Implementing graph transformations in the bulk synchronous parallel model

Abstract

Big data becomes a challenge in more and more domains. In many areas, such as in social networks, the entities of interest have relational references to each other and thereby form large-scale graphs (in the order of billions of vertices). At the same time, querying and updating these data structures is a key requirement. Complex queries and updates demand expressive high-level languages which can still be efficiently executed on these large-scale graphs.We use graph transformation rules and units as a high-level modeling language with declarative and operational features for transforming graph structures. To apply them to large-scale graphs, we introduce a method to distribute and parallelize graph transformations by mapping them to the Bulk Synchronous Parallel model. Our tool support builds on Henshin as modeling tool and consists of a code generator for Apache Giraph. We evaluated our approach with the IMDb movie database on a cluster with 24 servers with 8 cores each.