Ad Hoc & Sensor Wireless Networks Vol. 8, pp. 141–159 Reprints available directly from the publisher Photocopying permitted by license only ©2009 Old City Publishing, Inc. Published by license under the OCP Science imprint, a member of the Old City Publishing Group Trade-off Between Power Consumption and Performance in Bluetooth Juan-Carlos Cano1 , José-Manuel Cano2 , Carlos T. Calafate1 , Eva González2 and Pietro Manzoni1 1 Department of Computer Engineering, Technical University of Valencia, Spain Technology Department, University of Malaga, Spain E-mail: jucano@disca.upv.es 2 Electronics Received: December 15, 2007. Accepted: December 24, 2008. Advancements in communication technologies, coupled with an increase of computer’s performance, are allowing us to enter the age of ubiquitous computing. The Bluetooth communication standard appears as a major solution in this new arena. Bluetooth was designed targeting lowpower systems from the beginning, and this design criteria should be maintained for Bluetooth to be widely accepted in the marketplace. In this paper we investigate the power characteristics of the Bluetooth technology when supporting low-power modes. We provide accurate power consumption measurements for different Bluetooth operating modes. Such information could be used to drive technical decisions on battery type and design of Bluetooth-based end systems. Finally, we examine the trade-off between power consumption and performance for a commercial off-theshelf Bluetooth device. We find that the use of the sniff mode could be quite compatible with the use of multi-slot data packets. However, when the channel conditions require selecting single slot data packets, the sniff mode could have an impact on performance, and so the power/delay trade-off must be taken into consideration. Keywords: Bluetooth, power consumption, performance evaluation. 1 INTRODUCTION The widespread advent of wireless networks and the rapid proliferation of handheld computing devices have made mobile computing possible. In addition, the miniaturization of devices and low power wireless communication standards have paved the path towards a pervasive computing environment [1]. 141 “aswin123” — 2009/5/14 — 14:21 — page 141 — #1 142 J.-C. Cano et al. Bluetooth technology [2] is a major communication standard in this new arena of advancement that offers seamless wireless connectivity between Bluetooth-enabled computing devices. Bluetooth allows devices to communicate using short range radio links, and it is characterized by low complexity, low cost, and low power consumption. Bluetooth was designed for an inherently low-power operation, criteria which should be maintained in order for Bluetooth to be widely accepted in the marketplace. The emphasis on low power operation stems from portable devices, which strongly depend on the efficient use of their battery. In fact, Bluetooth allows making a trade-off between data transfer rates and power consumption. Energy-aware system design and evaluation of network protocols for an ubiquitous networking environment require practical knowledge of the energy consumption behavior of commercial wireless devices. Moreover, it may be worthy investigating the impact that the use of energy efficient operation modes will have over the overall system performance. From the user’s point of view there is little or no knowledge about Bluetooth power consumption. Even more, confusion exits at the technical level as well. Data sheets usually show power consumption values while transmitting and receiving, which represent average values for static modes of operation, missing relevant information for representative operating states such as startup, idle state and inquiry state, as well as for low-power modes, i.e., sniff, hold and park, as defined by the Bluetooth standard [2]. There are few published measurements of the power consumption of Bluetooth network interfaces. In [3] Kasten and Langheinrich report some experiments measuring the power consumption of a Bluetooth-based sensor node. These authors claim that, in small devices such as autonomous sensor nodes, Bluetooth interfaces will consume most of the power. The study focuses on power consumption for the transmission, reception, and inquiry operating modes. However, since the selected Bluetooth modules lacked many of the standard Bluetooth features, the authors do not cover low-power modes. A mathematical framework for the analysis of energy efficiency in Bluetooth systems is presented in [4]. This work presents some analytical results of the system behavior using different packet types and under several scenarios. The work is not based on experimental measurements, but rather on theoretical assumptions. Finally, other Bluetooth-related works employ rather old and inadequate power models that were derived from other wireless technologies [5,6]. In this paper we perform a full power characterization of the Bluetooth technology, providing accurate current and power consumption measurements for different operating modes. The study has been experimentally validated for our platform prototype [7], a Bluetooth-based wireless node designed to support spontaneous and ubiquitous computing. Based on this power characterization, we proceed by studying the trade-off between power consumption “aswin123” — 2009/5/14 — 14:21 — page 142 — #2 Performance in Bluetooth 143 and performance. In particular, in this paper we focus on the power-delay and power-throughput trade-off offered by a commercial off-the-shelf Bluetooth device using the sniff mode, and under a wide range of scenarios using both UDP and TCP traffic. We evaluate the impact of the packet type and distance between nodes on the observed network performance. The rest of this paper is organized as follows: Section 2 offers a brief overview of the Bluetooth standard. Section 3 investigates the power consumption of the Bluetooth technology, with a special focus on the low-power modes defined by the standard. Based on this power characterization, Section 4 examines the trade-off that grants the lowest power consumption while meeting some performance requirements. Finally Section 5 concludes this paper. 2 AN OVERVIEW OF BLUETOOTH Recently, Bluetooth technology has emerged as a promising platform for short range wireless networking. The Bluetooth standard defines a short range and low cost wireless radio system; its purpose is offering an alternative to connect portable devices without requiring drawing cables between them. It operates in the 2.4 GHz ISM band, and is the baseline approach for the IEEE 802.15.1 [8] Wireless Personal Area Network (WPAN) standard. Bluetooth uses a polling scheme where a single master coordinates the access to the medium of up to 7 active slaves, forming a piconet. It also considers the possibility of interconnecting multiple piconets in the same area to form a scatternet. Bluetooth is based on a connection-oriented scheme. Nodes that are close-by can find their neighbors using the inquiry procedure. After discovering close-by devices, a node can decide to page them to start a connection. A dedicated protocol called Service Discovery Protocol (SDP) can be used to interchange information about all the available services at each node. The Bluetooth specification defines two different types of links: Synchronous Connection-Oriented (SCO) and Asynchronous Connection-Less (ACL). The first one handles real time traffic, like voice, while the latter is commonly used for data transmission. An ACL link is parameterized by packet size and data encoding. Each ACL link allows using 1, 3, or 5 slot data packets, where a slot is 625 µs long. Additionally, it optionally allows using Forward Error Correction (FEC). According to these parameters, Bluetooth offers 6 different data packets at the baseband layer that can be classified in two main groups: Data Medium Rate packets (DM ), which provide a 2/3 FEC Hamming code, and Data High Rate packets (DH ), which provide no FEC coding at all. In general, DH packets can be good candidates for more efficient transmissions at the cost of reducing the probability of successful transmission under poor channel conditions. A major challenge of Bluetooth is allowing for an inherently low power design. The Bluetooth specification integrates various methods for achieving “aswin123” — 2009/5/14 — 14:21 — page 143 — #3 144 J.-C. Cano et al. low power consumption. Also, Bluetooth devices can adjust the power depending on the range of communication. Devices of categories 1, 2 and 3 cover a distance of 100, 10 and 1 meters, respectively. Although the standard specifies up to 100 m outdoors, few companies include such products in their market lines because such distances are rarely required/appropriate. A second method of reducing power consumption deals with dynamically adjusting the transmitted power. According to the receiver signal strength indicator (RSSI) a Bluetooth device can request an increase or a decrease of the other device’s transmitted power. The higher the transmit power level selected, the greater the communication range (though interference is increased as well). Finally, Bluetooth also allows a device to regulate power consumption depending on its activity. A Bluetooth node can be in the active mode when actively participating on the channel; otherwise it can reduce consumption by eventually using the low-power modes (sniff, hold, and park). We now proceed to study the power consumption of a class 1 Bluetooth module which offers all power modes, and that has been configured for a maximum target distance of 10 meters. 3 BLUETOOTH POWER CONSUMPTION CHARACTERIZATION The Bluetooth controller operates in two major states, i.e., standby and connection. The standby state is the default low power state in Bluetooth, where only the native clock is running and there is no possible interaction with any other device. A Bluetooth node uses the Inquiry and Page procedures to transit from the standby to the connection state, where nodes can exchange packets. A Bluetooth node in connection state can be in the active mode to actively participate on the channel, or it can also reduce consumption by switching to low-power modes (sniff, hold, and park), as defined by the Bluetooth standard. In this section we perform a detailed analysis of the power characteristics of the Mitsumi WML C11 [9] class 1 Bluetooth module, which is based on the CSR BlueCore 2 chipset, and is a fully qualified transceiver compliant with the Bluetooth 1.1 Specification [2]. 3.1 Measurements setup To obtain the power-consumption measurements we used the test circuit shown in Figure 1. The circuit is based upon the INA193 [10], a high side current monitor integrated circuit from Texas Instruments. In our test circuit, a shunt resistor (Rs) is used to cause a small voltage drop proportional to the current flowing into the Bluetooth module (Ibias ). This voltage is amplified 20 times by the INA193 circuitry, providing an output voltage that represents a measure of the bias current. The output of the INA193 is connected to a PC based data acquisition system that digitizes the signal. The circuit also incorporates a first-order low pass filter with cutoff “aswin123” — 2009/5/14 — 14:21 — page 144 — #4 145 Performance in Bluetooth RS 1 Vcc=3.3V RF C Ibias Ibias RF MITSUMI WML-C11 Bluetooth Module RS<<RF<<5k GND 5k 5k Vout RL=100k Rs RL I bias 5k Cutoff Frequency 1 fc 2 ·2 RF C PC based data acquisition and representation system + TI INA193 FIGURE 1 Test circuit for Bluetooth power measurements. frequency fc = 1/2RC, equal to 200 Hz, which allows an accurate sampling process while avoiding aliasing. To improve the accuracy, the measurement circuit has been calibrated using a 6.5 digits digital multimeter. The data acquisition system was built using an AduC812 microcontroller [11] from Analog Device, attached to a Personal Computer thought a serial port. We select a sampling rate of 1 kHz (one thousand samples per second), as it provides both accuracy and legible representation. Since each sample measures the instantaneous input current, we estimate the average cur rent consumption by: I = N i[n]/N , where N represents the total number n=1 of samples and i[n] represents the value of sample n. 3.2 Experimental power measurements We now proceed by studying the power characteristics of our Bluetooth module. Below we describe the power measurements for the following representative operating states: Bluetooth startup, standby, inquiry, page, active, hold, park, and sniff. For each state we trace the fluctuations in terms of current consumption during a period that roughly encompasses the given state. The bottom and left axes represent time and the corresponding instantaneous current consumption. 3.2.1 Startup and standby We first investigate the behavior of our Bluetooth module during startup and initialization. We are specially interested in knowing how much time and energy are required to switch from power down to the active state. Figure 2 shows the current consumption during startup and initialization. We observe that the Bluetooth module needs around 2 seconds to finish initialization and “aswin123” — 2009/5/14 — 14:21 — page 145 — #5 146 J.-C. Cano et al. Consumption (mA) Average Consumption (mA) Startup Standby Time (s) FIGURE 2 Current consumption (mA) during startup and initialization. then enter into the standby state. During this time the average current consumption is about 20 mA which, using a system voltage of 3.3V, represents a power consumption of 66 mW. After initialization the Bluetooth module enters into the standby state, which represents the default state for the Bluetooth technology. In this state the module is in a low-power mode where there can be no connections open and all components, except the internal clock, are switched off. The current consumption during the standby state fluctuates from 1.4 mA to 3 mA, and so the average consumption is about 2.2 mA (7.3 mW). Even though the average power consumption during the standby state is relatively low, it is expected that power consumption will be significant since nodes must maintain their Bluetooth interface in this mode for longer periods of time. Applications can periodically switch off the Bluetooth module to reduce the power consumption in the standby state; however, they should take into account the time and energy required for the startup process. 3.2.2 Inquiry and page The inquiry is an asymmetric procedure during which the inquiring node and the inquired one use the complementary inquiry and inquiry scan states. A node that has been configured into the inquiry state continuously sends out inquiry messages to find other nearby nodes. Similarly, Bluetooth nodes that have been put in the inquiry scan state reply to the inquirer as soon as they detect an inquiry message. We found that the power consumption is about 231 mW (70 mA) and 139 mW (42 mA) for the inquiry and the inquiry scan states, respectively. So, the power consumption in either state is considerably high with respect to the “aswin123” — 2009/5/14 — 14:21 — page 146 — #6 Performance in Bluetooth 147 standby state. A node in the inquiry state continuously sends inquiry messages during the entire interval. However, a node in the inquiry scan state does not need to be continuously listening to inquiry messages, and so it can alternate between the standby and the inquiry scan states in order to save power. Once the inquiring device discovers other close-by Bluetooth devices it can switch to the page state to setup a new connection. Nearby nodes in the page scan state will answer to page messages in order to share some essential information and, finally, both enter into the connection state. The consumption values are similar to those obtained previously, i.e., 208 mW (63 mA) and 149 mW (45 mA) for the page and the page scan states, respectively. The main difference of the page procedure with respect to the inquiry procedure is that the duration of the scan window is smaller than the inquiry one. After the page procedure completes both nodes are in connect state, and so data transmission can start almost instantaneously. 3.2.3 Connection state A Bluetooth device in the Connection state can be in the active mode or using any of the low-power modes defined in the standard. In the active mode both master and slave are kept synchronized with each other to participate actively on the channel by listening, transmitting or receiving packets. We observe that, when the ACL link is established, the consumption values fluctuate, leveling out at around 21 mA for the master node and at 41 mA for the slave node. The master node can reduce the average consumption since it knows the exact time when a packet transmission is going to occur, adjusting its transceiver accordingly. On the other hand, the slave has no other option but keeping itself synchronized and active at all times. In addition, when the master and the slave have data to transmit, the current consumption increases by up to 34% and 30% for the slave and the master, respectively. We show that, in active mode, data transmission can start almost instantaneously, but at the expense of increased power consumption. 3.2.4 Low-power modes: Hold, park, and sniff states Bluetooth nodes can reduce consumption by recurring to the low-power modes defined by the Bluetooth standard, i.e., sniff, hold and park. When low-power operation is activated, all the nodes within a piconet can reduce their duty cycle. Hold mode In this mode a node temporarily shuts down its radio interface, taking some time-off to save power. Prior to entering hold mode, master and slave should agree on when to return to active mode again. Figure 3 shows the current consumption of a node entering the hold mode. After the hold time has been configured, a node can enter this mode, where consumption is reduced near to 3 mA. However, the node can not exit from this state before the hold time expires. “aswin123” — 2009/5/14 — 14:21 — page 147 — #7 148 J.-C. Cano et al. Consumption (mA) Consumption (mA) Consumption Average Active Active Hold Time (Con urable) Time (s) FIGURE 3 Current consumption (mA) during the Hold state. Park mode When a slave node does not need to participate on the piconet, but still wants to remain synchronized to the master, it can select the park mode. The average current consumption for the park mode is about 4 mA. When in the park mode the node should periodically listen to the channel to remain synchronized within the piconet. In our measurements, when the parked slave listens to the channel to hear the beacon packet from the master, it increases the current consumption to around 35 mA. Sniff mode In this mode a slave device, rather than listening on every slot for the master’s beacons, only listens to the channel at specified times, agreed upon with the master. To enter the sniff mode, master and slave negotiate a sniff interval, Tsniff and a sniff window, Twin . A slave node will thus listen to the piconet at regular intervals (Tsniff ) for a short period (Twin ). Figure 4 shows the current consumption using a Tsniff of 1.5 seconds and a Twin equal to 200 ms. The average consumption is around 9 mA. The power consumption will remain low as long as Tsniff is large compared to Twin . In fact, when we modify the Twin parameter to 20 ms, the average consumption decreases to around 4 mA (similar to the park state). In addition, we check that the sniff mode allows, using the Ttimeout parameter, that a node with enough data to be transmitted is completely active during several consecutive Tsniff slots in order to reduce the data latency. Of course, during this period the consumption will be similar to the consumption during the active mode. Summary Table 1 summarizes the power consumption of the Mitsumi Bluetooth module [9] we have used. “aswin123” — 2009/5/14 — 14:21 — page 148 — #8 149 Performance in Bluetooth Active Consumption (mA) Consumption (mA) Consumption Average Sn Time (s) FIGURE 4 Current consumption (mA) during the Sniff state. Tsniff = 1.5 s and Twin = 200 ms. Bluetooth mode Current Power Startup and initialization Standby Inquiry Inquiry Scan Page Page Scan Active (no transmission) (Master) Active (no transmission) (Slave) Active (transmission) (Master) Active (transmission) (Slave) Hold Park Sniff 20 mA 2 mA 70 mA 42 mA 63 mA 45 mA 21 mA 41 mA 27 mA 55 mA 3 mA 4 mA 9 mA 66 mW 6.6 mW 231 mW 139 mW 208 mW 149 mW 69 mW 135 mW 89 mW 181 mW 10 mW 13 mW 30 mW TABLE 1 Average current and power consumption. System voltage is 3.3 V 4 THE POWER-PERFORMANCE TRADE-OFF IN BLUETOOTH The previous experimental study showed that Bluetooth modules implementing low-power modes could significantly alleviate the power consumption in Bluetooth. In particular, the hold mode could be extremely useful to reduce the power consumption for those applications using a predictable scheduling. Applications with a high number of peers within a piconet, and using traffic “aswin123” — 2009/5/14 — 14:21 — page 149 — #9 150 J.-C. Cano et al. FIGURE 5 Experimental test-bed configuration. patterns with a regular scheduling, could benefit from using parked slaves. Of special interest is the sniff mode, which can offer a trade-off between low power consumption and low transmission delay. Moreover, the sniff mode is the only low-power mode allowing a slave node to transmit information while still remaining in a low-power state. In this section we focus on the sniff mode to study the trade-off between power consumption and performance for a Bluetooth enabled device. We built a small test-bed using a set of laptops and commercial off-the-shelf Bluetooth devices to test Bluetooth connections. Figure 5 shows our test-bed network configuration, where each laptop runs on an Intel Pentium IV 2000 Mhz TOSHIBA Satellite 1900-303 based on the Suse Linux 8.1 distribution and with a Belkin Bluetooth USB Dongle card. We used the BlueZ [12] protocol stack to configure each Bluetooth device using the Bluetooth Personal Area Network (PAN) profile. BlueZ is the official Bluetooth protocol stack for Linux, providing a set of monitoring and traffic tools to deal with the Bluetooth technology. We study the impact that the use of the sniff low power mode has over the performance of UDP data traffic and TCP data traffic. Our experiments focused on evaluating the power-delay and power-throughput trade-off experimented by UDP and TCP traffic, respectively. We also performed a sensitivity analysis to evaluate how distance and packet type affect power consumption, throughput and delay. Each node in our test-bed has been configured using two different network interfaces: the Bluetooth interface and the standard Ethernet one. This configuration allows redirecting all the traffic between the two interfaces, allowing each node to evaluate the average packet delay through the feedback channel (Ethernet), avoiding having to synchronize the system clocks of the two hosts. With this setting we achieve a high accuracy in our measurements since the delay introduced by Bluetooth transmissions is several orders of magnitude “aswin123” — 2009/5/14 — 14:21 — page 150 — #10 151 Performance in Bluetooth higher than the delay caused by host processing and loopback transmission via Ethernet. To achieve the desired functionality we used the iptables tool, part of the netfilter framework, to redirect the input/output traffic according to the corresponding source and destination address of the test nodes. As an example, the following command would configure Host B in Figure 5 so that all the UDP packets received from Host A via the Bluetooth interface are sent back to it through the Ethernet interface. iptables – t nat – A PREROUTING – d 10.0.0.254 – p udp – j DNAT – to – destination 10.0.3.1 4.1 Power-delay trade-off analysis We first evaluate the effect that the use of the sniff mode has over the power consumption and the packet delay when the data traffic consists of UDP connections. Each UDP connection generates 1 packet/second, with a packet size of 1000 bytes. In our experiments we vary the packet type and the distance between nodes. 4.1.1 Power consumption evaluation Figures 6 and 7 allow comparing the current consumption in active and sniff modes during a data packet transmission, and using a DM1 type of packet. The configuration of the sniff mode depends on the application’s requirements, and the definition of the best Tsniff represents a compromise between power consumption and delay. The advantage of having a large Tsniff is low consumption (see Figure 4), and the inconvenience is high delay when sending data packets. In our experiments we select a relatively small Tsniff to avoid increasing the packet delay, and we fixed Twin to only one slot in order to reduce the power consumption. 35 35 Packet transmission (DM1) Packet transmission (DM1) 30 Consumption (mA) Consumption (mA) 30 25 20 15 10 5 25 20 15 10 0 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 Time (s) (a) Master in active mode 0.8 0.9 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 1 Time (s) (b) Master in sniff mode FIGURE 6 Current consumption (mA) of a packet transmission in active (a) and sniff mode (b) on a master device using a DM1 packet type. Distance between master and slave is of 6m. (Tsniff = 20 slots, Twin = 1 slot, Ttimeout = 1). “aswin123” — 2009/5/14 — 14:21 — page 151 — #11 152 J.-C. Cano et al. 35 Packet transmission (DM1) 35 Consumption (mA) Consumption (mA) 25 20 15 10 25 20 15 10 5 5 0 Packet transmission (DM1) 30 30 0 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 Time (s) (a) Slave in active mode 0.8 0.9 1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 Time (s) (b) Slave in sniff mode FIGURE 7 Current consumption (mA) during a packet transmission in active (a) and sniff mode (b) on a slave device using a DM1 packet type. Distance between master and slave is of 6 m. (Tsniff = 20 slots, Twin = 1 slot, Ttimeout = 1). Figures 6 and 7 highlight the bursty behavior of the sniff mode. As expected, the Bluetooth interface has periodic peaks every Tsniff slot pairs, and the baseline value is similar to the standby mode, which is lower that the active master and active slave ones. If we compare the active mode operation between master and slave we confirm that the consumption at the slave is significantly higher than the one at the master (just above 20 mA). This effect is due to the continuous listening activity a slave is required to perform. When using the sniff mode, the slave can reduce the current consumption with respect to the master’s. We repeated all the tests performed by varying the packet type from DM (5, 3, 1) to DH (5, 3, 1). When selecting multi-slot packets, Bluetooth reduces the time we need to send each data packet, thereby reducing the current consumption. We also confirmed that distance does not affect the power consumption of our Bluetooth devices (power regulation is off). Figure 8 shows the current consumption in sniff mode during a data packet transmission using a more efficient DH5 multi-slot packet, fixing the distance between master and slave to about 9 meters. 4.1.2 Packet delay evaluation We now evaluate the impact on packet delay of varying the ACL packet type and the distance between devices. For each test we obtain the average packet delay of 100 independent data packets. Figure 9 shows the results obtained when the master node sends data to the slave using either DHx or DM x packets, using a similar scenario that the one showed in Figure 5. We observe that, independently of the packet type chosen, Bluetooth offers a relatively stable packet delay up to 10 meters. When surpassing the 10 m limit Bluetooth still works without a sharp performance reduction, but when arriving at the border of 15 meters, performance degradation starts to be noticeable. “aswin123” — 2009/5/14 — 14:21 — page 152 — #12 153 Performance in Bluetooth 35 Consumption (mA) Packet transmission (DH5) 30 25 20 15 10 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 Time (s) FIGURE 8 Current consumption (mA) during a packet transmission in sniff mode on a master device using a DH5 packet type. Spatial distance between master and slave is 9 m. (Tsniff = 20 slots, Twin = 1 slot, Ttimeout = 1). 200 Data Hi gh Rate (DH5 ) 175 Data Hi gh Rate (DH3 ) Data Hi gh Rate (DH1 ) End to end delay (ms) 150 125 Data Hi g (DH5 ) Data Hi g (DH3 ) Data Hi g (DH1 ) 100 75 50 25 0 0m 3m 6m 9m 12 m 15 m 12 m 15 m Distance (m) 200 Data Medium Rate (DM5) Data Medium Rate (DM3) Data Medium Rate (DM1) (DM5) (DM3) (DM1) 175 End to end delay (ms) 150 125 100 75 50 25 0 0m 3m 6m 9m Distance (m) FIGURE 9 Comparison of the UDP packet delay in active and sniff mode under different DHx and DMx packets with respect to distance. “aswin123” — 2009/5/14 — 14:21 — page 153 — #13 154 J.-C. Cano et al. 30 100% 25 Histo gram delay ( 20 ) Cumulati ve delay (DH1 ) Cumulati ve delay ( ) 60% 15 40% 10 Cumulati ve distribution function Histogram frequency (# Tests) 80% Histo gram delay (DH1 ) 20% 5 0 0% 0 25 50 75 100 1 25 150 1 75 Time (ms) FIGURE 10 Comparison of the cumulative UDP packet delay in active and sniff mode for 100 data packets. Packet type is DH 1 . Distance between devices is of 6m. As expected, results confirm that the use of DHx packets increases efficiency. Moreover, when using the multi-slot DH 3 or DH 5 packets, the selected sniff mode, i.e., Tsniff = 20 slots, Twin = 1 slot, and Ttimeout = 1, allows to reduce the power consumption without significantly increasing the delay. However, when we select the sniff mode and use either DH 1 or DM 1 data packets, the observed delay increases significantly (44% and 52%, respectively). According to the obtained results we conclude that the use of the sniff mode could be quite compatible with the more efficient multi slot data packets. However, when single-slot data packets must be used, the sniff mode does have a significant impact on performance, and so the power/delay trade-off should be considered. Figure 10 shows the histogram distribution of the packet delay as a function of time for the 100 tests, at a distance of 6 m. The results show the impact that the use of the sniff mode has on the packet delay when using single slot DH 1 packets. Besides we find that the median is of 62 ms when the sniff mode is not used, and it grows to 85 ms when the sniff mode is activated. Moreover, we achieve the 95% percentile with 70 ms and 120 ms when using or not the sniff mode, respectively. Finally, when the slave node acts as the sender and the master node acts as the receiver, packet delay increases around 10% for all the different data packet types. When the master node acts as the sender, it can reduce the average packet delay since it knows the exact time when a packet transmission is going to take place. 4.2 Power-throughput trade-off analysis We now evaluate the impact that the use of the sniff mode has on throughput and power consumption when the data traffic consists of TCP connections. We “aswin123” — 2009/5/14 — 14:21 — page 154 — #14 155 Performance in Bluetooth configured the system to continuously download a data block of 512 kbytes varying the packet type and the spatial distance among the master and the slaves. We configured each TCP connection to generate data packet of 1500 bytes including the TCP and IP headers. 4.2.1 Power consumption evaluation Figure 11 compares the current consumption in active and sniff modes for a slave Bluetooth device sending data using different packet types. 6 TCP connection varying data typ e DM1 40 DM3 DM5 DH3 DH1 DH5 35 Active, no transmission Consumption (mA) 30 25 20 15 10 5 0 0 20 40 60 80 100 120 140 Time (s) (a) Slave in active mode DM3 6 TCP connection varying data typ e 40 DM5 DH3 DH5 DH1 DM1 Consumption (mA) 35 30 25 20 Active, no transmission 15 10 5 0 0 20 40 60 80 100 120 140 160 Time (s) (b) Slave in sniff mode FIGURE 11 Current consumption (mA) for 6 consecutive TCP connections in active (a) and sniff mode (b) on a slave device varying the packet type. The data block is of 512 kbytes and the distance between master and slave is of 6 m. (Tsniff = 20 slots, Twin = 1 slot, Ttimeout = 1). “aswin123” — 2009/5/14 — 14:21 — page 155 — #15 156 J.-C. Cano et al. Firstly, we observe that by using multi-slot Bluetooth packets there is a slight increase of the average current consumption with respect to the use of the DM 1 or DH 1 data packets. On the other hand, their use allows to drastically reduce the transfer time, thereby reducing the average current consumption. We also observe that by using the sniff mode the reduction on average current consumption is limited to 5% with respect to the active mode. We can further reduce current consumption by increasing the Tsniff parameter of the sniff mode; however, the observed throughput will also decrease. We now evaluate the impact of the Ttimeout parameter. Figure 12 shows a portion of a TCP connection when using either Ttimeout = 0 or Ttimeout = 1. To 40 35 Consumption (mA) 30 TCP connection in sni Timeout = 0 de, 25 20 15 10 5 0 0 0.5 1 1.5 2 Time (s) 40 35 Consumption(mA) 30 TCP connection in sni Timeout = 1 25 de, 20 15 10 5 0 0 0.5 1 1.5 2 Time (s) FIGURE 12 Comparison of the current consumption (mA) when varying the Ttimeout parameter. (Tsniff = 100 slots, Twin = 20 slot. “aswin123” — 2009/5/14 — 14:21 — page 156 — #16 157 Performance in Bluetooth Master to slave, active Master to slave, sniff Slave to Master, active Slave to master, sniff DM 1 DM 3 DM 5 DH 1 DH 3 DH 5 98.4 kb/s 333.6 kb/s 410.4 kb/s 160.8 kb/s 460.0 kb/s 564.0 kb/s 80.8 kb/s 278.4 kb/s 283.2 kb/s 129.6 kb/s 401.6 kb/s 424.0 kb/s 100.0 kb/s 336.0 kb/s 416.8 kb/s 154.4 kb/s 486.4 kb/s 577.6 kb/s 80.0 kb/s 272.0 kb/s 382.4 kb/s 128.0 kb/s 392.0 kb/s 520.0 kb/s TABLE 2 Comparison of the TCP throughput in active and sniff modes under different DHx and DMx . Spatial distance between master and slave is 6m and Ttimeout = 1 better appreciate the difference, in this test we have fixed the Tsniff to 100 slots, and the Twin to 20 slots. We observe that, when setting Ttimeout = 1, a node with enough data to be transmitted can be active during several consecutive Tsniff slots. As observed, the current consumption increases but the observed performance will be highly improved. 4.2.2 Throughput evaluation We now evaluate the impact that the use of the sniff mode has on TCP throughput with varying packet types. We evaluated the differences of having the master or the slave node acting as the data source. Table 2 shows the obtained results when the distance between the master and the slaves is of 6 meters. When we select the more efficient DH data packets, throughput always increases with respect to using DM packets. In active mode, DH packets outperform DM ones by 60%, 45% and 35% for 1, 3, and 5 slot packets, respectively. When the use of the sniff mode is selected the behavior is quite similar, though the throughput decreases by about 20%. With respect to throughput, there are no noticeable differences between the two selected case studies, i.e., the master or the slave acting as the data source. Finally, we observed that all the results obtained at this distance are below the maximum throughput stated in Bluetooth specification. The Bluetooth standard provides the following reference values: 723.2 kbit/s for DH5, 585.6 kbit/s, for DH3 and 172.8 kbit/s for DH1; 477.8 kbit/s for DM5, 387.2 kbit/s, for DM3 and 108.8 kbit/s for DM1. A master in active mode gets a 78%, 79%, and 93% of the maximum throughput provided in the standard for the DH 5,3,1 packets respectively. When using the more conservative DM 5,3,1 packets, these percentages are of 85%, 86%, and 90%. Such results confirm that ideal conditions can not always be achieved due to distance and noise; hence, bandwidth limitations and fluctuation should be considered. “aswin123” — 2009/5/14 — 14:21 — page 157 — #17 158 J.-C. Cano et al. 5 CONCLUSIONS In this paper we investigate the power characteristics of the Bluetooth technology when supporting low-power modes. We provide accurate current consumption measurements for different operating modes of the Bluetooth technology, and provide useful information to drive technical decisions on battery type for Bluetooth-based end systems. We find that the low-power modes stated in the standard could significantly alleviate the power consumption of Bluetooth nodes in active mode. We found that the hold mode could be extremely useful to reduce the power consumption for those applications using a predictable scheduling. Applications with a high number of peers and using data traffic patterns with a regular scheduling could benefit from using parked slaves. Of special interest is the sniff mode, which can offer a trade-off between low power consumption and low transmission delay. We have shown that the definition of the best Tsniff parameter for the sniff mode represents a compromise between power consumption and delay. The advantage of having a large Tsniff is low consumption, and the inconvenience is high delay and low throughput when sending data packets. Our study includes a performance evaluation of the Bluetooth technology to characterize the trade-off between power consumption and performance. We evaluated the impact that the use of the sniff mode has on network performance. Our experiments focused on evaluating the power-delay and the power-throughput trade-off experimented by UDP and TCP traffic, respectively. We also performed a sensitivity analysis to evaluate how distance and packet type affect power consumption, throughput and delay. We observed that Bluetooth offers a relatively steady packet delay and throughput up to 10 meters, independently of the selected packet type. We also observed that the use of the sniff mode could be quite compatible with the more efficient multi-slot data packets. However, when the channel conditions require selecting single-slot data packet, the sniff mode could impact performance and so the power/delay trade-off should be considered. With respect to throughput, we observed that a master node in active mode gets a 78%, 79%, and 93% of the maximum throughput provided in the standard for DH 5,3,1 packets, respectively. When activating the sniff mode, the behavior is rather quite similar with respect to power consumption, though throughput decreases by about 20%. Finally, our experiments also confirm that delay and throughput do not increase significantly with distance. Overall, the results obtained provide useful insight to determine whether or not to activate the sniff mode according to the selected packet type and the application’s requirements. As future work we would like to obtain some clear relationships among performance and the usage of Bluetooth low power modes, evaluating the impact that the size of the piconet has over the trade-off between performance and power consumption. “aswin123” — 2009/5/14 — 14:21 — page 158 — #18 Performance in Bluetooth 159 ACKNOWLEDGMENTS This work was partially supported by the Ministerio de Educación y Ciencia, Spain, under Grants TIN2008-06441-C02-01, and TEC2006-12211-C02-01. REFERENCES [1] Mark Weiser, The computer for the 21st century. Scientific American 256 (1991), 94–104. [2] Promoter Members of Bluetooth SIG, Specification of the Bluetooth System – Core. Version 1.1, Bluetooth SIG, Inc., February 2001. [3] O. Kasten and M. Langheinrich, First experiences with bluetooth in the smart-its distributed sensor network, in Workshop on Ubiquitous Computing and Communications, October 2001. [4] A. Zanella, D. Miorandi and S. Pupolin, Mathematical analysis of bluetooth energy efficiency, in Proceedings of the International Symposium on Wireless Personal Multimedia Communications, October 2003. [5] R. L. Ashok, R. Duggirala and D. P. Agrawal, Energy efficient bridge management policies for inter-piconet communication in bluetooth scatternets, in Vehicular Technology Conference, 2003. [6] H. Zhu, G. Cao, G. Kesidis and C. Das, An adaptive power-conserving service discipline for bluetooth, in IEEE International Conference on Communication. Volume 1, 2002. [7] J. C. Cano, J. M. Cano, E. Gonzlez, C. Calafate and P. Manzoni, Power characterization of a bluetooth-based wireless node for ubiquitous computing, in International Conference on Wireless and Mobile Communications, July 2006. [8] IEEE 802.15 Working Group for WPANs, IEEE 802.15 web site, http://ieee802.org/15/ index.html, May 2001. [9] Mitsumi Electric, Bluetooth Module WML-C11 Class 1, http://www.mitsumi.co.jp/english/. [10] Texas Instruments Inc, The INA193 Current Shunt Monitor (Rev. D), http://focus.ti.com/docs/ prod/folders/print/ina193.html. [11] Analog Devices, MicroConvertor Technical Note uC003: The AduC812 MicroConverter as an IEEE 1451.2 Compatible Smart Transducer Interface, Version 1.0, Analog Devices, 1999. [12] Maxim Krasnyansky, BlueZ: Official linux bluetooth protocol stack, http://bluez.sourceforge. net/, 2003. “aswin123” — 2009/5/14 — 14:21 — page 159 — #19 “aswin123” — 2009/5/14 — 14:21 — page 160 — #20