On Sun, September 7, 2014 9:55 pm, Keith Keller wrote: > On 2014-09-08, Valeri Galtsev <galtsev at kicp.uchicago.edu> wrote: >> >> Indeed, lucky me. As of this moment I have 6 of 9650 in production >> boxes. >> For at least 6 years. During which time none of them ever failed on me >> (including any trouble with arrays). Knocking on wood. > > You totally jinxed them! You'll probably have three of them fail in the > next month. ;-) > >> I must say though >> that I do prefer the most reliable drives. And I always have arrays >> checked at least once a week through 3ware scheduler (this causes walk >> through the whole surface of each of drives, thus ensuring bad blocks if >> any do not stay undiscovered...). > > I decided to do verifies once a month, instead of the default once a > week. My thinking was that hitting all of every drive so frequently > might be a wear factor, but periodic scrubs are still important; plus, > for my larger arrays, having performance slightly degraded 1/7 of the > time was not so desirable. So now I do 12 verifies a year instead of > 52. I think the verify has picked up maybe two errors in ~10 years. Then you too manage to stick with reliable drives! Performance... well, I have array verify scheduled to start 23:00 on Saturday... also, on newer controllers you can choose policy with highest priority for IO and lowest for rebuild etc. I'm kind of sceptical about extra wear to mechanical drives added by RAID check: my drives are always spinning full speed (no spin down crap, thank you), and heads are not touching the platters... So, I'm not certain there is extra wear. The only wear apart from bearings I know about is when arm hits the stopper which only happens when you power off the drive, which not the case here too. What do I miss? Valeri > (It does sometimes expose when a drive is failing, which is handy.) > > The LSI MegaRAID cards have their own scheduling, and I haven't had > time enough to read the manual to figure out how to set this (or indeed, > even to figure out what the schedule is; the MegaRAID UI is much more > arcane than the 3ware). > > --keith > > -- > kkeller at wombat.san-francisco.ca.us > > > _______________________________________________ > CentOS mailing list > CentOS at centos.org > http://lists.centos.org/mailman/listinfo/centos > ++++++++++++++++++++++++++++++++++++++++ Valeri Galtsev Sr System Administrator Department of Astronomy and Astrophysics Kavli Institute for Cosmological Physics University of Chicago Phone: 773-702-4247 ++++++++++++++++++++++++++++++++++++++++