[openib-general] CM header file

Sean Hefty mshefty at ichips.intel.com
Thu Dec 16 15:35:18 PST 2004


Libor Michalek wrote:
>>I have given some thought to how retries should work.  I've thought 
>>about adding a new call, ib_retry_cm_send() - or something like that, 
>>that resends the last message sent.  Or the callback could indicate to 
>>retry.
> 
>   Oh, so the CM would have enough information to generate the retry, but
> the consumer would need to notify the CM to do so?

I think it makes sense in the case of a timeout for the CM to optimize 
for the retry case.  (Such as keeping the last sent MAD around for 
retransmission.)  The argument used against putting retries into the 
MAD layer was to allow the consumer to set the retry policy, such as 
increasing the timeout value.

A major difference between the MAD API and the CM API is that the CM 
abstracts the MAD from the user, making the retry issue a more 
difficult to do efficiently.  So, while I think it would be nice to let 
  the user initiate retries, I want to make sure that it doesn't have a 
substantial impact on performance.

- Sean





More information about the general mailing list