*BSD News Article 7989


Return to BSD News archive

Xref: sserve comp.unix.bsd:8042 comp.os.linux:16420
Path: sserve!manuel.anu.edu.au!munnari.oz.au!network.ucsd.edu!usc!zaphod.mps.ohio-state.edu!cs.utexas.edu!sun-barr!olivea!uunet!math.fu-berlin.de!mailgzrz.TU-Berlin.DE!gmdtub!bigfoot!tmh
From: tmh@bigfoot.first.gmd.de (Thomas Hoberg)
Newsgroups: comp.unix.bsd,comp.os.linux
Subject: Re: bonnie i/o test results
Message-ID: <TMH.92Nov21093831@bigfoot.first.gmd.de>
Date: 21 Nov 92 08:38:31 GMT
References: <1992Nov6.144749.26760@ntuix.ntu.ac.sg> <1992Nov7.041653.3731@ntuix.ntu.ac.sg>
Sender: news@bigfoot.first.gmd.de
Followup-To: comp.unix.bsd
Organization: GMD-FIRST, Berlin
Lines: 37
In-reply-to: eoahmad@ntuix.ntu.ac.sg's message of 7 Nov 92 04:16:53 GMT

In article <1992Nov7.041653.3731@ntuix.ntu.ac.sg> eoahmad@ntuix.ntu.ac.sg (Othman Ahmad) writes:

[bonnie results omitted...]

bonnie was one of the benchmarks I always wanted to run on 386BSD.
Unfortunately it was also on one of these QIC525 tapes that were
written under ISC with a 256k block size and that I jast can't read
back in. I wanted to use it especially to benchmark Julian's and
Pace's drivers against each other (the former having all kinds of
features, the latter being simpler (but rather easier to understand :-)) 

What strikes me, though, are the rather low block i/o figures. Under
ISC HPFS (high performance after all) I was used to getting around
800k/sec on reads (Wren VII and Fujitsu 600MB disks) and between 400
and 600k/sec on writes (the 1.2GB Wren drive being somewhat slower
here). Those figures were done on 50MB files, I believe, as to obviate
the buffer cache's delayed writes on data blocks. The slower write
figures on the Wren drive might have had another reason, though. While
the file system still had around 100MB free, it was still a 600MB file
system and on V.3 Unix with file system hardening that means long
travels to the i-area or to indirect blocks. The Fujitsu was basically
empty...

The HPFS went some way to make the monstrously slow System V disk i/o
a bit faster by gathering multi-block requests into batches for the
device driver and reducing fragmentation, but I would have thought
that the BSD fast file system should provide faster or at least
comparable i/o than ISC's HPFS (not to be confused with OS/2' HPFS).

So where is the performance going under 386BSD? I know the hardware
can do far better.
---
Thomas M. Hoberg   | Internet: tmh@first.gmd.de
1000 Berlin 41     |           tmh@cs.tu-berlin.de
Wielandstr. 4      |
Germany            | BITNET:   tmh@tub.bitnet 
+49-30-851-50-21   |