Study on Case Based Parallel Programing

本文在虚拟共享内存并行机KSR-1上开发了一种基于案例的并行编程系统，并对该系统进行了实例并行编程实验。(1)基于案例的并行编程系统的开发我们开发了一个索引部分，它分析给定的问题，并通过与用户交谈来生成显示问题特征的索引。然后，我们开发了一个案例检索器和一个案例适配器。案例库有22个并行程序。(2)基于案例并行编程的实验与评价。以背包问题、字符串模式匹配（BM方法）、Romberg规则、Viginere密码和运行长度压缩为实验对象。系统使用索引检索每个问题的相关案例。通过对相关案例的框架进行改编，最终开发出一个并行程序。我们从线程、同步、任务划分和单元计算四个方面对案例的可重用性进行了评估，结果表明线程和同步几乎可以从骨架中重用，而单元计算则可以从串行程序中提取。尽管任务划分的一部分可以从框架中重用，但它们中的大多数应该由用户进行调整。由此得出的结论是，并行程序的框架部分可以从相关案例中重用，从而减少并行编程的负担。

This research developed a case-based parallel programming system on a virtual shared memory parallel machine KSR-1 and then evaluated the system by experimenting with case-based parallel programming for several problems.(1)Development of a case-based parallel programming systemWe developed an indexing part which analyzes a given problem and makes indices that show the characteristics of a problem by talking with a user. Then, we developed a case retriever and a case adaptator. The case-base has 22 parallel programs.(2)Experimentation and evaluation of case-based parallel programmingKnapsack problem, string pattern matching(the BM method), Romberg rule, Viginere cipher and run length compression were the target of the experimentation of case based parallel programming.A relevant case for each problem was retrieved by the system using the indices. A final parallel program was developed by adapting the skeleton of the relevant case by the user.We evaluated the re-usability of the case in terms of threads, synchronization, task division and unit calculations, The results show that threads and synchronization are almost reused from skeletons, and unit calculations are extracted from serial programs. Although a part of task division can be reused from the skeletons, most of them should be adapted by the user. This concludes that the skeletal part of parallel programs can be reused from relevant cases so that the burden of parallel programming can be reduced.