Empirical Evaluation of a Run-Time Dynamic Adaptable Framework

Run-time dynamic adaptation is becoming a required feature for many different types of applications due to the overall complexity and distinction of environments where these applications run. SMART (State Model Adaptive Run Time), a new framework for the design of these adaptive systems, is presented here. SMART is based on the mathematics of control theory and system identification techniques allowing accurate predictions of constraint violations, such as memory overflow. Once a constraint violation is predicted, a functionally equivalent component that better copes with the resource under stress is selected. A swap to the selected component is done prior to the actual violation consequently increasing system robustness. Two examples of the application of SMART are presented to show the need and applicability of the framework. Experimental results with a prototype of SMART demonstrate the framework has an 86% accuracy in predicting and averting memory constraint violations for the fisrt case study and a 95% accuracy for the second one. These results indicate the SMART Framework is feasible and has the potential to be a useful design solution for dynamic adaptable systems. Implementation issues such as the overhead in the prediction and run-time data collection are analyzed and improvements to the framework are proposed as future work.