Re: 2.6.0 NFS-server low to 0 performance

[James Pearson]

Trond Myklebust <trond.myklebust@xxxxxxxxxx> writes:
> På lau , 10/01/2004 klokka 11:08, skreiv Andi Kleen:
> > Trond Myklebust <trond.myklebust@xxxxxxxxxx> writes:
> > 
> > > The correct solution to this problem is (b). I.e. we convert mount to
> > > use TCP as the default if it is available. That is consistent with what
> > > all other modern implementations do.
> > 
> > Please do that. Fragmented UDP with 16bit ipid is just russian roulette at 
> > today's network speeds.
> 
> I fully agree.
> 
> Chuck Lever recently sent an update for the NFS 'mount' utility to
> Andries. Among other things, that update changes this default. We're
> still waiting for his comments.

If mount defaults to trying TCP first then UDP if the TCP mount fails,
should there be separate options for [rw]size depending on what type of
mount actually takes place? e.g. 'ursize' and 'uwsize' for UDP and
'trsize' and 'twsize' for TCP ?

James Pearson
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/