[OpenAFS] Network becomes terribly slow when cache manager flushes updates over xDSL

Douglas E. Engert deengert@anl.gov
Wed, 08 Jul 2009 10:53:33 -0500


Harald Barth wrote:
>> I have to say that ifconfig still shows a MTU of 1500.
> 
> Yes, your local interface does not know anything about the network further down the line.
> 
>> By the way, I didn't get the final point about the "-nojumbo" and
>> "-rxmaxmtu" params : will they appear in a release soon? And if yes,
>> they will replace the change to RX_MACK_FRAGS I had to make?
> 
> No, these three do control different aspects of "how to pack your packets".

The -rxmaxmtu will be in 1.4.11. The last pre-release was last week.
So it should be out very soon.

> 
> I'd set -rxmaxmtu to the lowest value you got with your ping experiments.

Actually 56 bytes less (see below).

You can then see what AFS is doing using:

rxdebug host.name.of.client -port 7001 -peers

The will show for each AFS server the client has a session with a line like:
     ifMTU 1444      natMTU 1444     maxMTU 1444

This was done on a client without the -rxmaxmtu being set, and AFS is
using the interfaces MTU minus 56.

On a client with -rxmaxmtu 1244 the line looks like:
     ifMTU 1244      natMTU 1244     maxMTU 1244

So set the -rxmaxmtu to at least 56 bytes less then what the max MTU
ping will sho as not fragmenting. (I think it is actually 48 as the ping
may be showing 8 bytes less because of the UDP header.)

> 
> Harald.
> _______________________________________________
> OpenAFS-info mailing list
> OpenAFS-info@openafs.org
> https://lists.openafs.org/mailman/listinfo/openafs-info
> 
> 

-- 

  Douglas E. Engert  <DEEngert@anl.gov>
  Argonne National Laboratory
  9700 South Cass Avenue
  Argonne, Illinois  60439
  (630) 252-5444