Modifying CloudSim to accurately simulate interactive services for cloud autoscaling

Summary We study the problem of autoscaling in cloud deployments and issues related to autoscaling strategies and their formal evaluation. With this goal, we propose modifications of the CloudAnalyst distribution of CloudSim for improved autoscaler simulations. We extend the CloudSim platform with virtual machine addition/removal at simulation run time as well as more granular load reporting and implement a simple autoscaler simulation. We point out and fix multiple problems in CloudSim job duration computation methods and make it consistent with both theory and reality, improving the relative errors of simulation from as high as 333% to single‐figure percentages. Our main contribution is a formally evaluated autoscaling strategy and improved version of CloudAnalyst capable of realistic evaluations.