| Summary: | 碩士 === 中華大學 === 資訊工程學系碩士班 === 93 === The data redistribution problems on multi-computers had been extensively studied. Irregular data redistribution has been paid attention recently since it can distribute different size of data segment of each processor to processors according to their own computation capability. High Performance Fortran Version 2 (HPF-2) provides GEN_BLOCK data distribution method for generating irregular data distribution. In this thesis, we develop an efficient scheduling algorithm, Smallest Conflict Points Algorithm (SCPA), to schedule HPF2 irregular array redistribution. SCPA is a near optimal scheduling algorithm, which satisfies the minimal number of steps and minimal total messages size of steps for irregular data redistribution. The algorithm intends to decrease the computation costs by firstly scheduling the conflict messages with maximum degree in the first phase of data redistribution process and then followed by arranging the rest messages of processors in a non-decreasing order of message size into the schedule of data redistribution. To evaluate the performance of our proposed algorithm, we have implemented SCPA along with the divide-and-conquer algorithm. The simulation results show that SCPA has significant improvement of communication costs compared with the divide-and-conquer algorithm.
|
|---|
