Access frequency based energy efficiency optimization in data centers

Abstract

With the rapid growth of various data, large-scale data centers are built to store and analyze them. These data centers are consuming enormous energy every year, while most of them are wasted due to the low utilization of servers. In recent years, many efforts are carried out to reduce the energy cost as well as ensure the QoS, while they do not distinguish different data with different access frequency. In this chapter, an efficient energy optimization and management system is proposed based on the concept of different data weights caused by their access frequency. We collected the data's frequency of being requested in large-scale data centers and found that some data are frequently visited and always in an active state, while some data are seldom visited. We assume that data with different access frequencies have different weights, thus different management strategies will be adopted to save energy. In order to guarantee the availability of data, two redundancy policies: Replication and erasure coding will be discussed in this chapter. At last, we bring forward a policy about how to avoid hot spot caused by our method. © 2012 Springer Science+Business Media B.V.