Using Static Code Analysis to Improve Performance of GridRPC Applications

Abstract

The paper presents an approach to improve performance of GridRPC applications by statically analysing dynamic workflows. An extension to GridRPC API is used to specify the region of code to apply static code analysis to, during the compilation phase. The information collected is then used at runtime for building a graph of dependencies between tasks, which is analysed to assign servers to tasks in an optimal way, minimising the time of computation and communication. This approach handles branching and looping correctly by building an extended dependency graph, which covers all branches of the code. The experimental results are provided to show that in many practically important cases this approach leads to better results than individual mapping of tasks or run-time task discovery.