>>We currently have nearly 100,000 accounts (not all of them active) - Cyrus
is
>>rapidly becoming the bottleneck on the system.
> In what way?
Currently the most-heavily-loaded machines are the ones running the servlet
engines, however we are about to upgrade to a later JVM which will reduce
the load there significantly. The next most highly loaded box is the one
running Cyrus (a beefy PC running FreeBSD, BTW), at the busiest time it is
running at 50% CPU utilisation and is not disk-bound. Ergo we expect that to
be the next bottleneck.
Toens wrote:
>Do you have all those users on one box?
At present, yes.
>To scale cyrus you
>most probably need an access-layer (IMAP-Proxy,
>WWW-Proxy behind load-balancers), which sends the user to
>the right mailstore (probably after asking an LDAP server
>for its location).
>With several mailstores you multiply the probability of
>failure - on the other hand most of the times not all
>users will be affected by a crash.
That's what we plan to do IF we stay with Cyrus - we already have the LDAP
setup, it would just be a question of bringing another Cyrus server onstream
and putting new users on it. But, AIUI, that wouldn't help us with
redundancy.
Thanks,
Barbara
