关键词: 并行任务, 云计算, 相似驱动, 任务划分

Abstract: Cloud computing is further developed by parallel computing, distributed computing and grid computing. Its characters of heterogeneity and on-demand are new challenge for high performance computing. Based on typical features of cloud computing，this paper proposes a similarity-driven parallel task partitioning method. Firstly, it describes parallel tasks and cloud architecture by graph and builds a graph-similarity definition. Secondly, it provides the problem intended to solve in cloud computing, gives the method by local similarity and minimizing global similarity deviation, and realizes heterogeneous matching between parallel task and architecture and on-demand require. Thirdly, by F-degree labels it proposes a similarity-driven task partitioning method. Finally, compared with other methods, the experimental results verify the advantages of the proposed method.

Key words: parallel task, cloud computing, similarity-driven, task partitioning