An ontology-based approach to concern-specific dynamic software structure monitoring

Software reliability has not kept pace with computing hardware. Despite the use reliability improvement techniques and methods, faults remain that lead to software errors and failures. Runtime monitoring can improve software reliability by detecting certain errors before failures occur. Monitoring is also useful for online and electronic services, where resource management directly impacts reliability and quality. For example, resource ownership errors can accumulate over time (e.g., as resource leaks) and result in software aging. Early detection of errors allows more time for corrective action before failures or service outages occur. In addition, the ability to monitor individual software concerns, such as application resource ownership structure, can help support autonomic computing for self-healing, self-adapting and self-optimizing software.

This thesis introduces ResOwn - an application resource ownership ontology for interactive session-oriented services. ResOwn provides software monitoring with enriched concepts of application resource ownership borrowed from real-world legal and ownership ontologies. ResOwn is formally defined in OWL-DL (Web Ontology Language Description Logic), verified using an off-the-shelf reasoner, and tested using the call processing software for a small private branch exchange (PBX). The ResOwn Prime Directive states that every object in an operational software system is a resource, an owner, or both simultaneously. Resources produce benefits. Beneficiary owners may receive resource benefits. Nonbeneficiary owners may only manage resources. This approach distinguishes resource ownership use from management and supports the ability to detect when a resource’s role-based runtime capacity has been exceeded.

This thesis also presents a greybox approach to concern-specific, dynamic software structure monitoring including a monitor architecture, greybox interpreter, and algorithms for deriving monitoring model from a monitored target’s formal specifications. The target’s requirements and design are assumed to be specified in SDL, a formalism based on communicating extended finite state machines. Greybox abstraction, applicable to both behavior and structure, provides direction on what parts, and how much of the target to instrument and what types of resource errors to detect.

The approach was manually evaluated using a number of resource allocation and ownership scenarios. These scenarios were obtained by collecting actual call traces from an instrumented PBX. The results of an analytical evaluation of ResOwn and the monitoring approach are presented in a discussion of key advantages and known limitations. Conclusions and recommended future work are discussed at the end of the thesis.