A Debugging Tool for Distributed Systems

This paper describes parts of the design of a debugger for a distributed real-time multimedia system. Emphasis lies on the distributed aspect of debugging, which means that attention is paid to the external behaviour of the processes. This type of debugging is useful to find communication or synchronization errors. However, experience shows that this is not enough: the debugger must also provide hooks for the user to use traditional sequential debuggers. This type of debugging focuses on the internal behaviour - or internal logic - of processes. For the sequential debugging part a normal debugger like GDB can be taken. Three key elements of the debugger are events, filters and recognizers. By definition events are the lowest level of system activity that may be observed by the event debugger. Filters are applied to remove events from the stream of events produced by the debuggee that are of no interest for the programmer. Recognizers are used to recognize behaviour -right or wrong- of the system. By combining events, different levels of abstraction are introduced, thus alleviating the task of the programme