From owner-freebsd-hardware Wed Feb 24 17:42:33 1999 Delivered-To: freebsd-hardware@freebsd.org Received: from panzer.plutotech.com (panzer.plutotech.com [206.168.67.125]) by hub.freebsd.org (Postfix) with ESMTP id 6F7D814C91 for ; Wed, 24 Feb 1999 17:42:29 -0800 (PST) (envelope-from ken@panzer.plutotech.com) Received: (from ken@localhost) by panzer.plutotech.com (8.9.2/8.8.5) id SAA42508; Wed, 24 Feb 1999 18:41:12 -0700 (MST) From: "Kenneth D. Merry" Message-Id: <199902250141.SAA42508@panzer.plutotech.com> Subject: Re: Tagged queueing and write cache on WDE SCSI drives In-Reply-To: <99Feb25.121437est.40397@border.alcanet.com.au> from Peter Jeremy at "Feb 25, 1999 12:25:32 pm" To: peter.jeremy@auss2.alcatel.com.au (Peter Jeremy) Date: Wed, 24 Feb 1999 18:41:12 -0700 (MST) Cc: ru@ucb.crimea.ua, hardware@FreeBSD.ORG X-Mailer: ELM [version 2.4ME+ PL43 (25)] MIME-Version: 1.0 Content-Type: text/plain; charset=US-ASCII Content-Transfer-Encoding: 7bit Sender: owner-freebsd-hardware@FreeBSD.ORG Precedence: bulk X-Loop: FreeBSD.org Peter Jeremy wrote... > Ruslan Ermilov wrote: > >I have tested all four combinations for Tagged Queueing (TQ) and > >Write Cache option (WC). > > Tagged Queueing is only relevant where a device has a number of > outstanding I/O requests. Since your test is single-threaded, > it doesn't test TQ. It may be single threaded, but since it goes through the filesystem, there can be multiple outstanding I/O requests. For instance: I just did the following test: # dd if=/dev/zero of=foo.tmp bs=64k count=8000 And then, while it was going: # ./camcontrol tags -v (pass0:ahc0:0:1:0): dev_openings 1 (pass0:ahc0:0:1:0): dev_active 62 (pass0:ahc0:0:1:0): devq_openings 0 (pass0:ahc0:0:1:0): devq_queued 0 (pass0:ahc0:0:1:0): mintags 2 (pass0:ahc0:0:1:0): maxtags 255 There was nothing else touching that drive at the time. The number of active commands is 62, which is pretty good. The disk in question is: da0 at ahc0 bus 0 target 1 lun 0 da0: Fixed Direct Access SCSI-2 device da0: 20.000MB/s transfers (20.000MHz, offset 15), Tagged Queueing Enabled da0: 2049MB (4197405 512 byte sectors: 255H 63S/T 261C) It's on a 7880. > For a realistic TQ test, you need to have a substantial number > of processes performing I/O - unfortunately, I don't know of > any suitable benchmark. The `seeker' part of bonnie would be > the closest. An alternative would be to run a lot of dd's > (maybe 20) in parallel. > > >WC TQ MB/s > >======= ======= ==== > >OFF OFF 3.9 > >OFF ON 4.2 > >ON OFF 7.9 > >ON ON 4.2 > > I would expect that enabling WC would improve write speed (which is > what dd is testing). I'm surprised by the impact of TQ: Since this is > a single-threaded test, it should have negligible impact. It almost > looks like something is wrong with the TQ handling in the driver, > controller or drive. I think the problem is the way the Western Digital drives handle tagged queueing. From what Ruslan has said, and from what Andrew Gallatin has said, I think the Western Digital drives probably perform best with tagged queueing turned off, and write caching turned on. Ken -- Kenneth Merry ken@plutotech.com To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-hardware" in the body of the message