Investigating MAC-layer Schemes to Promote Doze Mode in 802.11-based WLANs V. Baiamonte and C.-F. Chiasserini CERCOM - Dipartimento di Elettronica Politecnico di Torino Torino, Italy Email: baiamonte,chiasserini @polito.it Abstract This paper addresses the energy saving issue in 802.11-based WLANs. Previous work has pointed out that the power management function in the IEEE 802.11 standard presents some inefficiencies, thus other solutions to energy saving are needed. We focus on an infrastructure WLAN and explore the possibility to increase the time period that a wireless station spends in the low-power operational state, the so-called doze state. We present an energy-saving, MAC-layer scheme which is derived from the 802.11 DCF. The proposed mechanism enables a station to enter the doze state during channel contention by exploiting the virtual carrier sense mechanism and the backoff function. We compare the performance obtained through the proposed scheme with the results attained through both the standard DCF and a simple active/doze state switching mechanism. I. INTRODUCTION Today Wireless Local Area Networks (WLANs) based on the IEEE 802.11 standard [1] are widely deployed, since they both support user mobility and provide high data rates. Nevertheless, current 802.11 technology still has some limitations, such as the lack of quality of service support, the unfairness of the Medium Access Control (MAC) function and the high energy consumption experienced by battery-powered wireless stations (WSTAs) [2], [3]. In this work we focus on the energy efficiency issue, which is critical to the performance of the WLAN technology. Indeed, typical values for the power consumption of a 802.11 WLAN card are as follows: =1.65 W in transmit mode, =1.4 W in receive mode, =1.15 W in idle mode (i.e., when the channel is sensed without actively receiving), and =0.045 W in doze mode (i.e., when the RF circuitry is turned off) [3], [4]. These values of power consumption suggest that, in order to save energy, we need to reduce the time spent by a This work was partially funded by MIUR (Italian Ministry of Research) through the PRIMO project. WSTA in idle mode and increase as much as possible its time in doze mode. The IEEE 802.11 standard specifies a Power Management (PM) function, which allows a WSTA to switch to the doze mode when the interface is idle [1]. Several studies however show that the 802.11 PM scheme presents some inefficiencies and propose solutions to overcome the 802.11 PM limitations. In particular, in [3] the authors observe the 802.11 PM inefficiency due to the fixed beacon interval duration and propose a scheme controlling the transition to the doze state in the middle of a beacon interval. Solutions to energy saving in 802.11-based WLANs have also been proposed in [5], [6]. In this work, we devise a novel scheme at the MAC layer, called the drowsy scheme, which allows WSTAs to enter the doze state while participating in the network activity. By applying the drowsy technique, a WSTA greatly reduces its channel-sensing activity thus saving energy. However there exists a trade-off between energy saving and traffic delivery delay. We investigate this trade-off and compare the performance of our scheme with the results obtained through the 802.11 standard Distributed Coordination Function (DCF) as well as through a simple active/doze switching mechanism. The study is performed by using the network simulator ns [7]. We highlight that the drowsy technique can be used either as an alternative to the 802.11 PM function or jointly with the standard PM function. The remainder of the paper is organized as follows. In Section II we describe the network scenario under study. Section III first describes the proposed energy-efficient technique, then it presents a simple active/doze state switching mechanism that we consider for comparison to our scheme. In Section IV we show some performance results comparing the drowsy technique to the standard DCF function and to the simple active/doze switching mechanism. Section V concludes the paper and points at some aspects that will be the subject of future research.
II. REFERENCE SCENARIO We consider an 802.11b-based WLAN, which includes an Access Point (AP) and several stationary WSTAs using a bit rate over the wireless channel of 11 Mbit/s. When no modifications are introduced in the access scheme, the WSTAs employ the IEEE 802.11 DCF [1]. The DCF operates as follows. It exploits both a physical and a virtual channel sensing; virtual sensing is implemented by including in all transmitted frames an indication of their duration so that the non-destination WSTAs overhearing a transmission can set their NAV accordingly. Once a WSTA has set its NAV, it is in idle state [2]. When a WSTA has to transmit a frame, the physical and virtual carrier sense mechanisms are checked. If within an interval of DIFS (or EIFS if the previous frame was received in error) either the physical or virtual carrier sense mechanisms detect the channel as busy, the WSTA selects a backoff interval, that is decremented only during idle channel periods. Again, a WSTA can be considered as in idle state during the backoff time. With regard to the wireless channel behavior, we consider an independent error model for each communicating pair of nodes. The error model is represented by a three-state discrete-time Markov chain. The Markov chain time slot is equal to the 802.11b time slot duration. Errors over the channel occur in the states long bad (LB) and short bad (SB), while the good (G) state is error-free. Thus, a frame transmission is successful only if the error model is in state G for all slots it takes the frame to be transmitted, while it fails otherwise. The difference between the long bad and short bad states is the time correlation of errors: corresponds to long bursts of errors, SB to short ones. The probability that the Markov chain moves to the state given that it leaves the state, i.e., the probability that an error burst is long, is set to 0.05. We assume that the average time duration of a burst of bad slots experienced when the states and are entered, are respectively equal to 1.3 s and 0.04 s. The average number of consecutive error-free slots is set to 1 s. III. ENERGY-EFFICIENT TECHNIQUES In this section, we first introduce the proposed energyefficient technique, called the drowsy scheme; then we briefly describe a simple active/doze state switching scheme that will be used as term of comparison in the performance study in Section IV. A. The Scheme Our aim is to reduce the time spent by a WSTA in idle state and enable the WSTA to enter the doze state Other WSTAs Backoff (( """"" NAV &&&& ''' $$$$$$$ %%%%%% (( ))!!!!!!!""""" ##### WSTA &&&& ''' $$$$$$$ %%%%%% ))!!!!!!!##### NAV NAV DIFS WSTA Fig. 1. NAV Backoff Data Sensing and backoff decreasing (idle state) Idle state Dozing Backoff freezing (idle state) Modified backoff counter management in the drowsy scheme instead. To widen the time of inactivity of a WSTA so as to be able to move into the doze state and save energy, we act on two different aspects: (i) we exploit the virtual carrier sense mechanism, and (ii) we change the backoff management technique. First, consider that the MAC layer of a WSTA receives data to transmit from the upper layers and starts listening to the channel while a frame transmission is taking place. Assume that the WSTA can read the frame duration field in the frame header and set its NAV. According to the drowsy scheme, the WSTA also computes a backoff interval and sets its backoff counter to this value. While in the 802.11 standard the backoff is decremented only during idle channel periods, the drowsy WSTA continuously decreases its backoff counter with time. This implies that the WSTA does not need to sense the channel while being in backoff but it can be in doze state. When the backoff counter reaches zero, the drowsy WSTA behaves following the DCF scheme as specified in the IEEE 802.11 standard. The drowsy WSTA mechanism is presented and compared with the 802.11 DCF scheme in Figure 1. Note that, if the WSTA cannot read the frame duration field in the frame header and, hence, cannot set its NAV, it waits a time interval equal to EIFS as in the 802.11 DCF. Then, after the WSTA computes its backoff time, it applies the drowsy backoff management technique described above. We remark that, since the drowsy WSTA decrements the backoff counter continuously, the Contention Window (CW) should be enlarged in order to extend the backoff interval. B. A Active/Doze Switching Mechanism We consider a simple and ideal mechanism which controls the WSTA transition from the active state (where the WSTA can transmit/receive) to the doze state
Fig. 2. MAX_WAITING_TIME Doze Idle DOZE_TIME TX/RX Start My_RX/TX TX/RX End TX/RX State machine for the simple active/doze switching scheme TABLE I 802.11 PARAMETER SETTING CW CW RTS Threshold 400 bytes Slot time SIFS time DIFS time EIFS Short Retry Limit 7 Long Retry Limit 4 Preamble Length 144 bits PCLP length 48 bits ACK Frame Length 112 bits and vice-versa. When this scheme is implemented, a WSTA operates according to the state machine shown in Figure 2. Consider that the WSTA is in idle state. If the WSTA remains idle for a time period equal to MAX WAITING TIME, then it moves into the doze state. The maximum time spent by the WSTA in doze state is equal to DOZE TIME, afterwards the WSTA switches back to the idle state. However, if the WSTA needs to wake up to receive or transmit a frame while being in doze state, it moves into the TX/RX state. As a frame transmission/reception ends, the WSTA moves into the idle state. Clearly, this mechanism assumes an ideal behavior of WSTAs since WSTAs should be able to switch from the doze state to the TX/RX state as they are about to receive a frame. In actual implementations, this is unfeasible without using a low-power circuit, which allows us to wake up WSTAs when needed. IV. NUMERICAL RESULTS We derive the system performance via simulation, by extending the network simulator ns [7]. We assume that each WSTA has one uplink traffic connection with the AP. Indeed, considering uplink traffic only eases the implementation of the active/doze switching mechanism described in Section III-B, without affecting the performance of the drowsy scheme. The traffic offered to the network by each WSTA is generated by UDP flows exhibiting an on-off behavior. During off periods no traffic is generated; the average duration of the off period, denoted by off, is taken as a constant parameter and is set to 6 s. On the contrary, during on periods, the WSTA generates traffic at a constant rate of 256 kbit/s; the average duration of the on period is a configurable parameter that we vary in our simulations. Thus, we can act on the system load by tuning the value of the on period as well as the number of WSTAs in the network. The wireless channel is modeled as described in Section II, while the 802.11 parameter setting that we adopt is presented in Table I. As for the power consumption in the WSTA operational states, we consider the following values: =1.65 W, =1.4 W, =1.15 W, and =0.045 W. The plots that are shown in the following present the average energy consumption per successful packet (i.e., successfully delivered packet to the AP), and the average total packet delay that is computed as the sum of queueing delay and service delay at the MAC layer. The results obtained through the drowsy scheme are compared to the performance of the 802.11 DCF (tagged standard ) and of the simple active/doze switching mechanism (tagged simple ). In the case of the simple active/doze switching mechanism, we set MAX WAITING TIME and DOZE TIME to 2 and 5 s, respectively. In the case of the drowsy scheme, we present results for two different CW settings: CW =31 and CW =1023 (tagged drowsy ), and CW =63 and CW =2047 (tagged drowsy large CW ). Figure 3 presents the energy consumption per successful packet as a function of the ratio on off, i.e., of the total offered traffic load, when the number of WSTAs is equal to 10. Note that under the assumptions introduced above on off =0.25 corresponds to a total traffic generation rate equal to 512 kbit/s, while on off =2.5 corresponds to a total generation rate of 1.82 Mbit/s. Figure 3 shows that the drowsy scheme outperforms both the standard and the simple mechanism, regardless of the CW size and of the value of traffic load. As for the simple scheme, one would expect a significant improvement in energy consumption with respect to the standard case. Instead, the difference between the simple and the standard scheme behavior is negligible. To let such a difference become evident, smaller values of traffic load should be considered. As a last remark, we observe that the energy consumption per successful
Energy per successful packet (J) 8 6 4 2 0.08 0.06 large CW Energy per successful packet (J) 5 4 3 2 1 0.09 0.08 large CW 0.04 0.07 0.02 0 0.5 1 1.5 2 2.5 Ton/Toff Fig. 3. Average energy consumption per successful packet versus the ratio on off, when off s and the number of WSTAs in the system is equal to 10. The performance of the drowsy scheme with different CW size, of the standard function, and of the simple active/doze switching mechanism are compared 0.06 4 6 8 10 12 14 16 18 Wireless Stations Fig. 5. Average energy consumption per successful packet versus the number of wireless stations, for on s and off s. The performance of the drowsy scheme with different CW size, of the standard function, and of the simple active/doze switching mechanism are compared 0.5 0.4 large CW 10 1 large CW Average total delay (s) 0.3 0.2 Average total delay (s) 10 0 10 1 10 2 0 0 0.5 1 1.5 2 2.5 Ton/Toff Fig. 4. Average total packet delay as a function of the ratio on off, when off s and the number of WSTAs in the system is equal to 10. The performance of the drowsy scheme with different CW size, of the standard function, and of the simple active/doze switching mechanism are compared 10 3 4 6 8 10 12 14 16 18 Wireless Stations Fig. 6. Average total packet delay as a function of the number of wireless stations, for on s and off s. The performance of the drowsy scheme with different CW size, of the standard function, and of the simple active/doze switching mechanism are compared packet decreases as the ratio on off increases. This is because as the on period increases, the time spent in idle state by WSTAs becomes shorter. This suggests that the contribution of the idle state to energy consumption is still quite significant in all of the considered schemes, and that there is still some margin of improvement. Figure 4 shows the average total packet delay as a function of the ratio on off, in the case of 10 WSTAs. As expected, our scheme implies larger packet delay, especially in the case of large values of on off. Looking at Figures 3 and 4, we can see the energy/delay trade-off that can be achieved through the drowsy mechanism. Figures 5 and 6 present the average energy consumption per successful packet and the average total packet delay, respectively, as the number of WSTAs in the system varies. Results are derived by fixing on at 3 s. Figure 5 shows that the drowsy mechanism with a larger CW always gives better performances than both the standard and the simple technique. Whereas, the performance of the drowsy scheme with a smaller CW degrades as the number of WSTAs exceeds 13 because of the high number of collisions. As shown in Figure 6,
the average packet delay grows with the increase in the number of contending WSTAs, and, as expected, the drowsy mechanism behaves worse than the standard and the simple scheme. Clearly, such a degradation becomes more remarkable as the number of WSTAs grows. These results suggest that the drowsy scheme gives excellent performance in terms of energy consumption, even under high traffic load conditions. On the contrary, the difference between the performance of the simple mechanism and of the standard scheme is negligible unless a very low traffic load is considered. However, in the drowsy scheme the remarkable energy saving is obtained at the cost of a larger packet delay. Furthermore, in the case of the drowsy scheme we need to enlarge the CW as the number of WSTAs increases in order to keep the collision probability small. [6] L. Bononi, M. Conti, L. Donatiello, A Distributed Mechanism for Power Saving in IEEE 802.11 Wireless LANs, ACM Mobile Networks and Applications (MONET), vol. 6, no. 3, pp. 211-22, 2001. [7] UCB/LBNL/VINT, Network Simulator ns version 2.26, URL: http://www.isi.edu/nsnam/ns, 2002. V. CONCLUSIONS AND FUTURE WORK This paper addressed the issue of energy efficiency in WLANs. We proposed a novel energy-saving, MAClayer scheme which is derived from the DCF specified in the IEEE 802.11 standard. The proposed scheme enables a wireless station to enter the low-power operational state, the so-called doze state, while the NAV is set as well as during the backoff time interval. We investigated through simulation the trade-off existing between energy saving and traffic delivery delay, and compared the performance of the proposed scheme with those obtained through both the 802.11 standard DCF and a simple active/doze state switching mechanism. Further improvements of the proposed enery-saving scheme are currently under study. Using the proposed technique does not preclude using the power management function in the IEEE 802.11 standard. Rather, the two schemes could be jointly used. In our future work, we will study the system performance when the 802.11 power management function is implemented. REFERENCES [1] IEEE 802.11 WG, Part 11: Wireless LAN Medium Access Control (MAC) and Physical Layer (PHY) specifications, 1999. [2] L. M. Feeney, M. Nilsson, Investigating the energy consumption of a wireless network interface in an ad hoc networking environment, INFOCOM 2001, Anchorage, Alaska, 2001. [3] E.-S. Jung, N. H. Vaidya, An energy efficient MAC protocol for wireless LANs, INFOCOM 2002, New York, NY, 2002. [4] A. Kamerman, L. Monteban, WaveLAN-II: A High Performance Wireless LAN for the unlicensed band, Bell Labs Technical Journal, vol. 2, no. 3, 1997. [5] Y.-C. Tseng, C.-S. Hsu, T.-Y. Hsieh, Power-Saving Protocols for IEEE 802.11-Based Multi-Hop Ad Hoc Networks, Infocom 2002, New York, June 2002.