Rift Status PagesRouting In Fat Trees (Active WG)
Rtg Area: Alvaro Retana, Deborah Brungard, Martin Vigoureux | 2018-Feb-09 —Chairs:
IETF-108 rift minutes
Session 2020-07-29 1300-1350: Room 4 - Audio stream - rift chatroom
RIFT WG IETF 108
draft-rift-rift (Tony P)
Tony presents interesting new topics for RIFT
- horizontal links
Notes that the RIFT draft is still 12 so no changes. Issues going Python 3 (hashable problem)
Bruno: solved the problem in Python 3 some other way
RIFT-Python (Bruno R.): Playing prerecorded video.
Bruno presents disaggregation, positive then (for multiplane) negative
Questions on disaggregation:
- what’s the difference with withdrawing a route? The short answer is that the negative substracts from the encompassing aggregation, whereas withdrawal just removes the exact match.
- If the encompassing aggregation is withdrawn, do we need to substract the negative from the new encompassing aggregation over it? Andswer is yes.
Base model in good shape, very close to done. Sandy is cutoff,
Multicast (Jeffrey and Pascal T)
Jeffrey does the intro. Explains the use of a tree and PIM BIDIR/
Pascal presents how a spanning tree can be built to from a virtual Rendez-vous Point link (RPL) between the ToF and the subToF; The result is a panning tree that joins either all leaves or all nodes.
Pascal shows an optimization where the root is in the subtof which divides evenmy the replication work between ToF and subToF
When we want to use many trees for load balance, hash can be used
Tony P asks about consistency. Pascal agrees that the hash computation must be the same by all so all need the same information. Ideally packets would be tagged with the selected tree so there can be no loop
Jeffrey indicates that there can be no loop. Pascal agrees for outside of the RPL if forwarding is up and then through the RPL and then down. But inside the RPL there could be loops, (also with BIDIR if the packet is injected at mid point)