Development of Fault Tolerant Work Stealing Based Computing Framework

2016 Fiscal Year Final Research Report

• Project/Area Number: 25730041
• Research Category: Grant-in-Aid for Young Scientists (B)
• Allocation Type: Multi-year Fund
• Research Field: Software
• Research Institution: Kyoto University
• Principal Investigator: Hiraishi Tasuku 京都大学, 学術情報メディアセンター, 助教 (60528222)
• Project Period (FY): 2013-04-01 – 2017-03-31
• Keywords: ハイパフォーマンス・コン ピューティング / プログラミング言語 / 計算機システム / ソフトウェア開発効率化・安定化 / ネットワーク

Outline of Final Research Achievements

We improved a dynamic load balancing framework called Tascell from the perspective of practicalicity and fault tolerance. We implemented a practical parallel algorithm for graph mining and ensured that it achieves good parallel performance. In addition, we implemented additional functionalities of Tascell such as safe abortion of running tasks. Because we found that the execution model of Tascell can hardly solve the single point of failure issue, we proposed a novel execution model, which realizes dynamic load balancing and fault tolerance by letting all computing workers execute a single program redundantly in arbitrary order and exchange partial results to omit completed parts of the computation. We conducted preliminary evaluations using a prototype implementation of this model.

Free Research Field

計算機科学