Adaptive Feedforward and Feedback Control for Cloud Services

The use of cloud services is becoming increasingly common As the cost of these services is continuously decreasing, service performance is becoming a key differentiator between providers. Solutions that aim to guarantee Service Level Objectives (SLO) in term of performance by controlling cluster size are already used by cloud providers. However most of these control solutions are based on static if-then rules, they are therefore inefficient in handling the highly varying service dynamics of cloud environments. Client concurrency, network bottlenecks or non homogeneity of resources are just a few of the many causes that make the behavior of cloud services highly non linear and time varying. In this paper a novel control theoretical approach realizing resource allocation is presented that is robust to these phenomena. It consists of PI and feedforward controller adapted online. A stability analysis of the adaptive control configuration is provided. Simulations using a cloud service model taken from the literature illustrate the performance of the system under various conditions. The use of adaptation significantly improves control efficiency and robustness with respect to variations in the dynamic of the plant. (C) 2017, IFAC (International Federation of Automatic Control) Hosting by Elsevier Ltd. All rights reserved.