Why Network Diameter Is Still Important

It has become fashionable to argue against research on new interconnection networks by claiming that with wormhole switching, the dominant communication method in modern parallel computers, routing latency is insensitive to the diameter and other topological parameters of the network; hence, the inference that network topology is unimportant and the edict that we should adhere to 2D and 3D mesh networks that offer low-cost implementations, simple algorithms, and ease of expansion. This view, if left unchallenged, could be harmful to further developments in the Jleld of parallel computer architecture. In this paper, we present various types of evidence that show diameter to be quite important when networks are compared under realistic andfair conditions that include some form of aggregate bandwidth and/or cost normalization