Improving the performance of graph transformation execution in the BSP model - By improving a code generator for EMF Henshin

Abstract

Model-Driven Engineering is widely used within Software Engineering to increase the level of abstraction and thereby increase the efficiency and maintainability of software projects. Combining the notion of Graph Transformations with the Bulk Synchronous Parallel (BSP); big-data problems can be solved using a model-driven approach. However, such solutions are still in their early phases and do not provide the required performance. This presents the performance issues with a current approach for executing Henshin transformation models on a Hadoop cluster utilizing the BSP implementation Apache Giraph. A solution, in the form of an improved code generator, is developed in order to address the problems. The improved approach is evaluated using a purposefully designed set of synthetic graphs, transformations rules and a real data example. The real data example is the IMDB movie database and the evaluation is performed on a compute cluster. The results indicate that the new solution performs better and uses less memory for the synthetic graphs and rules with particular characteristics while it in this particular real data example shows no significant difference.