[ofa-general] Re: [PATCH draft, untested] ehca srq emulation (for IPoIB CM)

[Bernard King-Smith]

"Sean Hefty" <sean.hefty at intel.com> wrote on 06/15/2007 03:00:04 PM:

> 
> > Basically, I think that because of lack of SW level flow control,
> > generally IPoIB CM without SRQ does not make sense because of
> > the scalability problems.
> 
> Most clusters are only 16-32 nodes.  If IPoIB CM without SRQ can support 

> this number of systems and outperforms IPoIB UD mode, then I do believe 
> that it makes sense.  IPoIB CM support, with or without SRQ, is less 
> scalable than IPoIB UD mode, but it was still added because it provided 
> a benefit under most conditions.

I think Pradeep has been making this very clear all along and that scaling 
is a restriction we can make. Since SRQ is not a required part of the 
spec, then having support for non-SRQ in the IPoIB-CM driver supports the 
minimal requirements. I think it is typical that any driver that supports 
enhancements from a basic spec has exception handling for both cases ( 
base and enhanced ) in the layer in question (ipoib). Putting it in the 
device driver splits the non-SRQ IPoIB support to two layers which is not 
a good idea.

We are already running with the non-SRQ patch here and the results are 
very good. Changing to a different approach is not the right thing to do 
at this time. Emulating in the device driver will only increase the amount 
of work everyone will have to do to get this out, and runs the risk of 
uncovering more complex problems.

Can we close on the last few issues and get this lined up for OFED 1.3?

> 
> - Sean

Regards.

Bernie King-Smith 
IBM Corporation
Server Group
Cluster System Performance 
wombat2 at us.ibm.com    (845)433-8483
Tie. 293-8483 or wombat2 on NOTES 

"We are not responsible for the world we are born into, only for the world 
we leave when we die.
So we have to accept what has gone before us and work to change the only 
thing we can,
-- The Future." William Shatner
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openfabrics.org/pipermail/general/attachments/20070615/e6c878b4/attachment.html>