Managing energy consumption and quality of service in data centers

Abstract

The main goal of this paper is to manage the switching on/off of servers in a data center during time to adapt the system with incoming traffic changes to ensure a good performance and a reasonable energy consumption. In this work, the system is modeled by a queue then, an optimization algorithm is designed to manage energy consumption and quality of service in the data center. For several systems, the algorithm is tested by numerical analysis under various types of job arrivals: arrivals with constant rate, arrivals defined by an constant discrete distribution, arrivals specified by a variable discrete distribution over time, and arrivals modeled by discrete distributions obtained from real traffic traces. The optimization algorithm that we suggest, adapts and adjusts dynamically the number of operational servers according to: traffic variation, workload, cost of keeping a job in the buffer, cost of losing a job, and energetic cost for serving a job.