On Sat, 2004-12-18 at 10:46, Roland Dreier wrote: > Hal> From Woody's MPI data (logs in > Hal> > http://openib.org/pipermail/openib-general/2004-December/007167.html), > Hal> there appear to be some numbers above 1x to me. 1x is 2.5 > Hal> Gbps minus signalling is really a 2 Gbps data rate. There are > Hal> numbers in the log in excess of 350 MBytes/sec which is 2.6 > Hal> Gbps. > > Fair enough but I still don't think there are many fabrics right now > where not setting static rate for IPoIB will even be required, let > alone have a practical effect.
Agreed. > I don't object in principle to doing this right, it just seems like a > fair bit of work to write code to retrieve the local PortInfo to get > the active link speed and link width so that we can calculate our > local rate and get the right IPD. Link speed active (other than locked at 2.5 Gbps) was added at IBA 1.2. > To solve this, do you think it makes sense to add the active link > speed and link width to struct ib_port_attr so that it's easy for ULPs > to get them? Otherwise ULPs would have to do their own PortInfo queries. That makes sense to me. There's one catch: since link width/speed active can change under the covers, this info cannot be cached by the application or there needs to be a tie in to an event for either of these (1.2) or link width active (1.1) changing. -- Hal _______________________________________________ openib-general mailing list [EMAIL PROTECTED] http://openib.org/mailman/listinfo/openib-general To unsubscribe, please visit http://openib.org/mailman/listinfo/openib-general
