Application Problem Size Sequential Exec. Time(seconds) Exec. Time on 8 nodes (seconds) Speedup
FFT 1024x1024 17 3 5.5
Barnes 32K molecules 38 21 1.8
LU 2048x2048 208 28 7.4
Ocean 514x514 19 4 4.9
Water-Spatial 32K molecules 2033 370 5.5
Water-Nsquared 4K molecules 363 60 6.1
Radix 8M keys 16 4 4.3

Results were obtained on a cluster of 8 nodes (each node being an SMP with dual Pentium II 300 MHz processors and 512MB RAM). One of the processors was used to handle messages, removing the interrupt overhead from the main processor. Although our HLRC protocol (over VIA) can execute on single-processor PC nodes, we chose the above configuration to determine an upper bound on the performance.