Complier-Directed Parallelization of a Multilevel Solver Algorithm for EDP

EDP (Electronic Data Processing Accounting) develops rapidly with the booming of computer industry. Solving a continuous time Markov Chain to determine its steady state is an important problem in EDP. Solution methods such as Gauss-Seidel (GS), Successive Overrelaxation (SOR) and Jacobi are simple to implement, but may require an excessive amount of time to compute a result. In this paper, we study the Multilevel (ML) algorithm. We explore the available parallelism in the ML algorithm from both a theoretic and a practical viewpoint.