TSM: Efficient Thermal and Server Management for Greening Data Centers
Jianxiong Wan, Gefei Zhang, Limin Liu
School of Information Engineering, Inner Mongolia
University of Technology, Hohhot, China
e-mail: {jxwan, gfzhang, lmliu}@imut.edu.cn
Ran Zhang
Ordos UniCloud Technology co., Ltd. P.R. China
zhang.ran129@zte.com.cn
Abstract—Energy minimization is a crucial step to reduce the
operational cost in a data center. In this paper, we purpose a
novel thermal and workload management strategy, named
TSM, to cut down the energy consumption of a data center.
TSM, which is composed of a Thermal Control Coordinator
(TCC) and several Application Agents (AAs), is designed based
on the primal decomposition of the Total Energy Minimization
(TEM) problem. The advantage of the TSM is that it reduces
the total energy consumption while explicitly incorporating
with QoS and temperature constraints. Simulation results
show that TSM not only saves a significant amount of cooling
energy, but also cuts down the overall energy consumption of a
data center.
Keywords-Energy minimization, Thermal and workload
management, QoS control
I. INTRODUCTION
The past decade has experienced an unparalleled
development in the Internet services like Email, social networks,
E-commerce, live streaming, and online games, etc. The cloud
computing technology plays a key role in advancing these
Internet services. Using large scale data centers as the
underlying physical infrastructure, a cloud provides a reliable
and scalable platform for hosting various forms of Internet
services.
Unfortunately, large scale data centers suffer from the high
energy cost problem. Modern data centers for large enterprises
like Google, Microsoft, and Facebook consist of hundreds of
thousands of servers which consume significant amount of
power. It is reported that the energy consumption of data centers
accounts for 1
г
3% of the global electricity usage [1].
Literature [2] found that the energy consumption of data centers
grew by 56% from 2005 to 2010, and a recent study by Data
Centre Dynamics revealed that data center power usage
increased by 63% in 2012 [3]. Therefore, the need for
optimized power management is urgent to save the energy
consumption in worldwide data centers.
There have been lots of research efforts concentrating on
amortizing the energy consumption of data centers, and many
of which incorporate workload management into their problem
modeling and algorithm design since the resource utilization
plays a significant role in the server energy consumption model
[4]–[6]. However, the energy models in these works only
considered the energy consumption of servers and overlooked
the effect of the cooling systems. These models are insufficient
to embody the overall energy consumption of a data center,
since the electricity used to power the cooling system can take
up to half of the total electricity consumption [7], and even in a
well-designed data center, cooling still accounts for
approximately 37 percent of total electricity usage [8].
The cooling system reduces the temperature of servers by
extracting the air in the machine room, cooling it down, and
blowing the cold air back through the server racks. Liu. et al.
purposed an approach for cutting down the electricity bill by
leveraging various cooling techniques including chiller and
outside air cooling [10]. Li et al. explicitly derived a thermal
model and presented a closed form solution for the computing
and cooling energy management problem [11]. Other examples
like T* [12] and JETC [13] also used model-based approaches
to reduce the cooling energy consumption. These works, while
considering both server and cooling energy, did not encompass
the Quality-of-Service (QoS) issue.
In this paper, we purpose TSM, an efficient Thermal and
Server Management strategy, to green the data center while
ensuring QoS (in terms of server response time) and thermal
constraints. TSM is a distributed server management framework
coordinated by a central thermal control unit aiming at
minimizing total data center energy. In the following we present
TSM in detail: Section II presents the QoS model, thermal
model, and Total Energy Minimization (TEM) problem; Section
III discusses the solution method to the TEM problem and
proposes the TSM framework; Section IV demonstrates the
performance of TSM via simulations; Section V concludes the
paper.
II. SYSTEM MODEL
The data center in our consideration is shown in Fig. 1. It
consists of a set of servers and an air chiller. The data center
hosts a number of applications, each of which occupies a subset
of servers to meet their QoS requirement. We assume that all
servers belonged to a certain application share a same service
queue.
A working server generates waste heat, which can be carried
away by sucking cold air supplied by the air chiller form a
nearby vent tile and rejecting the hot air to the machine room.
Inside the machine room there is an air conditioner which
manipulates the indoor temperature to a predetermined set point
T
SP
. The air chiller is responsible for cooling down the air from
T
SP
to
T
c
(which is the cold air temperature for cooling the
servers). This scenario is a typical data center environment
considered by many researches.
2015 International Conference on Cyber-Enabled Distributed Computing and Knowledge Discovery
978-1-4673-9200-6/15 $31.00 © 2015 IEEE
DOI 10.1109/CyberC.2015.27
213
2015 International Conference on Cyber-Enabled Distributed Computing and Knowledge Discovery
978-1-4673-9200-6/15 $31.00 © 2015 IEEE
DOI 10.1109/CyberC.2015.27
213