Machine learning applied to accelerate energy consumption models in computing simulators
Castañé, Gabriel G.
Calderón Mateos, Alejandro
The ever-increasing growth of data centres and fog resources makes difficult for current simulation frameworks to model large computing infrastructures. Therefore, a major trade-off for simulators is the balance between abstraction level of the models, the scalability, and the performance of the executions. In order to balance better these, early forays can be found in the literature in which AI techniques are applied, but either lack of generality or are tailored to specific simulation frameworks. This paper describes the methodology to integrate memoization as a technique of supervised learning into any computing simulators framework. In this process, a bespoke kernel was constructed for the analysis of the energy models used in most well known computing simulators -cloud and fog-, but also to avoid simulation overhead. Finally, a detailed evaluation of energy models and its performance is presented showing the impact of applying supervised learning to computing simulator, showing performance improvements when models are more accurate and computations are dense.
Computer simulation , Machine learning , Memoization , Simulation
Castañéa, G. G. and Calderón Mateos, A. (2020) 'Machine learning applied to accelerate energy consumption models in computing simulators', Simulation Modelling Practice and Theory, 102, 102012 (16 pp). doi: 10.1016/j.simpat.2019.102012