From owner-freebsd-hardware Fri Aug 6 12:20:44 1999 Delivered-To: freebsd-hardware@freebsd.org Received: from sammy.tibco.com (sammy.tibco.com [192.216.111.146]) by hub.freebsd.org (Postfix) with ESMTP id 9117215681 for ; Fri, 6 Aug 1999 12:20:40 -0700 (PDT) (envelope-from aram@tibco.com) Received: from osgood.tibco.com (osgood.tibco.com [160.101.240.42]) by sammy.tibco.com (8.9.3/8.9.3) with ESMTP id MAA04917 for ; Fri, 6 Aug 1999 12:20:34 -0700 (PDT) Received: from venus.tibco.com (venus.tibco.com [160.101.240.40]) by osgood.tibco.com (8.9.3/8.9.3) with ESMTP id MAA29409 for ; Fri, 6 Aug 1999 12:20:53 -0700 (PDT) Received: from tibco.com ([160.101.22.192]) by venus.tibco.com (Netscape Messaging Server 3.6) with ESMTP id AAA6528; Fri, 6 Aug 1999 12:16:30 -0700 Message-ID: <37AB3571.5E1745CD@tibco.com> Date: Fri, 06 Aug 1999 12:20:17 -0700 From: "Aram Compeau" Organization: TIBCO X-Mailer: Mozilla 4.5 [en] (WinNT; I) X-Accept-Language: en MIME-Version: 1.0 To: Brian McGroarty Cc: freebsd-hardware@freebsd.org Subject: Re: The $500 Performance Question References: Content-Type: text/plain; charset=us-ascii Content-Transfer-Encoding: 7bit Sender: owner-freebsd-hardware@FreeBSD.ORG Precedence: bulk X-Loop: FreeBSD.org I'm sure you're correct in assuming that the compiles are disk I/O limited. I don't think this will come close to sliding under the $500 mark, or if you want to risk it, but tying two SCSI HD's together at RAID lvl 0 is a nice disk I/O speedup. :-) This has some rather obvious inherent risks involved with it... Not to mention a hefty "opportunity cost" in replacing "already paid for" IDE disks with SCSI disks... I think you could get 2 4.5GB IBM 9ES drives (DDRS-34560, I think?) for something like $150 ea. plus the controller? So maybe a 9GB build volume for $700 or so? Anyway, just guessing. I'll be trying a similar rig (dual CPU, SCSI RAID lvl 0) myself shortly, and I'll report back the "make world -j 40" time for reference. Barring such a solution, you might consider a slightly faster HD? There wasn't enough detail below to discern, but perhaps one of those nice IBM UDMA disks with very high areal densities and GMR heads *might* give you a performance boost? I believe if you compile in single user mode, you gain a slight performance boost as well (that's free, of course! :-). Aram Brian McGroarty wrote: > I've got a PC used primarily for programming. Projects tend to be large (8-12 > megs of C++ source), so build time is a concern. I'd like ideas on where the > best place to sink $500 would be to boost performance. > > Relevant in the current configuration: > > o (2) Celeron 300a (on socket converters, overclocked to 500mhz) > o Tyan Tiger 100 motherboard (Dual CPU) > o 512mb 100mhz RAM > > EIDE controller: > o 14 gig 7200 EIDE (/usr,/,swap) > o 28 gig 7200 EIDE (/tobackup,/cvs) > > EIDE controller 1: > o 14 gig 7200 EIDE (/home) > o 2/8x CDRW/CD-ROM > > For a familiar benchmark, a FreeBSD 'make world -j 40' takes about an hour > and ten minutes. This may be slewed against your ssytem by the inclusion of > -O3 optimization. > > The CPUs realize a lot of idle time; upward of 60%. I expect then that I/O is > my main bottleneck. > > The drives are Ultra-66 capable, but I don't believe FreeBSD supports this at > current. Thus, I don't see a way to enhance what I've got. (I'm already > enabling 32-bit and DMA on the controllers via flags). > > So what's my best bet? Is there a fast and economical SCSI-2 controller and > drive I should try? Any supported IDE RAID controllers? Or is there an > Ultra-66 controller FreeBSD merely sees as really fast EIDE? > > Or is this time being spent in the huge kernel lock? Would CAS2 capable RAM > then perhaps speed the buffer transfers noticably and get the CPUs back to > unmanaged portions more quickly? > > To Unsubscribe: send mail to majordomo@FreeBSD.org > with "unsubscribe freebsd-hardware" in the body of the message To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-hardware" in the body of the message