Improving heterogeneous workload performance in server virtualization based on user behaviors

Abstract

Large enterprises have many servers which must be managed and operated to support heterogeneous service. The emphasis, however, is to reduce power consumption of these servers for IT applications. To cope with such dilemma, the virtualization technique is employed which has gained attention in recent years. Server virtualization can improve hardware utilization by consolidate servers and also reduce physical space and power consumption in data center. However, the challenge is management of heterogeneous workloads in this system. This research proposes a new concept for managing workloads based on user’s behavior. The approach is divided into two parts. First, exploring the trend of user’s behavior in utilizing the service of each server with the help of data mining technique. Next, predicting hardware resources, CPU, and memory based on their actual operating profile using various prediction algorithms such as association rule discovery, ARIMA model, and exponential smoothing technique. A prediction performance model is devised to demonstrate between utilization and response time. The results show that user’s behavior are different in each type of service which affects the distribution of workload and time. The benefits obtained from the virtualization model permit assignment of resources from the prediction, thereby resource allocation can be carried out efficiently.