[Openib-windows] Fixes to the new low level driver to support WSD

Tzachi Dar tzachid at mellanox.co.il
Tue Apr 11 02:50:36 PDT 2006


Hi Fab,

It seems that our happiness what somewhat too early. 

It seems that your code is using one CQ for all QP's and depends on the
resize_cq functionality of the HCA. Currently this function is not
supported and therefore we were falling down to ipoib, so the 5 threads
test wasn't running on WSD after all.

One possible work around for this is to create the initial CQ with 101
QP * 2 CQ per QP * 16 entries per CQ (or 3232 entries), and therefore
supporting up to 100 sockets.

This allows the 50 threads test to pass. Although sometimes it seems to
be hang (I'm still checking on this).

So it seems that you can continue your tests while we still have to find
a solution for the problem of resize_cq is not supported. (either change
WSD or implement the function).

Thanks
Tzachi

> -----Original Message-----
> From: ftillier.sst at gmail.com [mailto:ftillier.sst at gmail.com] 
> On Behalf Of Fabian Tillier
> Sent: Monday, April 10, 2006 7:30 PM
> To: Tzachi Dar
> Cc: openib-windows at openib.org
> Subject: Re: [Openib-windows] Fixes to the new low level 
> driver to support WSD
> 
> Hi Tzachi,
> 
> On 4/10/06, Tzachi Dar <tzachid at mellanox.co.il> wrote:
> >
> > Hi Fab,
> >
> > In checkin 303 to open IB Leonid has updated the mthca driver to 
> > support RDMA reads of WSD.
> >
> > With this checkin, we can see that basic tests of WSD passes (5 
> > threads), while there are still failures in higher 
> stress(50 threads).
> >
> > We will continue to look for errors on our side, but I believe that 
> > you can also continue to run tests on your side.
> >
> > Please also let us know of any issue that you find, so that we can 
> > handle it (and also that we will not waste time on the same issue).
> 
> This is excellent.  I will resume testing with this driver, 
> and will let you know of issues.
> 
> - Fab
> 
> 
> 



More information about the ofw mailing list