A Coordinated Reactive and Predictive Approach to Cloud Elasticity

Abstract

Abstract—Based on pay-per-use service-oriented architec-tures, the cloud computing paradigm promises cost-efficient IT solutions. To meet fluctuating demands efficiently, Platform-as-a-Service solutions offer shared environments with on-demand scalability. It remains an open challenge for service providers to implement elastic scalability mechanisms capable of optimally utilizing resource whilst simultaneously guaranteeing that ap-plication performance continues to meet Quality of Service metrics. Typically, cloud providers offer only reactive rule-based mechanisms for triggering scaling actions. We introduce a new elasticity management framework that combines reactive and predictive controllers. Our elasticity controller builds pre-dictive models online based on the reactive rules, representing a natural extension to the common offering. We discuss the underlying architecture of the framework and describe how the controllers operate in tandem and complement each other. We present a case study based on real datasets that demonstrates the feasibility of our real-time cloud capacity framework. Keywords-elasticity; predictive; auto-scaling; platform-as-a-service. I.