[Kolab-devel] Distributed Kolab Setup / IMAP question

[Michael Leupold]

Am Donnerstag, 27. April 2006 16:09 schrieb Fabio Pietrosanti:
> Michael Leupold wrote:
> > Plugging it into kolab might mean adding some new admin code to the
> > webinterface if you want to separate frontend-servers from
> > backend-servers. Changes might be only few if every kolab-server was a
> > frontend and a backend at the same time.
> mmm, theorically the murder should not require any other webinterface
> admin code because kolab already have a design with 1 master and
> multiple slave so it's ok to implement it trasparently.
> The only possibile bad thing is "what if" the master murder server
> crash? The slave server will continue offering their services without
> trouble?

According to the cyrus documentation the backend-servers (who provide the 
actual data) will provide their services independently of the 
frontend-servers (who provide the aggregator features). The only critical 
component is the mupdate-master-server who propagates changes to mailboxes to 
all frontend-servers. If the mupdate-master is down, no propagation is done, 
but the frontends should provide their services nonetheless (even if 
outdated).

A feasible setup (eg. when supporting servers distributed among several sites) 
might be to implement a frontend AND a backend on each of the servers (kolab 
master and slaves) and to setup the mupdate-master on the kolab-master. That 
way only a minimum number of changes should have to be done and it should 
perform well if master and slaves are connect with a thin wire.

I'll try to get such a setup working tomorrow to see how it performs. I'm 
interested in the extra-load the new frontend-part adds to the server as well 
as the bandwidth consumed my the mupdate propagation on mailbox changes 
(which might be a factor for distributed setups).

I'll tell you about my results.

Regards,
Michael