<html><body>
<p><tt>Hello Michael,</tt><br>
<br>
<tt>> > My point of view this, if we run out of resource, there is nothingwe can do.<br>
> <br>
> Yes. So we need to fall back to datagram before running out of resources.<br>
<br>
Only high-end PPC supports eHCA, in the high-end PPC cluster, each node will configure enough memory to handle the connections within the cluster. So this patch will work OK. We would like to have IPoIB-CM w/o SRQ available to allow IPoIB-CM mode to be turned on as default soon.</tt><br>
<br>
<tt>That would be good on handling running out of resouce, but it is not a simple effort. Let's have an independent patch to handle resouce running out later without blocking this patch to be upper stream. I hope you can agree with this.</tt><br>
<br>
Thanks<br>
Shirley Ma</body></html>