From owner-freebsd-scsi Wed Oct 9 08:56:55 1996 Return-Path: owner-freebsd-scsi Received: (from root@localhost) by freefall.freebsd.org (8.7.5/8.7.3) id IAA04445 for freebsd-scsi-outgoing; Wed, 9 Oct 1996 08:56:55 -0700 (PDT) Received: from GndRsh.aac.dev.com (GndRsh.aac.dev.com [198.145.92.241]) by freefall.freebsd.org (8.7.5/8.7.3) with ESMTP id IAA04429; Wed, 9 Oct 1996 08:56:48 -0700 (PDT) Received: (from rgrimes@localhost) by GndRsh.aac.dev.com (8.7.5/8.7.3) id IAA06884; Wed, 9 Oct 1996 08:56:30 -0700 (PDT) From: "Rodney W. Grimes" Message-Id: <199610091556.IAA06884@GndRsh.aac.dev.com> Subject: Re: striping/mirroring? In-Reply-To: <199610091003.DAA04113@silvia.HIP.Berkeley.EDU> from Satoshi Asami at "Oct 9, 96 03:03:34 am" To: asami@FreeBSD.org (Satoshi Asami) Date: Wed, 9 Oct 1996 08:56:30 -0700 (PDT) Cc: bsdscsi@shadows.aeon.net, pjchilds@imforei.apana.org.au, freebsd-scsi@FreeBSD.org X-Mailer: ELM [version 2.4ME+ PL25 (25)] MIME-Version: 1.0 Content-Type: text/plain; charset=US-ASCII Content-Transfer-Encoding: 7bit Sender: owner-freebsd-scsi@FreeBSD.org X-Loop: FreeBSD.org Precedence: bulk > * does this work already? (i cant experiment yet, disks are on purchase > * list i have to get signed before i get the disks) > > You mean mirroring? Yes it's working, has been for a few months. > > * and am i right assuming (some old post listed these) that 128 is the > * interleave value that gives me most speed for reads? > > Depends on what kind of reads you are talking about. For large > sequential reads (here the size of individual read()s don't matter -- > the "large" refers to the total size that's read in succession) and > many disks, something a little smaller is usually better, e.g., 32 or > 64. > > For random reads, it should probably be the size of the read, i.e., if > your reads are 16K, then 32 would do the best. (Unless the read sizes > are very large, say 1M or so...in which case, treat this as > "sequential" as described above.) Of course, I'm assuming the reads > occur at offsets that are integer multiples of the read sizes -- if > not, I guess you want something a little larger to reduce the chance > of a single read falling between two disks. > > However, since 128 does reasonably well for both reads and writes, > that is the size I would recommend for normal (read/write) workloads. Could you please start recommending CG size interleaves (65536 or there abouts) for people using this for news spools. I have had several clients contact me about abizmal performance and they where using 16 to 128 block interleaves :-( :-( :-(. -- Rod Grimes rgrimes@gndrsh.aac.dev.com