IEEE-International Conference on Recent Trends in Information Technology, ICRTIT 2011 MIT, Anna University, Chennai. June 3-5, 20111
WSN based Temperature Monitoring for High Performance Computing Cluster D.Baghyalakshmi, Jemimah Ebenezer, S.A.V. Satyamurty Indira Gandhi Centre for Atomic Research, Kalpakkam.
[email protected] Abstract— Wireless Sensor Network (WSN) supports different kinds of applications in distinct areas, such as military, health care, agriculture, home or industry automation and others. Generally, there are three models of WSN: continuous, ondemand and event-driven. In continuous model, sensors send data periodically to the sink. In on-demand model, sensors sense continuously, store the data and sends only when requested. In event driven model the sensors send data only when certain events occur. In this paper we have presented the implementation details of WSN based temperature monitoring application. The main feature of our network is to continuously monitor the temperature in the 128 node High Performance Computing Cluster for its smooth functioning. The wireless sensor node senses and transmits the current value of temperature to the base station. This paper explains about the various steps involved in the experimental implementation and maintenance of the temperature monitoring network for High Performance Computing cluster at Computer centre, IGCAR. The performance analysis of the network is also discussed. Keywords - WSN, Temperature monitoring, TinyOS, LabVIEW
I. INTRODUCTION Wireless Sensor Network is an emerging technology in which a large number of sensor nodes organize and operate autonomously. These nodes are usually small in size with limited processing power, limited memory and limited energy source. Each sensor node comprises of sensing unit, microcontroller, ADC, memory, transceiver and power unit. Data collected at the wireless sensor node is transmitted to the gateway or base station (BS) for further processing. The transmission can be in a single hop or in multiple hops through routing / forwarding nodes. According to the magnitude of the network and the coverage area, the multihopped network needs multiple routes for forwarding the data to base station. WSN have a wide range of applications such as military, health care, agriculture, home or industry automation and others. Generally, there are three types of WSN application: continuous, on-demand and event-driven. In continuous monitoring, the sensors sense continuously and send data periodically to the sink. In on-demand, the sensors sense continuously and store the data in its memory. They will transmit the sensed data only when they receive a request from the base station. In event driven, the sensors send data only when certain events occur. Many general routing protocols are developed for wireless sensor networks. These protocols can be grouped as Location-based, Data-centric, Hierarchical, Mobility-based, Multipath-based, Heterogeneitybased and QoS-based protocols. The Architecture of WSN is
978-1-4577-0590-8/11/$26.00 ©2011 IEEE
generally classified as flat type and hierarchical type. In flat type, each WSN node transmits its data to the near by node for forwarding and all the nodes have equal priority. In hierarchical architecture, a cluster head will be elected for a cluster of nodes. It collects the data from all the nodes present in the cluster and forwards it to base station. Even though there are many routing protocols and architectures proposed, wireless sensor network is mainly application specific. For our experimentation, we have used the flat type architecture. IEEE 802.15.4 is the proposed standard for WSN. It focuses on low cost of deployment, low complexity and low power consumption. The data rate is 250Kbps at 2.4GHz. The star and peer-to-peer (mesh network) topology were allowed for communication between network devices. In the star topology, the communication is established between devices and a single central controller. In peer-to-peer topology, any device can communicate with any other device which is in range. The paper is organized as follows: Section II discusses about the High Performance Computing cluster and the need for temperature monitoring. Section III elaborates on the hardware used, the design of WSN, the implementation details and performance analysis. The conclusion and future enhancement are discussed in Section IV. II. NEED FOR HPC CLUSTER TEMPERATURE MONITORING A High Performance Computing (HPC) cluster is a parallel processing system, which consists of a collection of interconnected stand-alone computers cooperatively working together as a single, integrated computing resource. Such a cluster typically has a large number of computers (often called nodes) interconnected by high-speed, low-latency network. The HPC Cluster commissioned at IGCAR comprises of 128 Compute nodes and a Management node interconnected by high-performance Infiniband and Gigabit networks as shown in fig. 1. Each cluster node is powered by dual-processor, quad-core 64-bit Intel Xeon processor with the clock speed of 3.16 GHz. Each node has 16GB DDR2 Fully Buffered Memory and two SATA 160 GB Internal HDDs. The cluster system has three interconnect networks. The primary network meant for inter process communication (IPC) is based on the Infiniband architecture. The Infiniband switch supports upto 144 DDR 4X ports each with speed of 20 Gbps. The Administration network meant for cluster management and monitoring is based on 1Gbps Gigabit Ethernet. The Management network meant for hardware remote management and console access of nodes using Intelligent Platform Management Interface (IPMI) is based on Gigabit
1105
IEEE-ICRTIT 2011 Ethernet. The cluster has four storage nodes each with 6 TB of disk capacity. The management / master node is the head node of the cluster which is used for cluster administration and providing user interface for job submission and management. The cluster delivers a sustained performance of 9.5 TeraFLOPS with industry-standard HPL benchmark.
Datarate RadioCurrent drawn Radio outdoor range indoor
250Kbps 13mA 16mA 300m
50m
Tx at -3dBm Rx 1/4wave dipole antenna, LOS
IRIS node is having a provision for connecting various sensor boards, like MTS300, MTS310, MTS 400, MTS420 and MTS510. Each sensor board manufactured by Crossbow Technology has various combinations of sensors such as Temperature, Pressure, Vibration, Accelerometer, Magnetometer etc. For our experimentation, we have used MTS300 sensor board which has light, temperature, acoustic sensors, as well as a sounder as shown in Fig. 2. The temperature sensor used in this board is a NTC 10 Kȍ thermistor (Panasonic ERT-J1VR103J), which is a surface mountable component. It is configured in a simple voltage divider circuit. The output of the temperature sensor circuit is available at ADC1. The ADC output can be converted to degree Celsius using the following formula over 0 -50 °C. Fig. 1. HPC Cluster
The temperature in the 128 node High Performance Computing Cluster has to be maintained in 16˸ C for its smooth functioning. If the cooling is not proper and temperature raises beyond 20˸ C, the HPC cluster will auto shutdown due to overheating. Hence continuous monitoring of temperature is essential. The data collected at base station is made available to the administrators of HPC for necessary action. III. TEMPERATURE MONITORING
1/T(K)= a + b . ln(R t ) + c . (ln(R t))3 Where, R t = R1.(1023 - ADC)/ADC a = 1.30705 x 10 -3 b = 2.14381 x 10 -4 c = 9.3 x 10 -8 R1= 10kȍ ADC = output voltage from Mote’s ADC Microphone Temperature sensor
A. Hardware used for temperature monitoring A wireless sensor network has been established using IRIS nodes, whose specifications are given in Table 1. It has been developed by researchers at the University of California, Berkeley. It is operating at 2.4 GHz ISM band.
Light sensor
TABLE 1 SPECIFICATIONS OF IRIS
Processor/Radio
IRIS
CPU
Atmega 1281
Program memory Data memory AD Convertor Processor current draw Radio frequency Battery External Power
128KB 512KB 10 bit 8mA 8ȝA 2405MHz to 2480MHz 2 x AA batteries 2.7 V - 3.3 V
Sounder
Fig. 2. MTS300 sensor board
Remarks
B. User Interface and Application Tools
8 channel, 0-3V input Active Sleep
Molexconnector provided
1) Graphical User Interface: The temperature measured by the sensors need to be displayed at the PC connected to BS. For that GUI has been developed using LabVIEW. Laboratory Virtual Instrumentation Engineering Workbench (LabVIEW) is a programming language with Graphical Language developed by National Instruments. It is built for the design, simulation, modification, and compilation of digital instrumentation systems. The basic unit of the resulting program is the virtual instrument (VI) that consists of executable code controlled via a graphical front panel on the screen similar to a real instrument. In contrast to conventional programming languages, it is programmed on the basis of block diagrams and front panel elements. These elements are
1106
WSN based Temperature Monitoring for High Performance Computing Cluster connected by means of a wiring tool. After having tested a virtual instrument, the graphical language built from an application, compiles standalone executable code. fig. 3. Shows the web enabled front panel of HPC cluster room temperature monitoring application.
Fig. 3. Snap shot of GUI
fig.4. Shows the block digram of the corresponding front panel and it explains the internal interfacing using LabVIEW. This block diagram connects the hardware, PC and the LabVIEW system. It controls and receives the data from the WSN hardware. The function Create stream ( ) helps LabVIEW to control the USB equipments.
The USB port no. has to be given as input to the Create stream ( ) to identify the hardware. The packet size and access period are defined using Stream property ( ). Sub VI for data logging is also been shown in fig. 4. The raw analog voltage to temperature conversion equation has been incorporated in the Sub VI. 2) TinyOS Programming: TinyOS is a distributed, opensource operating system which supports large scale, selfconfiguring sensor networks. TinyOS includes radio messaging, message hopping from Mote to Mote, sensor measurements and signal processing. nesC is used as the programming language for TinyOS. The call graph generated by the nesC compiler is shown in the fig. 5. The call graph has two components namely nodes and labeled arrows which indicates the nesC components and interfaces The outgoing arrow denotes that the component is using the interface labeled on the flow, whereas an incoming arrow denotes that the component implements the labeled interface. XMTS300M is the main module of the application. It uses the interface ADC. It is implemented by the components Voltage, PhotoTemp which are used to read the battery voltage and temperature from the MTS300 sensor board respectively. It uses the Led interface for indicating packet transmission. The timer from TimerC is used to manage the delays required between setting up the hardware and reading the data.
1107
IEEE-ICRTIT 2011
Fig. 4. Snap shot of block diagram
sensor nodes with temperature sensor, two routing nodes and a BS. The computer centre layout and deployment of nodes were shown in the Fig. 6.
Generic Comm Promiscous Std con trol
Receive Msg
Std
l tro con
MULTIHOPROUTER
C te ou R
con tr
XMTS300M
M sg
Std control
Timer
Se nd
Timer C
Std
Std
ol tr on
Std control
Main
Voltage
C AD
Mhop Send
St d
ol contr
BaseStation
AD C
co nt ro l
Router1
PhotoTemp
Router2
ol
Sensor 2
Queue Send
Sensor1
Fig. 5. Call graph Fig. 6. Deployment of nodes
In our experiment, whenever the OTAP is done, the interval at which the Timer getting fired is altered. This in turn varies the sampling rate which helps to study and improve the performance of the network. C. Design of WSN The flat architecture has been chosen for WSN for HPC cluster temperature monitoring. This network consists of two
Initially the site survey has been done using a handheld spectrum analyzer to identify the interference at 2.4GHz spectrum range with in computer centre. All nodes were configured to work at 2410MHz frequency in channel 12 with 3dbm power level. The group ID was also configured. The ¼ wave dipole antenna is used. Two sensor nodes are placed inside and outside HPC rack to monitor the outlet and inlet air temperature of HPC cluster respectively as shown in fig. 7. The signal strength has been measured for identifying the proper position of router nodes till the signal reaches BS. The
1108
WSN based Temperature Monitoring for High Performance Computing Cluster base station is connected to the HPC administrator’s PC where the data is logged. Base Station Sensor node to measure the outlet temperature of cluster
The deployment of nodes remains the same as that of initial network setup. Instead of battery power, each node was equipped with a 3volt DC continuous power supply. A small PCB has been designed to generate 3volt DC output from AC mains as shown in the Fig. 9. This solves the problem of keeping the data rate high and it is working smoothly.
Sensor node to measure the inlet temperature of cluster
Fig. 7. Sensor nodes placed at Cluster rack and room
D. Experimental Implementation Fig. 9. PCB
WSN has been established with five nodes. Initially all the nodes in the network were programmed with default sampling interval (2sec). Whenever the timer gets fired the nodes transmit the data to its parent node for forwarding. Then the data collected at the router is transmitted to the gateway or BS for further processing. The nodes in the network were equipped with 3Volt battery power. In order to study the performance of network with respect to battery life of the nodes, the sampling rate of all the nodes were varied using the Over The Air Programming (OTAP) feature. For different sampling interval from 15sec to 240sec, the experiment was carried out repeatedly. The battery value was also transmitted to BS along with temperature values and it is been logged in database. The performance of the network is explained using the battery life versus Time graph plotted with logged data as shown in the Fig. 8. From the graph it has been observed that the nodes battery life gets exhausted soon when the data rate was set high. This results in replacement of batteries manually in a routine basis. But in our temperature monitoring application, it is not possible to fix the data rate very low. Based on this performance analysis, the upgraded implementation has been done as given below.
The WSN based Temperature Monitoring for High Performance Computing cluster application was discussed. The performance analysis of network with respect to battery life of the nodes was also performed. Since Wireless Sensor Network is very much specific to application, we have made modification in the source of power. Instead of battery power we preferred continuous power supply to node for our experimentation. Still there are some issues in continuous functioning of the network because of low RSSI and low link quality. In order to improve the network performance it has been planned to replace the existing nodes with in-house developed WSN node. ACKNOWLEDGMENT The work reported herein was done in collaboration with Anna University, Chennai. REFERENCES [1]
3.5 Node Battery voltage (Volts)
IV. CONCLUSION AND FUTURE WORK
3 15 S
2.5
30 S 45S
2
[2]
60S 1.5
120S 180S
1
[3]
240S
0.5 0 0
20
40
60
80
100
[4]
Time (Hours)
Fig. 8. Performance analysis
1109
Shio Kumar Singh, M P Singh, and D K Singh, ” A Survey of EnergyEfficient Hierarchical Cluster-Based Routing in Wireless Sensor Networks, ” Int. J. of Advanced Networking and Applications, Volume: 02, Issue: 02, Pages: 570-580 (2010) Sridevi Veerasingam, Saurabh Karodi Sapna Shukla, Mehar Chaitanya Yeleti, “Design of Wireless Sensor Network Node on ZigBee for Temperature Monitoring,” 2009 International Conference on Advances in Computing, Control, and Telecommunication Technologies Nair, M.K. Desai, A. Kumar and N. Gopalakrishna, V. “Temperature monitoring using Sun SPOTS applied to vermiculture,” Wireless Communication and Sensor Computing, 2010. ICWCSC 2010. International Conference on 2-4 Jan. 2010. Angela, D., Ghenghea, M. and Bogdan,I,” Supporting environmental surveillance by using wireless sensor networks,” Electrical and Electronics Engineering (ISEEE), 2010 3rd International Symposium on
IEEE-ICRTIT 2011
[9] [5]
[6] [7] [8]
Ajay Jangra, Swati, Richa, Priyanka,’’ Wireless Sensor Network (WSN): Architectural,” Ajay Jangra et al. / (IJCSE) International Journal on Computer Science and Engineering, Vol. 02, No. 09, 2010, 3089-3094 N. Kurata, Tokyo, M. Ruiz-Sandoval “Risk Monitoring of Buildings Using Wireless Sensor Network, “ Sasha Slijepcevic, Ranjit Iyer, Michael Panossian,” A Survey of Wireless Sensor Networks” I.F. Akyildiz, W.Su, Y.Sankarasubramaniam, E. Cayirci, “Wireless Sensor Networks: A Survey”, IEEE commun. Mag., published by Elsevier Science B.V. in 2002.
[10] [11]
[12]
1110
Jennifer Yick, Biswanath Mukherjee, Dipak Ghosal “Wireless sensor network survey’’, Computer Networks 52 (2008) 2292---2330
István Matijevics, and Simon János, “Control of the Greenhouse’s Microclimatic Condition using Wireless Sensor Network” Matijevics István, Simon János, „ Comparison of various wireless sensor networks and their implementation”, Proceedings of the Conference SIP 2009, pp 1-3, Pécs, Hungary, 2009 Horton, M.A., Glaser, S. & Sitar, N. 2002. Wireless Networks for Structural Health Monitoring and Hazard Mitigation. Proc. of the US-Europe Workshop on Sensors and Smart Structures Technology, 19-23.