Autonomic Characterization of Workloads Using Workload Fingerprinting

Abstract

In a cloud service management environment, service level agreements (SLA) define the expectation of quality (Quality-of-Service) for managing performance loss in a given service-hosting environment comprising of a pool of compute resources. Typically, complexity of resource inter-dependencies in a server system often results to sub-optimal behaviors leading to performance loss. A well behaved model can anticipate the demand patterns and proactively react to the dynamic stresses in a timely and well optimized manner. Dynamic characterization methods can synthesize self-correcting workload fingerprint code-book that facilitates phase prediction to achieve continuous tuning through proactive workload-allocation and load-balancing. In this paper we introduce the methodology that facilitates the coordinated tuning of the system resources through phase-assisted dynamic characterization. We describe the method to develop a multi-variate phase model by learning and classifying the run-time behavior of workloads. We demonstrate the workload phase forecasting method using phase extraction using a combination of machine learning approach. Results show the new model is about 98% accurate in phase identification and 97.15% accurate in forecasting the compute demands.