Steve Wise wrote:
Seems like the retry functionality is perhaps useful enough then to add it into the rdma-cm itself? IE have a retry count param to rdma_resolve_route() maybe? Just thinking aloud...
Steve, Sean
As I just wrote, with IB, there's a lower level mechanism (mad) to enforce timeout and implement retries for route lookup. Since with iWARP route resolution is no-op, its seems we are done in that respect.
As for address resolution, the kernel addr module implements a timeout mechanism.
I wonder if one of the neighbour subsystem sysctls/params, could be used to implement retries, looking on arp(7) I came a cross the retrans_time sysctl which stands for "The number of jiffies to delay before retransmitting a request. Defaults to 1 second", but I wasn't sure if it stands for itself or has to be used in conjunction with other sysctls. Basically, when an arp is a byproduct of the IP stack wanting to send a packet, there's a rich state-machine of retries, etc, I assume it should/could be usable for the RDMA stack.
Or. _______________________________________________ general mailing list [email protected] http://lists.openfabrics.org/cgi-bin/mailman/listinfo/general To unsubscribe, please visit http://openib.org/mailman/listinfo/openib-general
