From owner-freebsd-smp Wed Apr 23 12:02:21 1997 Return-Path: Received: (from root@localhost) by hub.freebsd.org (8.8.5/8.8.5) id MAA17184 for smp-outgoing; Wed, 23 Apr 1997 12:02:21 -0700 (PDT) Received: from bagpuss.visint.co.uk (bagpuss.vis.net.uk [194.207.134.1]) by hub.freebsd.org (8.8.5/8.8.5) with ESMTP id MAA17176 for ; Wed, 23 Apr 1997 12:02:17 -0700 (PDT) Received: from bagpuss.visint.co.uk (bagpuss.vis.net.uk [194.207.134.1]) by bagpuss.visint.co.uk (8.7.5/8.7.3) with SMTP id UAA14534; Wed, 23 Apr 1997 20:10:06 +0100 (BST) Date: Wed, 23 Apr 1997 20:10:06 +0100 (BST) From: Stephen Roome To: Steve Passe cc: freebsd-smp@freebsd.org Subject: Re: hardware list In-Reply-To: <199704231651.KAA22820@Ilsa.StevesCafe.com> Message-ID: MIME-Version: 1.0 Content-Type: TEXT/PLAIN; charset=US-ASCII Sender: owner-smp@freebsd.org X-Loop: FreeBSD.org Precedence: bulk On Wed, 23 Apr 1997, Steve Passe wrote: > Hi, > > What sort of kernel compile times should be expected, with my last compile > > > > make -j8 it took 3 minutes and 48 seconds for a Generic kernel with > > the sysv stuff for X, and none of the drivers I'm not using. > > (i.e. one net card a 2940 and 2 processors, nothing much else but 1 com > > port for a mouse) > > > > and make -j16 took 3 minutes 19 seconds on the same > > kernel. > > > > So, are these good times then ? > > they're better than I got from my dual P5-133 with slow SCSI disk: > make -j8 4min 20sec > > for a fairly dramatic speedup add this line to your /etc/make.conf file: > > COPTFLAGS= -pipe (Oh dear, you've got me going on the 'benchmark your computer' thing again) I have this at the moment CFLAGS=-O6 -pipe -m486 #COPTFLAGS=-O -pipe notice this is commented out. HAVE_FPU=yes right now, and that's what the last two compiles ran under, I'd be surprised though if your disk is slower than this Quantum Fireball here. with iozone 51 I get: IOZONE performance measurements: 2894335 bytes/second for writing the file 4024164 bytes/second for reading the file That's one Quantum Fireball 3.5 series disk, which isn't much good by my standards. Besides, I haven't mounted it async,noatime yet, I'm just testing these speed as a plain drive.. I'll try freaky mount options later =! (I'll be adding a 2gb seagate hawk as well soon.) Anyway, I just chucked COMPAT_LINUX in as well and this is what I get ------ CFLAGS =-06 -pipe -m486 COPTFLAGS commented out of make.conf, HAVE_FPU=yes Wed Apr 23 19:12:59 BST 1997 start time = 55 second make depend Wed Apr 23 19:13:54 BST 1997 end of make depend = 3 minutes 36 seconds for make -j8 Wed Apr 23 19:17:30 BST 1997 end of make -j8 ----- This was interesting so I put the COPTFLAGS back in the make.conf and tried again. This time: ----- CFLAGS =-O6 -pipe -m486 COPTFLAGS = -O -pipe HAVE_FPU =yes Wed Apr 23 19:24:50 BST 1997 start make = 51 second make depend Wed Apr 23 19:25:41 BST 1997 end of make depend = 3 minutes 24 seconds for make -j8 Wed Apr 23 19:29:05 BST 1997 end of make -j8 Not a huge improvement, only -slightly- (worth it?) better. ----- Now, this (a make -j16) is interesting though: CFLAGS =-O6 -pipe -m486 COPTFLAGS=-O -pipe HAVE_FPU=yes Wed Apr 23 19:34:14 BST 1997 start make = 1 minute 1 second for a make depend Wed Apr 23 19:35:05 BST 1997 end of make depend = 3 minutes 27 second for make -j16 Wed Apr 23 19:38:33 BST 1997 -j16 doesn't seem to make any difference ? I can't find anything to say make doesn't accept more than 8 jobs max.. So I wondered what -j64 would do !? I assume that for a machine with 512MB of memory -jBIGNUMBER would be good.. Anyway, I'm trying these things out, if you have some more interesting tests please let me know! This is from a make -j64, obviously the -j64 happened, and was too much for this machine on memory (can't easily check this while timing it.) ----- Wed Apr 23 19:41:43 BST 1997 start make = 52 seconds for the make depend Wed Apr 23 19:42:35 BST 1997 end make depend = 4 minutes 42 seconds for a make -j64 Wed Apr 23 19:47:17 BST 1997 ----- I'm gonna write a script that just does this with as many different settings and see if I get the improvements I expect. (could do some nice graphs perhaps.) Then it'll be obivous where too little memory becomes the major factor and where disk access is slowing it down much. (Although it won't be tonight as I'm going to leave my other 4 FreeBSD machines running scripts to get random files (quickly!) through httpd from this server to see if I can tweak the settings for and number of httpd's to run!) Just for fun make -j200 took 53 secs for make depend, 5 min 50 secs for the make =) (mmm maybe it's not memeory slowing this down?) Perhaps make should have a -J option saying start as many processes as possible without getting the swap space usage to increase, and with some limit on the amount of process swapping that ends up happening. Steve Roome (Anyone interested in a unix benchmark site or can point me at one?) -- Steve Roome Technical Systems Manager, Vision Interactive Ltd. E: steve@visint.co.uk M: +44 (0) 976 241 342 T: +44 (0) 117 973 0597 F: +44 (0) 117 923 8522