Efficient and Scalable Graph Parallel Processing With Symbolic Execution

Abstract

Existing graph processing essentially relies on the underlying iterative execution with synchronous (Sync) and/or asynchronous (Async) engine. Nevertheless, they both suffer from a wide class of inherent serialization arising from data interdependencies within a graph.In this article, we present SymGraph, a judicious graph engine with symbolic iteration that enables the parallelism of dependent computation on vertices. SymGraph allows using abstract symbolic value (instead of the concrete value) for the computation if the desired data is unavailable. To maximize the potential of symbolic iteration, we propose a chain of tailored sophisticated techniques, enabling SymGraph to scale out with a new milestone of efficiency for large-scale graph processing. We evaluate SymGraph in comparison to Sync, Async, and a hybrid of Sync and Async engines. Our results on 12 nodes show that SymGraph outperforms all three graph engines by 1.93x (vs. Sync), 1.98x (vs. Async), and 1.57x (vs. Hybrid) on average. In particular, the performance for PageRank on 32 nodes can be dramatically improved by 16.5x (vs. Sync), 23.3x (vs. Async), and 12.1x (vs. Hybrid), respectively. The efficiency of SymGraph is also validated with at least one order of magnitude improvement in contrast to three specialized graph systems (Naiad, GraphX, and PGX.D).

AsynGraph

“…AsynGraph is compared with four existing systems, i.e., Gunrock v1.1 [29], Groute v1.0 [4], Tigr [12], and Di-Graph [35]. Besides, AsynGraph is also compared with SymGraph-G, TP-G, FBSGraph-G, and HATS-G, which are the versions of the approaches proposed in SymGraph [37], TP-X [17], FBS-Graph [34], and HATS [25], and are extended by us for GPU-based graph processing, respectively. We also extend Enterprise [19] to support PageRank and SSSP based on its supported BFS.…”

“…Hub 2 -Labeling method [14] is proposed to reduce the search space for the query of k-degree shortest path. SymGraph [37] tries to overlap the communication delay between nodes and the computation on the nodes through generating unknown symbols for the vertices without available states of their neighbors. However, when using SymGraph for iterative graph processing on the GPU, high runtime overhead is incurred to generate many unknown symbols and it also needs high reprocessing cost to update all related vertices according to the arrived new vertex state.…”

A Survey on Graph Processing Accelerators: Challenges and Opportunities

“…Compiler supporting is an effective way to fill the gap between high-level programming and low-level graph iteration. Symbolic execution is used to parallelize the dependent computations of vertices for achieving compelling performance results on general-purpose processors [132] . Execution parallelism can be also explored for irregular applications by aggressively scheduling execution dependencies at compile time [133] .…”