From owner-freebsd-current Mon Aug 25 16:29:33 1997 Return-Path: Received: (from root@localhost) by hub.freebsd.org (8.8.7/8.8.7) id QAA20079 for current-outgoing; Mon, 25 Aug 1997 16:29:33 -0700 (PDT) Received: from sendero-ppp.i-connect.net (sendero-ppp.i-Connect.Net [206.190.143.100]) by hub.freebsd.org (8.8.7/8.8.7) with SMTP id QAA20068 for ; Mon, 25 Aug 1997 16:29:30 -0700 (PDT) Received: (qmail 24609 invoked by uid 1000); 25 Aug 1997 23:29:46 -0000 Message-ID: X-Mailer: XFMail 1.2-alpha [p0] on FreeBSD X-Priority: 3 (Normal) Content-Type: text/plain; charset=us-ascii Content-Transfer-Encoding: 7bit MIME-Version: 1.0 In-Reply-To: <199708252251.QAA18963@Ilsa.StevesCafe.com> Date: Mon, 25 Aug 1997 16:29:46 -0700 (PDT) Organization: Atlas Telecom From: Simon Shapiro To: Steve Passe Subject: Re: Make and SMP - what can be done ? Cc: nnd@itfs.nsk.su, current@FreeBSD.ORG Sender: owner-freebsd-current@FreeBSD.ORG X-Loop: FreeBSD.org Precedence: bulk Hi Steve Passe; On 25-Aug-97 you wrote: > Hi, > > > > As a result my last successfull 'make -j12 buildworld' > > > produced 170.9% processor's usage and takes 2:54:00 as > > > opposed to 104.1% and 4:34:23 for 'make buildworld' > > > (without any patches). > > > > I can't help it :-)) It takes only between 1:26 and 2:36 to do make > > world > > here. On a P6-200 UP. > > first, why the wide range? Depends if I am doing make in the kernel, another make rleease, dump the entire system over NFS, or something ugly like that. > extrapolating these numbers the 2:36 would go down to 1:39, and the 1:26 > would go down to 0:55! then when we turn a 4 CPU box loose on it... I do not remember the (fat thumb) rules for compilation, but for RBMS work, figure out 1MB memoy, 1MB/Sec memory bandwidth and 0.5MB/Sec disk I/O per 1 MIPS. Sorry, they are now called SpecInt or something like that. So, even with 2 CPU's we will probably be diskbound. To eliviate that, we need to integrate the DPT code that allows RAID arrays to span controllers, or at least configure ccd over two RAID-0 arrays. Another possibility is to make my up and coming DBFS a real filesystem. It has the aility to span across devices. With some work, it could stripe across devices. How would one manage such configuration, is a mystery. Simon