On Oct 6, 2011 9:06 PM, "Mark Knecht" wrote: > > On Thu, Oct 6, 2011 at 1:28 PM, Michael Mol wrote: > > On Thu, Oct 6, 2011 at 4:21 PM, Mark Knecht wrote: > >> On Thu, Oct 6, 2011 at 1:03 PM, Paul Hartman > >> My worry was that if the mdraid daemon saw one drive gone - either > >> when starting to spin down or when one spins up slowly - and if mdraid > >> didn't understand that all this stuff was taking place intentionally > >> then it might mark that drive as having failed. > > > > Does mdraid even have an awareness of timeouts, or does it leave that > > to lower drivers? I think the latter condition is more likely. > > > > I suspect, though, that if your disk fails to spin up reasonably > > quickly, it's already failed. > > > > In general I agree. However drives that are designed for RAID have a > feature known as Time Limited Error Recovery (TLER) which supposedly > guarantees that they'll get the drive back to responding fast enough > to not have it marked as failed in the RAID array: > > http://en.wikipedia.org/wiki/Time-Limited_Error_Recovery > > When I built my first RAID I bought some WD 1TB green drives, built > the RAID and immediately had drives failing because they didn't have > this sort of feature. I replaced them with RAID Edition drives that > have the TLER feature and have never had a problem since. (Well, I > actually bought all new drives and kept the six 1TB drives which I'd > mostly used up for other things like external eSATA backup drives, > etc...) > > Anyway, I'm possibly over sensitized to this sort of timing problem > specifically in a RAID which is why I asked the question of Paul in > the first place. My first RAID was with three Seagate economy 1.5TB drives in RAID 5, shortly followed by three 1TB WD black drives in RAID 0. I never had the problems you describe, though I rebuit the RAID5 several times as I was figuring things out. (the 3TB RAID0 was for some heavy duty scratch space.)