Next: Performance Measurement for Up: Performance and Numerical Previous: Performance and Numerical

Performance and Scalability Analysis

In this section we analyze the performance and scalability of parallel algorithms and their implementation. We demonstrate scalability empirically using speed-up curves and show expressions for memory, computation, I/O, and communication costs versus problem size and machine size. The performance is measured in terms of the run time, speed-up, and Mflops. The first performance measurement is the run time for fixed machine size versus problem size. The time is reported for most components of the implementation on the CM-5 which gives the overall picture of the role of each component in terms of run time requirements within the code. We present the run times for three machine sizes for two Intel machines and two machine sizes for the IBM SP-1 computer. The section consists of three sub-sections. In Section , the CPU time is reported on a certain number of nodes of partition of CM-5, Intel Touchstone Delta, Intel Paragon, and IBM SP-1 of the particular architecture. In Section , the performance measurement in terms of speed-up is presented by speed-up curves. The scalability analysis for each portion of parallel algorithm in the parallel code is given in Section .



xshen@
Sat Dec 3 17:51:03 EST 1994