[ofa-general] [PATCH] perftest: don't attach the sender QP

[Or Gerlitz]

Oren Meron wrote:
> What about the send_lat test?
latency tests are typically based on RTT/2 measures which means that 
both sides do send and receive.... if this apply to the send_lat test, 
please don't apply the same patch over there. As for the bandwidth test, 
my patch I think has a defect that makes the client not to join also for 
the bidirectional test, where in that case it needs to, sorry.

Also, I don't think to ever managing to see the server side statistics 
printed, this means that I could only see the sender bandwidth which is 
not necessarily the receiver bandwidth, the importance of seeing it BTW 
applies also the the unicast UD tests.

One nice enhancement which you might want to look at would be to have a 
some sort of MGID supplied from the command line and attach to this MGID 
instead of the current implementation. This would allow to have > 1 
receivers

Or.