mhz cpu; unknown ram; SCSI Jbod
HP e3K 969KS-220:
HP e3000 969KS; 2 x 120mhz cpu; 2gb ram; SCSI Jbod
HP e3K 987/150:
HP e3000 987/150; 96mhz cpu; unknown ram; SCSI Jbod
HP e3K A500-200-140:
HP e3000 A500; 2 x 140mhz cpu; unknown ram; SCSI Jbod
HP e3K A400-110:
HP e3000 A400; 110mhz cpu; 2gb ram; SCSI Jbod (72gb)
HP e3K A400-150:
HP e3000 A400; 150mhz cpu; 2gb ram; SCSI Jbod (72gb)
HP e3K 928/RX:
HP e3000 928; 48mhz cpu; unknown ram; SCSI Jbod
HP e3K 918/LX:
HP e3000 918; 34mhz cpu; unknown ram; SCSI Jbod
Per HP e3000 Business Servers Configuration Guide - 09/01 the
clock speeds of these systems are affected by software: A400, A500,
N4K-220, N4K-330, 918, 929KS/030, 939KS/030.
Benchmark Commentary
Cobol Compilers
Except where noted the Fujitsu timings were made with the final shipping
version of NetCOBOL for Linux version 7 or NetCOBOL for HP-UX version
20.
The KOBOL timings were made with a pre-release version of the compiler.
We would expect to see improvements when running these tests against the
final shipping version of KOBOL.
Prime
Computes all prime numbers up to 50,000. Time is reported in cpu seconds.
Note: MPE timings were made using the 6 and 7 version
of the HP cobol compiler. Starting with 6.5 (MPE) the COBOL compiler
began using a new 'back-end' that was designed for handling 64-bit
integers and is slower than previous versions of the 'back-end'.
We understand you can get a fix to the 6.5+ versions of the compiler
which use the previous version of the 'back-end' since HP COBOL has
no need for 64-bit integers.
Linux/NetCOBOL timings were done with 'comp' reported as [x] and with
'comp-5' reported as [y]. NetCOBOL 'comp' is always big-endian and
'comp-5' is aligned with the target system architecture.
File Build
Fills a new file with 300,000 100-byte records
of which the first 10-bytes are a key built by rotating the letters
of the alphabet. The remaining 90-bytes are the same. Time is reported
in cpu seconds.
File Sort
Reads the file from (2) and does a 'release'
and 'return' and creates a new output file. Time is reported in
cpu seconds. Many open systems compilers have fast sort routines
which can be optionally licensed. We tested with the default sort, not
with any optional fast sort routine. Note: QSS did NOT do the rp7410
test and did not get any information on the whether the test included an
optional fast sort routine for the MicroFocus compiler.
DB
For MPE/Image: A loop of 10 dbfinds using a
2-byte search path followed by dbmode5 dbgets until end-of-chain.
There are 3 logic tests for each record and all 'kept' records are
output to a temporary file. 73873 records are processed. Timings are
listed as cpu seconds followed by wall seconds. We used multiple
samplings and the wall time is reported as an average.
For Linux/PostgreSQL: A loop of 10 selects using the same
2-byte values as the MPE/Image test with this field defined as an index.
There are 3 logic tests for each record and all 'kept' records are
output to a file. 73873 records are processed. Timings are listed as
cpu seconds followed by wall seconds. We used multiple samplings and
the wall time is reported as an average. PostgreSQL was version 7.1.3
for this test and we used the pre-release version of NetCOBOL. We have
been working on changes to our DB architecture so we cannot do a fair
test at this time with the final shipping version of NetCOBOL for Linux.
Warning: The cpu timing is ONLY the cpu cost for the
COBOL program accessing the database and does not include the db-engine
cpu time expended to satisfy the SQL request. We will attempt to
quantify this, but until then you should be very careful when comparing
MPE/Image vs PostgreSQL using cpu time.
5 x DB
This is the same benchmark as DB, but was performed
with five (5) simultaneous processes. The idea behind this test is
to try to validate performance for more than one process accessing
the database. The wall time is the time for the five (5) processes
to finish when they all started at the same time and the cpu time is
the computed average.
This information brought to you by QSS. We tried our best to be
good testers, but we are human and could have goofed by accident.
Remember, the value of information is usually directly related to how
much you paid. This site powered by QWEBS - the only known web server
written in COBOL.