Green data centers: Using hierarchies for scalable energy efficiency in large web clusters

Abstract

The growth in the demand for web services requires increasing processing capacity to maintain adequate response time to customer requests and, consequently, results in an increase in the energy consumption to support this infrastructure. This work is directed to the energy saving in large scale web server clusters, towards “green” data center construction. Energy consumption reduction is fundamental due to the economic and environmental aspects involved: energy generation has high costs and produces millions of tons of carbon. However, while saving energy, the quality of service offered to the customers should be maintained above an acceptable minimum level. Our solution involves optimization techniques, the use of Dynamic Voltage and Frequency Scaling technology (DVFS), and the application of a hierarchical architecture that uses a heuristic approach to define the cluster configuration at each instant.

Introduction

The main objective of this work is the reduction of the energy consumed by server clusters of web content providers, while maintaining a contracted SLA (Service Level Agreement). The growth of the demand for these services imposes a larger processing capacity to assist them and, consequently, an increase in the energy consumption to support the necessary infrastructure. It is essential to reduce this consumption due to the economic and environmental aspects involved: energy generation produces millions of tons of carbon at high costs [7].

The energy consumption management of servers can be done in several components, such as: processor, main memory, and disk storage system [7]. We deal with the problem by managing the consumption of the processors using a DVFS (Dynamic Voltage and Frequency Scaling) based strategy, and by turning on and off servers. DVFS uses the capacity of varying the supplied voltage and the operating frequency of the processor. In a brief way, DVFS allows power savings that are approximately proportional to the square or cube of the supply voltage. For more details, see [17].

Server clusters are usually designed to support the applications at their top demand. Therefore, most of the time, servers stay idle and consuming energy [3], [11]. To apply our strategy, the utilization of each one of the processorsʼ cores is periodically collected (this is designated in the remaining of the text as “utilization”). This data is used to define the frequency configuration of the processorsʼ cores, which should maintain the utilization below a certain limit, in a way to offer an appropriate QoS (Quality of Service) and to minimize the energy consumed. However, an exact on-line solution suffers from scalability problems [18]. In this work, we present an alternative to solve this problem, through a hierarchical architecture and a heuristic strategy in the choice of the frequency configuration.

For the validation of our proposal, we made experiments and the results are compared with the optimal solution, obtained through the GLPK package (GNU Linear Programming Kit) [8], and, also, with a well-known off-the-shelf DVFS technique, the OnDemand strategy implemented in the Linux distribution [13].

In the next section we present the contributions of our proposal. The remainder of this paper is as follows: in Section 3 we present related work. In Section 4 we present the problem definition. In Section 5 we describe the hierarchical architecture proposed. In Section 6 we introduce our local DVFS policy. In Section 7 we introduce the heuristic developed for the problem solution. Section 8 was divided in two subsections: in Section 8.1, with deal with the scale problem in the traditional architectures, and in Section 8.2 we show the results for experiments made in a real web cluster. Finally, in Section 9 we present our conclusions and indicate directions for future research.