Coda File System

Re: Some questions...

From: Jan Harkes <jaharkes_at_cs.cmu.edu>
Date: Wed, 25 Sep 2002 14:14:58 -0400
On Wed, Sep 25, 2002 at 11:01:33AM -0700, Derek Simkowiak wrote:
> > try just a cluster{ed|ing} file system.
> 
> 	That's what I'm after.  How is CODA not a "cluster{ed|ing}"
> filesystem?

Coda is a distributed filesystem. The main difference (in my mind) is
that a distributed filesystem assumes a WAN, while with a cluster you
assume typically better than LAN environment. (i.e. all nodes connected
through a switch, or which special high speed interconnects).

So a distributed filesystem would need more aggressive caching and lock
avoidance to avoid the latency hit of the network. This can lead to
consistency problems in case of replication.

While a clustered filesystem doesn't need to cache, because it is
relatively cheap to refetch the data and can use locking to guarantee
node-to-node consistency.

Jan
Received on 2002-09-25 14:16:03