On 2/22/06, amith rajith mamidala <[EMAIL PROTECTED]> wrote: > > I have a question related to the mixture of HCA bandwidths > in the fabric. For an upper layer, like MVAPICH, "negotiating" > for a rate so that all the ports are "involved" can be quite > expensive, especially if the code falls in the critical path. > Can any additional support be provided by the underlying > SA interface, so that the upper protocal layers can do the job > in a minimum time. This kind of support can be used not only for > MPI but for other stacks as well,
All the rate information can be figured out at process launch time. If at launch time each MPI process used the SA to query for path information to connect to all the other processes, then the application could also look at the returned set of paths and figure out what the slowest rate in the fabric is and use that for multicast traffic. Note that point-to-point traffice, whether reliable connected or unreliable datagram, would use the rate in the path record returned by the SA. Only multicast traffic must be aware of the rate limitations of the whole job. Hopefully that helps. - Fab _______________________________________________ openib-general mailing list [email protected] http://openib.org/mailman/listinfo/openib-general To unsubscribe, please visit http://openib.org/mailman/listinfo/openib-general
