Abstract
Autonomic computing revolutionised the commonplace understanding of proactiveness in the digital world by introducing self-managing systems. Built on top of IBM’s structural and functional recommendations for implementing intelligent control, autonomic systems are meant to pursue high level goals, while adequately responding to changes in the environment, with a minimum amount of human intervention. One of the lead challenges related to implementing this type of behaviour in practical situations stems from the way autonomic systems manage their inner representation of the world. Specifically, all the components involved in the control loop have shared access to the system’s knowledge, which, for a seamless cooperation, needs to be kept consistent at all times.A possible solution lies with another popular technology of the 21st century, the Semantic Web,and the knowledge representation media it fosters, ontologies. These formal yet flexible descriptions of the problem domain are equipped with reasoners, inference tools that, among other functions, check knowledge consistency. The immediate application of reasoners in an autonomic context is to ensure that all components share and operate on a logically correct and coherent “view” of the world. At the same time, ontology change management is a difficult task to complete with semantic technologies alone, especially if little to no human supervision is available. This invites the idea of delegating change management to an autonomic manager, as the intelligent control loop it implements is engineered specifically for that purpose.Despite the inherent compatibility between autonomic computing and semantic technologies,their integration is non-trivial and insufficiently investigated in the literature. This gap represents the main motivation for this thesis. Moreover, existing attempts at provisioning autonomic architectures with semantic engines represent bespoke solutions for specific problems (load balancing in autonomic networking, deconflicting high level policies, informing the process of correlating diverse enterprise data are just a few examples). The main drawback of these efforts is that they only provide limited scope for reuse and cross-domain analysis (design guidelines, useful architectural models that would scale well across different applications and modular components that could be integrated in other systems seem to be poorly represented). This work proposes KAS (Knowledge-centric Autonomic System), a hybrid architecture combining semantic tools such as:• an ontology to capture domain knowledge,• a reasoner to maintain domain knowledge consistent as well as infer new knowledge,
• a semantic querying engine,• a tool for semantic annotation analysis with a customised autonomic control loop featuring:
• a novel algorithm for extracting knowledge authored by the domain expert,
• “software sensors” to monitor user requests and environment changes,
• a new algorithm for analysing the monitored changes, matching them against known patterns and producing plans for taking the necessary actions,
• “software effectors” to implement the planned changes and modify the ontology accordingly.
The purpose of KAS is to act as a blueprint for the implementation of autonomic systems harvesting semantic power to improve self-management. To this end, two KAS instances were built and deployed in two different problem domains, namely self-adaptive document rendering and autonomic decision2support for career management. The former case study is intended as a desktop application, whereas the latter is a large scale, web-based system built to capture and manage knowledge sourced by an entire (relevant) community. The two problems are representative for their own application classes –namely desktop tools required to respond in real time and, respectively, online decision support platforms expected to process large volumes of data undergoing continuous transformation – therefore, they were selected to demonstrate the cross-domain applicability (that state of the art approaches tend to lack) of the proposed architecture. Moreover, analysing KAS behaviour in these two applications enabled the distillation of design guidelines and of lessons learnt from practical implementation experience while building on and adapting state of the art tools and methodologies from both fields.KAS is described and analysed from design through to implementation. The design is evaluated using ATAM (Architecture Trade off Analysis Method) whereas the performance of the two practical realisations is measured both globally as well as deconstructed in an attempt to isolate the impact of each autonomic and semantic component. This last type of evaluation employs state of the art metrics for each of the two domains. The experimental findings show that both instances of the proposed hybrid architecture successfully meet the prescribed high-level goals and that the semantic components have a positive influence on the system’s autonomic behaviour.
| Date of Award | 25 Aug 2017 |
|---|---|
| Original language | English |
| Supervisor | Hai Wang (Supervisor) |