Προσαρμοστική Παρακολούθηση Συστημάτων Λογισμικού

Abstract

The main goal of this diploma thesis is the design and implementation of a goal driven framework that allows for adaptive monitoring of software systems. The increasing complexity of today's software systems complexity has necessitated the development of adaptive and self-managed software systems. The development of self-managed systems will facilitate their management by hiding the systems' complexity from the administrators and engineers. However a self-managed system requires continious monitoring and analysis of the data collected in order to detect, verify and solve problems. Such a solution would impose a heavy load on system resources and influence the overall system's performance. Towards a solution to this problem we propose the development of adaptive monitoring systems, able to determine at the runtime, which data should be collected and analyzed, minimizing that way the monitoring cost. Our approach to this challenge utilizes goal trees to model and define hypotheses about the system's state. These hypotheses get activated upon detecting significant patterns of events in the event stream, and then their verification and root-cause analysis is attempted. Verifying a hypothesis may require collecting additional data that were not being collected before. This results in the adaptation of the monitoring level. This way only the set of data, necessary to detect a problem and determine it's root cause, is being monitored resulting in lower monitoring cost and faster problem determination since less data are being processed. For the purposes of this work, we defined the architecture of a framework that allows for adaptive monitoring and developed a prototype of that framework. The architecture is based on the Blackboard architectural pattern and supports indirect calls between its components by using a publisher/subscriber service. Furthermore we defined a model for the hypotheses, based on the goal trees model, and we provide a graphical user interface so as to define them. Finally we transform the goal trees in CNF and propose the use of SAT-Solvers to analyze them.