Abstract

This thesis is a part of the whole project called CDB, which involves a team of graduate students whose collective aim is to design and implement a distributed debugger. Unlike many other implementations of distributed debuggers, CDB is an "integrated system". It integrates the concepts of non-deterministic replay, distributed breakpoints, rollback and recovery, white-box to black-box approach, and the user interface to support interactive debugging. Several theses have been written and the work is still ongoing, with regards to CDB. Every concept discussed in this thesis is implemented under Mach, tested, and documented. Thus one of the contributions of this thesis is the implementation of ideas for conducting experimental research. Based on the research reported in literature, a specification language called PDL (Predicate Definition Language) is proposed in this thesis. Distributed breakpoints specified using PDL can be detected by the present implementation and the distributed computation can be halted. PDL assists the user to specify breakpoints at different levels of granularity, thus facilitating white-box or black-box type inspection of the computation. We have implemented two subsystems of CDB: breakpoint and user interface. They are designed such that future breakpoint and user interface methods can be incorporated. In developing these two subsystems, we have identified building blocks which can be re-used in building other CDB subsystems. The two building blocks, vector clock server and local debugger are designed in such a way that their re-usability is maximized.