Question: How exactly do we get O(1) latency in the crossbar topology? To go from one corner to another, don't we need to get through 2N switches?
This comment was marked helpful 0 times.
chaihf
@rokhinip. With Crossbar, every node has a direct path to other nodes without getting through switches like Mesh, so the latency is almost O(1). And that's why it has a higher cost than other topologies.
This comment was marked helpful 0 times.
DunkMaster
O(sqrt(N)) is an average latency
This comment was marked helpful 0 times.
rokhinip
@chaihf, the comments on the slide for graph seem to suggest otherwise. But I agree with you for we won't get O(1) latency.
Question: How exactly do we get O(1) latency in the crossbar topology? To go from one corner to another, don't we need to get through 2N switches?
This comment was marked helpful 0 times.
@rokhinip. With Crossbar, every node has a direct path to other nodes without getting through switches like Mesh, so the latency is almost O(1). And that's why it has a higher cost than other topologies.
This comment was marked helpful 0 times.
O(sqrt(N)) is an average latency
This comment was marked helpful 0 times.
@chaihf, the comments on the slide for graph seem to suggest otherwise. But I agree with you for we won't get O(1) latency.
This comment was marked helpful 0 times.
@chaihf, why is the latency almost O(1)?
This comment was marked helpful 0 times.