Date: Fri, 12 Jan 2001 19:16:53 -0600 (CST) From: Mike Meyer <mwm@mired.org> To: "Matt Wilbur" <matt@efs.org> Cc: questions@freebsd.org Subject: Re: FreeBSD SMP benchmarks / processing cluster Message-ID: <14943.44165.918182.911115@guru.mired.org> In-Reply-To: <19204041@toto.iv>
next in thread | previous in thread | raw e-mail | index | archive | help
Matt Wilbur <matt@efs.org> types: > Hello, > > We're costing out building a N node processing cluster (values of N starting > between 20 and 30) and are planning to use FreeBSD as the OS for each node. > One question we're dealing with is whether we're better off with N 1 CPU > systems, N 2 CPU systems, or N/2 2 cpu systems.. I don't think SMP systems > will give enough performance increase to warrant halving the number of > nodes.. Can anyone point me towards *any* benchmarks of FreeBSD's SMP > performance, be it FreeBSD vs. linux, FreeBSD SMP vs FreeBSD 1 CPU .. ? There have been a couple mentioned on the list; you might try checking the archives if you haven't already. > The codes we've ported over so far (from Irix) are scientific applications > we plan to run serially ... no PVM, no MPI, just scripts sending the jobs > out and validating/collecting the results .... We have this all running on > 4 1GHz athlons in standard cheese-o $950 desktop systems now.. It's sort of hard to comment on which is going to be better without knowing about the application. If it's CPU bound, and you can get enough memory in the box to keep it CPU bound with more processors, then you should be able to run N copies in one box in the same time it would take to run N copies in N boxes, which would appear to be a win. Setiathome is a good example. I run two copies on a dual CPU box, and each one runs as fast as if it were the only one on a single-processor box. I suspect I could run one copy/CPU until I started running out of RAM. Another issue to look at is the CPU cache. I was amazed at how much improvement I got out of setiathome when I went from 512K to 2M of cache on each CPU. The running time dropped from ~10.5 hours to ~6 hours. Ok, part of it was due to going to 450MHz from 400MHz, but that's only a small part of that increase. For 20 or 30 of them, it might be worthwhile benchmarking it on systems with more cache to see how things change. <mike -- Mike Meyer <mwm@mired.org> http://www.mired.org/home/mwm/ Independent WWW/Perforce/FreeBSD/Unix consultant, email for more information. To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-questions" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?14943.44165.918182.911115>