[openib-general] scaling issues, was: uDAPL cma: add support for address and route retries, call disconnect when recving dreq

[Sean Hefty]

> Maybe -- I wonder how scalable our RMPP implementation is though.
> What will happen on the SM node with 256 RMPP requests returning 256
> paths each?  How about 1024 * 1024?

The RMPP implementation doesn't expect to receive lots of simultaneous RMPP 
transactions, and expects them to be reasonably behaved.  It uses linear 
lookups, so there may be room for improvement here.  However, I would expect 
most scalability issues to be on the end receiving an RMPP message, rather than 
on the send side.

- Sean