*BSD News Article 62358


Return to BSD News archive

Path: euryale.cc.adfa.oz.au!newshost.anu.edu.au!newshost.telstra.net!news.ci.com.au!wabbit.cc.uow.edu.au!metro!metro!inferno.mpx.com.au!news.mel.aone.net.au!imci4!newsfeed.internetmci.com!news.mathworks.com!fu-berlin.de!news.belwue.de!news.uni-stuttgart.de!news.rhrz.uni-bonn.de!RRZ.Uni-Koeln.DE!zpr.uni-koeln.de!se
From: se@ZPR.Uni-Koeln.DE (Stefan Esser)
Newsgroups: comp.unix.bsd.freebsd.misc
Subject: Re: Serious performance problems with NCR PCI SCSI card
Date: 21 Feb 1996 23:32:31 GMT
Organization: Institute for Mathematics, University of Cologne, Germany
Lines: 48
Sender: se@Sysiphos (Stefan Esser)
Message-ID: <4gga2f$61d@news.rrz.uni-koeln.de>
References: <4gf23e$1vv@fozzie.sun3.iaf.nl>
NNTP-Posting-Host: sysiphos.mi.uni-koeln.de
To: geert@fozzie.sun3.iaf.nl (Geert Bosch)

In article <4gf23e$1vv@fozzie.sun3.iaf.nl>, geert@fozzie.sun3.iaf.nl (Geert Bosch) writes:
|> When I wanted to benchmark a new IBM Spitfire 1 GB drive yesterday,
|> I noticed some really strange performance problems. What I wanted
|> to test was throughput of sequential reads, these were the results
|> of both FreeBSD (using dd, reading raw filesystem sd0a) and OS/2 
|> (directly calling the OS/2 API from a small testprogram, reading 
|> raw filesystem). Both tests ran on the same system (486DX33 w/ 
|> 16 MB RAM) on same disk.
|> 
|>                    FreeBSD		      OS/2
|>                 (kB)    (kB/s)		(kB)	(kB/sec)
|> Blocksize (kB)	Total 	Speed		Total	Speed
|> 	1	39996	  102		51200	 1430
|> 	4	39996	  403		51200    1269
|> 	8	39996	  805		51200	 1278
|>        16       39996	 1611		51200    3096
|>        32	39996	 3218		51200	 3997
|>        48	39996    4240		(not measured)
|> 
|> The very strange thing is that FreeBSD or the NCR driver or dd
|> seems to be limited to 100 reads/second (which *is* really bad).
|> There is no problem with the timing, the FreeBSD test with 1 kB
|> blocksize *really* took 8 mins. Even my ethernet can do much 
|> faster than that with 1 kB packet size.
|> 
|> Isn't the standard timeslice of many unices 10 ms? Is there any
|> code which might require waiting till next clock tick? I checked
|> that the NCR driver did find the right IRQ, but I've included
|> the output of the dmesg command, because there might be clues in
|> it (although I couldn't find them).

Well, but I could.

Please assign another interrupt to the NCR,
not IRQ 15 as it is currently set up ...

Guess you assigned the IDE compatibility 
interrupt to the NCR. I've seen strange 
things like this happen before in this case.

How about using 9, 11 or 12 ?

Regards, STefan
-- 
 Stefan Esser, Zentrum fuer Paralleles Rechnen		Tel:	+49 221 4706021
 Universitaet zu Koeln, Weyertal 80, 50931 Koeln	FAX:	+49 221 4705160
 ==============================================================================
 http://www.zpr.uni-koeln.de/~se			  <se@ZPR.Uni-Koeln.DE>