Substream-Centric Maximum Matchings on FPGA

Abstract

Developing high-performance and energy-efficient algorithms for maximum matchings is becoming increasingly important in social network analysis, computational sciences, scheduling, and others. In this work, we propose the first maximum matching algorithm designed for FPGAs; it is energy-efficient and has provable guarantees on accuracy, performance, and storage utilization. To achieve this, we forego popular graph processing paradigms, such as vertex-centric programming, that often entail large communication costs. Instead, we propose a substream-centric approach, in which the input stream of data is divided into substreams processed independently to enable more parallelism while lowering communication costs. We base our work on the theory of streaming graph algorithms and analyze 14 models and 28 algorithms. We use this analysis to provide theoretical underpinning that matches the physical constraints of FPGA platforms. Our algorithm delivers high performance (more than 4× speedup over tuned parallel CPU variants), low memory, high accuracy, and effective usage of FPGA resources. The substream-centric approach could easily be extended to other algorithms to offer low-power and high-performance graph processing on FPGAs.

FAST: FPGA-based Subgraph Matching on Massive Graphs

“…This is reflected by the recent interests in developing various graph algorithms and graph processing frameworks on FPGAs. For examples, [10] applies FPGAs to speed up Maximum Matching and [41] utilizes FPGAs to accelerate the process of the Single-Source-Shortest-Paths. In addition to these specific graph algorithms on FPGAs, a lot of effort was devoted to design generic frameworks for facilitating the implementation of graph algorithms on FPGAs [15], [24], [40].…”

“…FPGAs have also been rolled out by major cloud service providers such as Amazon Web Services [2], Alibaba [3], Tencent [4], Huawei [5], and Nimbix [6]. In academia, it has become a promising trend to use FPGAs to speed up different research problems including many graph processing problems [10], [15], [24], [40], [41]. Nevertheless, subgraph matching algorithms using FPGAs have not been developed in the literature.…”

Pebbles, Graphs, and a Pinch of Combinatorics: Towards Tight I/O Lower Bounds for Statically Analyzable Programs

“…Ballard et al [45] present an extensive collection of linear algebra algorithms. Moreover, a large body of work exists for minimizing communication in irregular algorithms [46,47], such as Betweenness Centrality [5], min cuts [48], BFS [49], matchings [50], vertex similarity coefficients [51], or general graph computations [52,53,53]. Many of them use linear algebra based formulations [54].…”

High-Performance Parallel Graph Coloring with Strong Guarantees on Work, Depth, and Quality

“…Some works impose additional restrictions, for example coloring balance, which limits differences between numbers of vertices with different colors [138]- [140]. Other lines of related work also exist, for example on edge coloring [141], dynamic or streaming coloring [142]- [148], k-distance-coloring and other generalizations [149]- [151], and sequential exact coloring [152]- [154]. There are even works on solving graph coloring with evolutionary and genetic algorithms [155]- [157] and with machine learning methods [158]- [163].…”