Hierarchical autonomic management of structured grid applications can be efficiently implemented using production rule engines. Rules of the form “precondition→ action” can be used to model the behaviour of autonomicmanagers in such a way that the autonomic control and the application management strategy are kept separate. This simplifies the manager design as well as user customization of autonomic manager policies. We briefly introduce rule-based autonomic managers. Then we discuss an implementation of a GCM-like behavioural skeleton – a composite component modelling a standard parallelism exploitation pattern with its own autonomic controller – in SCA/Tuscany. The implementation uses the JBoss rules engine to provide an autonomic behavioural skeleton component and services to expose the component functionality to the standard service framework. Performance results are discussed and finally similarities and differences with respect to the ProActive-based reference GCM implementation are discussed briefly

Abstract. Behavioural skeletons have been introduced as a suitable way to model autonomic management of parallel, distributed (grid) applications. A behavioural skeleton is basically a skeleton with an associated autonomic manager taking care of non-functional issues related to skeleton implementation. Here we discuss an implementation of a task farm behavioural skeleton exploiting SCA, the Service Component Architecture recently introduced by IBM. This implementation is meant to provide plain service/SCA users some efficient skeleton modelling common parallel application pattern and also to investigate the advantages and the problems relative to skeletons in the service world. Experimental results are eventually discussed.