Agent-Based Green Web Service Selection and Dynamic Speed Scaling
Jiwei Huang
†‡
and Chuang Lin
†
†
Tsinghua National Laboratory for Information Science and Technology
Department of Computer Science and Technology, Tsinghua University, Beijing 100084, China
‡
School of Computer Science, Georgia Institute of Technology, Atlanta, Georgia 30332, USA
Email: huangjw05@gmail.com, chlin@tsinghua.edu.cn
Abstract—With the increase of the energy consumption
associated with IT systems and services, energy efficiency
is becoming a critical concern in the design, development
and management of web service systems. In this paper, both
the web service selection and server dynamic speed scaling
are optimized by maximizing the quality of service (QoS)
revenue and minimizing energy costs. Stochastic models of
web service systems are proposed, and quantitative analysis
of the performance and energy consumption is carried out.
In addition, the service selection and speed scaling problem
is formulated as a Markov Decision Process (MDP) problem,
and algorithms to solve it are introduced. Furthermore, we
propose an agent-based optimization framework and design
related algorithms to solve the service selection and speed
scaling problem in large-scale web service systems. Finally,
their effectiveness is validated by simulation results.
Keywords-Web service; energy efficiency; service selection;
multi-agent; Markov reward model.
I. INTRODUCTION
Besides high performance, energy efficiency is an impor-
tant consideration during the design and development of IT
systems. It is reported that the average power consump-
tion per server class is increasing every year [1], and the
problem of energy consumption is even worse for large-
scale computing infrastructures such as web service systems,
cloud systems and data centers [2]. It was estimated that
it took about 61 billion kWh for the total electricity and
cost about 4.5 billion dollars for running IT infrastructures
in the US in 2006 [3]. In 2009, Google reported that
0.0003 kWh of energy was consumed per Google search,
resulting in more than 32 million kWh being consumed per
year [4]. Recently, more and more attention has been paid
to power consumption and energy efficiency in computer
system design and optimization.
As services computing becomes more and more popu-
lar, with the growing popularity of third-party commod-
ity clusters and cloud environments, there is a significant
opportunity promoting energy efficiency through effective
distribution and management of geographically-dispersed
web services [5]. It helps facilitate effectiveness of the
use of national resources and make more profits to the
service providers. Therefore, the energy efficiency in web
services and cloud computing has become a hot topic in
both academia and industry.
In web service systems, there are many categories of
services, each of which may have multiple instanced ser-
vices deployed on different web servers. An effective way
to achieve high energy efficiency is the efficient service
allocation or selection in service level which makes use of
distributed services placed in different servers [6]. Choosing
the appropriate services in different physical servers can
help improve the performance of services as well as reduce
power consumption of physical servers especially when the
servers are in sleeping or stand-by states before the service
placement. Another way is to dynamically adjust the CPU
frequencies of the servers in hardware level to effectively
reduce power consumption in web servers [2]. Each of
the two methods can be formulated as an optimization
problem, and has been studied separately in related literature.
However, the combination of the two problems, which can
help improve the energy efficiency of the whole web service
systems, is unexplored.
In this work, we fill this gap, and jointly study the service
selection and dynamic server speed scaling in web service
systems to achieve high energy efficiency while meeting the
performance requirements. We propose stochastic models to
describe the web service systems, and give mathematical
analysis of them to evaluate the energy efficiency. Then we
formulate both the service selection problem and dynamic
speed scaling problem using Markov Decision Process
(MDP) and introduce two relevant algorithms to solve them.
We theoretically prove that the MDP could optimize the
steady-state rewards in the stochastic models. Furthermore,
we design a framework and related algorithms to solve
the problem in large-scale web service systems effectively
and efficiently using multi-agent techniques. The simulation
results show that our proposed approach can significantly
and dynamically reduce the energy consumption of the web
service systems.
The remainder of this paper is organized as follows.
Section II introduces the service selection and speed scal-
ing problems in web service systems. In Section III, a
continuous-time Markov reward model and its embedded
discrete-time Markov model are built up to evaluate the
energy efficiency of the web service systems, and detailed
mathematical analysis is proposed. Based on the model, we
formulate the speed scaling as an MDP problem and give its
solution methodology in Section IV. Furthermore, an agent-
2013 IEEE 20th International Conference on Web Services
978-0-7695-5025-1/13 $26.00 © 2013 IEEE
DOI 10.1109/ICWS.2013.22
91