Granula: Toward Fine-grained Performance Analysis of Large-scale Graph Processing Platforms

Abstract

Big Data processing has become an integral part of many applications that are vital to our industry, academic endeavors, and society at large. To cope with the data deluge, existing Big Data platforms require significant conceptual and engineering advances. In particular, Big Data platforms for large-scale graph processing require in-depth performance analysis to continue to support the broad applicability of linked data processing. However, in-depth performance analysis of such platforms remains challenging due to many factors, among which the inherent complexity of the platforms, the limited insight provided by coarse-grained "black-box" and inefficiency of fine-grained analysis, and the lack of reusability of results. In this work, we propose Granula, a performance analysis system for Big Data platforms that focuses on graph processing. Granula facilitates the complex, end-to-end processes of fine-grained performance modeling, monitoring, archiving, and visualization. It offers a comprehensive evaluation process that can be iteratively tuned to deliver more fine-grained performance information. We showcase with a prototype of Granula how it can provide meaningful insights into the operation of two large-scale graph processing platforms, Giraph and PowerGraph.