Return to BSD News archive
Path: euryale.cc.adfa.oz.au!newshost.anu.edu.au!harbinger.cc.monash.edu.au!news.rmit.EDU.AU!news.unimelb.EDU.AU!munnari.OZ.AU!spool.mu.edu!uwm.edu!lll-winken.llnl.gov!enews.sgi.com!sgigate.sgi.com!uhog.mit.edu!news.mathworks.com!newsfeed.internetmci.com!uunet!not-for-mail
From: Heiko.Blume@bln.sel.alcatel.de (H. Blume von Contributed)
Newsgroups: comp.unix.bsd.freebsd.misc,news.admin.technical
Subject: Re: Poor performance with INN and FreeBSD.
Date: 16 Apr 1996 23:34:43 -0400
Organization: Alcatel SEL
Lines: 35
Sender: zorch@ftp.UU.NET
Approved: zorch@uunet.UU.NET
Distribution: world
Message-ID: <k891g27wjc.fsf@slbh03.bln.sel.alcatel.de>
References: <311F8C62.4BC4@pluto.njcc.com> <4go23v$8hp@fozzie.sun3.iaf.nl> <DpHI0o.2A6@unixhub.SLAC.Stanford.EDU> <4kea27$6f0@kelewan.dandelion.com>
NNTP-Posting-Host: ftp.uu.net
Xref: euryale.cc.adfa.oz.au comp.unix.bsd.freebsd.misc:17566 news.admin.technical:1403
>>>>> "Leonard" == Leonard N Zubkoff <lnz@dandelion.com> writes:
In article <4kea27$6f0@kelewan.dandelion.com> lnz@dandelion.com (Leonard N. Zubkoff) writes:
Leonard> Indeed, striping for news is a win if done properly. You do not want
Leonard> concatenation. Consider a 14GB news spool composed of 7 2GB disks that's
Leonard> only 85% full. With a concentated file system, a random I/O load will be
Leonard> spread only across the first 6 disks since file systems will generally
Leonard> optimize for locality. If we stripe those 7 disks instead, we are
Leonard> guaranteed to be spreading the random I/O load across all the spindles.
Leonard> The key point is to make sure the stripe width is set large enough that the
Leonard> typical I/O transaction only needs to access one of the 7 drives.
i'm currently playing with three 2GB seagate barracuda disks, that i
combined into a single 6GB filesystem with the ccd-960131 driver on FreeBSD.
works so far.
judging from TFM i should use the size of a single track in units
of DEV_BSIZE (normally 512). however, the drives have variable
track sizes. dividing the number of sectors on the disk by number
of heads and cylinders (4194058/11/3520) yields 108.3 sectors/track
on average. so i use 108 as the interleave now, but i think its too
high.
i enabled the write caches in the drives, too.
i'm now going to do some performance measurements.
opinions? experiences?
hb
--
email: src@contrib.net / Heiko.Blume@bln.sel.alcatel.de
TCP/IP GmbH, Woertherstr. 16, 10405 Berlin
Tel: +49.30/443366-0, Fax: +49.30/443366-15