Performance evaluation of data management layer by data sharing patterns for grid RPC applications

Grid RPC applications, typically master-slave type of applications, often need to share large size of data among workers. For efficient and flexible data sharing among a master and workers, we have designed and developed a data management layer called OmniStorage. This paper enhances the OmniStorage functionality to accommodate several data transfer methods and to specify a hint for data sharing patterns, and develops a set of synthetic benchmarks based on data sharing patterns required by grid RPC applications to evaluate the performance and characteristics of each data transfer method. The performance evaluation and the hint help to select a suitable data transfer method, which improves the efficiency and also scalability of grid RPC applications that need to share large size of data among a master and workers.