Towards automatic tuning of adaptive computations in autonomic middleware

An autonomic middleware performs adaptive computations on the fly that bring benefits to the system while consuming additional resources such as CPU and memory. These computations can sometimes interfere with normal business functions of the system due to resource competition, especially when under heavy load. In this paper, we propose an approach to tuning the computation levels and thus controlling the resource costs of the adaptive computations in an autonomic middleware. The tuning (i.e., upgrading or degrading) of the computation levels is performed automatically based on the varying workloads, and the features and gains of the adaptive computations. The essence of our approach is to enable a flexible tradeoff between business functions and adaptive computations by executing the latter dynamically when resources are limited and competed. We present tuning policies and mechanisms to suit different adaptive computations, and implement an automatic tuning framework to investigate our approach. The experiment on the framework indicates that it is effective and efficient to improve the performance of the middleware system.