On Thu, 8 Jul 2004, Timo Sirainen wrote:
Doesn't that mean that a user is located only in a single server, so in
case it breaks, the user can't read mail until admin has fixed the
problem by restoring mails from backups and moved the accounts to new
server?

As opposed to having all users located in a single NFS server, so in case it breaks, no user can't read mail until admin has fixed the problem by restoring mail from backup and built a new NFS server.


Nobody said that you can't have mirroring, RAID, and the like. The point is that those backup/robustness facilities should be co-located with the IMAP server processes.

Unless the account is mirrored between multiple servers in realtime as
well (how?), user can also lose mails. Better than everyone's mail
breaking, but I'd prefer transparent failovers without any data loss.

Why do you believe that an IMAP server system is less robust than an NFS server system? What magic exists that makes a machine running NFS be reliable and a machine running IMAP unreliable?


I claim that there is no such magic; and that an IMAP server system is at least as reliable as an NFS server system. In fact, I claim that a small IMAP server system that serves a fraction of the users of the NFS monolith will be much more reliable.

I'm serious. I consider your basic assumptions to be faulty, and suspect that you've never questioned them.

Empirical evidence supports my claim. We have had very few failures. Typically, we know when hardware is one the verge of failure, and can move people to other hardware before the failure becomes solid. We had many facility-wide failures with the late unlamented NFS monolith.

Oracle works with GFS, I bet IMAP server could work with it too.

I will believe it when I see it. I have heard enough bogus claims over the years that I require proof.


-- Mark --

http://staff.washington.edu/mrc
Science does not emerge from voting, party politics, or public debate.
Si vis pacem, para bellum.

Reply via email to