Coda File System

Re: Questions of architecture...

From: Ivan Popov <pin_at_medic.chalmers.se>
Date: Thu, 12 Aug 2004 16:31:20 +0200
Hello Marc,

On Thu, Aug 12, 2004 at 02:38:34PM +0200, EXT / FOCAL CASSUTO Marc wrote:
> I'd like to set up a cluster of 2 servers with shared data.
> Datas can be modified by both clients
> Just one volume would be used.

a word of caution first, any given file should not be modified
by both clients at the same time, otherwise you get a conflict
(and you do not want that).

> Which is the good configuration ?
> - 1 server and two clients
> - 2 servers and 2 clients

Wonder who (which processes / applications) will access the data.
I assume you are going to run some applications on two hosts, and all
you have is two physical machines, so that you can not run you Coda servers
separately.

In any case, you are going to have two Coda clients.
Then you can have one, two or more Coda servers, on your hosts running
the applications or on separate machines.

> What is the more reliable method :
> - using server/server replications
> - using venus fonctionnalities (connected/disconnected mode)

In your case I would go for just one Coda server. During its downtime
the clients would not be able to fetch files missing in the caches,
nor see each other's updates, but on the positive side you never get
server-server conflicts which can be tricky to resolve.
Local-global conflicts can be "solved" in the worst case by reinitializing
the client cache, losing some updates, but easily.

With other words, administration of a single server setup is simpler.
Harder to do a "disaster recovery" if the hardware breaks,
but you should have working backups at hand anyway.

My 2c,
--
Ivan
Received on 2004-08-12 10:33:28