NFS not negotiating MTU correctly?

DISCUSSION Chris Worley 2 participants 2 messages May 25, 2004

Subsystems (2)

NFS client NFS server

lore

Share & tools

lore

Chris Worley

author

2004-05-25 14:23 UTC #0 lore raw

I don't know why this is an NFS specific problem (I'd think this would be part of IP)...

I run 9K MTU's on all my NFS servers and most of my clients. Great performance (the servers are also channel bonded... but that's irrelevant).

Clients that can't do 9K MTU's work with the 9K MTU systems for all sorts of network tasks, except NFS.

To get NFS to work, I either have to use NFS over TCP, or set my w/rsize below 1500 bytes. Both cause performance problems.

In grilling my switch vendor, they assure me that it's the client & servers job to negotiate properly fragmented packets based on the lessor MTU.

Chris

-------------------------------------------------------
This SF.Net email is sponsored by: Oracle 10g
Get certified on the hottest thing ever to hit the market... Oracle 10g. Take an Oracle 10g class now, and we'll give you the exam FREE.
http://ads.osdn.com/?ad_id=3149&alloc_id=8166&op=click
_______________________________________________
NFS maillist - [email protected]
https://lists.sourceforge.net/lists/listinfo/nfs

Trond Myklebust

2004-05-25 15:04 UTC #1 lore raw

P=E5 ty , 25/05/2004 klokka 09:18, skreiv Chris Worley:

Quoted 3 lines

In grilling my switch vendor, they assure me that it's the client & servers job to negotiate properly fragmented packets based on the lessor MTU.

That sounds like a piss-poor switch to me. AFAIK all routers are supposed to automatically fragment datagrams when forwarding between networks that have a different MTU. Automatic MTU discovery is not part of the IP protocol (although methods for doing so under certain circumstances do exist).

In any case, dealing with MTUs is entirely the networking layer's job. NFS just deals with sockets. I suggest that you check that the ip_no_pmtu_disc sysctl is set correctly, and then check with the Linux networking mailing list ([email protected]) if there are further problems.

Cheers,
Trond