Enabling High Performance Computing For Java Applications Using The Message-Passing Interface

In the last year a number of the applications written in Java programming language have been developed which will benefit from running on High Performance Computing (HPC) systems. For some of them, such as coming from the Semantic Web domain, use of HPC is a key factor in achieving the required scalability and performance characteristics. Parallelisation enables such applications to utilize HPC most efficiently. The Message-Passing Interface (MPI) is one of the most popular implementation standards for parallel applications. Despite several implementations of MPI for Java, such as mpiJava and MPJ Express, the MPI standard has been for a long time MPI out of scope of the real-life Java applications, due to the design features pertained to this programming language (such as Garbage Collector) and lack of use cases for HPC support. However, the tremendously growing amount of the data processed by the Semantic Web applications makes parallelisation the major challenge also for Java applications. In this paper, we describe a parallelisation technique based on MPI and show its application for a case study Java application, used for extracting semantically similar words from the co-occurrence statistics of the words in text data. We show that use of MPI allows a Java application to achieve good performance and scalability on several compute architectures and with different MPI implementations.