Self-organising agent communities for autonomic computing

Abstract

Efficient resource management is one of key problems associated with large-scale distributed computational systems. Taking into account their increasing complexity, inherent distribution and dynamism, such systems are required to adjust and adapt resources market that is offered by them at run-time and with minimal cost. However, as observed by major IT vendors such as IBM, SUN or HP, the very nature of such systems prevents any reliable and efficient control over their functioning through human administration.

For this reason, autonomic system architectures capable of regulating their own functioning are suggested as the alternative solution to looming software complexity crisis. Here, large-scale infrastructures are assumed to comprise myriads of autonomic elements, each acting, learning or evolving separately in response to interactions in their local environments. The self-regulation of the whole system, in turn, becomes a product of local adaptations and interactions between system elements.

Although many researchers suggest the application of multi-agent systems that are suitable for realising this vision, not much is known about regulatory mechanisms that are capable to achieve efficient organisation within a system comprising a population of locally and autonomously interacting agents.

To address this problem, the aim of the work presented in this thesis was to understand how global system control can emerge out of such local interactions of individual system elements and to develop decentralised decision control mechanisms that are capable to employ this bottom-up self-organisation in order to preserve efficient resource management in dynamic and unpredictable system functioning conditions. To do so, we have identified the study of complex natural systems and their self-organising properties as an area of research that may deliver novel control solutions within the context of autonomic computing.

In such a setting, a central challenge for the construction of distributed computational systems was to develop an engineering methodology that can exploit self-organising principles observed in natural systems. This, in particular, required to identify conditions and local mechanisms that give rise to useful self-organisation of interacting elements into structures that support required system functionality. To achieve this, we proposed an autonomic system model exploiting self-organising algorithms and its thermodynamic interpretation, providing a general understanding of self-organising processes that need to be taken into account within artificial systems exploiting self-organisation.

More information

Identifiers

Catalogue record

Export record