Return to BSD News archive
Path: sserve!newshost.anu.edu.au!harbinger.cc.monash.edu.au!msuinfo!agate!howland.reston.ans.net!xlink.net!zib-berlin.de!rs1-hrz.uni-duisburg.de!rrz.uni-koeln.de!RRZ.Uni-Koeln.DE!RRZ.Uni-Koeln.DE!news
From: se@fileserv1.MI.Uni-Koeln.DE (Stefan Esser)
Newsgroups: comp.os.386bsd.misc
Subject: File System Performance (was: FreeBSD platform)
Date: 27 Jun 1994 14:02:52 GMT
Organization: Institute of Nuclear Physics, University of Cologne, Germany
Lines: 67
Distribution: world
Message-ID: <2umm6cINN15a5@rs1.rrz.Uni-Koeln.DE>
References: <2uacoc$gfi@reuter.cse.ogi.edu> <2ukb2g$r7b@huey.cc.utexas.edu> <Cs19IJ.B07@mozo.cc.purdue.edu>
NNTP-Posting-Host: fileserv1.mi.uni-koeln.de
Keywords: configuration
In article <Cs19IJ.B07@mozo.cc.purdue.edu>, bj@staff.cc.purdue.edu (Ben Jackson) writes:
|> Right now I'd have to recommend you avoid the Adaptec 1542Cf. I'm
|> getting horrible performance from it. The board can DMA at 8M/s, the
|> SCS2 disks can talk to it at 3-5M/s, and somehow on the other side of
|> my filesystem layer, I get 650k/s. Bleah.
You can have at least 2 to 3 times that performance through the file
system ! (See BONNIE results below.
Since you wrote you have a PCI graphics board, you might be interested
in knowing that there is a driver for the NCR 53c810 PCI SCSI chip,
that comes builtin on several PCI motherboards, or that is <$100 as
a PCI board.
The following is the performance we get from a Seagate ST31200N using
this driver on a 486DX2/66 running FreeBSD-1.1R (ASUS motherboard).
We wanted to check the achievable performance boost due to tagged
command queueing, not maximum absolute througput in this tests.
The performace will go further up under FreeBSD-1.1.5 due to block
clustering. We get best performance when using the driver with 8 tags
(commands sent to the drive simultaneously).
From the better performance of the 100MB test cases you can see, that
the 10MB values are too low because of the >70% filled filesystem
(such tests are generally done on free partitions to get repeatable
results.)
We expect >3.5MB/s through the filesystem under FreeBSD 1.1.5 with this
configuration and our driver.
In case you want to try it, you can send mail to Wolfgang Stanglmeier
<wolf@dentaro.GUN.de> or to my address. We will make it available via
ftp soon, but there is a last round of beta testing starting now, to
wedge out portability problems ...
(The driver is in use for a few month now, but feature have been added
until recently. Its out to some 20 beta testers with no unresolved
installation problem we know of.)
-------Sequential Output-------- ---Sequential Input-- --Random--
-Per Char- --Block--- -Rewrite-- -Per Char- --Block--- --Seeks---
Machine MB K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU /sec %CPU
async 10 640 66.2 548 4.2 281 3.7 1235 96.7 2086 22.0 81.4 7.3
fast.1 10 650 59.1 548 4.0 317 4.9 1274 99.0 2234 26.2 82.7 7.8
fast.2 10 900 99.4 912 7.5 519 7.8 1273 99.2 2318 24.0 82.3 7.9
fast.4 10 895 99.3 1359 12.9 514 9.4 1259 99.2 2235 22.0 83.6 8.2
fast.8 *** 10 889 98.3 1747 16.6 526 9.8 1276 99.2 2318 25.4 82.7 7.4
fast.16 10 902 99.3 1289 12.0 526 8.9 1281 99.4 2054 21.9 83.0 7.9
fast.1 100 591 61.9 583 4.2 342 4.7 1259 97.8 2900 36.3 47.9 4.9
fast.16 100 955 99.3 1402 12.5 583 10.4 1274 98.1 2895 37.2 48.6 4.8
All tests on FreeBSD-1.1R with NCR 53c810 and Seagate ST32100N (only 30% free!)
ASYNC SCSI and FAST SYNC without, with 2, 4, 8 and 16 Tags on 10MB file.
FAST SYNC without and with 16 Tags on 100MB file.
--
Stefan Esser Internet: <se@MI.Uni-Koeln.DE>
Mathematisches Institut Tel: +49 221 4706010
Universitaet zu Koeln FAX: +49 221 4705160
Weyertal 80
50931 Koeln