From owner-chemistry # - at - # ccl.net Fri Jul 21 12:11:43 1995 Received: from gatekeeper.tripos.com for matt%!at!%metis.tripos.com by www.ccl.net (8.6.10/930601.1506) id MAA29210; Fri, 21 Jul 1995 12:05:40 -0400 Received: (from news&$at$&localhost) by gatekeeper.tripos.com (8.6.12/8.6.12) id LAA09931; Fri, 21 Jul 1995 11:04:36 -0500 Received: from tripos.tripos.com(192.160.145.1) by gatekeeper.tripos.com via smap (V1.3) id sma009929; Fri Jul 21 11:04:23 1995 Received: from metis.UUCP (metis.tripos.com [192.160.145.41]) by tripos.tripos.com (8.6.12/8.6.12) with SMTP id LAA17363; Fri, 21 Jul 1995 11:02:38 -0500 Received: by metis.UUCP (4.0/4.7) id AA10901; Fri, 21 Jul 95 11:02:32 CDT From: matt ^at^ metis.tripos.com (Matt Clark) Message-Id: <9507211602.AA10901-!at!-metis.UUCP> To: Charles Letner Cc: Computational Chemistry List , matt -8 at 8- tripos.tripos.com Subject: Re: CCL:Benchmarks In-Reply-To: Your message of Thu, 20 Jul 95 18:31:47 -0400. Date: Fri, 21 Jul 95 11:02:32 EDT Regarding what measurement to use in a benchmark, I feel you should measure what you are really interested in. Namely "how long does it take to run", or wall clock time. If you use CPU time you would add up the time for each individual CPU, and that isn't a very useful measure of the performance increase of a parallel computer. Wall clock time takes into account any paging/swapping/io that the program causes, which are important practical issues.