> On 20 Jul 2016, at 10:04, Ahmed Bashandy (bashandy) <[email protected]> > wrote: > > Stewart, > > You are stating the most obvious:) > For a feature to work, e.g. encapsulating into an RSVP tunnel, the entry > point of the packet must be able to support that feature. That is not a > problem. That how the entire world works:)
So are you saying that your scheme only works in an rsvp only network, or in an SR only network. If so that constraint was not clear to me. If you are saying that this works in an ordinary hop by hop network, then it needs to be made much clear to the WG that EVERY ingress router needs to apply the loop avoidance two phase strategy. The diagram and presentation you used yesterday used too simple a topology to make that clear. > > SR-based ti-uloop avoidance is no different. So if a packet arrives at node > "D" for example, then node "D" must be able to steer the packet such that it > guarantees loop-freeness. > Once "D" source-routes the packet into the loop-free path towards the > destination "Z" , downstream routers from "D" towards "Z" need NOT have the > ability to do uloop avoidance or even know that packets was source-routed by > node "D" > > As for the need for using "strict" vs "loose" source routing, that is > topology dependent. The same applies for ti-lfa, rLFA, or even directly > connected LFA. > > One last thing, if you think that source routing can reduced to a "tunnel" we > might as well shutdown SPRING WG:) > SR in an mpls network can be viewed as a set of concatenated tunnels. The complexity is all in the naming of the tunnel end points and the control plane. Stewart > Ahmed > > > > >> On 7/19/2016 7:48 AM, Stewart Bryant wrote: >> Appologies if this is a duplicate. >> >> The RFC that had all of the generic methods that were known at the time of >> publication is RFC5715. >> >> The two phase methods were in section 6.2 (near-side) and section 6.3 >> (far-side). The first describing tunnelling the traffic towards the repair >> (and continue to use the repair), the second describing tunnelling traffic >> towards the destination. For the purposes of this discussion source routing >> of all flavours can be considered a type of tunnel. >> >> If this approach is a new genetric two phase method, it would be useful to >> articulate it in general terms. >> >> The type of topology that I was trying to explain in the meeting is as >> follows >> >> A-B-C-D-E-F-H-I-J.....W >> | | >> | | >> +------X-Y-Z----------+ >> >> All costs are 1 and Y has failed. >> >> Traffic to Z can enter enywhere, and is protected by X. >> >> When the network starts to converge ALL the routers A..J will need to update >> their fib to forward towards Z via W rather than towards X via A. >> >> If they do this in a random order as would be the case without LF >> convergence then you may precipitate microlooping. >> >> What you need to do is to force the packets toward either X or Z using a >> tunnel, or a source routed path, and as far as I can see you need to do that >> at every point of potential entry into the network, in the above case A..J, >> else you risk a microloop. >> >> Now I suppose that if ALL packets were source routed, then you could >> consider that the network was constantly in the first phase, but I think >> that you would need to use strict source routing, rather than loose source >> routing else it reduces the the problem I describe above. >> >> >> - Stewart >> >> _______________________________________________ >> rtgwg mailing list >> [email protected] >> https://www.ietf.org/mailman/listinfo/rtgwg > _______________________________________________ rtgwg mailing list [email protected] https://www.ietf.org/mailman/listinfo/rtgwg
