A Replay-Oriented Software Architecture for Easy Debugging

This paper shows how event-driven software architecture may facilitate easy debugging. Most existing replay-based debugging techniques involve tool developments. Drawbacks for such techniques are two-fold: (1) extra cost for tool development. (2) lack of availability, tools and techniques for one platform might not be available or applicable on other platforms. This paper tries to improve the situation by facilitating replay at software architecture level. Programs conforming to the proposed architecture are replayable by construction. There is no need for tool development and the technique is available as long as the used programming language is available on the target platform. Two notions, namely pure property and pure entity, are proposed. It is proved that bugs in pure entities and pure properties can be systematically found with the support of the underlying replay-oriented architecture. The architecture is applied to the development of a typical distributed system, which is used to manage a collection of directory servers. Experience with this application shows that more than half of user developed code can be characterized as pure entities and pure properties. Our experience also shows that a large portion of impure properties can still be debugged using this architecture. It is also explained why this replay-oriented software architecture is generally applicable to other applications.