Coda File System

Re: High availability

From: Jan Harkes <jaharkes_at_cs.cmu.edu>
Date: Wed, 22 Sep 1999 12:31:02 -0400
On Wed, Sep 22, 1999 at 02:32:50PM +0200, Antoine Brenner wrote:
> 
> Hi,
> 
> I understand that the Read-Write Replicated Volumes can be used to provide
> high data availability in case of failure of a server.
> But I could see nowhere what happens if the cell SCM server fails.

The only way in which the SCM is special, is that it is the server where
all others are fetching updated volume location and user databases from.
So the SCM is only special in the cases where new users/volumes are
added. And by replacing the name of the SCM in /vice/db/scm, and
restarting the update daemons on all servers it is possible to switch to
a completely different SCM.

As all servers have their own local copy of the volume location and user
databases, no server relies on the SCM during normal operation, and
clients can authenticate and find volumes in the cell by asking _any_
available server. Point a client to a single server in the cell, and it
will be able to find all others.

> Are the replicated volumes still shared ?

I don't understand that question. Shared with what?

Replication in coda works as follows:
- When a client writes to a replicated volume the update is sent to all
  accessible servers in the replication group.
- When a file's attributes are fetched, the version vectors from all
  accessible replica's are collected and compared. If there is a
  difference, some server has missed updates, and a server-server
  resolution is triggered.
- The file is fetched from any of the available replicas.

Jan
Received on 1999-09-22 12:32:14