1993 International Conference on Parallel Processing EMPIRICAL EVALUATION OF INCOMPLETE HYPERCUBE SYSTEMS
Nian-Feng Tzeng · 1993
The incomplete hypercube provides far better incremental flexibility than the complete hyper¬ cube, whose size is restricted to exactly a power of 2. In this paper, the performance of incomplete hypercube sys¬ tems is evaluated empirically. The simulation results reveal that mean latency for delivering messages is roughly the same in an incomplete hypercube as in a com¬ patible complete hypercube, unless the message genera¬ tion rate is extremely high (≥ 0.9). It is also found that mean latency for messages traversing links with heavy traffic can be appreciably larger than the mean latency of overall messages. When the link communication capabil¬ ity is doubled, mean message latency becomes virtually uniform no matter whether or not the message traverses a link with heavy traffic, confirming that the incomplete hypercube can be made congestion-free easily to guaran¬ tee high performance under any traffic load.