Date: Sun, 11 Apr 2004 06:35:08 +0100 From: Colin Percival <colin.percival@wadham.ox.ac.uk> To: Poul-Henning Kamp <phk@phk.freebsd.dk> Cc: current@freebsd.org Subject: Re: Optimizing RCng execution speed ? Message-ID: <6.0.1.1.1.20040411061504.03923940@imap.sfu.ca> In-Reply-To: <11095.1081621779@critter.freebsd.dk> References: <11095.1081621779@critter.freebsd.dk>
next in thread | previous in thread | raw e-mail | index | archive | help
At 19:29 10/04/2004, Poul-Henning Kamp wrote: >Is anybody working on optimizing RCng execution speed ? > >I think our boottime is getting a bit too slow these days... One obvious optimization would be to avoid processing /etc/rc.subr many times; on my 2.8GHz system running FreeBSD 5.2.1, this reduces RCng time from 4.8 seconds down to 4.0 seconds: Original times (with http://www.daemonology.net/tmp/src.rc.cp_time.diff) : Starting RC scripts: kern.cp_time: 2 0 69 0 62 RC scripts done: kern.cp_time: 74 0 442 4 863 After applying http://www.daemonology.net/tmp/src.rc.subr.diff : Starting RC scripts: kern.cp_time: 2 0 70 1 61 RC scripts done: kern.cp_time: 60 0 400 3 715 Most of the idle time is due to HTT, but there is still 1.5 seconds when both logical processors are idle (presumably while waiting for data from IDE); launching rcNG scripts in parallel would presumably help with this. But it's a bit silly to optimize a 4 second startup time; could someone running 5.x on slower hardware send me their cp_time values with and without the second patch? Colin Percival
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?6.0.1.1.1.20040411061504.03923940>