chen - Energy-Aware Server Provisioning and Load...

Info iconThis preview shows pages 1–2. Sign up to view the full content.

View Full Document Right Arrow Icon
Energy-Aware Server Provisioning and Load Dispatching for Connection-Intensive Internet Services Gong Chen , Wenbo He ± , Jie Liu , Suman Nath , Leonidas Rigas , Lin Xiao , Feng Zhao Dept. of Statistics, University of California, Los Angeles, CA 90095 ± Dept. of Computer Science, University of Illinois, Urbana-Champaign, IL 61801 Microsoft Research, One Microsoft Way, Redmond, WA 98052 gchen@stat.ucla.edu, webohe@uiuc.edu { liuj, sumann, leonr, lixiao, zhao } @microsoft.com Abstract Energy consumption in hosting Internet services is be- coming a pressing issue as these services scale up. Dy- namic server provisioning techniques are effective in turning off unnecessary servers to save energy. Such techniques, mostly studied for request-response services, face challenges in the context of connection servers that host a large number of long-lived TCP connections. In this paper, we characterize unique properties, perfor- mance, and power models of connection servers, based on a real data trace collected from the deployed Windows Live Messenger. Using the models, we design server provisioning and load dispatching algorithms and study subtle interactions between them. We show that our al- gorithms can save a signi±cant amount of energy without sacrificing user experiences. 1 Introduction Internet services such as search, web-mail, online chat- ting, and online gaming, have become part of people’s everyday life. Such services are expected to scale well, to guarantee performance (e.g., small latency), and to be highly available. To achieve these goals, these services are typically deployed in clusters of massive number of servers hosted in dedicated data centers. Each data cen- ter houses a large number of heterogeneous components for computing, storage, and networking, together with an infrastructure to distribute power and provide cooling. Viewed from the outside, a data center is a “black box” that responds to a stream of requests from the Internet, while consuming power from the electrical grid and pro- ducing waste heat. As the demand on Internet services drastically increases in recent years, the energy used by data centers, directly related to the number of hosted servers and their workload, has been skyrocketing [8]. In 2006, U.S. data centers consumed an estimated 61 bil- lion kilowatt-hours (kWh) of energy, enough to power 5.8 million average US households. Data center energy savings can come from a number of places: on the hardware and facility side, e.g., by de- signing energy-ef±cient servers and data center infras- tructures, and on the software side, e.g., through resource management. In this paper, we take a software-based approach, consisting of two interdependent techniques: dynamic provisioning that dynamically turns on a mini- mum number of servers required to satisfy application- specific quality of service, and load dispatching that dis- tributes current load among the running machines. Our approach is motivated by two observations from real data sets collected from operating Internet services. First, the
Background image of page 1

Info iconThis preview has intentionally blurred sections. Sign up to view the full version.

View Full DocumentRight Arrow Icon
Image of page 2
This is the end of the preview. Sign up to access the rest of the document.

Page1 / 14

chen - Energy-Aware Server Provisioning and Load...

This preview shows document pages 1 - 2. Sign up to view the full document.

View Full Document Right Arrow Icon
Ask a homework question - tutors are online