[Beowulf] MPI and Redhat9 NFS slow down

[Joachim Worringen]

Jack Chen wrote:
> I have tried to mount the drive using different parameters (rw,soft
> and rw,hard,bg,intr,noac) and increased the nfsd daemon from 8 to 16
> on the NSF server, but nothing change.

MPI-IO (are you doing this?) via NFS is generally not for large amounts 
of I/O. Two things:
- you need to mount with the noac option to ensure a certain degree of 
consistency between nodes
- increasing the NFS block size to 8kB can increase performance.

> If you have any idea on what is going on, please help!

First, you should determine the I/O performance that can be achieved for 
the different NFS mounted directories with a tool like bonnie and see 
what magnitude of performance you can expect. Than you need to see what 
the application is actually doing for I/O.

And check the ethernet configuration - it might also be a full/half 
duplex problem between those nodes.

You also might consider using a more appropiate (parallel) file system 
like PVFS.

  Joachim

-- 
Joachim Worringen - NEC C&C research lab St.Augustin
fon +49-2241-9252.20 - fax .99 - http://www.ccrl-nece.de