From owner-freebsd-hackers Fri Jun 2 22:41:09 1995 Return-Path: hackers-owner Received: (from majordom@localhost) by freefall.cdrom.com (8.6.10/8.6.6) id WAA00425 for hackers-outgoing; Fri, 2 Jun 1995 22:41:09 -0700 Received: from gndrsh.aac.dev.com (gndrsh.aac.dev.com [198.145.92.241]) by freefall.cdrom.com (8.6.10/8.6.6) with ESMTP id WAA00369 for ; Fri, 2 Jun 1995 22:41:01 -0700 Received: (from rgrimes@localhost) by gndrsh.aac.dev.com (8.6.11/8.6.9) id WAA09481; Fri, 2 Jun 1995 22:40:33 -0700 From: "Rodney W. Grimes" Message-Id: <199506030540.WAA09481@gndrsh.aac.dev.com> Subject: Re: A performance mystery To: davidg@Root.COM Date: Fri, 2 Jun 1995 22:40:33 -0700 (PDT) Cc: henrich@crh.cl.msu.edu, freebsd-hackers@freebsd.org In-Reply-To: <199506030517.WAA00138@corbin.Root.COM> from "David Greenman" at Jun 2, 95 10:17:02 pm X-Mailer: ELM [version 2.4 PL24] MIME-Version: 1.0 Content-Type: text/plain; charset=US-ASCII Content-Transfer-Encoding: 7bit Content-Length: 1627 Sender: hackers-owner@freebsd.org Precedence: bulk > > >> > > 16 8192 3360694 2890287 > > ^^^^^^^ > > > >How come *this* number and the number below are different by a factor > >of 7??? It should have been the same. ``iozone 16 8192'' and the > >above line from ``iozone auto'' do exactly the same things yet we > >are seeing this *huge* descrepancy. > > Because there isn't enough 'free' memory at the time that he ran the test > to cache the entire file. Since it's read sequentially and the cache policy > for something like this is more or less LRU, it get pushed out of the cache > before it is read back. Then I have doubts about all of his benchmark results if he has not truely been dedicating the machine to the task being benchmarked :-(. He wants to know why one machine is slow than the other, and every set of numbers he produces says the opposite should be true, until these numbers came in. Note that he also say very low numbers all the way down the test, so I do suspect you are right, he had something else sucking up his memory. Both machines where heavy on memory (one was 64MB the other 32MB) so he should *easily* be able to get a 16MB fast iozone results, and infact his singal run of ``iozone 16 8192'' did show the good number. I think I am just going to leave this to him to sort out, as he can not produce meaniful benchmark results if he does not duplicate environments correctly :-(. Thanks for playing :-) :-) -- Rod Grimes rgrimes@gndrsh.aac.dev.com Accurate Automation Company Custom computers for FreeBSD