From owner-freebsd-hackers Mon Apr 13 02:55:17 1998 Return-Path: Received: (from majordom@localhost) by hub.freebsd.org (8.8.8/8.8.8) id CAA18322 for freebsd-hackers-outgoing; Mon, 13 Apr 1998 02:55:17 -0700 (PDT) (envelope-from owner-freebsd-hackers@FreeBSD.ORG) Received: from dt050n33.san.rr.com (@dt050n33.san.rr.com [204.210.31.51]) by hub.freebsd.org (8.8.8/8.8.8) with ESMTP id CAA18302 for ; Mon, 13 Apr 1998 02:55:13 -0700 (PDT) (envelope-from Studded@san.rr.com) Received: from san.rr.com (Studded@localhost [127.0.0.1]) by dt050n33.san.rr.com (8.8.8/8.8.8) with ESMTP id CAA06038 for ; Mon, 13 Apr 1998 02:55:13 -0700 (PDT) (envelope-from Studded@san.rr.com) Message-ID: <3531E101.591C98CE@san.rr.com> Date: Mon, 13 Apr 1998 02:55:13 -0700 From: Studded Organization: Triborough Bridge & Tunnel Authority X-Mailer: Mozilla 4.05 [en] (X11; I; FreeBSD 2.2.6-STABLE i386) MIME-Version: 1.0 To: freebsd-hackers@FreeBSD.ORG Subject: "unofficial" buildworld stats with IDE + SCSI Content-Type: text/plain; charset=us-ascii Content-Transfer-Encoding: 7bit Sender: owner-freebsd-hackers@FreeBSD.ORG Precedence: bulk X-Loop: FreeBSD.ORG Without revisiting the previous debate regarding the proper way to set up a worldstone test, I thought I'd report the results of a test I did here locally involving relative disk speeds in case anyone's interested. I run -Stable, and am trying to decide how to spread my system across two disks since I just freed up some space on the old IDE disk. I have a WD Caviar 1.6G disk and an IBM DCAS 34300 4G SCSI UW drive. My main question was whether putting /usr/src or /usr/obj on the "slow" IDE disk would yield the greatest buildworld performance. I deleted my old sources (for an unrelated test, but the timing was good :), cvsup'ed -Stable and did the local modifications that I usually do. I don't build perl, xntpd (I use more up to date versions), qcam, uucp, or profiled libraries. I build everything with -O -pipe. I did the test the same way each time, with the same partitions on the IDE and SCSI disks as obj or src depending. Always starting with a clean /usr/obj I did 'make -DNOCLEAN -DCLOBBER buildworld.' I didn't use any log file for the build to avoid the extra variable. I have all the performance options for AHC in my kernel, /usr/src was mounted noatime and /usr/obj was mounted noasync,noatime. I used the same exact sources for each test, mv'ing them as needed. I did the test 3 ways: With /usr/src on IDE and /usr/obj on SCSI 2 hours 8 minutes With /usr/src on SCSI and /usr/obj on IDE 2 hours 6 minutes With both on SCSI 2 hours 14 minutes I imagine that if I had two fast SCSI disks those numbers would be a lot different. :) However I thought it was pretty interesting that it doesn't matter which is on the "slow" disk. In case anyone's wondering I did one installworld after the last buildworld and it took 14 minutes. That total time is virtually identical to the total make world time I've been seeing with mounting /usr/obj on the IDE disk for the last few weeks. I hope this is of use to someone. More details are available on request. Doug -- *** Chief Operations Officer, DALnet IRC network *** *** Proud operator, designer and maintainer of the world's largest *** Internet Relay Chat server. 5,328 clients and still growing. *** Try spider.dal.net on ports 6662-4 (Powered by FreeBSD) To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-hackers" in the body of the message