Kolab numbers.

[Luca Villani]

Alle 09:06, mercoledì 7 aprile 2004, Bo Thorsen ha scritto:


> > Did any of you tested Kolab in a clustered/HA architecture?
> > Did you think Kolab working fine in a such environment?
>
> There is no support for clustering in Kolab1.

This is a bad news, I like kolab.


> One of the things we're working on for Kolab2 is to add multilocation
> capabilities of the mail and LDAP parts. Multilocation is not exactly
> clustering, I know. 

No, it isn't the same thing.


> Kolab2 do not have any plans that will help you with failover
> capabilities, but clustering should not be necessary.

Clustering is a must in an HA scenario.

Let me explain a possible failover architecture.

We have two machine (kolab-a and kolab-b), everyone tith two ethernet cards.
The firts ethernet card of every server (kolab-a.eht1 and kolab-b.eth1) is 
exposed on big internet, and every server has a public ip address (x.y.z.1 
and x.y.z.2) but *only* the main server has a second public ip address 
(x.y.z.3), the adress of the workgroup server.
The first server (kolab-a) is up adn running, and *all* kolab components are 
up and running. The second server is up and running but have *only* openldap 
up and running, slurping user data from the first server.
The totality of kolab istallation (without openldap database!) is replicated 
via a network file system like Intermezzo (http://www.inter-mezzo.org/) on 
the *second* ethernet card of every server (kolab-a.eth2, kolab-b.eth2), ie 
in a not internet exposed channel.
Finally, we have heartbeat on all server.

If the first server go down, heartbeat fier up a full kolab server on the 
second server, and attahch the main ip address (x.y.z.3) to the first 
ethernet card of the second server.

In this scenario, can cyrus work on an Intermezzo shared storage? I know it 
don't work on NFS...

And in this scenario (cyrus working on top of a network shared storage), why 
can't I running two *separate* server working toghether?



> A fairly new Kolab server will happily handle 10.000 users, so even though
> your machines might have a bit of age, 300 users is just not a problem.

You don't know my commercial/marketing cow-workers... :-)
2 MByte emails is a normal job, only for asking "Hi, how are you?"... :-(



> It is generally a problem for Kolab migration from Exchange, that
> companies try to use their Exchange requirements directly on Kolab.

We are not an exchange based company, only commercial/marketing guys are on 
top of an exchange server.
But we are planning to migrate *all* our users (exchange users ad postfix 
users) on top of a groupware server: our actual mail server is a single dual 
processor machine, with an average load of 0.90, and our exchange server is 
an identical machine, slow down all the time.



-- 
Luca Villani                Wireless Solutions S.p.A.  -  Dada Group
NOC Manager                 Europe HQ: via Castiglione 25, 40124 Bologna Italy
http://www.wseurope.com     Tel: +39 051 2966826     Fax: +39 051 2966800
GPG public key available    Mobile: +39 348 5298542  UIN: 76272621