A Machine Learning Middleware For On Demand Grid Services Engineering and Support
Wail M. Omar, Azzelarabe Taleb-Bendiab, Yasir Karam · 2005
Abstract. Over the coming years, many are anticipating grid computing infrastructure, utilities and services to become an integral part of future socioeconomical fabric. Though, the realisation of such a vision will be very much affected by a host of factors including; cost of access, reliability, dependability and security of grid services. In earnest, autonomic computing model of systems ’ self-adaptation, self-management and self-protection has attracted much interest to improving grid computing technology dependability, security whilst reducing cost of operation. A prevailing design model of autonomic computing systems is one of a goal-oriented and model-based architecture, where rules elicited from domain expert knowledge, domain analysis or data mining are embedded in software management systems to provide autonomic systems functions including; self-tuning and/or self-healing. In this paper, however, we argue for the need for unsupervised machine learning utility and associated middleware to capture knowledge sources to improve deliberative reasoning of autonomic middleware and/or grid infrastructure operation. In particular, the paper presents a machine learning middleware service using the well-known Self-Organising Maps (SOM), which is illustrated through a casestudy scenario-- intelligent connected home. The SOM service is used to classify types of users and their respective networked appliances usage model (patterns). The models are accessed by our experimental self-managing infrastructure to provide Just-in-Time deployment and activation of required services in line with learnt usage models and baseline architecture of specified services assemblies. The paper concludes with an evaluation and general concluding remarks.