I've read several artical about it, but can't seem to understand this. What sort of routing algorithm is typically used in the fat-tree topology?
5
-
What do you mean? This is a fixed sort of design/configuration. For instance, Cisco recommends a 20:1 access to distribution bandwidth ratio, and a 4:1 distribution to core ratio. This was developed after studying traffic. This is a classic fat tree, but there is no real algorithm.– Ron Maupin ♦Commented May 27, 2016 at 13:34
-
Well, I'm attempting to implement a fat-tree in an SDN environment. I've made static routes to everything, but not entirely certain of how changes in link states should propagate up the tree.– TheSjiraffen123Commented May 27, 2016 at 13:38
-
You are going to need to provide a lot more information if you want help with that. Basically, you want to give the endpoints enough bandwidth, but, as you get closer to the root, you need to increase the bandwidth to a level where traffic drops are acceptable. How much you need to increase it by will depend on what sort of traffic you have. Trying for a 1:1 ratio is ideal, but it is impractical and expensive. Cisco has the general recommendation based on traffic studies, but you could tweak it to your specific needs.– Ron Maupin ♦Commented May 27, 2016 at 13:47
-
fat trees. ...? Sounds like weed to me.– Ronnie SmithCommented May 27, 2016 at 17:36
-
Did any answer help you? If so, you should accept the answer so that the question doesn't keep popping up forever, looking for an answer. Alternatively, you could provide and accept your own answer.– Ron Maupin ♦Commented Aug 14, 2017 at 4:31
Add a comment
|
2 Answers
It depends if your tree is layer 2 or layer 3. Either way, you need a protocol that provides equal-cost multi-path (ECMP), so you can use parallel paths from source to destination.
Multi-path Routing:
Ex. ECMP:
Without it, the largest cluster = 1,280 nodes. It helps to perform static load splitting among data flows. It Leads to over subscription for simple communication patterns. Routing table entries grows multiplicatively with number of paths, cost ++, lookup latency ++.
