Date: Sun, 14 Oct 2001 12:43:45 -0500 (CDT) From: Mike Silbersack <silby@silby.com> To: Dag-Erling Smorgrav <des@ofug.org> Cc: Matt Dillon <dillon@earth.backplane.com>, Peter Wemm <peter@wemm.org>, Alfred Perlstein <bright@mu.org>, Mikhail Teterin <mi@aldan.algebra.com>, <cvs-committers@FreeBSD.ORG>, <cvs-all@FreeBSD.ORG> Subject: Re: noatimedir? (Re: cvs commit: src/sys/vm vnode_pager.c) Message-ID: <20011014123818.J38047-100000@achilles.silby.com> In-Reply-To: <xzp7ktycmw4.fsf@flood.ping.uio.no>
next in thread | previous in thread | raw e-mail | index | archive | help
On 14 Oct 2001, Dag-Erling Smorgrav wrote: > Mike Silbersack <silby@silby.com> writes: > > An effective way to implement this might be to tune it based on the > > sysctl, but enable it only on partitions mounted softupdates or async. In > > either of those cases, you clearly can't count on *any* disk activity > > occuring on a predictable basis, so losing perfect accuracy on atimes is > > no big deal. I'd imagine that a setting of 60 seconds would provide > > reasonable accuracy while reducing atime updates a *lot*. > > Only on directories. If you're like me and update your CVS repo every > hour, and run 'cvs update' maybe a couple of times every day, a 60- > second hysteresis isn't going to make any difference to file atimes. > > DES I'm thinking more about the cvs server you and others are hitting simultaneously. For something like that (or an ftp server, or an http server), the effect might be more drastic. Hrm, have any atime vs noatime benchmarks been done? It's always mentioned in tuning guides, but I've never thought about how much of a performance difference it makes. Mike "Silby" Silbersack To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe cvs-all" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20011014123818.J38047-100000>