From owner-freebsd-current Fri Apr 18 22:48:12 1997 Return-Path: Received: (from root@localhost) by freefall.freebsd.org (8.8.5/8.8.5) id WAA27499 for current-outgoing; Fri, 18 Apr 1997 22:48:12 -0700 (PDT) Received: from mx.serv.net (mx.serv.net [199.201.191.10]) by freefall.freebsd.org (8.8.5/8.8.5) with ESMTP id WAA27490 for ; Fri, 18 Apr 1997 22:48:07 -0700 (PDT) Received: from MindBender.serv.net by mx.serv.net (8.7.5/SERV Revision: 2.30) id WAA29042; Fri, 18 Apr 1997 22:48:11 -0700 (PDT) Received: from localhost.HeadCandy.com (michaelv@localhost.HeadCandy.com [127.0.0.1]) by MindBender.serv.net (8.7.5/8.7.3) with SMTP id WAA17328; Fri, 18 Apr 1997 22:47:50 -0700 (PDT) Message-Id: <199704190547.WAA17328@MindBender.serv.net> X-Authentication-Warning: MindBender.serv.net: Host michaelv@localhost.HeadCandy.com [127.0.0.1] didn't use HELO protocol To: Warner Losh cc: current@freebsd.org Subject: Re: Speed deamons: How to build a build box? In-reply-to: Your message of Fri, 18 Apr 97 13:16:41 -0600. Date: Fri, 18 Apr 1997 22:47:44 -0700 From: "Michael L. VanLoon -- HeadCandy.com" Sender: owner-current@freebsd.org X-Loop: FreeBSD.org Precedence: bulk > I have a PPro 180 machine that is currently building world in about >5 hours +-. This time is from a "clean" state, not from a previously >built state. I've been told that it maybe possible to do this in 1:21. Unless there is something really huge in your source tree that isn't in mine, yours is taking waaay too long. > How do I morph the machine I have into a faster build box? >I have 64M memory, a SuperMicro 6SNS motherboard, 60ns memory, and one >SCSI disk (a Quantum Fireball 10.5ms 4500). A second disk (a jaz drive, >12ms 5400) didn't seem to make the build faster or slower. The MB has >built in SCSI "2940UW". The fireball is the "ultra" version only. If I remember right, there were some of these that were actually much less capable than their stand-alone counterparts (only allowed three outstanding transactions, or something like that). Wonder if you got one of those. >As I see it, I have the following avenues to make it faster: > 1) Buy more memory You have more than enough already. > 2) Buy 50ns memory (the board can cope) That shouldn't matter much. > 3) Buy a faster disk, ultrawide scsi 7200 or faster disk. This could help. But you'd be better off buying more smaller disks (4500-5400 rpm, not wide) and striping them. Do some people not listen? (See Joe Greco's millions of posts on this.) > 4) Overclock the 180 to 200 or 233 (I have a good heat sink and fan) Somebody wasn't paying attention when I said don't buy a multiple of 30. Sigh. A 166 would most likely have run faster. On the upside, you can probably overclock that chip to 200MHz. That will make a significant difference (though not more than 10%). > 5) Get a second scsi bus Couldn't hurt. > 6) ccd? Get the religion, brother! >Can those that have tweaked for this sort of thing help me out here? I'm >running -current as of April 17 (ctm cvs-cur 3222). >Also, what is the 1:21 number really? I'm at least one of the people who claimed a 1:21 time. Here's what I did... It was on NetBSD just before 1.2 was released -- not FreeBSD. There shouldn't be a significant difference, though. I did it on a 200MHz Pentium Pro, running in an Asus P6NP5 (Natoma/440FX) motherboard. Using 64MB of 60ns EDO SIMMs. I was running an Adaptec 2940UW (the real card, not a built-in chip), with tagged-command-queuing turned on in the kernel. All my binaries and sources were on one Seagate 2GB Barracuda -- very fast, and the object tree was writing to a mostly empty old Seagate 1GB 31200N -- not super fast (both drives were borrowed). My filesystems were async mounted. All my binaries, including the kernel, had been built previously with these flags (under gcc 2.7.2): -O6 -m486 -pipe -fno-strength-reduce. I also did the test build with the same flags. The object directory was totally empty at the beginning of the build. I let the build do all the standard parts (build standard, profiled, and shared libraries, format man pages, etc.), including make install. The build output was directed to a text file, and I checked the text file after the build to verify that there were no failures during the build which might have affected the time (and there were none). I was not running X, but was running only a text console during the build. I can't think of anything else relevant... Although I was running on borrowed, non-striped drives during these tests, my "production" system that I use for day-to-day work is ccd striped over four SCSI drives. FWIW... ----------------------------------------------------------------------------- Michael L. VanLoon michaelv@MindBender.serv.net --< Free your mind and your machine -- NetBSD free un*x >-- NetBSD working ports: 386+PC, Mac 68k, Amiga, Atari 68k, HP300, Sun3, Sun4/4c/4m, DEC MIPS, DEC Alpha, PC532, VAX, MVME68k, arm32... NetBSD ports in progress: PICA, others... -----------------------------------------------------------------------------