Combining Self-Organisation and Autonomic Computing in CASs with Aggregate-MAPE

Abstract

Aggregate computing is a recently proposed framework to build CASs (collective adaptive systems) by focussing on direct programming of ensembles so as to abstract away from individual devices and their single interaction acts: this approach is shown to streamline the identification of highly reusable block components, and support reasoning about their resiliency properties. Following this paradigm, in this paper we present a framework for bridging the gap between the MAPE (Monitor-Analyse-Plan-Execute) loop of autonomic computing managers, and fully-distributed self-organising CASs. This is achieved by seeing the collection of M components of each agent as an aggregate, amenable to a direct specification as overall CAS Monitoring behaviour, and similarly for A, P and E. As a result, a self-organising CAS can be programmed by clearly separating the M, A, P, and E parts of it, though each is expressed in terms of a collective behaviour. The proposed approach is exemplified with an application scenario of crowd dispersal in a large-scale smart-mobility application.