On Mon, 2007-04-23 at 18:05, Sean Hefty wrote: > >Has anyone thought about using replication rather than caching to > >solve this problem? It seems to me it would be alot faster for some > >single process in the network to fetch and keep a copy of the entire > >SA route database, format it into a binary format and use RC RDMA to > >transfer it to every node each time it changes. > > I have given thought to using RC RDMA to distribute the data to all nodes, > especially to eliminate the MAD protocol overhead. There are a couple issues > with this: > > To work with existing SAs, we need to working within the defined SA interface > (i.e. SA MADs), so something still needs to query for all path records. > > The GetTable query requires an SGID, which means that whatever node collects > the > path records must first collect all the GIDs. (And the most efficient way > I've > found to obtain a list of all GIDs is via a GetTable path record query...) > This > also means that the node collecting the path records will generate 1 query per > GID. This has the same impact on the SA as each node issuing their own query. > And the impact on the subnet is higher, since we still need to distribute that > data to the end nodes. > > In short, until we can standardize on some new SA interface, or we have a > distributed SA, I don't see where we can do much better than caching GetTable > responses at the end nodes.
Or perhaps an additional standard for replication. I think distributed SA is a separate related issue. -- Hal > - Sean > _______________________________________________ > 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 _______________________________________________ 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
