From owner-freebsd-hackers Sun Oct 14 1:49: 6 2001 Delivered-To: freebsd-hackers@freebsd.org Received: from beppo.feral.com (beppo.feral.com [192.67.166.79]) by hub.freebsd.org (Postfix) with ESMTP id C057237B403; Sun, 14 Oct 2001 01:49:02 -0700 (PDT) Received: from mailhost.feral.com (mjacob@mailhost.feral.com [192.67.166.1]) by beppo.feral.com (8.11.3/8.11.3) with ESMTP id f9E8mnH04937; Sun, 14 Oct 2001 01:48:50 -0700 (PDT) (envelope-from mjacob@feral.com) Date: Sun, 14 Oct 2001 01:48:46 -0700 (PDT) From: Matthew Jacob X-Sender: mjacob@beppo Reply-To: mjacob@feral.com To: =?X-UNKNOWN?Q?S=F8ren_Schmidt?= Cc: Jordan Hubbard , jrossiter@symantec.com, sos@FreeBSD.ORG, freebsd-hackers@FreeBSD.ORG Subject: Re: Severe I/O Problems In-Reply-To: <200110140750.f9E7oZi05487@freebsd.dk> Message-ID: MIME-Version: 1.0 Content-Type: TEXT/PLAIN; charset=US-ASCII Sender: owner-freebsd-hackers@FreeBSD.ORG Precedence: bulk List-ID: List-Archive: (Web Archive) List-Help: (List Instructions) List-Subscribe: List-Unsubscribe: X-Loop: FreeBSD.ORG > > All parameters were not totally lab controlled, but pretty close. I'd call it > > significant enough that I'll leave WC off for this machine. > > Well I tried something semilar here, CVS tree on one disk, then newfs and > cvs co src on another (both DTLA's on via686b if that matters) to keep > things as equal as possible, singleuser mode of cause... > > -SU: softupdates -wc writecache enabled -tags tagged queing enabled > > res: 33.973u 19.425s 27:27.88 3.2% 583+7328k 47219+160828io 37pf+0w > res-SU: 33.552u 16.235s 9:53.49 8.3% 586+7244k 47180+864io 6pf+0w > res-wc: 32.959u 20.104s 3:38.23 24.3% 586+7317k 47232+160828io 37pf+0w > res-wc-SU: 33.343u 16.909s 2:52.81 29.0% 585+7269k 47184+863io 6pf+0w > res-wc-tags: 33.891u 29.705s 3:40.62 28.8% 587+7595k 47225+160548io 20pf+0w > res-wc-tags-SU: 33.946u 20.759s 2:48.97 32.3% 578+7234k 47183+863io 5pf+0w > > That behaves more or less as I would expect... > > It is worth nothing that there is close to an order of magnitude in > difference from the slowest to the fastest, and that WC is the single > option that can make that difference on its own... > I don't doubt this. I just would suggest that there is such a spread of h/w and configurations that sometimes turning on WC is fantastic, and sometimes not so fantastic. Now- before y'all get huffy about this and suggest that I don't have the right h/w- I'll agree- but when people run into performance issues this is certainly one thing to check. D'ya think you could come up with a little tester program that could predict whether WC would make sense for a particular h/w configuration or not? That'd be darned usefull. -matt To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-hackers" in the body of the message