Rich Graves wrote:
>Mailbox replication is nearly impossible due to locking problems and other
>race conditions. You can route around this in two ways: get clever down the
>stack with a snapshotting filesystem (LVM or if you have the means, get a
>pair of NetApps with SnapMirror), or get clever up the stack by rethinking
>the application.
>
I don't think it's right to call it "nearly impossible". If you dump
your e-mail into an Oracle DB, you can use Oracle to keep each
transaction against several datastores in sync. And I can't claim to be
a DBA or anything but I've been told that Oracle can keep several copies
of a database in-sync with fall-back servers and all.
I don't know of one...but I wouldn't be suprised if Oracle didn't have
an IMap client that uses the database as the storage. And it should be
preaty trival to do.
>But do you really need full replication, or can you reduce the problem
>space to "never lose email"?
>
Yes this is the key question. Properaly made and maintained systems can
have uptimes in the 6-12 month range, normaly brought down for patching,
and nothing else (I'll argue that if you don't apply patches every 6-12
months you aren't mainting the system.) So is a 3-4 hours of planned
outage every 6 months a big deal? (8hours per year is 99.9% uptime.)
With some careful planning, you can most likely drop that to <1 hour
per year, 99.99% uptime.
It's true you want battery backed up systems, nightly disk backups, and
a good raid system for the mailboxes. I would also put in a backup mx
host.
Remeber most of what brings mailsystems down these days is load/use
related problems, 2 mailbox servers would RARELY help the problem.
Most of the time you get hit by some virus or something and have
replicating/bouncing e-mail all over the place, or someone filling up
the mail spool.
johno
---
Send mail for the `bblisa' mailing list to `[EMAIL PROTECTED]'.
Mail administrative requests to `[EMAIL PROTECTED]'.
