Improving Rule-Based Elasticity Control by Adapting the Sensitivity of the Auto-Scaling Decision Timeframe

Cloud computing offers the opportunity to improve efficiency with cloud providers offering consumers the ability to automatically scale their applications to meet exact demands. However, “auto-scaling” is usually provided to consumers in the form of metric threshold rules which are not capable of determining whether a scaling alert is issued due to an actual change in the demand of the application or due to short-lived bursts evident in monitoring data. The latter, can lead to unjustified scaling actions and thus, significant costs. In this paper, we introduce AdaFrame, a novel library which supports the decision-making of rulebased elasticity controllers to timely detect actual runtime changes in the monitorable load of cloud services. Results on real-life testbeds deployed on AWS, show that AdaFrame is able to correctly identify scaling actions and in contrast to the AWS auto-scaler, is able to lower detection delay by at least 63%.