A middleware for autonomic QoS management based on learning

In any system, applications compete for a limited amount of resources. As long as there are enough resources, resource sharing based on a best effort policy is satisfactory. When resources become scarce, the system gives rise to uncontrol-lable degradations. From a global view of the system and according to the degrees of freedom of applications, Quality of Service (QoS) managers aim to adapt application behavior to deal with overload effects.This paper proposes a middleware for autonomic QoS management of a system in a dynamic environment. It associates a reinforcement learning technique with a control mechanism to improve and adapt the QoS management policy in an execution context that changes unexpectedly. The simulation of the QoS management of a set of heterogeneous applications illustrates our results.