Jumbo frames from 1750 to NetApp filer

Dan Pritts danno at internet2.edu
Thu May 20 15:26:00 CDT 2004


On Tue, May 18, 2004 at 10:28:31AM -0500, Matt Domsch wrote:
> On Tue, May 18, 2004 at 03:19:57PM +0100, David Bloomfield wrote:
> > With everything configured for an MTU of 9000, NFS access from the
> 
> <snip>
> 
> > rw,proto=tcp,hard,bg,intr,wsize=32768,rsize=32768
> 
> You're telling NFS to use 32KB datagrams, with an MTU of 9000 so every
> UDP datagram will have to be fragmented, which doesn't appear to work.
> Have you tried with wsize and rsize=8192?

I'd be shocked if this were the problem (not that i wouldn't try it, 
mind you).

First, he's using TCP, not UDP.  

Second, if fragmentation were the issue why does it start working again
when the MTU drops to 1500?

I'd be interested to know what happens if you use UDP instead of TCP -
on a LAN you should get better performance with UDP, the advantage of
TCP is on lossy links.

I'd also wonder what happens if you use, say, 4500 byte packets.

danno
--
dan pritts                                       danno at internet2.edu
systems administrator                            734/352-4953 office
internet2                                        734/834-7224 mobile




More information about the Linux-PowerEdge mailing list