Challenges in Parallel Graph Processing

Abstract

Graph algorithms are becoming increasingly important for solving many problems in scientific computing, data mining and other domains. As these problems grow in scale, parallel computing resources are required to meet their computational and memory requirements. Unfortunately, the algorithms, software, and hardware that have worked well for developing mainstream parallel scientific applications are not necessarily effective for large-scale graph problems. In this paper we present the inter-relationships between graph problems, software, and parallel hardware in the current state of the art and discuss how those issues present inherent challenges in solving large-scale graph problems. The range of these challenges suggests a research agenda for the development of scalable high-performance software for graph problems.

Parallel Breadth-First Search on Distributed Memory Systems

“…The absolute performance numbers we achieve on the large-scale parallel systems Hopper and Franklin at NERSC are significantly higher than prior work. The performance results, coupled with our analysis of communication and memory access costs of the two algorithms, challenges conventional wisdom that fine-grained communication is inherent in parallel graph algorithms and necessary for achieving high performance [25]. We list below optimizations that we intend to explore in future work, and some open questions related to design of distributed-memory graph algorithms.…”

“…There is thus an additional all-to-all communication step at the end of each frontier expansion. Interprocessor communication is considered a significant performance bottleneck in prior work on distributed graph algorithms [10,25]. The relative costs of inter-processor communication and local computation depends on the quality of the graph partitioning and the topological characteristics of the interconnection network.…”

Locality Exists in Graph Processing: Workload Characterization on an Ivy Bridge Server

“…A survey [29] of both hardware and software concerns for parallel graph processing lists 'poor locality' as one of its chief concerns. Although it is cognizant of the greater cost of heavily multithreaded systems, it argues they are better for graph algorithms due to their memory latency tolerance and support for fine-grained dynamic threading.…”

OpenCL Parallel Blocked Approach for Solving All Pairs Shortest Path Problem on GPU

“…This approach is in-place and recursive in nature. Challanges in parallel graph processing is discussed in [3] Our computations involve matrices and therefore some fast matrix multiplication algorithms such as [19][6] are also our area of concern. As CPU implementations have several limitations of performance so some cache optimization techniques and cache friendly implementations are given in [2]and [5] using recursion for dense graphs.…”