Just to close off this threadm, even if it's old. This is "only" a private/testing box, and I've been busy, so I've only been able to test stuff every now and then.
A few runs of memtest86 found no errors. I turned to the "md5sums of parts of disks" approach. If I read large chunks (5 GB), from different places on the disks, with 5+ iterations with each chunk, I got errors occasionally (diverging md5sums). But this is 10 disks across two controllers and all but two gave errors several times, albeit seldomly.
I started swapping hardware, and with different RAM I am OK. I guess clean runs of memtest shouldn't be trusted 100%. I can even say, the way these errors have crept up on me gradually over months(!), it means the RAM stick(s) have failed gradually, without being touched or anything. Scary!
-gaute --On 29. April 2009 08:19 +0200 Gaute Lund <gaute idrift no> wrote:
Thanks, and also to others who gave feedback. The approach with md5summing devices came from another source too, and I'll try a systematic approach as soon as time allows.