Protecting SLAs with surrogate models

Abstract

In this paper, we propose the use of surrogate models to avoid or limit violations of the service level agreements (protect SLAs) of enterprise applications executed within virtualized data centers (VDCs). Modern enterprise services are delivered along with service level agreements (SLAs) that formalize the expected quality of service, and define penalties in case of violations. By deploying enterprise applications within VDCs, providers can dynamically change the execution configuration of the services to react to unplanned environmental conditions, like sudden changes in the workload mix and intensity, with the goal of avoiding SLA violations while reducing operational costs with respect to traditional over-provisioning solutions. Surrogate models are successfully used in modern engineering to approximate systems' behavior, and thus support a wide scope of activities, especially design optimization. In this paper, we show that by reducing the problem of protecting SLAs in VDCs to an optimization problem, we can adapt surrogate models to this new framework and implement SLA protection controller components. In the paper, we present the main ideas, we illustrate how surrogate models can be used to protect SLAs, and we discuss preliminary results obtained on a case study deployed in an industrial virtualized infrastructure.