Compositionally adaptive mobile software agents for pervasive environments

Abstract

The vision of pervasive computing foresees environments filled with multiple computing devices unobtrusively aiding humans in their daily activities. With recent advances in hardware and communication technologies, computing capacity is becoming increasingly pervasive and invisible in our environment. The continuing challenge is to achieve seamless and graceful integration of these pervasive technologies with our daily activities, which is now the increasing focus of research within this area. Mobile agents are lightweight software entities that move from one node to another in a network autonomously and continue execution of their own accord. Mobile software agents are seen as a suitable enabling technology for pervasive computing because they bring flexibility, scalability and the ability to simplify tasks by movement and delegation for a range of pervasive applications. However, mobile agent based applications have limited adaptability, and this is an inhibiting factor in pervasive environments. The inherent nature of pervasive environments is that they are complex, heterogeneous and highly dynamic. This makes it extremely hard to design and program mobile agent systems that cater for all possible situations that could be encountered in such environments. This thesis investigates, proposes and develops a novel platform for building adaptive software agents which combines mobility with the ability to dynamically change the internal structure and capabilities, as a strategy to harness the potential of mobile agents for pervasive environments. We term this compositional adaptation of mobile agents and focus on enhancing the adaptation potential of mobile agents for the purpose of building applications and services for pervasive environments. The first contribution of this thesis is our proposed VErsatile Self-adaptive AGents (VERSAG) framework. The framework is centred on a component-based agent model where an agent is a lightweight mobile entity whose application-specific functionality is provided in the form of reusable software components termed capabilities. Furthermore, an agent can increase its autonomy at runtime through acquisition (and potential discarding) of capabilities which improve its core functions such as reasoning ability and context-awareness. Our second contribution is the peer capability sharing feature of VERSAG agents. That is, when an agent does not contain required capabilities, it is able to search for, select and acquire them from other agents who are willing to share their capabilities. Our third contribution achieves cost-efficient adaptation of agents using a proposed cost model that allows agents to make adaptation decisions by taking into consideration multiple cost criteria. These cost criteria can be either user/organization specified or based on contextual constraints. We use time, network usage, computational resource consumption and accuracy of results, as representative cost criteria. In this thesis we propose, implement and evaluate the conceptual framework of VERSAG and its theoretical foundations. We have developed a prototype implementation of VERSAG to demonstrate its feasibility and to conduct experimental evaluation of the concepts that underpin the proposed framework. The feasibility and functional benefits achieved from this approach are demonstrated using an application case study. Our experimental evaluation also demonstrates that VERSAG agents bring about performance benefits over conventional mobile agents through the use of their capability sharing feature. Through extensive experimental evaluation we also establish the scalability of component sharing mobile agents and the effective performance of the decision making cost model. Thus, this thesis takes a step forward in realising the potential of mobile software agents enhanced with compositional adaptation for pervasive computing. The research work described in this thesis has resulted in one international journal article and seven peer-reviewed international conference papers.