Hi Jason, Sean, Your mail triggered me to double check the IPoIB for the sentence requiring the SL and rate to follow the MCG. I was sure I have seen this in the IETF IPoIB spec and that was the reason for my comments. But I can't find it there.
So I must admit your argument regarding having the SL and rate be calculated by the PathRecord for the unicast traffic seems reasonable to me. Eitan > -----Original Message----- > From: Jason Gunthorpe [mailto:[EMAIL PROTECTED] > Sent: Tuesday, August 07, 2007 8:35 PM > To: Eitan Zahavi > Cc: Sean Hefty; OpenIB > Subject: Re: [ofa-general] [PATCH 1/4] ib/ipoib: specify > TClass and FlowLabelwith PR queries for QoS support > > On Tue, Aug 07, 2007 at 07:59:46PM +0300, Eitan Zahavi wrote: > > > > Can you clarify why you think the SL and rate should be provided, > > > versus just the traffic class and flow label? > > > The broadcast group of the IPoIB subnet should dictate the > parameters > > to be used for that subnet. > > There is no reason a unicast IPoIB path should be restricted > to having the same SL as the broadcast group. That limits the > possible topologies that can be routed efficiently. Sean's > use of the tclass to derive the SL seems correct to me. > Except in unusual cases the SL should not be specified in a PR query. > > Rate and MTU should probably be copied over. Not sure about > flowlabel though.. > > Jason > _______________________________________________ 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
