Doctor of Philosophy (Ph.D.)
Monitoring a program's execution is fundamental to the debugging, testing and maintenance phases of program development. This research addresses the issue of monitoring the execution of a distributed program. In particular, we are concerned with efficient techniques for evaluating global state predicates for distributed programs. The global state of a distributed program is not well-defined, making the monitoring task complex compared to that of a sequential programs. Processes of a distributed program execute concurrently, and the events of the program cannot be totally ordered. Each process has its own local memory, and the local memories are physically separate.;Despite the difficulties of defining a distributed computation's states, monitoring a distributed program requires reasoning about constituent processes' execution as a single collective entity. We have extrapolated the semantics of the sequential program's assert statement into the distributed context. A distributed assert statement is a global predicate that is anchored at a control point of one processes, and that is evaluated when that process executes the assert.;We have developed a runtime method for monitoring both stable and unstable properties that does not disrupt the computation of the distributed system. A distributed assert statement is evaluated with that statement's causal global state which incorporates the state of the system as a whole as it may have causal impact upon the assert statement. A runtime protocol has been implemented that constructs the causal global state and evaluates the assert statement. No additional synchronization or message passing is imposed on the distributed application although some message sizes are increased to propagate state information. The causal global state is immediately available providing real-time feedback.
© The Author
Simmons, Sharon J., "Causal distributed assert statements" (1999). Dissertations, Theses, and Masters Projects. Paper 1539623962.