From owner-freebsd-smp Tue Jun 22 10: 2: 5 1999 Delivered-To: freebsd-smp@freebsd.org Received: from alpha.netaccess.on.ca (alpha.netaccess.on.ca [199.243.225.10]) by hub.freebsd.org (Postfix) with ESMTP id 29E6E152D2 for ; Tue, 22 Jun 1999 10:02:01 -0700 (PDT) (envelope-from rob@ControlQ.com) Received: from fatlady.controlq.com (dial203.nas.net [199.243.225.203]) by alpha.netaccess.on.ca (8.9.0/8.9.0) with SMTP id NAA23040; Tue, 22 Jun 1999 13:01:57 -0400 (EDT) Date: Tue, 22 Jun 1999 13:02:07 -0400 (EDT) From: "Robert S. Sciuk" To: Ian Grigg Cc: freebsd-smp@FreeBSD.ORG Subject: Re: Multi processor support? In-Reply-To: <199906221618.MAA15934@systemics.com> Message-ID: MIME-Version: 1.0 Content-Type: TEXT/PLAIN; charset=US-ASCII Sender: owner-freebsd-smp@FreeBSD.ORG Precedence: bulk X-Loop: FreeBSD.org On Tue, 22 Jun 1999, Ian Grigg wrote: > > > > > Heh. That wouldn't be very symmetric... > > > > > > No, but having processor affinity would be quite nice. There's > > > nothing wrong with MP. > > > > Processor affinity is more than nice. Work I've done on large HP boxes > > indicate that by putting producer/consumer processes on the same processor > > one avoids a lot of cache coherency issues, and increases bandwidth > > considerably. Admittedly, this is not true of all applications, but > > having the ability to have two or more processes share a processor can go > > a long way towards tuning application performance. > > Are you making the assumption that the two processes > are sharing memory for IPC? Or, maybe kernel threads? > > If not, I don't understand how you improve cache coherency > with separate data sets. > I'm not sure that IP type socket IPC would be appreciably helped here though it is possible that Unix _domain sockets_ may be implemented using mapped memory -- I'm not sure about FreeBSD, but it is possible (likely?) Guess I could check the sources, but I don't have time right now ... in the case of applications sharing mmap'ed memory ... certainly there is a speedup on certain platforms. > > I would suggest an > > ioctl or other API type interface, with a userland tool to assist ... > > But, I agree with the general context. Processor affinity > is nice, but should only be an option on top of SMP. I'd > rather have a good SMP than an MP any day. I can't disagree 8-). Cheers, Rob. -=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=- Robert S. Sciuk 1032 Howard Rd. PO Box 6A Ph:905 632-2466 Control-Q Research Burlington, Ont. Canada Fx:905 632-7417 rob@ControlQ.com L7R 3X5 http://www.ControlQ.com To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-smp" in the body of the message