Debugging the communication behavior of distributed programs in a message-based system
Michael S. Kenniston · 1986
Conventional sequential debuggers are inadequate for debugging application programs in a distributed system. This dissertation considers the facilities that should be provided by a distributed debugger and discusses the experience gained from the implementation of such a debugger. We assume a message-based distributed system, i.e. one in which there are multiple processors, each running multiple processes, and in which all communication is performed with messages rather than shared memory. The four principal debugging facilities treated are: suspension of a distributed program in a transparent or nearly transparent way, display of the program's execution history using space-time diagrams, tracing of events, and setting breakpoints. A number of other issues are also considered, including performance measurements and operating system features that would facilitate implementation of a distributed debugger. Finally we recommend a specific list of facilities that a distributed debugger should have. Appendixes give details of the prototype implementation that demonstrates the feasibility of many of these ideas.