Graphalytics: from Benchmarking to Performance Engineering

Keynote, PELGA workshop at Euro-Par

Download PDF Slides


Graphs model social networks, human knowledge, and other vital information for business, governance, and academic practice. Although both industry and academia are developing and tuning many graph-processing algorithms and platforms, the performance of graph-processing platforms has never been explored or compared in-depth. Moreover, graph processing exposes new bottlenecks in traditional HPC systems (see differences in Top500 and Graph500 rankings).

In this talk, we introduce the LDBC Graphalytics benchmark, which focuses on batch full-graph analytics. Attendees will learn about methods and tools for performance evaluation and optimization for graph processing platforms, and hear about our view that the performance forms a dependency triangle Platform-Algorithm-Dataset. We will present real-world experiences with commonly used systems, in particular with graph-processing platforms such as Giraph, Graphmat, OpenG, PGX.D, and PowerGraph. Moving towards performance engineering, and in particular bottleneck analysis and performance diagnosis, we further focus on Granula, a component of LDBC Graphalytics for in-depth performance analysis for graph-processing systems with various data processing models.