Evaluating PCJ library for graph problems - Graph500 in PCJ

In the paper we present evaluation of the PCJ (Parallel Computations in Java) library for graph problems, namely for the first kernel in Graph500 benchmark. PCJ is a library for parallel computations in Java based on PGAS (Partitioned Global Address Space) paradigm. PCJ hides communication details such as threads or network programming, which allows to develop distributed applications easily. In the library all variables are local to the program threads however some variables can be marked as shared, which makes them accessible from different tasks via one-sided communication. We used the PCJ library to implement the first kernel of Graph 500 benchmark. In the first kernel from the input tuples list an undirected graph is constructed and compressed into CSR (Compressed Sparse Row) structure. The Graph500 benchmark includes a scalable data generator (Kronecker generator) that we used to produce edge tuples containing the start vertex and end vertex for each edge. We present the evaluation of the performance of the PCJ library for the first kernel with division into data distribution, computation and data gathering. Computation and data gathering shows good scalability of the PCJ.