Coda File System

Re: so called "multihoming" / volumes

From: Jan Harkes <jaharkes_at_cs.cmu.edu>
Date: Sat, 19 Jul 2003 12:29:03 -0400
On Fri, Jul 18, 2003 at 03:23:42PM -0400, Pascal Bourguignon wrote:
> The documentation says that: "Coda does not support multiple-homing 
> cells at this time."
> 
> One problem is that "this time" seems to be quite old, up to five years, 
> while I'm trying to use coda-6.0.1. Isn't there any up-to-date 
> documentation?

Well, in a way 'at this time' is still a valid comment. When a Coda
clients asks for something like volume location, the server only sends
back a single ipv4 address.

As a multi-homed server has multiple possible addresses this will
ofcourse not work right for most setups.

> Now, actually I don't want coda to work on more than one interface.  

Ah, but that is a setup that 'should' work (in theory). You should be
able to use the 'hostname=' option in /etc/coda/server.conf. This tells
the server to resolve the ip-address of that specific hostname and bind
it's listening socket to that specific address.

It will also use that address in any volume location queries. It becomes
a little more complex for multiple servers because the volume location
database is built on the SCM so it will have to get the right (private)
addresses of other servers.

> I've got CPUs with three ethernet interface each, one for the extranet, 
> one for the intranet, and one reserved for the intra-cluster trafic, 
> where I would want to direct all coda trafic.  That's why I specify 
> plainly the name of this interface (janus-1.janus.infra.afaa.asso.fr) in 
> the realms, vicetab and servers configuration files. 
> 
> But coda still refuses to work correctly...  Does it test explicitely 
> for several interfaces and refuses to please me on purpose?  Could it 
> not just ignore the other interfaces?

It should work if you set 'hostname=janus-1.janus.infra.afaa.asso.fr' in
/etc/coda/server.conf you are forcing the server to not use 'hostname()'
to figure out where it is located.

> I've got a 20GB partition where I put (symlinks for) vice and vicepa 
> (and the /etc/coda and /usr/coda directories too).  Then I want to 
> create volumes.  Do I need to create a directory for each coda volume or 
> can I use /vicepa for all?  Does each createvol_rep need a diferent 
> /vicepX ?

/vicepa will be used to store container files of all volumes, the data
in RVM will contain the actual 'object' to 'container file' mapping.

Jan
Received on 2003-07-19 12:30:32