[Gllug] Two Queries

John Hearns john.hearns at clustervision.com
Mon Jul 5 10:05:03 UTC 2004

On Mon, 2004-07-05 at 10:42, Huw Lynes wrote:

> we've always found that firmware revisions on the disks make a bigger
> difference than filesystem type. Disks are going to fail whatever you do. Plan
> for it and move on. Plus disks are cheap. You are going to be better off
> picking the filesystem that gives you the best performance for your particular
> usage than worrying about whether it is going to wear out your disks.
I agree. On a server you should be having mirrored system disks.

For render nodes, or nodes in a compute farm cheap IDE disks are fine.
Just arrange for the swap process to be quick and painless.
In our case, bring down the node, swap the disk and a network install
will format and install to the new disk.

At the other end, think of large storage arrays. I once went to a
seminar by someone in Compaq's storage division, who said hsi dream was
to see a wall of disk drives, and when one failed there is a whirring
noise and the disk pops out. Push in a new one and the free space starts
being used by the array.

Gllug mailing list  -  Gllug at gllug.org.uk

More information about the GLLUG mailing list