SISA: Set-Centric Instruction Set Architecture for Graph Mining on Processing-in-Memory Systems

Abstract

Simple graph algorithms such as PageRank have recently been the target of numerous hardware accelerators. Yet, there also exist much more complex graph mining algorithms for problems such as clustering or maximal clique listing. These algorithms are memory-bound and thus could be accelerated by hardware techniques such as Processing-in-Memory (PIM). However, they also come with non-straightforward parallelism and complicated memory access patterns. In this work, we address this with a simple yet surprisingly powerful observation: operations on sets of vertices, such as intersection or union, form a large part of many complex graph mining algorithms, and can offer rich and simple parallelism at multiple levels. This observation drives our cross-layer design, in which we (1) expose set operations using a novel programming paradigm, (2) express and execute these operations efficiently with carefully designed set-centric ISA extensions called SISA, and (3) use PIM to accelerate SISA instructions. The key design idea is to alleviate the bandwidth needs of SISA instructions by mapping set operations to two types of PIM: in-DRAM bulk bitwise computing for bitvectors representing high-degree vertices, and near-memory logic layers for integer arrays representing low-degree vertices. Set-centric SISA-enhanced algorithms are efficient and outperform hand-tuned baselines, offering more than 10× speedup over the established Bron-Kerbosch algorithm for listing maximal cliques. We deliver more than 10 SISA set-centric algorithm formulations, illustrating SISA's wide applicability.

Motif Prediction with Graph Neural Networks

“…However, recent works argue the importance of higher-order graph organization [6], where one focuses on finding and analyzing small recurring subgraphs called motifs (sometimes referred to as graphlets or graph patterns) instead of individual links. Motifs are central to many graph mining problems in computational biology, chemistry, and a plethora of other fields [9,12,13,18,20,27,30]. Specifically, motifs are building blocks of different networks, including transcriptional regulation graphs, social networks, brain graphs, or air traffic patterns [6].…”

“…Simultaneously, heuristics do not need any training. Finally, SEAM also successfully predicts more arbitrary communities or clusters [12,13,25,36]. They differ from motifs as they do not have a very specific fixed structure (such as a star) but simply have the edge density above a certain threshold.…”

Motif Prediction with Graph Neural Networks

“…However, recent works argue the importance of higher-order graph organization [10], where one focuses on finding and analyzing small recurring subgraphs called motifs (sometimes referred to as graphlets or graph patterns) instead of individual links. Motifs are central to many graph mining problems in computational biology, chemistry, and a plethora of other fields [16,22,23,28,30,41,46]. Specifically, motifs are building blocks of different networks, including transcriptional regulation graphs, social networks, brain graphs, or air traffic patterns [10].…”

“…Finally, SEAM also successfully predicts more arbitrary communities or clusters [16,22,39,54]. They differ from motifs as they do not have a very specific fixed structure (such as a star) but simply have the edge density above a certain threshold.…”

Parallel Algorithms for Finding Large Cliques in Sparse Graphs

“…Finding large cliques has many applications in the social sciences, bioinformatics, computational chemistry, and others [1,8,12,23,33,34,43,48,51,52]. As the problem is NP-hard and remains hard even when parameterized by the size of the clique 𝑘 [28], it makes sense to consider special families of graphs for which the problem is tractable.…”