ASPIRE: Exploiting Asynchronous Parallelism in Iterative Algorithms using a Relaxed Consistency based DSM

作者：Vora Keval^*; Koduru Sai Charan; Gupta Rajiv

来源：ACM Sigplan Notices, 2014, 49(10): 861-878.

DOI：10.1145/2660193.2660227

• Distributed Shared Memory
• Communication Latency
• Bounded Staleness
• Best Effort Refresh
• PDE Solvers
• Graph Mining
• Graph Analytics

Many vertex-centric graph algorithms can be expressed using asynchronous parallelism by relaxing certain read-after-write data dependences and allowing threads to compute vertex values using stale (i.e., not the most recent) values of their neighboring vertices. We observe that on distributed shared memory systems, by converting synchronous algorithms into their asynchronous counterparts, algorithms can be made tolerant to high inter-node communication latency. However, high inter-node communication latency can lead to excessive use of stale values causing an increase in the number of iterations required by the algorithms to converge. Although by using bounded staleness we can restrict the slow-down in the rate of convergence, this also restricts the ability to tolerate communication latency. In this paper we design a relaxed memory consistency model and consistency protocol that simultaneously tolerate communication latency and minimize the use of stale values. This is achieved via a coordinated use of best effort refresh policy and bounded staleness. We demonstrate that for a range of asynchronous graph algorithms and PDE solvers, on an average, our approach outperforms algorithms based upon: prior relaxed memory models that allow stale values by at least 2.27x; and Bulk Synchronous Parallel (BSP) model by 4.2x. We also show that our approach frequently outperforms GraphLab, a popular distributed graph processing framework.

• 出版日期2014-10