[lug] Reliable SATA server?

[Rob Nagler]

Hi Dan,

> that more than 10% of IO wait continuously means you have a problem.

This is a disk intensive process (backups), that is, rm, tar, and/or
cp.  The only thing it should be doing is waiting on disks.
Controller caches can't improve the performance, because simultaneous,
continuous, non-sequential reads/writes are running as fast as the
controller/disks/bus can go. Caches would most likely get in the way
(we've tested this and write back has no effect).

If you're curious, the code we're using is here:

http://www.bivio.biz/f/bOP/lib/Bivio/Util/Backup.pm

There's another part of the process which involves rsync and cp --link
--archive which hits a lot of directories, which is not in the above
code.  The rsync isn't the problem, because it's running over the
network, and slow enough that it doesn't busy the disks too much.

> Do you use hardware host bus adapters with write back caching, or are
> you using plain sata drives and mdadm?

This machine is not using mdadm.  It's a 3ware 9550 with an Intel
motherboard with one CPU running CentOS 5.7.  I have another machine
which is using mdadm.  The latter has been crashing pretty frequently,
and I'm looking for a replacement for it, which is why I emailed this
list.

Thanks,
Rob