2011 Fifth International Conference on Sensing Technology

Wireless Sensor Network for Data-Center

Environmental Monitoring

Michael G. Rodriguez, Luis E. Ortiz Uriarte, Yi Jia Kazutomo Yoshii, Robert Ross, Peter H. Beckman
Laboratory for Integrated Sensing Technologies Mathematics and Computer Science Division
Department of Mechanical Engineering Argonne National Laboratory
University of Puerto Rico - Mayaguez 9700 S Cass Avenue
Mayaguez, PR 00681-9045, USA Argonne, IL 60439, USA
yi.jia@upr.edu kazutomo@mcs.anl.gov

Abstract— Data centers’ energy consumption has attracted to values upwards of 6 kW per square foot of equipment for
global attention because of the fast growth of the information compute servers and over 10 kW per square foot of
technology (IT) industry. Up to 60% of the energy consumed in a communication equipment by 2014, based on EPA projections
data center is used for cooling in wasteful ways as a result of lack of [3].
environmental information and overcompensated cooling systems. In
this project, a wireless sensor network for data-enter environmental Cooling in data centers is a challenge that cannot be solved
monitoring was developed to improve energy efficiency and to with regular air conditioning methods. It has therefore become
optimize data-center performance. The sensor network consists of a customary to employ specialized air-conditioning units that
suite of sensor nodes for data sensing, a router node to relay sensed supply the computer racks with cold air from a raised floor
data, and a coordinator node to establish a network, receive the data, plenum. Vents are located on the tiles, which are perforated
and process the data. The prototype sensor network was built on
and placed directly next to the rack, such that air flow may
Arduino open source hardware with a seamlessly integrated XBee RF
module and configured to operate within the ZigBee mesh network travel through it. The rack rows are placed in what is known as
standard. A 24-hour test run at Argonne’s data center demonstrated a hot aisle/cold aisle arrangement, consisting of aisles where
that the wireless networked environmental monitoring solution is air flow alternates between the cold air coming out of the floor
easy to integrate and manage with the existing IT infrastructure, and hot air exhaust exiting the room through ceiling return
while delivering better visibility into the data center’s 3D vents. The objective of this approach is to prevent the mixing
temperature and humidity distribution and substantial improvements of the cold and hot air streams, which would reduce the
in energy efficiency. efficiency of the cooling process.
Keywords- wireless sensor network, data center, environmental Much of the work performed for data-center efficiency
monitoring, ZigBee, energy efficiency management. improvement so far has focused on the development of
analytical and numerical models to ensure that sufficient
I. INTRODUCTION cooling is provided to all the facilities involved. This work
includes case studies to model thermal maps after computer
Data centers’ energy consumption has attracted global
room air conditioning (CRAC) units become faulty [4]. Such
attention because of the fast growth of the information
analytical research is used to power tools that control and
technology (IT) industry. According to a U.S. Environmental
manage the cooling resources in computer machine rooms.
Protection Agency (EPA) report, U.S. data centers consumed
Furthermore, studies have been made of the effect of the
61 billion kilowatt-hours (kWh), or 1.5 percent of total U.S.
placement of the CRAC units on a data center's thermal map
electricity consumption in 2006, and this amount could
[5]. It has been established as well that the efficiency of data-
potentially double by 2011 [1]. This forecast indicates that
center cooling processes can be increased by implementing
unless energy efficiency is improved beyond current trends, the
automatic controls that respond to feedback from temperature
federal government’s electricity cost for servers and data
sensors located at key locations throughout the data center [6].
centers could be nearly $740 million annually by 2011. Around
The data collected from sensors, as well as knowledge of how
60 percent of the energy consumed in a data center goes to
environmental variables affect the conditions in the room, can
cooling in wasteful ways as a result of lack of environmental
be used to design control systems that can adjust the cooling
information and overcompensated cooling systems [2].
resources, such as the fans and outlet temperatures, to
Another trend at data centers is that the heat density of maintain the room in its operating range. The Microsoft
computing systems has increased at an accelerated rate. This Research Group [7], for example, has changed wireless sensor
increase in heat density is brought about by the increasing networks for data centers using a string of sensors as workers
density of computing resources, which yields more computing to a transmitter. The workers were all wired to the master,
power consumption. Such consumption is expected to increase

978-1-4577-0167-2/11/$26.00 ©2011 IEEE 533

which in turn polls the different sensors under it and
broadcasts the readings to a base station.
Wireless sensor networks provide an optimal and
integrated solution for distributed data collection, delivery and
analysis in many applications including data-center
environmental monitoring. In terms of a data-center
application, the wireless sensor nodes are to be deployed in the
preset locations, which include the computer racks and
structural areas through the room (e.g., walls, tables, or other b. Multihop Multipoint-to-Point Network
stationary locations) and at the air conditioner inlets and Fig.1 Architecture Diagram of Proposed Wireless Network.
outlets. Also, mobile nodes may be placed for useful free-
point data collection. These data collection points collect data
according to which information is relevant in their specific III. NETWORK HARDWARE DESIGN
location. For cooling purposes, temperature and humidity data
will be collected, providing a real-time, high-resolution As the core of the wireless sensor network (WSN)
thermal map of the rack environment that could be used for hardware for data center environmental monitoring, the WSN
dynamic controlling of the data center. node is usually composed of a series of basic sensing elements,
sensor interface circuit, power supply, and a radio
The objective of this research is to develop a wireless communication module.
sensor network for data-center environmental monitoring in
order to improve its energy efficiency and optimal data-center A. Sensing Unit
performance at Argonne National Laboratory. This real-time The prototype wireless sensor network is used to monitor
networked environmental monitoring solution should be easy environmental data relevant to the cooling processes at a data
to integrate and manage within existing IT infrastructure while center. For this purpose, Sensirion SHT15 temperature and
delivering better visibility into the data center’s thermal and humidity sensors are selected. This sensor employs a 2-wire
power consumption profiles. interface and has a response time of less than 4 seconds and
around 40 μW of power consumption.
II. NETWORK ARCHITECTURE DESIGN
B. Sensor Interface Unit
A wireless sensor network consists of sensor nodes, The Arduino Fio open source hardware is adopted as the
interface circuit, power supply, and RF radio module. sensor interface. This platform is based on AVR's ATMega
Compared with many short-range wireless communication 328P microcontroller, and the board houses 8 analog pins and
protocols such as Bluetooth and WiFi, the ZigBee standard is 14 digital pins for input and output operations, supporting SPI
more suitable for data-center environmental monitoring and I2C communication on certain pins. The role of the
applications. In a data center, there are two areas of interest: the microcontroller in the wireless sensor network is to poll the on-
open area in the room and the rack inlet and outlet temperatures board sensor’s data, arrange it using the developed packet
and humidity. The wireless sensor network in this study is protocol, and send it to the RF module to be transmitted to the
designed having two different wireless network architectures, route nodes. The data can be time stamped on board if
as shown in Figure 1. A multihop multipoint-to-point network necessary.
design is to monitor the large area environmental profile inside
the entire data center, and a multihop cluster network C. Wireless Communication Unit
architecture is used to capture temperature and humidity The wireless sensor network is developed based on XBee
information at a concentrated area of the rack. Each network is 2.5 RF modules, which were engineered to operate within the
composed of a suite of sensor nodes to sense environmental IEEE 802.15.4/ZigBee protocol and support the needs of low-
information, a router node for relaying the sensed data, and a cost, low-power wireless sensor networks. ZigBee defines
coordinator node to start the network, receive the data, and three device types. The coordinator is responsible for
process the data. Software has been developed for managing establishing the operating channel and personal area network
the Arduino sensor interface, ZigBee wireless communication, (PAN) ID for an entire network. Once established, the
and data management. coordinator can form a network by allowing routers and end
devices to join to it. The router maintains network information
and uses this information to determine the best route for a data
packet. The end device always interacts with the parent nodes
to receive or transmit data. The router can relay messages and
allow the child node to connect to it. End nodes are intended to
sleep periodically and therefore have no routing capacity.
XBee 2.5 RF modules support the use of the ZigBee
standard, which provides a robust mesh networking
architecture needed for the multihop multipoint-to-point
network and multihop cluster hierarchy presented. This
a. Multihop Cluster Network wireless module integrates seamlessly with the Arduino Fio

534
board used as well. The RF module operatees in the 2.4 GHz saves the data into the databbase file. The second program
band with 16 direct sequence channels. Alsoo, since the ZigBee organizes the database into sepparate temperature and humidity
standard is stacked on top of the IEEE 8022.11.4 standard, it data files. The third program viisualizes the sensed data. This is
supports 16-bit and 64-bit addressing for eaach node. In total, an easy-to-use graphic interfacce with a drop-down menu for
each channel has 65,000 unique networkk addresses while selecting sensor nodes, and tem mperature or humidity files, and
supporting up to 250 kbps data rate. XBee modules consume the program has a button to reffresh the screen in order to show
38-40 mA when transmitting and 35-40 mA when in receiving all the data in the same screen.
mode. Thus, the radio module is the most crittical component of
the wireless node from a power consumption viewpoint. V. RESULTS AND
A DISCUSSIONS
D. Power Supply Figure 2 shows 10 protottype networked wireless nodes
The prototype sensor node is powered by a 2000 mAh, with temperature and humidity sensors.
3.7V lithium polymer battery. This battery was selected A. Key Features of the Developped Wireless Sensor Network
because of its long battery life when takingg into account the
application and the fact that the Arduino Fio includes a lithium 1) High Performance
polymer battery port. Also, lithium polym mer batteries are - Up to 100 m/indoor annd 1.6 km/outdoor
rechargeable, which is of interest for continuued deployment in - RF data rate: 250 kbpss
the data center.
2) Low Power
IV. NETWORK SOFTWARE DESIGN
E
- Tx current: 295 mA @ 3.3 V
Three software products were developedd for this wireless - Rx current: 45 mA @ 3.3 V
sensor network in order to establish thee sensor interface, - Power down current (IIdle mode): < 1 µA @25 ºC
configure mesh network, and manage thee sensed data for
receiving, storing and displaying data. 3) Advanced Networking Capability
C
A. Sensor Interface Software Design - 16 channels
The Arduino Fio open-source software reads temperature - 65,000 unique networkk address
and humidity strings from the sensors, convverts these strings - Self-routing, self-healing, fault-tolerant mesh
into numerical values with specific data format, manages network.
sensor ID, and develops a data packet to be b sent to the RF - 16-bit unique physicall address for each sensor node
module. Then, Arduino Fio is programmedd by the Arduino
Integrated Development Environment with FTDI USB cable. 4) Sleep Mode
The Arduino open source software providees sensor interface - Both pin sleep and cyclic sleep, allowing the RF
programming to facilitate application developpment. module to enter statees of low power consumption
B. Zigbee RF Module Configuration when not in use.
In order to realize the designed netw work architectures,
XBee modules were configured to behavee as coordinators,
routers, and end devices depending of the circumstances.
c The
XBee Series 2 is designed to add mesh networking
n to the
underlying 802.15.4 radio. The mesh netw work consists of a
coordinator node that receives data from m its surrounding
router/end nodes. The coordinator has the funnction of forming a
network and is responsible for establishiing the operating
channel and PAN ID. Once established, thhe coordinator can
form a network by allowing end devices to join
j to it. Since all
XBee modules come with a default PAN ID,, a unique PAN ID
was assigned to each XBee module in orderr to avoid having a
communication problem with other sensorr networks in the Fig. 2 Networked Wireless Nodes with Temperature and
same area. After choosing the correct function set by X-CTU, Humiddity Sensors.
which is the XBee configuration software proovided by the Digi B. System Operation Verification
website, the new configuration will be downlloaded into the RF
modules. The data-center metal case environment, humidity
conditions and arrangement of o electronic equipment could
C. Data Management System Software Desiggn affect the radio wave propagaation of the developed wireless
Three data management programs were developed
d by using sensor network. To validatee the reliable communication
the MATLAB programming language. The first fi program reads performance and packet loss rate,
r we conducted two 24-hour
real-time data strings from coordinator nodess, converts them to measurements. Figure 3 shows the deployment of the multihop
numerical values, detects types of informatioon (temperature or multipoint-to-point sensor nettwork at the data center and
humidity), puts time stamps on each receiveed sensor data, and Figure 4 illustrates the multtihop cluster network wireless
sensor node deployed on one raack.

535
 (b)
Fig.3 Deployment of Multihop Multipoinnt-to-Point Sensor
Network at the Argonne Data Center. Fig. 5 Examples of Temperratures (Red Line) and Humidity
(Blue Line) Readings froom Node 5 (a) and Node 9 (b).

Examples of preliminaryy temperature and humidity

monitoring results from node 5 and node 9 are given in Figure
5. The maximum data-center registered temperatures varied
from 15°C to 32°C, and humiddity values fluctuated from 28%
to 72%.
This study demonstrates that wireless sensor networks can
be an effective tool for enviironment monitoring in a data
center. Such a network offfers the advantage of easy
deployment throughout the com mputer racks because there is no
need of wiring for power and datad transmission. This network
also offer freedom in deploymeent, as the sensor modules can be
placed in locations where wireed sensors would be unfeasible
for technical or safety reasons.
The performance of the wireless sensor network prototype
Fig. 4 Wireless Sensor Nodes Deployed
D inside a rack during the 24-hoour run, from a data collection
at the Argonne Data Centter. point of view, was shown to bee effective. No data packets were
lost during transmission.
Although this particular test lasted for about 24 hours, it
was noticed that the temperrature and humidity variations
occurred between the zones. The results show that some areas
of the data center are being chhilled too much, while in others
the cooling is not enough. Signnificant changes in humidity and
temperature can cause timing, corrosion, and hygroscopic dust
failures, and electrostatic dischaarge in data center equipment.
VI. CONCLUSIONS
O

A prototype wireless sensor network for data-center

environmental monitoring has been b developed. A 24-hour test
run at Argonne’s data center haas demonstrated that the wireless
networked environmental moonitoring solution is easy to
(a) integrate and manage with the existing IT infrastructure, while
delivering better visibility into the
t data center’s 3D temperature
and humidity distribution. Any type of analog or digital sensors
can be added to this sensor netw work.
Collecting data to understtand temperature and humidity
distribution is a first step tow
ward improving a data-center’s
energy efficiency. The ultimatee goal of this project is to close
the loop between large-scale seensing and distributed actuation,
in order to dynamically changge the environmental conditions
and energy resource allocationn in data centers. The developed

536
wireless sensor network could also be used as a diagnostic tool [3] U.S. Environmental Protection Agency, Report to Congress on Server
to readily identify current data-center energy efficiency and Data Center Energy Efficiency, 25-32, 2007.
problems and help mitigate the cooling challenges of the data [4] Abdlmonem H. Beitelmal, Chandrakant D. Patel, Thermo-Fluids
Provisioning of a High Performance High Density Data Center,
center. Distributed and Parallel Databases, 21, 227–238, 2007.
ACKNOWLEDGMENT [5] Bedekar, V.; Karajgikar, S.; Agonafer, D.; Iyyengar, M.; Schmidt, R.,
Effect of CRAC Location on Fixed Rack Layout, ITHERM ’06, May
30-June 2, 2006, San Diego, CA.
This work was supported in part by the U.S. Dept. of
Energy under Contract DE-AC02-06CH11357 and in part by [6] Boucher, T., Auslander, D., Bash, C.E., Federspiel, C., Patel, C.D.,
Viability of Dynamic Cooling Control in a Data Center Environment,
National Science Foundation. We thank Rajesh Sankaran for ASME Journal of Electronic Packaging, 2006.
his contributions to this research work. [7] Jie Liu, Bodhi Priyantha, Feng Zhao, Chieh-Jan mike Liang, Qiang
Wang, Sean Jame, Towards discovering data center genome using
sensor nets, HotemNets’ 08, June 2-3, 2008, Charlottesville, VA.
[8] Chieh-Jan Mike Liang, Jie Liu, Liqian Luo, Andreas Terzis, Feng Zhao,
REFERENCES RACnet: A High Fidelity Data Center Sensing Network, SenSys’09,
November 4-6, 2009, Berkeley, CA.
[1] Eun Kyung Lee, Indraneel Kulkarni, Dario Pompili, Manish Parashar,
Proactive thermal management in green data centers, Journal of [9] Jeffrey Rambo, Yogendra Joshi: Modeling of data center airflow and
Supercomputing, 2010. heat transfer: State of the art and future trends. Distributed and Parallel
Databases, 193-225, 2007.
[2] Kenneth G. Brill, Heat Density Trends in Data Processing,Computer
Systems, and Telecommunications Equipment: Perspectives, [10] Hendrik F. Hamann, A measurement based method for improving data
Implications, and the Current Reality in Many Data Centers, 2006. center energy efficiency, 2008 IEEE International Conference on Sensor
Networks, Ubiquitous, and Trustworthy Computing.

537