Hi Lizhong, > not draft-ietf-tictoc-1588overmpls aware. You will note that this draft > also has some text on the non 1588aware LSRs. These LSRs could be doing > ECMP and without an entropy label, may cause PTP messages to be load > distributed across different paths. [Lizhong] then in section 13, it is said: "This capability MUST then be taken into account during path computation to prefer nodes that advertise themselves as 1588-aware, so that the PTP LSPs can be properly handled." How to understand "prefer nodes"? If there is a shortest path with some non 1588aware LSRs, and a less optimal path with all 1588aware LSRs, which path the ingress LER should take?
[Manav] There is nothing new here and it happens all the time while computing CSPF. You pick the path which satisfies your contraints. Now, if you want a path with all 1588aware LSRs then thats precisely what CSPF will return you. If you want to insist on a shorter path then you need to add other constraints in addition to this. > > > > > >The head end > > > when setting up the RSVP path would know this (by inspecting the IS- > > > IS or OSPF TE info) and could in such cases make use of the Entropy > > > label to ensure that all PTP packets follow the same path. > > [Lizhong] How to use entropy label to ensure all PTP packets follow the > > same path? > > Using an entropy label will ensure that all PTP packets in one direction > follow the same path. [Lizhong] since control word is required for 1588 over PW encapsulation, if without entropy label, all PTP packets in one direction will also follow the same path. [Manav] I think we must use a stronger language in Sec 5.2 regarding using control word. I think i will concur with you here that if CW is being used then we may not need an entropy label. However, i dont see a strong reason why we should restrict it. The current draft says that control word usage is optional and i dont find any issue with that. Do you? > > > If there is entropy label, it is possible that the forward > > and reverse path would be different. > > Since LSPs are unidirectional there is no guarantee that the two paths > would be symmetrical. So this really is an issue orthogonal to the use > of entropy labels for carrying PTP traffic. > > We should probably add some text in the draft that says that the two > paths must be symmetrical. [Lizhong] If there is a boundled interface between two LSRs, the control plane will only see one logical interface. Then it is still possible the forward and reverse path be different at the physical level in ECMP case. [Manav] Is the assumption that the two ends of LAG could be connected to two different systems, something like a multi-chassis lag? I am not sure if there is much that we can do about this except add a note against such a form of load distribution in the document. Cheers, Manav
_______________________________________________ TICTOC mailing list [email protected] https://www.ietf.org/mailman/listinfo/tictoc
