> I'm noticing an issue sending RMPP responses where the response message is > timing out before the requestor can send back an ACK. I don't know if this > is an issue with the RMPP timeouts that are used or some problem setting up > the MAD response.
I hope I'm looking at this wrong, but I don't see how RMPP can work reliably at all. If a user sends an RMPP response, they should set mad_element resp_expected=0, retry_cnt=0, and timeout_ms=0. The result is that the code will wait 0 ms after sending an RMPP segment for the corresponding ACK. If the ACK is not received by the time the timeout thread runs, the message will have timed out. With retries set to 0, the send will fail. I believe that this explains the behavior seen with umad. The ibal vendor library appears to work around this issue (accidentally?) by always setting the retry_cnt and timeout_ms fields. The header file states that those two fields are ignored if resp_expected is false, which is not the case. - Sean _______________________________________________ ofw mailing list [email protected] http://lists.openfabrics.org/cgi-bin/mailman/listinfo/ofw
