From owner-freebsd-hackers Mon Jul 24 19:26:20 1995 Return-Path: hackers-owner Received: (from majordom@localhost) by freefall.cdrom.com (8.6.11/8.6.6) id TAA01749 for hackers-outgoing; Mon, 24 Jul 1995 19:26:20 -0700 Received: from gndrsh.aac.dev.com (gndrsh.aac.dev.com [198.145.92.241]) by freefall.cdrom.com (8.6.11/8.6.6) with ESMTP id TAA01739 for ; Mon, 24 Jul 1995 19:26:15 -0700 Received: (from rgrimes@localhost) by gndrsh.aac.dev.com (8.6.11/8.6.9) id TAA20558; Mon, 24 Jul 1995 19:25:43 -0700 From: "Rodney W. Grimes" Message-Id: <199507250225.TAA20558@gndrsh.aac.dev.com> Subject: Re: Who killed the cache? To: ianh@mincom.oz.au Date: Mon, 24 Jul 1995 19:25:43 -0700 (PDT) Cc: freebsd-hackers@freebsd.org In-Reply-To: <199507242343.AA23551@saturn.mincom.oz.au> from "Ian Holland" at Jul 25, 95 09:43:12 am X-Mailer: ELM [version 2.4 PL24] MIME-Version: 1.0 Content-Type: text/plain; charset=US-ASCII Content-Transfer-Encoding: 7bit Content-Length: 3049 Sender: hackers-owner@freebsd.org Precedence: bulk > > > I thought I'd relate an experience that I've just had that may > help others, as well as illuminate the vagarities of the hardware > that FreeBSD has to put up with. The system I refer to is on > cheap (relatively) hardware running FreeBSD 1.1.5. > > A while ago, I asked the questions list for possible causes of > stray interrupts (especially during gcc execution) and occasional > system reboots (not panics). > > The concensus appeared to be bad cache chips, or even a poorly > designed motherboard. So, being the optimistic type, I tried > disabling the cache, and things appeared to work okay. > > After several months of procrastination, I decided that I'd replace > the cache, but wanted to totally convince myself that it *was* the > cache. So I disabled the cache and began to thrash the machine > (by repeatedly recompiling a shallow source tree). After six > recompilations, gcc fell over with an interrupt. This was a tad > surprising, and I must admit, a bit disappointing. > > By this stage I was beginning to resign myself to parting with > some "readies" for replacement components, when I remembered that > the system appeared more stable during winter (that's now folks). > So, off came the cover, out came the pedestal fan, and with a > large box and a bit off counterbalancing, I had an external cooling > system. > > Ran the tests again, and after 30 odd cycles, my confidence was growing. > Out came xv and xboard, still with the compilations in the background. > After 300+ recompilations I was feeling a bit cocky. The upshot being > that I thrashed the machine like it ain't been thrashed before, and it > just sailed on through with nary a wimper. > > Now, all I need to do is find a more permanent cooling system. If you have a good DMM check the output of your power supply, if it is not withing 4.875 to 5.125V on +5 and 11.5 to 12.5V on +12 replace it. Though the components are spec'ed +-5% VCC by the time you get power distributed accross the board it can be quite a ways from what you measure at the power connector. A power supply running at 5.25V causes excessive component heat, and with a marginal timing in the design can often cause really strange behavior. A power supply running at 4.75V pushes components towards there slow timing spec, and thus the operate better if the temp is kept down. So either a supply to high or to low can cause what appear to be temperature related problems. A high quality power supply will be within 4.95 to 5.05V and 11.9 to 12.1V, a good heavy 2oz copy power/ground plane pair in a MB will have less than 0.05V loss accross the whole board. One other really rare one I have seen (only at TRW where they have some rather large current drawing dual ethernet cards and they stack 4 of them in a box) is the power supply connector at the motherboard litterly burned due to current overload :-). -- Rod Grimes rgrimes@gndrsh.aac.dev.com Accurate Automation Company Reliable computers for FreeBSD