Dear WG: I received an unicast a question that was, bascally, what's the difference between MRT and ARCs.
>From the draft, hopefully it appears that an ARC is a new tool with new >properties, up to us to find interesting applications and maybe find new >solutions to existing problems. The core of that tool is the reversible link, >so that we form downwards U-shaped arcs as opposed to arrows for routing. But maybe the simpler approach is to make an exercise and see what we get - my current understanding here, Gábor please correct me if I'm wrong -. So, say we have this simple network. 2 --- 3 | | | | 1 --- 4 | | | | Destination I illustrated naming the routers with a value that matches my understanding of the values of the partial order, to make things simple. We end up with 2 ARCs and 2 ears, which are basically collocated. The ARC view is 2 <=> 3 | | where links 2 - 3 and 1 - 4 are reversible. V V 1 <=> 4 | | V V Destination (Omega) A first property of the ARCs can be seen immediately. ARCs are hierarchical routing friendly. We can collapse each ARC to simplify the representation which becomes: 2 | | V V 1 | | V V Destination If we break any link both ears and ARCs provide an alternate route, that's FRR for you. Now let us see what happens with more than one breakage. There are double breakages that isolate a piece of the network, say if links 2 - 1 and 3 - 4 both break, there is no solution in the world that can keep 3 connected to the Destination. Say now that 3 - 4 and 1 - Destination are both broken a 2 has a packet to send. 2 --- 3 | | V X 1 --- 4 | | X V -------------Destination (Omega) - With ears, using the increasing order 2 will pass to 3, and 3 to 4 will fail. So the packet is switched to decreasing order and the packet reaches 1 via 2. 1 fails to send to the destination and we are screwed. PPPPP>> <<ppppp 2 --- 3 PPP packet progressing in increasing order | | | | p | ppp packet progressing in decreasing order V X V X p X 1 --- 4 1 --- 4 V --- 4 When packet reaches 1 must be dropped | | | | | | because increasing again may cause a loop X V X V X V -----------------------------------------Destination (Omega) - With ARCS, say that bad luck the shortest path is via 3. 2 passes to 3 and 3 to 4 will fail. So 3 u-turns the packet along the ARC, back to 2. Some tagging indicates that the packet was u-turned which can happen only once in that ARC. 2--p->3 2<==p=3 2 --- 3 2 --- 3 2 --- 3 -p-> in shortest path | | | | p | | | | | =p=> vs. returned V X V X V X V X V X packet 1 --- 4 1 --- 4 1 --- 4 1==p=>4 1 --- 4 | | | | | | | | | p packet reaches Omega X V X V X V X V V V ------------------------------------------------------------------Destination (Omega) say that bad luck again the shortest path for 2 is via 1. 2 passes to 1 and since the ARC is exited the marking is removed. 1 fails to send to the destination, marks the packet and sends it along its ARC. Packet reaches Destination via 4. IOW, ARCs form isolated recovery domains and allow up to as many breakages as the ARC or Comb has exits, with no disruption. Also, ARCs can form multi-ended structures called combs with better recovery capabilities than a 2-ended structure. For bicasting, the difference will be a bit more subtle. In a biconnected graph, ears will build non congruent path every time, but they might be far away from shortest. ARCs will attempt to stay closer to shortest but may incur collisions, which are then resolved. Then ARCs can be employed for other purposes. We have an interesting approach to the Olympic rings issue for instance. Cheers, Pascal
_______________________________________________ rtgwg mailing list [email protected] https://www.ietf.org/mailman/listinfo/rtgwg
