An Efficient Communication Scheduling Method for the Processor Mapping Technique Applied Data Redistribution

Abstract

Array redistribution is usually required for more efficiently executing a data-parallel program on distributed memory multi-computers. In performing array redistribution using synchronous communication mode, data communications among the processors should be properly arranged to avoid incurring higher data transfer cost. Some efficient communication scheduling methods for the Block-Cyclic redistribution have been proposed. On the other hand, the processor mapping technique can help reduce the data transfer cost of redistribution. To avoid degrading the benefit of data transfer cost reduction, it is needed to construct optimal communication schedules for the redistribution in which the processor mapping technique is applied. In this paper, we present a unified approach to constructing optimal communication schedules for the processor mapping technique applied Block-Cyclic redistribution. The proposed method is founded on the processor mapping technique and can more efficiently construct the required communication schedules than other optimal scheduling methods.

Operators for Data Redistribution: Applications to the STL Library and RayTracing Algorithm

“…Our work focuses on the first approach, that is, static-scheduled distributed SPMD programs. Dataredistribution techniques for load balance have been studied in the context of data-parallel languages and parallel compilers, such as High Performance Fortran [5,6]. However, the algorithms developed in this context are dependent on specific data partition policies, fixed at compile time, and in general designed to derive homogeneous sizes for dense arrays.…”

A message passing strategy for array redistributions in a torus network

“…In [8][9][10] processor mapping techniques for dynamic data redistribution are described. The main idea of these techniques is to minimize the volume of data to be exchanged between processor sets by mapping logical processor IDs to redistributed data elements without violating the redistribution patterns.…”