Coda File System

Re: R: Hint on client cache

From: Jan Harkes <jaharkes_at_cs.cmu.edu>
Date: Fri, 10 Aug 2001 14:01:15 -0400
On Fri, Aug 10, 2001 at 06:41:57PM +0200, Andrea Cerrito wrote:
> > > And is it safe to force it? I'm on 100T switch, so network is
> > > good. Thank you
> > 
> > For all other cases 'cfs strong' avoids going into
> > write-disconnected mode due to (assumed) low bandwidth
> > connectivity.  
> 
> Mmmhhhh... so, the cfs strong just prevent to going into
> w/disconnected mode when there is a "assumed" low bandwidth case, or
> it does some sort of "network optimization" to venus work?

It simply ignores bandwidth estimates from the RPC2 layers. There is no
network optimization, otherwise I would have used those optimizations
for the cases where there is low bandwidth.

> I'm trying to boost venus performance, and I set my cache to 200Mb
> (full of about 170, so I think I've hit the right) for this reason.
> Now I'm trying to search other inputs (well accepted :). 

Forcing the client into write-disconnected state vastly improves
performance. On the other hand it also allows for those nasty
reintegration conflicts, i.e. it's a tradeoff.

Write-back caching is an experimental attempt at achieving fully
connection behaviour while having write disconnected performance.

Jan
Received on 2001-08-10 14:01:23