Return to BSD News archive
Path: euryale.cc.adfa.oz.au!newshost.anu.edu.au!harbinger.cc.monash.edu.au!bunyip.cc.uq.oz.au!munnari.OZ.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.os.linux.development.system,comp.unix.bsd.freebsd.misc
Subject: Re: The better (more suitable)Unix?? FreeBSD or Linux
Followup-To: comp.os.linux.development.system,comp.unix.bsd.freebsd.misc
Date: 19 Feb 1996 23:33:06 GMT
Organization: Institute for Mathematics, University of Cologne, Germany
Lines: 77
Sender: se@Sysiphos (Stefan Esser)
Message-ID: <4gb1bi$rbv@news.rrz.uni-koeln.de>
References: <4er9hp$5ng@orb.direct.ca> <31220995.C4C54C1@acm.org> <4g0sam$r6p@agate.berkeley.edu> <4g5cl6$q6k@news.rrz.uni-koeln.de> <4g8vl3$8h7@pell.pell.chi.il.us>
NNTP-Posting-Host: sysiphos.mi.uni-koeln.de
To: orc@pell.chi.il.us (Orc)
Bcc: se
Xref: euryale.cc.adfa.oz.au comp.os.linux.development.system:17633 comp.unix.bsd.freebsd.misc:14028
In article <4g8vl3$8h7@pell.pell.chi.il.us>, orc@pell.chi.il.us (Orc) writes:
|> In article <4g5cl6$q6k@news.rrz.uni-koeln.de>,
|> Stefan Esser <se@ZPR.Uni-Koeln.DE> wrote:
|> [a collection of statistics comparing Linux ext2fs to BSD ffs]
|>
|> I may have missed something, but aren't you leaving out the
|> statistics showing the performance of ext2fs on BSD vs ffs on BSD?
|> Comparing ext2fs behavior on Linux to ffs behavior on BSD is meaningless
|> unless you also factor in the speed of the buffer cache. It's like the
|> comparisons of native NCR810 drivers on freeBSD vs Linux; they make
|> wonderful propaganda, if you're only preaching to the converted; the
|> unconverted, on the other hand, will look at comparisons of the Linux
|> port of the BSD NCR810 driver vs the native Linux driver and then
|> disregard every comparison coming from the BSD camp, never mind if they
|> are real or mind candy.
Well, I compare EXT2FS on Linux to FFS on FreeBSD,
because it would be unfair (and useless) to compare
FFS and EXT2 on BSD ...
1) People are most likely to run FFS on BSD and
EXT2 on Linux. I'm giving numbers for what
people will see ...
2) EXT2 uses 1KB blocks, and can't easily be as
fast as FFS on FreeBSD, since FFS can just
identify blocks with pages (except for the
last partial block) while EXT2 doesn't fit
this model. EXT2 on Linux may well be faster
than on EXT2 on BSD, and thus I was doing it
a favour :)
Regarding the Linux and BSD NCR driver under
Linux: You don't see much of a difference, since
both perform with near zero overhead. Only the
buffer cache and file system code consumes CPU
cycles.
If you are looking for differences, then don't
look at the (very similar) performance, but at
the features each one offers ...
The following is what I see on my ASUS SP3G:
# time dd if=/dev/rsd0a of=/dev/null bs=1k
24576+0 records in
24576+0 records out
25165824 bytes transferred in 16 secs (1572864 bytes/sec)
16.04 real 0.21 user 3.83 sys
# time dd if=/dev/rsd0a of=/dev/null bs=32k
768+0 records in
768+0 records out
25165824 bytes transferred in 4 secs (6291456 bytes/sec)
3.81 real 0.00 user 0.18 sys
# time dd if=/dev/rsd0a of=/dev/null bs=64k
384+0 records in
384+0 records out
25165824 bytes transferred in 4 secs (6291456 bytes/sec)
3.60 real 0.00 user 0.12 sys
That gives 150 microseconds of overhead in the
1KB transfer case, and 300 for 64K, i.e.:
systime = cmds * 150us + KB * 2.4us
This indicates, that continously reading at 10MB/s
would make my 486 CPU spend far less than 3% in
the NCR driver ... This is near zero overhead :)
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>