I thought the whole point of RAID was so that you didn't get failures like that?

(Don't worry, I know what you mean. I've had my own share of bad experiences with stripe sets.)

I try to avoid striping-only RAID sets. OK, you get large continuous logical volumes at expense of (slightly worse than) linear decrease in reliability (measured as probability of your filesystem staying with you (as opposed to bidding you farewell) at any given time interval). Mirroring is waste of space, Level 5 and similar incur performance penalty.... As Doug says, tanstaafl!

I don't even use RAID arrays for database storage. I prefer to hand-allocate volumes on continouos regions of physical drives (taking into account that seek times vary across the disk), manually assigning tablespaces, logspaces etc to them, and let DBMS manage mirroring (of course, where I can). More often than not I find after a while several heavily (and mostly concurrrently) used tables allocated into the same physical chunk of storage, competing for read head's attention. Ah, one never learns ones application's secrets completely.

I think that something like Veritas volume manager (with not only drive spanning, striping, mirroring, but hotspot detection and online reconfiguration) should be as standard part of every OS as text editor .

Now, was it OT or what?

Cheers!


Dragi "Bonzi" Raos
Zagreb, Croatia
Q#5196, MkII#80000376, 18GB green