Similar findings here, where the first batch of drives in our hadoop clusters were WD greens. smartctl shows huge Load_Cycle_Count numbers for those drives which have been in service for a while (and they do indeed keep us busy with RMAs). Eventually we found this utility which can disable the absurdly short head-park timer from within linux:
http://idle3-tools.sourceforge.net/ which might help, but I wouldn't choose to buy these drives again for this purpose. Graham On Thu, Oct 04, 2012 at 11:07:03AM -0600, Doug Johnson wrote: > Greetings, > > I have built many RAID systems using desktop disks and they are > generally quite stable. One of the issues with WD drives are with their > Green drives. By default, they park the heads after ~8 seconds of > inactivity. This will cause them to drop out of the array. The disk > firmware can be configured to disable this feature. I have had very good > luck with Samsung and Seagate drives. I have never spent the money on > Enterprise level drives. > > BTW: The head parking feature is notoriously bad. If you calculate 1 > park every 8 seconds. At constant usage, the drive will die in about 3 > months. Perhaps this feature is fine for a desktop, but it is not for a > server or production environment. -- ------------------------------------------------------------------------- Graham Allan - I.T. Manager - al...@physics.umn.edu - (612) 624-5040 School of Physics and Astronomy - University of Minnesota -------------------------------------------------------------------------