PERFORMANCE EVALUATION OF PARALLEL SURFACE GENERATION

This quantitative study evaluates the performance of parallel surface modeling algorithms on actual/physical and virtual/cloud multi-core systems/machines; these kinds of parallel computing systems are underutilized, in general, and barely used for this application, in particular. Experiments were performed on these machines/systems and performance metrics, including execution time, speedup and efficiency, were obtained for various algorithms that implement various parallelization strategies of the incremental insertion technique of the Delaunay triangulation algorithm. The t-test was also employed to determine whether or not differences between the performance of these algorithms (with respect to the aforementioned performance metrics) on actual and virtual ¿versions¿ of these systems, as well as during different runs of the algorithms on both versions of the machines, are statistically significant. Results show that (1) The physical system is approximately twice faster than the cloud machine at executing the programs for the various parallelization strategies and (2) All parallelization strategies demonstrate poor scalability, as the number of threads/cores used increases.