并行任务动态派生的积极惰性化控制方法

A PRACTICAL EAGER-LAZY CONTROL METHOD FOR DYNAMIC DERIVATION OF PARALLEL TASKS

本文给出了并行任务派生的理想状态，分析和研究了积极任务派生（ＥＴＤ）方法和惰性任务派生（ＬＴＤ）方法，指出了这两种方法所具有的局限性，提出了一种新的并行任务派生的积极惰性化方法（ＥＬＤＴ）及其算法．初步研究表明ＥＬＤＴ方法可安全有效地增大计算粒度，在由多个商售单处理器构成的小规模并行系统上ＥＬＤＴ算法有效地控制计算粒度和任务派生，使并行任务的派生近似达到理想状态．

One of the most issues on exploiting effective parallelism is how to dynamically control parallel task derivation. In this paper,based on the ideal state of parallel tasks deriving, Eager Task Deriving (ETD)and Lazy Task Deriving(LDT) methods are analysed and discussed briefly, drawbacks of ETD and LTD are also presented. A practical Eager Lazy method is proposed for dynamically Deriving parallel Tasks (ELDT). The exprimental results have shown that ELDT can effectively control and increase the granularity of derived tasks whose granularity is finer than the ideal granularity that multiprocessor systems can exploit efficiently, and take an ideal or satisfied state of parallel task derivation.