From owner-cvs-all Sun Nov 25 20:13:32 2001 Delivered-To: cvs-all@freebsd.org Received: from mail.tgd.net (pop3-ssl.tgd.net [209.81.25.10]) by hub.freebsd.org (Postfix) with ESMTP id 2194637B41B; Sun, 25 Nov 2001 20:13:28 -0800 (PST) Received: by mail.tgd.net (Postfix, from userid 1001) id 638B420F0A; Sun, 25 Nov 2001 19:39:57 -0800 (PST) Date: Sun, 25 Nov 2001 19:39:57 -0800 From: Sean Chittenden To: Ian Dowse Cc: Kris Kennaway , cvs-committers@FreeBSD.org, cvs-all@FreeBSD.org Subject: Re: cvs commit: src/usr.bin/tail forward.c Message-ID: <20011125193957.O20895@ninja1.internal> References: <20011125153656.A94203@xor.obsecurity.org> <200111260038.aa52845@salmon.maths.tcd.ie> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline In-Reply-To: <200111260038.aa52845@salmon.maths.tcd.ie>; from "iedowse@maths.tcd.ie" on Mon, Nov 26, 2001 at = 12:38:47AM X-PGP-Key: 0x1EDDFAAD X-PGP-Fingerprint: C665 A17F 9A56 286C 5CFB 1DEA 9F4F 5CEF 1EDD FAAD X-Web-Homepage: http://sean.chittenden.org/ Sender: owner-cvs-all@FreeBSD.ORG Precedence: bulk List-ID: List-Archive: (Web Archive) List-Help: (List Instructions) List-Subscribe: List-Unsubscribe: X-Loop: FreeBSD.ORG > >I dunno..the major use of tail -F is probably on logfiles which get > >rotated infrequently. I can imagine that a site might want to > >simultaneously monitor a large number of logfiles, but I don't know > >what the CPU impact of that with your change would be..maybe not much. > >If you tail -F, say, 100 simultaneous files, how much CPU does that > >stat'ing use? > > Running 100 "tail -F"s requires a pretty negliglible amount of CPU > time (try it!), but kqueue does help by allowing the processes to > be swapped out during any long pauses between file updates. > > The reason for this change is simply to fix a particular use of > tail that was broken by the addition of kqueue support. The current > code gives us about half of the kqueue gains over the pre-kqueue > version for the -F case (changes noticed quickly, polling frequency > reduced by a factor of 4). Doing better is of course possible, but > it doesn't seem particularly easy. Maybe the author of this patch > (Maxim Konovalov) would be willing to have a go at it. Why not wrap this functionality in a new CLI argument? If I'm tailing over NFS or some non-kq enabled FS, then I'll turn on the flag manually (ex: -N). Seems like this is a step backwards to me... -sc -- Sean Chittenden To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe cvs-all" in the body of the message