[PVFS2-users] Low performance of PVFS2 on our small cluster

Rob Ross rross at mcs.anl.gov
Wed Oct 27 09:17:22 EDT 2004


Hi,

To start with, you are asking a lot from those nodes to run two compute 
processes and the pvfs2-client *and* the pvfs2-server.  What does 
performance look like if you only run 5 compute processes total instead of 
10?  Perhaps you're just seeing the impact of too many processes on a 
node?

In addition to Nathan's suggestions regarding syncing, your choice of 
local file system type can have a big impact, particularly in parallel 
performance.  The FAQ covers this issue to the extent that we understand 
it, and it has some suggestions.

Regards,

Rob

On Wed, 27 Oct 2004, Silvestre Zabala wrote:

> * Silvestre Zabala <silvestre at zabala.name> [2004-10-27 00:21]:
> 
> As I have left out some data and made a mistake with network, I'll try
> to correct my post here:
> 
> > Here is a quick description of our test cluster. It is built out of five
> > Dual-Xeons at 2GHz with 1G memory and interconnected with Fast Ethernet.
> > We are running the latest MPICH2 and PVFS2. The storage are is located
> > on the local HDDs which, according to bonnie++, are able to read or
> > write about 45 MB/s. So it's a pretty generic COTS-cluster.
> 
> Of course we are using Gigabit Ethernet, as Philipp Sadleder noted.
> And at the moment we are running one pvfs2-server per node, and on the
> same nodes two b_eff_io processes per node.
> However we will run a 5 IO-Node/5 Compute-Node-Configuration later.


More information about the PVFS2-users mailing list