On Thursday 16 August 2007 21:18:38 Tracy R Reed wrote: > Dexter Filmore wrote: > > Well... that Google report hands misunderstandings on a plate, alone > > about how temperature affects disk life. > > Have you done a study over thousands of drives like they have? Why > should we believe you over google who actually has data?
I don't say that the google data are wrong, I say that report has been misinterpreted quite often. That report for example can lead to believe that disk cooling was actually a bad idea, which obviously is bogus. > > In general one would think an array of cheap disks would provide more > > security since one failed disk in a raid5 can be replaced and all is well > > again, and that's just the point: this only works when you _know_ a disk > > failed and from my experience some ATA drives don't even recognize if you > > stab them in the eye, chop off a leg and kick them square in the balls. > > To help with this I have smartd do a short test on each disk every day > and a long test every Sunday morning. I haven't had it find any errors > yet but I do whatever I can to exercise the disk thoroughly. SMART isn't > perfect but it is better than nothing and there is a group over at UCSD > doing research to make it better. Well, in my case it wasn't better than nothing. (And in a couple of others as well btw.) You wanna be sure your disk is alright dd it to /dev/null every once in a while. > > (On another note I noticed that RAID has it tripwires elsewere and linux > > software raid particularly blows chunks big style. Check your disk for > > bad block regularly or you're in for a surprise when the day come you > > really need to resync.) > > I have had very good success with Linux software RAID. I have had many > disks fail under it over the years and never lost data. One thing that > the CMU paper tells us is that RAID 5 isn't as reliable as we would > think. There is a much greater chance of a double disk failure than we > would expect. And what you say about not being able to detect a bad disk > has merit here. If we have a disk failing but don't know it and then > have another disk fail which we do find out about we will have a big > problem when we replace the failed disk we do know about and try to > rebuild. With a mirror setup the changes of this are less likely. This > combined with RAID 5's inherent performance problems and I always go > with mirroring these days. An economic RAID5 array has 4-8 disks, beyond that the double failure risk increases again and raid6 would make more sense. When mirroring this we're at possibly 16 drives. That's not really fun. And here a hardware raid5 controller like the 3ware Escalade 9000 series is the better choice, on resyncing if encountering bad blocks it fills the missing bad blocks with zeroes. So you still lose data but at least the array is rebuilt - and you get rid of the perf issues with soft raid. Dex -- -----BEGIN GEEK CODE BLOCK----- Version: 3.12 GCS d--(+)@ s-:+ a- C++++ UL++ P+>++ L+++>++++ E-- W++ N o? K- w--(---) !O M+ V- PS+ PE Y++ PGP t++(---)@ 5 X+(++) R+(++) tv--(+)@ b++(+++) DI+++ D- G++ e* h>++ r* y? ------END GEEK CODE BLOCK------ http://www.stop1984.com http://www.againsttcpa.com -- [email protected] http://www.kernel-panic.org/cgi-bin/mailman/listinfo/kplug-list
