Measures of effectiveness (MOEs) for parallel architectures

This study uses commonly recognized architectural descriptive metrics as architectural measures of effectiveness. The architectural descriptive metrics studied were: (a) architectural diameter and latency, (b) number of communication links and cost of the system, (c) number of processors, and (d) response time behavior and speedup. These MOEs were applied to the (a) square mesh, (b) binary tree, (c) ring or linear array, and (d) hypercube topologies. The application used for the analysis was the matrix multiply. A discussion on each of the above metrics is presented, as well as the architectures studied. The results support the notion of matching an architecture to the problem. More significantly, many of the MOEs can be used to assess whether a better fit is being realized. Finally, it was found that architecture and algorithm granularity must be carefully tuned