Monitoring middleware for distributed applications

Abstract

With growing maturity Internet services are proving integral to the provision of computer services. To provide consistent end-user experiences these services are increasingly augmented with some notion of 'Quality-of-Service' (QoS), which typically requires the management of computing resources to maintain a predictable level of service performance. It is difficult to guarantee consistent servIce provision In dynamic and open environments such as the Internet. However service monitoring can be used to inform compensatory actions by collecting meaningful service performance data from strategic points in an active service environment. Due to the unpredictable nature of the Internet distributed monitoring mechanisms face challenges with respect to the various communication protocols, application languages, and monitoring requirements associated with a service environment. With the growing popularity of Internet services creation of monitoring solutions on a per- service basis becomes time-consuming and misses opportunities to re-use existing logic. Ideally monitoring solutions would be domain-agnostic, automatically generated and automatically deployed. This thesis progresses these ambitions by providing a generic, distributed monitoring and evaluation framework based on Metric Collector (MeCo) components. These components can transparently gather measurement data across a range of service technologies as used within E-Commerce service environments. MeCo components form part of a framework which can interpret Service Level Agreements (SLAs) to automatically provide tailored service monitoring. The evaluation paradigms of the Meeo Framework are re-appropriated for use in Distributed Virtual Environments (DYEs). Quantifiable QoS requirements are established for Interest Management mechanisms (which limit message production based on object localities within a DYE). These are then incorporated into a DVE Simulator application. This application allows DYE application developers to evaluate Interest Management configurations for their suitability. Extensions to the DVE Simulator are exhibited in the Evolutionary Optimisation Simulator (EOS), which provides automated optimisation capabilities for DVE configurations through utilisation of genetic algorithm techniques