On Wed, 15 Apr 2009, John Hearns wrote:
I politely disagree. We always built cluster head nodes with a mirrored disk pair for the systems disk, and a RAID 5 or 6 set with the rest of the disks in the chassis. If possible, you want a sensibly sized disk for your system disk - IMHO running terabyte-sized SATA drives as the OS drives on a system is inviting trouble. Get something small enough that you can readily back it up, and make sure it is mirrored. Then let the big data store look after itself - format it with XFS, run LVM on it, run RAID 6 on it. And if the worst happens and you lose your system disks, or indeed more likely you want to do a systems update, you UNLATCH A DRIVE AND PUT IT ON THE SHELF. Then you merrily update/patch/generally fool with the other drive (plus a replacement spare) and you have an instant fallback plan if things go pear-shaped. Sorry - I just don't buy having a RAID6 across all disks in one chassis. Unless you are booting over a SAN or something.
When management was breathing down my neck to maximize available storage, a RAID 6 across the six drives in the chassis seemed like the best option. There was certainly a lack of planning involved in this cluster, so many decisions were made in too much of a hurry.
-- Matt It's not what I know that counts. It's what I can remember in time to use. _______________________________________________ Beowulf mailing list, [email protected] sponsored by Penguin Computing To change your subscription (digest mode or unsubscribe) visit http://www.beowulf.org/mailman/listinfo/beowulf
