There is a growing need for the automation of the IT infrastructure of enterprises. Autonomic computing provided a theoretical support for the foundation of mechanisms for self-optimization of computational resources at all the levels of the IT infrastructure of the enterprise. As the Autonomic Computing paradigm requires collecting information in regards to specific parameters based on which a decision module will act, the architecture of an autonomic computing system is very much similar to a real-time control system. Thus the validation of the model used for the mathematical characterization of the autonomic computing processes is crucial. In this paper, starting from the model of autonomic computing processes an identification technique adapted to autonomic computing processe, is introduced. The identification is based on injecting pseudo random arrival rates into the autonomic system as disturbances. The observations are collected from sensors for CPU load, throughput, response time, etc implemented in the middleware over which applications were deployed. The identification process described in this paper determines first the sampling rate and then uses the Recursive Parameter Estimation technique (RPE)for Extended Kalman Filters, to obtain a model on which the whole control strategy relies upon. Experiments and results are described in the end of this paper.
展开▼
