[1297] in linux-scsi channel archive
Linux 2.0 SCSI performance
daemon@ATHENA.MIT.EDU (Daniel Bischof)
Thu Jan 23 12:25:54 1997
Date: Thu, 23 Jan 1997 09:50:17 +0100 (MET)
From: Daniel Bischof <daniel@papkas.ap-kas.ie.philips.com>
To: linux-scsi@vger.rutgers.edu
Dear Linux/SCSI-people,
we're running Linux 2.0 on Intel Boxes equipped with different SCSI
controllers. To determine the disk-I/O we're using the BONNIE benchmark
included in the iX-SSBA benchmark suite (source is available from
ftp://ftp.uni-paderborn.de/pub/ix/benches/ix_ssba/)
Compared to our previous system, SORIX SVR4, Linux is much
faster - except in random access to big files (>10MB).
Typical example:
BONNIE - Intel-Box with AHA2940 Controller - Linux 2.0
-------Sequential Output-------- ---Sequential Input--- -Random--
-Per Char- --Block--- -Rewrite-- -Per Char- --Block--- --Seeks---
MB K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU
10 3543 68.2 20078 100 1177 12.3 3368 80.3 6918 27.0 3921 98.0
30 3518 67.2 5278 29.7 1411 15.5 2139 44.4 3353 14.2 98.4 2.8
50 3452 66.8 3413 20.5 1390 14.7 2467 51.2 3326 14.4 57.5 2.4
70 3719 73.6 3241 18.1 1364 15.0 2510 51.4 3580 16.1 52.2 2.4
BONNIE - Intel-Box with AHA2940 Controller - SORIX SVR4
-------Sequential Output-------- ---Sequential Input-- --Random--
-Per Char- --Block--- -Rewrite-- -Per Char- --Block--- --Seeks---
MB K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU
10 1408 61.9 1896 17.4 1939 18.2 3160 97.5 32000 100 1408.5 50.7
30 1488 61.2 1951 16.0 1797 19.4 3216 100.0 31346 100 1433.7 49.1
50 1427 60.3 1713 17.3 659 11.5 1319 52.8 2130 23.0 117.7 7.8
70 1463 61.1 1788 15.5 648 12.4 1479 58.9 2110 22.1 73.7 6.4
As you can see, SORIX performance (random seeks) remains constant for
files > 10MB, whilst Linux breaks down to less than 100 KB/s.
Since we're running time-critical database applications, this behaviour
may become a problem. I've tried several hardware configurations
(SCSI-HA: AHA-2940, AHA-2920 (FutureDomain), ASUS SC-200 (NCR)
SCSI-Harddisks: IBM-DORS, Quantum Fireball TM);
all results show the same behaviour.
If you have any suggestions, please let me know.
Regards,
D.B.
+=====================+======================================+
| Daniel Bischof | e-mail: daniel@ap-kas.ie.philips.com |
+=====================+======================================+
