Developing Sparse Linear Algebra, Unstructured Grids and Dynamic Programming simulation benchmarks for MPJ Express

Abstract

MPJ Express is a message passing library in Java that allows application developers to write and execute parallel applications. These applications usually have high computation and memory requirements. The aim of this project is to create realistic benchmark applications that will help to evaluate the performance of MPJ Express. The objective is to develop a new benchmark suite for the MPJ Express software. This new suite is based upon 13 Dwarfs identified in a report titled “The Landscape of Parallel Computing Research: A View from Berkeley” (http://www.eecs.berkeley.edu/Pubs/TechRpts/2006/EECS-2006-183.html). A dwarf is an algorithmic method that captures a pattern of computation and communication. A unique feature of this new benchmark suite is that it closely represents real-world HPC applications. Traditionally, existing benchmarks like NPB-MPJ (http://www.nas.nasa.gov/publications/npb.html) are more specific in nature. It is therefore important to have practical benchmarks in place for parallel programming tools such as MPJ Express. The project goal is development of 4 dwarfs implementation mentioned below out of 13. Dynamic programming is a method of solving complex problem by breaking it down into simpler sub- problems. Backtrack and Branch-and-Bound algorithms are used to search a large space by discarding particular branches to find optimal solution. Sparse linear algebra involves the data set having large density of zeros. Unstructured grids are widely accepted in CFD because this gives us far more flexibility and computer resources are also efficiently utilized.