Autonomic Resource Management Handling Delayed Configuration Effects

Abstract

Today, cloud providers offer customers access to complex applications running on virtualized hardware. Nevertheless, big virtualized data centers become stochastic environments with performance fluctuations. The growing number of cloud services makes a manual steering impossible. An automatism on the provider side is needed. In this paper, we present a software solution located in the Software as a Service layer with autonomous agents that handle user requests. The agents allocate resources and configure applications to compensate performance fluctuations. They use a combination of Support Vector Machines and Model-Predictive Control to predict and plan future configurations. This allows them to handle configuration delays for requesting new virtual machines and to guarantee time-dependent service level objectives (SLOs). We evaluated our approach on a real cloud system with a high-performance software and a three-tier e-commerce application. The experiments show that the agents accurately configure the application and plan horizontal scalings to enforce SLO fulfillments even in the presence of noise.