CEDULE+: Resource Management for Burstable Cloud Instances Using Predictive Analytics

Nearly all principal cloud providers now provideburstable instances in their offerings. The main attraction ofthis type of instance is that it can boost its performance fora limited time to cope with workload variations. Althoughburstable instances are widely adopted, it is not clear how toefficiently manage them to avoid waste of resources. In thispaper, we use predictive data analytics to optimize the man-agement of burstable instances. We design CEDULE+, a data-driven framework that enables efficient resource managementfor burstable cloud instances by analyzing the system workloadand latency data. CEDULE+ selects the most profitable instancetype to process incoming requests and controls CPU, I/O, andnetwork usage to minimize the resource waste without violatingService Level Objectives (SLOs). CEDULE+ uses lightweightprofiling and quantile regression to build a data-driven predictionmodel that estimates system performance for all combinations ofinstance type, resource type, and system workload. CEDULE+ isevaluated on Amazon EC2, and its efficiency and high accuracyare assessed through real-case scenarios. CEDULE+ predictsapplication latency with errors less than 10%, extends themaximum performance period of a burstable instance up to 2.4times, and decreases deployment costs by more than 50%.