Exploiting Inherent Task-Based Parallelism in Object-Oriented Programming

Abstract

While object-oriented programming (OOP) and parallelism originated as separate areas, there have been many attempts to bring those paradigms together. Few of them, though, meet the challenge of programming for parallel architectures and distributed platforms: offering good development expressiveness while not hindering application performance. This work presents the introduction of OOP in a parallel programming model for Java applications which targets productivity. In this model, one can develop a Java application in a totally sequential fashion, without using any new library or language construct, thus favouring programmability. We show how this model offers a good trade-off between ease of programming and runtime performance. A comparison with other approaches is provided, evaluating the key aspects of the model and discussing some results for a set of the NAS parallel benchmarks.