*BSD News Article 12949


Return to BSD News archive

Newsgroups: comp.os.386bsd.bugs
Path: sserve!newshost.anu.edu.au!munnari.oz.au!news.Hawaii.Edu!ames!saimiri.primate.wisc.edu!usenet.coe.montana.edu!osyjm
From: osyjm@cs.montana.edu (Jaye Mathisen)
Subject: Re: IOZONE results
Message-ID: <1993Mar18.212127.21263@coe.montana.edu>
Sender: usenet@coe.montana.edu (USENET News System)
Organization: CS
References: <C43MBo.3owM@austin.ibm.com>
Date: Thu, 18 Mar 1993 21:21:27 GMT
Lines: 44

In article <C43MBo.3owM@austin.ibm.com> fredriks@austin.ibm.com (Lars Fredriksen) writes:
>
>
>Hi,
>	This is not a bug per se, as much as a request to see what
>other people are seeing.
>
>	I ran iozone auto on my machine and got the maximum of 500K/s
>throughtput. I saw a report from someone running Interactive Unix on
>a 50MHz 486 with a Bustek controller getting 4.5M/s.
>
>	My machine is a 33MHz 486 with 128K of cache, 16Mb of Memory,
>	with an Adaptec 174x controller. I have Julians scsi drivers,
>	and the drive is supposed to be capable of 5M/s.
>
>Anyone have a clue to why I am seeing a factor of 10 difference?  
>
>Where/how does the kernal determine who big of a buffer cache to use?

I just spend a fair amount of time with benchmarking some i/o stuff with
IOZONE:

Machine:  DECPC 486/50 (not dx2),EISA,2 Adaptec 1740's,DEC RZ25,Fuji 2624.
OS:  386BSD 0.1 with patchkit 2.1 and 84-110.

Most of my performance varied with the block/frag sizes and some tweaking
with tunefs, but the upshot was:

4k/512byte,            read:230k, write 250k.
8k/2k,tunefs -a 8      read:1.7M, write 1.3M.

The difference in speeds between the DEC drive and the Fuji drive wasn't much.
But the DEC was faster in all cases.

Any combination that I tried (and I think I tried all combo's of 2k-8kblocks,
and 512byte to 2k frags) other than 8k/2k were well under 700k/second.

I didn't try 16k block sizes, I don't know if they're supported.

So, check your disk partitioning.  your mileage may vary.
-- 
 Jaye Mathisen, COE Systems Manager                (406) 994-4780
 410 Roberts Hall,Dept. of Computer Science
 Montana State University,Bozeman MT 59717	osyjm@cs.montana.edu