摘要
并行程序的错误可分为进程间错误(interprocess fault)和进程内部错误(intraprocess fault).其中, 进程内部错误是由控制流错误造成的, 而进程间错误是由进程间的不合理时序关系造成的. 文中主要探讨了关于并行程序(以基于消息传递的分布式并行程序为主)时序测试中的一系列关键技术.为了能够简捷、较完备地反映并行程序的运行流程,文中首先构筑了一个同步序列模型,所有问题均在此模型的基础上进行讨论. 为了检测分布式并行程序同步序列的合法性,文中构筑了有效同步序列的形式规范.为使问题更加明了, 还提供了两个典型的聚类操作——树型广播与归并的有效同步序列的形式规范, 并总结了两个操作中的各同步事件间的时序约束规律.庞大的时序组合使得并行程序的测试难以实现, 因此文中提出了一种现实可行的测试策略——原子事件测试方法, 其思想方法是线性化同步序列使其仅由串行的原子同步事件组成。
Testing of parallel programs involves two parts——testing of control\|flow within the processes and testing of timing\|sequence. The paper here focuses on the latter,particularly on the timing\|sequence of message\|passing paradigms.In order to simply and precisely describe the execution of distributed programs,the coarse\|grained SYN\|sequence model is proposed,on which all of the topics discussed in the paper are based.In order to analyze the validity of observed SYN\|sequence,this paper constructed the formal specification (Backus normal form)of the valid SYN\|sequence.With two typical collective operation——combine and broadcast as instances, the paper further clarified the significance of formal specification for valid SYN\|sequence in the testing of distributed programs.In practice the number of the distinct timing\|sequence combination for distributed programs is so large that it is almost impossible to test them directly.Therefore, an efficient and practical testing strategy——atomic SYN\|event testing is proposed, which is to linearize the SYN\|sequence(enabling it to consist only of serial atomic SYN\|events)at first and then test each atomic SYN\|event independently.
出处
《计算机研究与发展》
EI
CSCD
北大核心
1999年第9期1062-1068,共7页
Journal of Computer Research and Development
基金
国家攀登计划项目基金
关键词
时序测试
并行程序
同步事件
软件工程
timing\|sequence test, parallel programs, SYN\|event, SYN\|sequence,atomic\|event test