On 31/07/09 19:58, Todd Rimmer wrote: >> Yossi Etigin wrote: > >> What if we query the remote port LMC once, when the path is resolved, >> and then >> use it to mask the LID until the path is refreshed again? > > Doing the LMC query when the PathRecord query is made makes, sense. > > I assume you are proposing a query of the SA for the PortInfoRecord? > > Note that a direct query of the remote SMA would be a bad idea and has at > least the following issues: > 1. There are limited VL15 queuing resources and no VL15 flow control, hence > if multiple nodes queried the same SMA at the same time, packet loss could be > significant which would impact SM operation as well (note how carefully the > recent opensm work for parallel queries has had to tune operations to SMAs) > 2. If MKey security is enabled, the client will not (and should not) know the > MKey of the remote SMA and hence can't make a query. > 3. The intent of the IB Architecture is for the SM to be the only entity > which interacts with SMAs. A handful of diagnostic tools (like ibdiagnet) > may be an acceptable exception, but it would not be recommended to make > non-SM queries of an SMA as part of a normal protocol. >
Actually I wasn't sure which is better, but given all these issues SM should be the one and not the SA. _______________________________________________ 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
