MRC at CAC.Washington.EDU
Fri Sep 28 16:34:00 PDT 2007
On Fri, 28 Sep 2007, Jeroen van Aart wrote:
> In the same line of thought, why would a SAN or DAS (or ata over ethernet,
> iscsi) be any less likely to crash than an imap server?
Yup. It just takes a while for people to conclude that.
We used an EMC SAN for our IMAP servers here at UW. We got rid of it due
to the cost. It didn't seem to make any difference.
> Which I guess leads to the logical conclusion that distributing email
> accounts over multiple servers is the better solution (that's the uw approach
> if I understood correctly). So if one server fails not everyone is affected.
> The replication idea would, in theory, just make recovering from that failure
> easier, by switching to a live backup holding the same mailboxes.
Yes. You then have to take a cost/benefit analysis to determine the
likely failures and how to recover.
It doesn't do much good to have a live backup to cover you from a power
failure if the live backup system is on the same circuit. Nor does it do
much good to have a live backup to cover you from natural disaster if the
live backup is in the same building...
If you're worried about failing hardware, a lot can be said for
undertaking efforts to maximize hardware reliability. Our policy is to
use new, inexpensive hardware, and retire it when it goes off warranty in
We keep hot spares, and monitor for failing hardware. It's relatively
uncommon for hardware to have a catastrophic failure with no advance
indication. For example, there is typically an extended period of
recoverable disk errors before total failure.
-- Mark --
Science does not emerge from voting, party politics, or public debate.
Si vis pacem, para bellum.
More information about the Imap-use