Subject: Re: nfs performance
To: Nathan J. Williams <nathanw@wasabisystems.com>
From: Johnny Billquist <bqt@update.uu.se>
List: current-users
Date: 02/10/2004 23:22:45
On Tue, 10 Feb 2004, Nathan J. Williams wrote:

> Daniel Carosone <dan@geek.com.au> writes:
>
> > While a 2k blocksize is certainly not optimal, it's also a strange
> > size wrt ethernet frames.  Out of curiosity, try something like 1400
>
> I think this number comes up because NICs tend to have power-of-2
> sized memory, and 2k is (was?) a common number for cheap NICs. So a
> full-sized packet and a fragment that fit into the 2k is less of a
> problem than a more-than-2k-total burst of fragments, and perhaps the
> fragmentation isn't much of a problem in the local network situation.

Well, 2k blocksize means you need more than 2k since you also have some
overhead.
On some NICs, 2k is an often used solution. The reason for this is that
the NIC can't handle bursts of packets, but almost all will handle atleast
two packets. People then tend to want to use block sizes in NFS that is a
power of 2, and 4k will mean atleast three packets, which some NICs once
again will fail on. So, from that point of view, 2k is pretty "optimal".

Another solution is (of course) to get a better ethernet card. :-)

	Johnny

Johnny Billquist                  || "I'm on a bus
                                  ||  on a psychedelic trip
email: bqt@update.uu.se           ||  Reading murder books
pdp is alive!                     ||  tryin' to stay hip" - B. Idol