Performance and Resource Prediction at High Utilization for N-Tier Service Systems in Cloud an Experiment Driven Approach

摘要

In cloud computing, one of its intensions is tomeet service level agreements (SLAs) under high resource utilization. Cloud service providers normally make policies for the provisioning of resource sharing and performance optimization. As a result, the prediction of performance and resource utilization prior to the implementation deems as a promising and efficient tool for cloud providers. It is a challenge work due to the fact that requests of resource are always interacting and complex. Hence, the provisioning of an accurate model which capturing interactions between requestshas become an impressing demand. Moreover, the dynamic feature of cloud environment brings more challenge to predictthe performance of a running query or workload. To this end, this work develops an efficient and highlyaccurate resource sharing and performance evaluation model of n-tier service systems that take into account the interactionsamong concurrently running requests. The developed modelextends experiment-driven approach is able to dynamicallyadapt changes and varieties in the workload and physicalresource allocation. The developed model is trained and testedin light of Multi-tier application benchmark. The numericalresults demonstrate that the developed models are accurate and effective.