[openib-general] Performance Degradation with OFED v. Voltaire(lustre)

[Or Gerlitz]

Bernadat, Philippe wrote:
> I didn't have time to re-run then non FMR cases.
> For FMR VIB and OFED are comparable.
> But I will e-run tests for all cases

my main concern is FMR/no-FMR for OFED, FMR should be at least good as 
no-FMR and if this is not the case, lets look into that.

> Right now I am fighting with ib_query_device() that crashes the kernel !
> Trying to use this to test the HCA type.

so your approach is:

	if (the-active-side-is-mlx-tavor)
		then set-path-mtu-to-1024

then this is a bug, since the only **active** side sets the path mtu, 
where the passive side (SFS ...) might be mlx-tavor and the active side 
can be something else, and the tavor mtu bug will hit you.

I am thinking what is the correct way to approach the problem, at the 
cma level, there will be probably some discussion here.

saying all the above - ib_query_device must not crash  the kernel!
make sure that in case there is some issue, please report it here.

Or.