Space-based approach to high-throughput computations in UNICORE 6 grids

Abstract

We explore a novel approach to high-throughput, embarassingly parallel applications in UNICORE 6 based Grids. This is an XML centric tuple space based approach inspired by JavaSpaces, with an implementation using the UNICORE 6 WSRF framework. Other approaches such as the layered workflow and data splitting architecture developed in the Chemomentum project, batch processing using the UNICORE commandline client UCC and concurrent programming using the HiLA Java API are discussed as well. Performance and scalability evaluations are presented, backed up by preliminary experimental results comparing our approach to the standard batch mode of the UNICORE commandline client. © Springer-Verlag Berlin Heidelberg 2009.