GraphMat

Abstract

Given the growing importance of large-scale graph analytics, there is a need to improve the performance of graph analysis frameworks without compromising on productivity. GraphMat is our solution to bridge this gap between a user-friendly graph analytics framework and native, hand-optimized code. GraphMat functions by taking vertex programs and mapping them to high performance sparse matrix operations in the backend. We thus get the productivity benefits of a vertex programming framework without sacrificing performance. GraphMat is a single-node multicore graph framework written in C++ which has enabled us to write a diverse set of graph algorithms with the same effort compared to other vertex programming frameworks. GraphMat performs 1.1-7X faster than high performance frameworks such as GraphLab, CombBLAS and Galois. GraphMat also matches the performance of MapGraph, a GPU-based graph framework, despite running on a CPU platform with significantly lower compute and bandwidth resources. It achieves better multicore scalability (13-15X on 24 cores) than other frameworks and is 1.2X off native, hand-optimized code on a variety of graph algorithms. Since GraphMat performance depends mainly on a few scalable and well-understood sparse matrix operations, GraphMat can naturally benefit from the trend of increasing parallelism in future hardware.

Exploiting NVM in large-scale graph analytics

“…The generator provides graphs that correspond to the structure of realworld graphs of interest and is widely used by the graph analytics community for system evaluation [29], [15], [28], [30].…”

“…For Pagerank and BFS, the input is a scale 26 graph with 2 26 nodes and 2 30 connections. Since BFS requires an undirected graph, we add reverse edges to the dataset.…”

“…Graphmat expresses computation as a vertex program but the operations are internally converted to sparse-vector matrix computations [30], resulting in a better compute time while maintaining a simple intuitive API.…”

“…We chose a representative subset based on the work in [29]. Section 5 shows the impact that different bandwidth and latency points of NVM have on Galois [21], Graphlab [15], Graphmat [30] and X-Stream [28] running Pagerank, Bread-First Search (BFS), Triangle Counting and Collaborative Filtering. We did not find a publicly available implementation of Collaborative Filtering for Galois, hence we do not evaluate Galois for this algorithm.…”

A Comparison of Parallel Graph Processing Implementations

“…Most performance analysis of graph processing systems come from the empirical results of each new library designer's publications. For example, GraphMat [29], Pow-erGraph [9], and GraphBIG [22] present a comparison of performance between their system and a selection of other software packages and in general any new approach will compare itself to existing implementations. We assume the performance results which accompany a new library are at risk of bias because the authors typically know their own package better than the competition and as such are aware of more optimizations.…”