On Thu, 21 Jul 2005, Erik Mouw wrote:
On Wed, Jul 20, 2005 at 11:32:33PM -0700, Juan Ortega wrote:Why dont you simply use memtest86?Because memtest86 doesn't catch all memory errors, while a kernel compile stresses the system in ways memtest86 doesn't. Some memory errors only happen when there is IO going on at the same time, which is exactly what a kernel compile does. For more information, see the sig11 FAQ at http://www.bitwizard.nl/sig11/ .
Well, to be fair to memtest86+, we have so far only found one repeatablecrash situation that we could pinpoint to faulty RAM that memtest86+ didn't find, and in that case it could just be a case of not running it for long enough (I had a service technician on site on other duty anyway).
This under a couple of dozen finds of faulty RAM. But if it is memory controller timings etc, they are much harder to find or trigger than pure bit errors.
One good other test to run is bonnie -f, because then all available ram will be used up as disk cache, at the same time as you have IO load.
/Mattias Wadenstein - with 1.5TB of ram, you learn to find errors..