Implement Cyrus IMAPD in High Load Enviromment

[Brian Awood]

On Tuesday 29 September 2009 @ 06:59, Bernd Petrovitsch wrote:
> On Mon, 2009-09-28 at 15:33 -0700, Vincent Fox wrote:
> [...]
>
> > Really I've looked at fsck too many times in my life and
> > don't ever want to again.  Anyone who tells me "oh yes but
>
> Especially not in the >100GB area.

We haven't looked at ZFS, though as Bron suggested, I doubt it will 
solve all filesystem issues.  We use ext3 on large partitions, 
ranging from 2-5TB.  While it takes 14-18hrs to fsck, that doesn't 
really matter if you have replication, we can promote a replica to a 
primary in about 15minutes.  

>
> How much performance do you gain and what are the risks?
> So - in a larger environment - buying a few disks more and/or
> faster disks and/or battery-backed controllers and more RAM usually
> outweighs the risk of losing reputation and (commercial) customers.
>
> The next question is: Why do I - as the techie/admin/.. - win by
> saving a few 100€ (or 2.000€) on the hardware (and how much is the
> total hardware cost?) for *my* decision to use $BRAND_NEW_FAST_FS
> instead of ext3 and what can I loose (like personal reputation or
> some sleepless nights and killed weekends in the future)?
>
> Does anyone has scripts/tools to - at least - simulate 1000s of
> (semi-realistic) parallel IMAP clients on a big setup?
>

With cyrus, murder+replication+comodity hardware is the least 
expensive and probably most scalable way to go.  We have some perl 
scripts for imap load testing, you can find links to them at the 
bottom of this page,  http://blackops.mail.umich.edu/cyrus  along 
with other info on our cyrus implementation. 

Brian