Subject: Re: Slow UDP traffic
To: None <netbsd-users@netbsd.org>
From: Thor Lancelot Simon <tls@rek.tjls.com>
List: netbsd-users
Date: 06/07/2004 13:37:04
On Mon, Jun 07, 2004 at 12:12:29AM -0400, Perry E. Metzger wrote:
> 
> Johnny Billquist <bqt@update.uu.se> writes:
> >> You mean UDP. Initially NFS used UDP, but it turns out that most of
> >> the time you're better off with a TCP based implementation anyway, and
> >> almost all modern NFSes do TCP.
> >
> > Most NFS implementations can do TCP, but I suspect that most still use UDP
> > anyway,
> 
> Not if you want decent performance.

That's not true, unless your path to the fileserver is many hops or your
local network is very congested.  In my benchmarks, in Jonathan's
benchmarks and, tellingly, in published SPECsfs runs, for clients on a 
local network UDP with 32K RPCs seems to win -- even with the resulting 
fragmentation.

Thor