A Study of Design Choices in Distributed Graph Processing Frameworks

Abstract

In the past decade, various distributed data processing systems have been developed. Among various challenging problems in big data processing, the importances of graph processing engines are being emphasized as it covers various use cases in social networks. Most of the proposed graph processing frameworks have common aspects such as BSP or GAS synchronization model, graph partitioning methods, network communication model, in-memory processing or disk-based processing, etc. These design factors have a significant impact on the overall performance and their usability. This thesis identifies these design choices of various distributed graph processing frameworks and it categorizes the frameworks according to them.