From owner-freebsd-current Wed Mar 11 19:08:44 1998 Return-Path: Received: (from majordom@localhost) by hub.freebsd.org (8.8.8/8.8.8) id TAA14309 for freebsd-current-outgoing; Wed, 11 Mar 1998 19:08:44 -0800 (PST) (envelope-from owner-freebsd-current@FreeBSD.ORG) Received: from godzilla.zeta.org.au (godzilla.zeta.org.au [203.2.228.19]) by hub.freebsd.org (8.8.8/8.8.8) with ESMTP id TAA14299; Wed, 11 Mar 1998 19:08:35 -0800 (PST) (envelope-from bde@godzilla.zeta.org.au) Received: (from bde@localhost) by godzilla.zeta.org.au (8.8.7/8.8.7) id OAA12807; Thu, 12 Mar 1998 14:04:22 +1100 Date: Thu, 12 Mar 1998 14:04:22 +1100 From: Bruce Evans Message-Id: <199803120304.OAA12807@godzilla.zeta.org.au> To: hasty@rah.star-gate.com, jkh@FreeBSD.ORG Subject: Re: my worldstone Cc: current@FreeBSD.ORG Sender: owner-freebsd-current@FreeBSD.ORG Precedence: bulk X-Loop: FreeBSD.ORG >So, in the spirit of getting specific, I'd like to suggest the >following set of prototype ground rules to anyone daring to post their >"worldstone rating" here: > >Before beginning a make world test, the following conditions must be >met: > >1. /usr/obj must be empty (no "clean" pass to add minutes to time). > >2. /usr/obj *must* be mounted async (too few mount it sync and it does > bestow a large advantage). It should be mounted -noatime too, unless you want to throw away "free" optimizations. >3. /usr/src must reside on a physically different disk (unless ccd > is in use, in which case this should be noted). I know that's hard > for some folks, but if you're running off a single drive then your > numbers just won't be competetive. /usr/src should probably be mounted -noatime too. My numbers are competitive (1:37 on a K6/233 FIC-2007 64MB with 1 Quantum ST IDE drive in DMA mode, and /usr/src/Makefile tweaked to build al tools static, and DESTDIR on the same file system as /usr/obj (async,noatime,not-the-root-fs)). makewhorldstone should not be disk intensive on systems with enough RAM. If I had more time, then I would repeat the benchmark with the drive in PIO mode to prove how unimportant DMA mode and fast disk seek times are for non-disk-intensive non-seek-intensive benchmarks like this. The more useful "makeahllstone" is disk intensive, since it usually spends most of its time traversing trees to find nothing to do. >4. -pipe must be added to CFLAGS in /etc/make.conf. *no other build > optimizations or omissions can be done!* Same goes for additions - > no kerberos bits enabled. > >5. Source tree must be -current to within 7 days of the other people > comparing numbers. No point in compiling old bits just before > perl5 entered the tree, or whatever, since that obviously skews the > numbers. Building an agreed-on old version would be better (check one out from cvs). This would defeat my Makefile optimizations :-). >... 8(?) $TMPDIR should probably be on an mfs. I haven't tried this yet. K6/233: 5273.18 real 3496.20 user 993.51 sys 8620 maximum resident set size 605 average shared memory size 608 average unshared data size 130 average unshared stack size 5923250 page reclaims 18460 page faults 0 swaps 50889 block input operations 28885 block output operations 0 messages sent 0 messages received 8 signals received 384427 voluntary context switches 250941 involuntary context switches Bruce To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-current" in the body of the message