The Consensus Object: Coordinating the Behavior of Independent Adaptive Systems

Nowadays the complexity of computing systems is skyrocketing. Programmers have to deal with extremely powerful computing systems that take time and considerable skills to be instructed to perform at their best. This work analyzes the stated problem and proposes a simple, yet powerful mechanism for optimizing performance through the coordination of the interaction of multiple, independent adaptive systems called services. In this scenario we developed the Consensus Object, a system-centralized decision engine based on reinforcement learning. The Consensus Object gathers information about the performance goals of the system and it can either turn services on or off. The Consensus Object analyzes the runtime impact of services and of their autonomous decision policies, looking for a combination of services that makes it possible to reach the given goals. The experiments that have been carried out show the ability of the Consensus Object to adapt to changing conditions, confirming the validity and the flexibility of the followed approach.