Performance Studies of a Parallel Processor for Large Linear Programming Problems

摘要

In order to speed up the solution of large linear programming (LP) problems a new parallel processor will be developed. For this processor a parallel algorithm is established and investigated. Several pivot searching strategies are described and compared using some LP models which have been made for simulation purposes. It appears that using k processors, the speed for solving an LP can be about k times faster (when there is no bus-congestion). It also appears that a little effort in the pivot searching phase can reduce the total number of iterations for solving an LP model and thus the total computing time.