[ofa-general] Re: IPoIB path caching

[Or Gerlitz]

Michael S. Tsirkin wrote:
>> I believe a better solution is for everyone to use cached records, if 
>> they exist, with a feedback mechanism from the CM that removes paths on 
>> a connection failure or path migration event.

> Ack timeout on an RC QP is also a good indication we should redo the lookup.

I am not following you.

The only way I know for IB SW to be aware to ACK timeout is if the RC QP 
retries are set to zero (actually also at this scheme the QP can move to 
error as of other reason).

Indeed IPoIB-CM uses RC and does this zero-retries settings, but we have 
already agreed that moving forward, IPoIB-CM default IB transport would 
change to UC, didn't we?

Or.