Overhead considerations for parallelizing conjugate gradient
Mark K. Seager · Communications in Applied Numerical Methods · 1986
Abstract A standard preconditioned conjugate gradient (PCG) algorithm for the solution of symmetric linear systems is studied in the context of multi‐processing. It has been observed that the overheads associated with parallel processing can be expensive. Several alternative approaches are taken in order to try to minimize this cost. Examples are given for Cray X‐MP class machines using CRI's standard multitasking library and a new micro‐tasking compiler preprocessor. These approaches to utilizing multiple CPU's are discussed and overhead measurements given. The quite different PCG implementations under multitasking and micro‐tasking are discussed and compared.