From owner-freebsd-current Mon Nov 17 09:57:33 1997 Return-Path: Received: (from root@localhost) by hub.freebsd.org (8.8.7/8.8.7) id JAA28007 for current-outgoing; Mon, 17 Nov 1997 09:57:33 -0800 (PST) (envelope-from owner-freebsd-current) Received: from rah.star-gate.com (rah.star-gate.com [204.188.121.18]) by hub.freebsd.org (8.8.7/8.8.7) with ESMTP id JAA28001 for ; Mon, 17 Nov 1997 09:57:30 -0800 (PST) (envelope-from hasty@rah.star-gate.com) Received: from rah.star-gate.com (localhost.v-site.net [127.0.0.1]) by rah.star-gate.com (8.8.8/8.8.5) with ESMTP id JAA02845; Mon, 17 Nov 1997 09:57:11 -0800 (PST) Message-Id: <199711171757.JAA02845@rah.star-gate.com> X-Mailer: exmh version 2.0gamma 1/27/96 To: Bob Bishop cc: current@FreeBSD.ORG Subject: Re: make world time???/ In-reply-to: Your message of "Mon, 17 Nov 1997 10:07:50 GMT." Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Date: Mon, 17 Nov 1997 09:57:10 -0800 From: Amancio Hasty Sender: owner-freebsd-current@FreeBSD.ORG X-Loop: FreeBSD.org Precedence: bulk Hi, I was just trying to find out about the scalability of doing make worlds . A couple of us are suspecting problems with the system: fast SMP systems VS. fast Uniprocessor systems with fast disk drives. Simon Shapiro first alluded to a problem and I wanted an independent confirmation . As for a benchmark or standard way of reporting "make world" it is a a good idea -- to track the system performance or code bloat in the standard distribution. We need a short readme file or FAQ entry: kernel configuration , in the case of scsi it is important to have the following options: options AHC_TAGENABLE options AHC_SCBPAGING_ENABLE options AHC_ALLOW_MEMIO If we have any options to enable dma on fast ide drives it should included as a suggestion on how to speed up ide drives. The partition(s) should be mounted async, noatime: mount -u -o async,noatime /obj mount -u -o async,noatime /usr Yes, there is a danger in setting up /usr async do a make clean -- to eliminate the variable of some cleaning an empty /obj vs a populated /obj Include the output of dmesg /etc/make.conf options: CFLAGS= -O -pipe NOPROFILE= true I never use the profiled libraries and it seems to me that this should be the default given that most users would probably not know how to use them anyhow. Since most of us prefer the above CFLAGS options it should be default or does anyone have any objects as to why "-O -pipe" should not be the default. Next, we should have a web page where people can post their benchmarks. This will help newbies to determine the best configurations at least with respect to a build machine. Amancio > >[Various people on what should comprise the definitive worldstone...] > > Unless everyone has identical configurations (which would kinda miss the > point), it's not going to be possible to level the playing field. Eg: > > - -pipe and/or -jx don't help unless you have enough real memory > > - multiple spindles seem to help a lot: I got well under 2h (not accurately > timed, unfortunately) burning in a K6-200 with 2x4.4G Quantum Atlas on a > 2940UW > > - do the rules allow turning off daemons to free up memory while building? > > Also, the target time (whatever that is) is going to change over time. I've > seeing significant changes in build time on the same machine with the same > options over a period of months: I may post something on this subject. > > > -- > Bob Bishop (0118) 977 4017 international code +44 118 > rb@gid.co.uk fax (0118) 989 4254 between 0800 and 1800 UK > >