Previous | Next --- Slide 10 of 34
Back to Lecture Thumbnails
mschervi

Question Could somebody explain how this picture shows that every node is connected to every other node? I see that there is a path from every node to every other node, but that is the case in the other topologies we looked at as well. From the picture it looks like a packet would have to travel through several switches to get to its destination, but I think in lecture Kayvon mentioned that the links could be thought of as instantaneous?

xiaowend

@mschervi In my opinion, each node is directly connected to all switches in its row and its column.

akashr

@Xiaowend, Based on your reasoning, would we expect n-1 wires from each node in the graph, with each wire going to a different switch? I am still confused about this as the slide is slightly confusing.

Also, can someone go over the difference between direct and indirect graphs in terms of this architecture setup.

askumar

@akashr This network is indirect because all of the nodes sit on the "outside" of the the network. That means, to pass data between any two nodes, the data has to go between different switches, but never has to pass through another node. If a graph has a node on the actual path of data transmission, then it is a direct network.

Mayank

I think @Xiaowend's comment is correct. Every switch connects a directed pair of node (i,j). Since, there are N^2 possible pairs, hence we need N^2 switches. This must be true to ensure:
1. O(1) latency : If there weren't wires from node i to each switch in its row and column, then we would not get O(1) latency (1 hop to reach any other node j)
2. Non-Blocking: Because of N^2 switches with different wires, any 2 nodes, i and j, can communicate to each other irrespective of any other communication that might be occurring.

monster

Here @Xiaowend's comment is correct. It is just like a complete directed graph and you put a switch on each edge in the graph.