Previous | Next --- Slide 14 of 34
Back to Lecture Thumbnails
Xelblade

Note that this is different from the crossbar because each intersection is an actual node instead of the nodes sitting on the outside. (It is a direct network rather than an indirect one.)

xiaowend

Here, we use fewer switches and have lower latency. But it is blocking and the switches in the middle can be busy and blocking when many nodes in left uppermost send packets to nodes in right undermost.

sjoyner

Average latency is O(sqrt N) because the width and height are both about sqrt N in terms of the number of nodes. In the worst case scenario, to get from node A to node B, we have to travel sqrt N nodes to the left or right to get to the correct column and sqrt N nodes to the top or bottom to get the correct row. 2*sqrt N = O(sqrt N)

mitraraman

Here we can see that every node has a switch and a fixed number of links (or wires) connected to it, so it has a blocking topology.

kayvonf

The bisection bandwidth of a mesh is O(sqrt(N)). Here's a simple way to think about it. In a square 2D mesh topology like the one above, there are N nodes, and any vertical line will bisect sqrt(N) links.