*BSD News Article 37283


Return to BSD News archive

Path: sserve!newshost.anu.edu.au!harbinger.cc.monash.edu.au!yarrina.connect.com.au!warrane.connect.com.au!godzilla.zeta.org.au!not-for-mail
From: bde@zeta.org.au (Bruce Evans)
Newsgroups: comp.os.386bsd.questions
Subject: Re: dismal file system performance and odd scsi stuff
Date: 2 Nov 1994 10:14:33 +1100
Organization: Kralizec Dialup Unix Sydney - +61-2-837-1183, v.32bis v.42bis
Lines: 35
Message-ID: <396i4p$o0@godzilla.zeta.org.au>
References: <oigcIge00iUzI28UlV@andrew.cmu.edu>
NNTP-Posting-Host: localhost.zeta.org.au

In article <oigcIge00iUzI28UlV@andrew.cmu.edu>,
Alex R.N. Wetmore <aw2t+@andrew.cmu.edu> wrote:
>I'm trying to figure out how to get the best file system performance out of
>my system (486sx/25, 250 meg WD IDE, 1gig Micropolis 4110 SCSI w/ AHA1542C).
>I know about tunefs, and have even played with it a bit, but can't get iozone
>speeds about 400k per second or so (on the SCSI, even worse on the IDE).  SCSI
>performance is what I'm looking for, since there isn't much on the IDE.
>
>Is there any way that I can figure out the ideal performance options for a
>drive?  Should I just play around.  I read the chapter on the filesystem in

Try "newfs -d 0 -n 1 ..." and the default of "newfs -d 4 ..." and values
in between.

>This also might be more of a hardware based question.  I've noticed that
>I get exactly one megabyte per second using dd to read data from my sd0a
>device.  (I used the command "dd if=/dev/sd0a of=/dev/null bs=4k
>count=1024).  This seems like a very odd number to get constantly, as if
>its being throttled that low for some reason.  Is there anything in my

It's throttled by the too-small value of 2048 for BLKDEV_IOSIZE in
<machine/param.h>.  On my system, the U34F SCSI controller seems to take
a minumum of 5ms per request, so it is limited to 400K/s through
/dev/sd0d and 100K/sec through /dev/rsd0d with a block size of 512
although it can read /dev/rsd0d at 2500K/sec with a block size of 64K.
Use the raw device to determine guaranteed-not-to-exceed speeds.  I
see speeds like the following:

	max raw speed			2500K/sec
	iozone				1250K/sec (read) 1700K/sec (write)
	read 250M file system		600K/sec
	copy 250M file system		150K/sec throughput
	copy /usr/share/zoneinfo	15K/sec throughput
-- 
Bruce Evans  bde@zeta.org.au