Previous | Next --- Slide 13 of 34
Back to Lecture Thumbnails
max

The ring approach connects each pair of vertices using $O(n)$ routing distance, so there is high latency. There are only a linear number of wires needed to connect each pair of nodes, so the ring is easily scalable. In modern processors, this ring approach is used, but there are multiple rings connecting a set of nodes. The rings may send the data in different directions (so node 0 might route to 1 in one of the rings, or 1 might route to 0 in another). This allows for easier scheduling, as the processor can choose to route a message to the left or to the right. Using multiple rings, the more connections also allows for higher bandwidth.

sjoyner

There is a scalability issue because if we cut the network in half, we will always have only 2 links crossing the cut. Therefore this has low bisection bandwidth since.

Mayank

The ring is directed topology (because communication between 2 nodes might have other nodes in between) and is blocking (because paths may collide during simultaneous communication)

rutwikparikh

Although it appears that the nodes are on the "outside", we must keep in mind that the switches have only one node connected to it. As a result, communication between two nodes might have switches (with one nodes) in between. This results in a directed topology.