[ofa-general] MPI IB Errors

[Jeff Squyres]

On Aug 22, 2007, at 5:32 PM, John Leidel wrote:

> Yeah, I'm sure the OFED release is the same.  I'm running ROCKS  
> 4.2.1, so all the node images are identical regarding the package  
> selection.  There could possibly be [well, probably] a difference  
> in the firmware releases of the HCAs and switches from the older  
> machines and the latest delivery.

Try running ibv_devinfo on your new nodes, which should show you the  
HCA(s) on your host.  I suspect that it will fail with a similar  
error (but am not 100% sure -- I'm the MPI guy, not the verbs guy :-) ).

If this is the case, then you've got a bigger issue that your IB  
drivers are not loading.  This will need to be fixed before you  
investigate the firmware level on your HCAs across the cluster.

(other IB stack experts feel free to chime in...)

-- 
Jeff Squyres
Cisco Systems