Debugging of Distributed Programs

这是对调试具有大量进程的分布式程序中的问题的调查。现有的分布式调试系统是顺序调试系统的简单扩展，不足以支持基于消息的分布式程序。由于不确定性、并发性和通信延迟的存在，分布式程序本质上不同于顺序程序。分布式调试的研究同时支持基于状态和基于事件的调试方法。基于状态的检测主要通过使用断点谓词检测来支持。用于指定断点谓词的逻辑描述了分布式环境中全局跟踪的最有用的属性。该逻辑被适当地定义为允许高效地检测全局谓词，而无需求助于全局快照的计算。已经开发了算法来检测表示为简单的合取、析取或局部谓词序列的全局谓词。这些机制可以检测逻辑定义的一大部分谓词，即使这些谓词是不稳定的。主要目标是将这些算法扩展到更一般的全局谓词。

This is an investigation of issues in debugging distributed programs with large numbers of processes. Existing distributed debugging systems are simple extensions of sequential debugging systems and are inadequate for supporting message-based distributed programs. Distributed programs are inherently different from sequential programs due to the presence of nondeterminism, concurrency and communication delays. The research in distributed debugging supports both state-based and event-based approaches for debugging. The state-based is supported mainly by use of breakpoint predicates detection. A logic for specification of breakpoint predicates characterizes the most useful properties of global traces in a distributed environment. The logic is suitably defined to permit efficient detection of global predicates with out resort to computation of global snapshots. Algorithms have been developed to detect global predicates expressed as simple conjunction, disjunction or sequence of local predicates. These mechanisms can detect a large subset of predicates defined by the logic, even if the predicates are unstable. The main objective is to extend these algorithms for applicability to more general global predicates.