A Novel Idle Mode Operation in IEEE 802.11 WLANs - MWNL

Download PDF
2 downloads 0 Views 703KB Size Report
Comment
Sep 25, 2006 - broadcast transmission rate even when a mobile host is in the PSM. ..... user replies with a Call-Setup-Response after successful pag- ing.
A Novel Idle Mode Operation in IEEE 802.11 WLANs: Prototype Implementation and Empirical Evaluation Sunggeun Jin

Kwanghun Han

Sunghyun Choi

School of Electrical Engineering and INMC Seoul National University, Seoul, Korea {sgjin,khhan}@mwnl.snu.ac.kr, [email protected]

ABSTRACT

1.

IEEE 802.11 Wireless Local Area Network (WLAN) became a prevailing technology for the broadband wireless Internet access, and new applications such as Voice over WLAN (VoWLAN) are fast emerging today. For the battery-powered VoWLAN devices, the standby time extension is a key concern for the market acceptance while today’s 802.11 is not optimized for such an operation. In this paper, we propose a novel Idle Mode operation, which comprises paging, idle handoff, and delayed handoff. Under the idle mode operation, a Mobile Host (MH) does not need to perform a handoff within a predefined Paging Area (PA). Only when the MH enters a new PA, an idle handoff is performed with a minimum level of signaling. Due to the absence of such an idle mode operation, both IP paging and Power Saving Mode (PSM) have been considered the alternatives so far even though they are not efficient approaches. We implement our proposed scheme in order to prove the feasibility. The implemented prototype demonstrates that the proposed scheme outperforms the legacy alternatives with respect to energy consumption, thus extending the standby time.

Recently, IEEE 802.11 Wireless Local Area Network (WLAN) became a prevailing technology for the broadband wireless Internet access. Along with that, new types of applications such as Internet Protocol (IP) telephony over WLAN are fast emerging today. In order to support IP telephony, the functionality to inform a mobile host of incoming calls becomes indispensable even if the user holding an idle mobile host moves around. Along with such a functionality, considering the battery-powered IP phone devices, the standby time extension is a key concern for the market acceptance while today’s 802.11 is not optimized for such an application. The reason is rooted in the fact that the 802.11 WLAN Medium Access Control (MAC) protocol [3] defines only two operational modes, which a mobile host can operate in, namely, Active Mode (AM) and Power Saving Mode (PSM). In both modes, a mobile host always has to stay associated with one of the APs even when there is no traffic to/from the mobile host. This implies two critical problems. One problem is that a mobile host should receive all broadcast and multicast packets even if those packets are really useless since the mobile host cannot differentiate the packets destined to it until receiving those packets. The energy consumption increases in proportion to multicast and broadcast transmission rate even when a mobile host is in the PSM. Through surprising measurement results of a commercial network, as be discussed in Section 2.2, we find out that multicast and broadcast packet interval arrival times under 10 ms occupies 67 %. This represents a mobile host should wake up to manage broadcast and multicast packets more frequently than our expectation, and hence, consume even more energy correspondingly. The other problem is that it has to perform a handoff at every AP cell boundary in order to maintain associated with an AP. The inevitable handoffs, at every AP cell boundary, cause a mobile host to consume redundant energy. As the handoff frequency, approximately proportional to the mobile host’s speed, increases, the mobile host consumes more energy. Even worse, when IEEE 802.11i [11, 10] is employed for security enhancement, a larger amount of message exchanges during the handoff operation are expected, and these incur more energy consumption. That is, IEEE 802.11 WLAN is naturally lack of an efficient support of the mobile host mobility when there is no traffic to be served for the mobile

Categories and Subject Descriptors C.5.m [Computer System Implementation]: Miscellaneous; C.2.2 [Computer-Communication Networks]: Network Protocols—applications, protocol architecture, protocol verification, routing protocols

General Terms MEASUREMENT, PERFORMANCE

Keywords Idle Mode, IEEE 802.11 WLAN, energy management, mobile computing, low-power computing

Permission to make digital or hard copies of all or part of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for profit or commercial advantage and that copies bear this notice and the full citation on the first page. To copy otherwise, to republish, to post on servers or to redistribute to lists, requires prior specific permission and/or a fee. WMASH’06, September 25, 2006, Los Angeles, California, USA. Copyright 2006 ACM 1-59593-484-7/06/0009 ...$5.00.

INTRODUCTION

hosts. Therefore, it is desired to have a new Idle Mode (IM), different from the AM and the PSM, in the 802.11. Due to the lack of the IM operation, the combination of the IP paging and the PSM have been considered an alternative to the IM operation so far [1] though the original aim of IP paging is to facilitate integrating of different wireless technologies. IP paging is independent of Layer-2 (L2) technologies, nevertheless, since the IP paging is considered one of the alternatives to the IM, it provides insufficient way to overcome the lack of the IM. For example, since the 802.11 MAC cannot differentiate IP paging-related packets (which are of broadcast type) from other packets, the mobile host’s MAC has to receive all the broadcast packets, and forward them to the IP layer, thus consuming considerable energy. In order to overcome these problems, which cause redundant energy consumption, we propose an IM operation, comprising paging, idle handoff, and delayed handoff, which can be used when an IEEE 802.11 WLAN standard-based mobile host does not have traffic or on-going sessions. Under the proposed IM operation, the mobile host can stay in the doze state consuming very little energy for an extended period, and performs less operation than in the PSM. Our proposal can be easily deployed with the IEEE 802.11 WLAN standard. In our scheme, a mobile host does not perform any handoff within a predefined Paging Area (PA). The handoff with minimum operation, called idle handoff, is performed only when a mobile host leaves a PA. The paging provides a way to inform a mobile host in the IM of a new packet arrival. The IP-level handoff should be deferred until a paging success in order to reduce redundant operations, and hence it is referred to as delayed handoff. In this paper, we show that our proposal provides an efficient way to page VoWLAN devices without IP paging. In order to prove that the proposed IM is feasible, we implement the prototype for both mobile host and AP. Utilizing the prototype, we evaluate the energy saving efficiency and show that the IM extends the standby time of an mobile host more than the PSM practically even when a mobile host is stationary. Before we go further, we summarize the acronyms for this paper as shown in Table 1.

2. 2.1

PROBLEMS TO SUPPORT IDLE MODE IN IEEE 802.11 WLANS Limitation of IEEE 802.11 PSM

According to [3], a WNIC can be in either of awake and doze states at a given moment. In awake state, a WNIC can transmit, receive or sense the physical channel, and it actually continues to sense the channel unless it either transmits or receives a frame. On the other hand, in doze state, a WNIC is not able to transmit or receive, and consumes very little energy. How a WNIC switches between these two states is determined by the WNIC’s power management mode, i.e., AM and PSM. When a WNIC is in the AM, the WNIC always operates in the awake state. On the other hand, when a WNIC is in the PSM, the WNIC can change its state between the awake and doze states depending on the traffic pattern. However, the PSM as an alternative to the IM has the following limitations. Since a WNIC running in the PSM should stay associated with an AP, the handoff should be performed at every AP cell boundary in order to maintain its association. During a

Table 1: Acronym summary ACPI AM AP BIOS CPU EAP-TLS FA HA HD IAPP IM IP L2 L3 MAC MH MRA-AP PA PAID PCM PDA P-MIP PM bit PSM RTP SIP TIM UDP VoIP WNIC WLAN

Advanced Configuration and Energy Interface specification Active Mode Access Point Basic Input Output System Central Processing Unit Extensible Authentication Protocol Transport Layer Security Foreign Agent Home Agent Handheld Device Inter-Access Point Protocol Idle Mode Internet Protocol Layer-2 Layer-3 Media Access Control Mobile Host Most Recently Associated AP Paging Area Paging Area IDentification number Pulse Code Modulation Personal Digital Assistance Paging extension for Mobile IP Power saving Mode bit Power Saving Mode Realtime Transport Protocol Session Initiation Protocol Traffic Indication Map User Datagram Protocol Voice over IP Wireless Network Interface Card Wireless Local Area Network

handoff procedure, the WNIC has to stay in the AM since the handoff can be severely delayed otherwise. The requirement for a WNIC to keep associated, even during performing handoff, exert a significant influence on the energy consumption of the mobile host. For example, we consider an IEEE 802.11i-compliant mobile host performs a handoff as illustrated in Fig. 1 [11, 10]. As shown in the figure, it takes a mobile host a long time to finish a handoff procedure to fulfill the operation related to security and Inter-Access Point Protocol (IAPP)1 . Making an 802.11 WLAN more secure from malicious attacks is very important, and hence we expect that the 802.11i will become more and more popular in the future. However, as we have seen above, the cost for more secure communication is more signaling for the mobile host, especially, during the handoff. For an idle mobile host without any active session, such a heavy signaling is not desirable at all in the energy consumption’s perspective. In the above procedure, the removal of Re-association delay comprising IAPP procedure, Extensible Authentication Protocol-Transport Layer Security (EAP-TLS) authentication, and handshake could contribute to a significant reduction of handoff processing time, thus reducing the energy consumption required for the operation. As to layer-3 (L3) handoff, it takes several seconds to perform an L3 handoff [20]. Note that mobility is harmful to power saving schemes since more energy is consumed as the 1

The IAPP, specified by IEEE 802.11f [5], is a protocol defining the communication among APs involved with a mobile host’s handoff.

Table 2: Statistics of multicast and broadcast packet inter-arrival time 0∼10 ms 67 %

Figure 1: IEEE 802.11i-based handoff operation frequency of handoffs increases. Moreover, inevitable handoffs cause additional problems such as signaling cost increase and session dropping. The purpose of paging is to locate an idle mobile host when an incoming call destined to it arrives. However, if an IP-based wireless access network does not offer a paging scheme, IP paging can be used as the substitutable scheme of the original access network level paging [1]. In the case of IEEE 802.11 WLANs, utilizing both the PSM and IP paging has been considered. However, since the PSM is developed without considering IP paging, the use of IP paging along with the PSM is an inefficient approach. In this section, we analyze the reason why the combined PSM and IP paging are not suitable as an alternative to the IM for IEEE 802.11 WLAN. An mobile host consists of Wireless Network Interface Card (WNIC) and Handheld Device (HD) such as Personal Digital Assistance (PDA) or smart phone. The WNICs implemented based on the IEEE 802.11 WLAN standard have a MAC layer while the IP layer as a part of Operating System (OS) is embedded in HDs. Therefore, in the perspective of functionality, WNIC plays a role of the MAC. For this reason, in this paper, we use the term of WNIC instead of station (STA) used in IEEE 802.11 WLAN standard. Similarly, we use the term of HD instead of IP layer.

2.2

Limitation of IP Paging

The original IP paging is targeted at the mobile hosts which do not have an on-going IP session. With the aid of IP paging, the network load and signaling cost to manage the mobility for the idle mobile hosts are reduced. However, most of the IP paging protocols [7, 8] are designed without

10∼100 ms 27 %

100ms∼1 s 6%

≥1 s 0%

considering the underlying MAC operation. This implies that the IP paging and MAC-specific paging operate independently. We can easily assume that a particular wireless access network has its own paging algorithm and IP paging is also adopted by the wireless access network. In such a case, the IP paging protocol actually generates redundant overheads since both MAC and IP provide the same functionality, i.e., paging. However, due to the lack of a paging scheme in IEEE 802.11 MAC, the IP paging would be a useful alternative to the MAC-level paging. However, the IP paging scheme as an alternative to the MAC-level paging in IEEE 802.11 WLANs has the following limitations. Since the IM is not defined in [3], a mobile host should be in the PSM instead of the IM, and has to get associated with an AP all the time. In order to support the IP paging, the mobile host, being associated with an AP, periodically listens to the broadcast paging packets for signaling, and performs IP paging-related operations. Despite the fact that the energy consumption should have been a major concern, the research efforts related to IP paging made so far have been limited only to the signaling cost reduction. Additionally, the IP layer, located right above the 802.11 WLAN MAC, receives all types of IP packets through the MAC. Such a tightly coupled relationship causes unexpected side effect to the PSM operation. For instance, if we assume that P-MIP is adopted to IEEE 802.11 WLAN, the routers or MIP agents broadcast advertisement messages periodically when some of mobile hosts are in the IM defined by [7]. Upon receiving the advertisement messages, the mobile hosts in the IM are forced to wake up and receive the advertisement messages to perform P-MIP-related operations. It is impossible for a WNIC to differentiate PMIP control packets from other broadcast packets. Therefore, the WNIC has to receive all broadcast packets, and after that, it is possible for the P-MIP layer to pick P-MIP control packets out of the received packets. From such a mechanism, one can easily imagine that a mobile host with both WNIC and HD having P-MIP consumes more energy due to the unnecessary broadcast packets. In order to verify the reasoning, we measure the interarrival time of broadcast and multicast packets for 24 hours in a large-scale commercial WLAN, called NESPOT [13], operated by Korea Telecom in Korea. As shown in Table 2, surprisingly the inter-arrival times under 10 ms represent the major portion. The average packet inter-arrival time is 22.7845 ms. This means that there will be 4 to 5 broadcast/multicast packets every beacon interval (assuming the typical 100 ms beacon interval) in average, and hence the WNIC has to wake up very often, e.g., every beacon interval, to receive these packets. Therefore, any mobile hosts adopting IP paging are compelled to consume their energy in vain in order to receive unnecessary packets. One can imagine IP paging message to wake up a mobile host in the IM by transmitting unicast IP packets. By doing so, the broadcast messages are not needed so that the overhead for the discrimination would decrease. However, it is not actually possible since the exact location of an idle

mobile host in terms of the associated AP is not known to the network in advance.

3.

PROPOSED IDLE MODE OPERATION

In order to overcome the problems discussed in the previous section, we define a new mode, i.e., Idle Mode (IM), for the 802.11 WLAN. We attempt to minimize required operations for the IM in order to minimize the energy consumption. When a WNIC is in the IM, it performs only essential operations to wake up in the future. The necessary operations for the IM are defined as follows: 1. A handoff does not occur at every cell boundary unlike a WNIC in the PSM. A handoff, called idle handoff, is performed only when a mobile host leaves a PA to enter another PA. 2. When a WNIC is in the IM, the WNIC is not associated with any AP. The only thing the WNIC in the IM has to do is to listen to the beacons periodically at every predefined interval in order to switch itself to the AM when a packet destined to itself arrives. The typical beacon listening interval for receiving beacons to wake up is set to be 1 second, while beacons are transmitted by APs every 100 ms typically. 3. Only a successful paging makes a WNIC in the IM enter the AM. The detailed idle mode operation is presented in the following.

3.1

Protocols for Idle Mode

Fig. 2 illustrates the PA structure for paging. A number of neighboring AP cells are grouped into a PA. The APs belonging to different routers can also be grouped into the same PA. The APs in the same PA have the same identifier, which is broadcast through the beacons via a newly-defined Paging Area Identifier (PAID) field. Each WNIC in the IM can differentiate a PA from the received PAID. We define a new procedure in order to support the IM. Fig. 3 shows the procedure when a WNIC enters and leaves the IM. After a session (e.g., a VoIP session) completion, the WNIC transmits a Disassociation-Request frame with Power saving Mode (PM) bit (in the frame control field) set to ‘1’ in order to enter the IM. After receiving the corresponding Disassociation-Response from its AP, the WNIC in the IM can move around within the same PA while the AP, which transmitted the Disassociation-Response, keeps the information regarding the WNIC to perform a handoff procedure in the future. This AP is referred to as Home-AP. In Fig. 2, AP2 take a role of Home-AP. After entering the Idle Mode, the WNIC starts listening to the beacons periodically (e.g., every 1 second). Even when a WNIC recognizes the change of AP cell through the beacon information, the WNIC keeps listening to only the beacons as long as the WNIC stays in the same PA. This continual beacon listening operation is called AP-Reselection. In this figure, the WNIC performs AP-Reselection until it reaches the boundary between PA1 and PA2. For an efficient AP-Reselection, there could be many optimization issues as addressed in [2, 4]. However, we do not consider the AP-Reselection issues

Figure 3: Procedure for the idle mode operation since they are beyond the scope of the paper. For simplicity, AP-Reselection is assumed to be performed without overhead, e.g., scanning, via optimization.2 When a packet destined to a particular WNIC in the IM arrives at the Home-AP, the Home-AP broadcasts a PageNotify message to all the APs, belonging to the same PA, which in turn start paging the destination WNIC. That is, the APs convey the paging information via their beacon frames. If a WNIC recognizes that it is paged by receiving such beacon(s) from an AP, it attempts to associate with the AP by transmitting a Reassociation-Request. In Fig. 2, AP7 pages the WNIC. After finishing all the preparations for serving the WNIC, the new AP replies to the WNIC with an Association-Response frame and broadcasts Paging-Success to the APs in the same PA to stop paging operations of these APs. At the same time, after successful paging for the WNIC, the mobile host having the WNIC starts to perform the delayed handoff operation. Delayed handoff is explained in Section 3.2. When a mobile host has a packet to transmit, the mobile host transmits a Reassociation-Request frame in the same manner as 802.11-compliant mobile host tries to perform an L2 handoff. The AP, which the mobile host tries to associate with, starts performing handoff operations along with Home-AP as specified in [5].

3.2

Idle and Delayed Handoffs

Idle handoff as a kind of handoff is performed whenever a WNIC in the IM moves across a PA boundary. As can be seen in Fig. 2, an idle handoff is performed when the mobile host moves across the boundary between PA1 and PA2. Fig. 4 shows the procedure for the idle handoff. Af2

For example, the Neighbor Report information from the emerging 802.11k [4] will make this possible in the near future.

Figure 2: New concept and terms

Figure 4: Idle handoff ter a WNIC enters a new PA, which can be identified by a newly-received beacon, it transmits a Reassociation-Request frame with the BSSID of its Home-AP in the “Current AP address” field. The PM bit of the Reassociation-Request frame is set to ‘1’ in order to discriminate ReassociationRequest frame used for the L2 handoff in the PSM or the AM. Upon receiving the Reassociation-Request frame, the new AP replies with a Reassociation-Response frame. Then, the WNIC transmits Disassociation-Request frame in the same manner as to initially enter the IM, i.e., with PM set to ‘1’. After the completion of a successful 3-way management

frame exchange, the WNIC resumes listening to the beacons periodically in order to receive the paging information. The AP, which is involved with the 3-way frame exchange, is referred to as Most Recently Associated AP (MRA-AP). In Fig. 2, once the mobile host enters PA2, it starts to perform AP-Reselection until paging or idle handoff. Home-AP corresponding to a idle mobile host keeps unchanged until successful paging while MRA-APs are changed whenever a mobile host enters a new PA. Now, the MRA-AP initiates to obtain the authentication information about the WNIC from the Home-AP. Through the operations, the MRA-AP performs a user validation check using the MAC address of the WNIC. Note that the user validation check is performed after the completion of the frame exchange with the WNIC in order to reduce the WNIC’s awake time as well as energy consumption. After a successful validation check, the Home-AP informs the previous MRA-AP in the previous PA, which the WNIC previously visited immediately before entering the new PA, by transmitting a Remove-STA message, that the WNIC moves to the new MRA-AP. After receiving the RemoveSTA, the previous MRA-AP removes the information about the WNIC. There could be a lot of security issues about our scheme. However, because of inefficient allowed pages, we do not deal with the issues in this paper. When there is at least one idle handoff, the Home-AP actually transmits a Page-Notify to the MRA-AP, which in turn forwards it to all the APs in the same PA. Since our proposed scheme enables IEEE 802.11 WLAN to keep track of the locations of the mobile hosts in the IM, the IP layerrelated operations including IP paging becomes redundant. That is, our proposed protocol replaces IP paging. Therefore, in our approach, the handoff operations related to the IP layer are postponed until the successful completion of a paging. For this reason, we call this handoff operation, which delays the reactivation of IP layer, a delayed handoff. While performing the delayed handoff, the operations to check the user validity are also performed.

4.

PROTOTYPE IMPLEMENTATION We have built a prototype system to demonstrate our pro-

Table 4: Source code versions of madwifi ATH PCI VERSION ATH HAL VERSION WLAN VERSION

0.9.6 0.9.14.9 0.8.6.0

the PSM, while there is no traffic, are expected to be very similar. For this reason, we decide to use the IM with the sleep duration set to 100 ms instead of the PSM for the measurement.

Figure 5: System configuration for the experiments Table 3: The employed resources of the mobile host Laptop model CPU WNIC model WLAN chip model

LG-IBM X40 Intel(R) Pentium(R) M processor 1.2 GHz MMC Tech. Inc. MW-500CB Atheros Communications, Inc. AR5212

posed IM operation. The system is composed of a mobile host, a Home-AP, and a VoIP system emulator. Both mobile host and Home-AP have a WNIC capable of performing the IM operation. We have developed the system on the Linux Operating System (OS). The kernel version of the Linux OS is 2.6.5-1.358. LG-IBM R32 laptop is used for a Home-AP and LG-IBM X40 laptop is used as a testbed for a mobile host. We install VoIP system emulator in both a PC server and the mobile host as shown in Fig. 5. The resources of the mobile host used in our implementation are shown in Table 3. As a WNIC, we choose an Atheros chipset-equipped WNIC since open source of its device driver is distributed under the name of madwifi project [17]. The versions of the device driver we employ are shown in Table 4. When the Home-AP receives a packet destined to a mobile host in the IM, it begins to locate the mobile host. Netfilter technology enables Home-AP to detect the packet arrival [16]. After the detection, Home-AP begins to transmit beacon conveying paging information. For this purpose, we define a new command in ioctl(), called PAGE-MH. If the device driver is requested for paging via PAGE-MH command of ioctl(), it prepares a beacon which contains the paging information comprising the element identity and the MAC address of the paged mobile host. When we started to develop the IM, madwifi did not support the PSM for a mobile host. In order to obtain the energy consumption rate when a mobile host is in the PSM, we choose the following approach. In case of the PSM, when a mobile host is in the PSM, Traffic Indication Map (TIM) is transferred to the mobile host when new packet arrives at an AP. On receiving the TIM, the mobile host transmits a frame with Power-Saving-Poll (PS-Poll) to its AP, and after receiving the PS-Poll, the AP starts to transmit buffered packet. The procedure, at least in the energy consumption’s perspective, is rather similar to the IM operation when a new packet arrives at an AP. Reminding that the paging beacon is transmitted to a mobile host in the IM, and then, the mobile host, receiving the paging beacon, wakes up to transmit a Reassociation-Request frame for the packet reception. Therefore, if we set the sleep duration for the IM to be 100 ms, the energy consumption rates of the IM and

5.

PERFORMANCE EVALUATION

We have two goals with our experiments. First, we prove the functionality provided by our new paging scheme, which is a part of the IM. Second, we evaluate the energy consumption efficiency provided by the paging. Since there has never been a paging scheme in IEEE 802.11 WLANs, it is important to prove the effectiveness. Our experiments consist of three parts. First, for the proof of the paging functionality, we observe the paging operations in detail to ensure that the designed functionality of the paging is achieved properly. During the experiments, we observe the change of the energy consumption rate when the paging operation is applied and capture the IEEE 802.11 WLAN frames related to the paging operation. Second, we measure the energy consumption of a WNIC in each mode when there is no traffic, and additionally, with traffic. The reason why we perform this experiment is to obtain the amount of energy consumed by a WNIC in each mode. Finally, we also measure the energy consumption rate of a PDA, HP3715 [19], in each active state and idle state, respectively, to estimate the standby time of the HP3715 when the IM is applied to the HP3715. From this experiment, we make an estimation about the standby time of a PDA when the IM is applied.

5.1

Experimental Environment

In order to evaluate the energy saving efficiency of the paging, we have built a testbed for a mobile host as presented in Section 4, and additionally, a VoIP system emulator. The VoIP system emulator has been developed to support a practical VoIP system framework. Fig. 5 illustrates the configuration for our experiments utilizing the VoIP system emulator. An user application running in a mobile host, emulates a mobile user using a wireless IP phone operated in the mobile host capable of performing the IM. The other agent is running in a PC server. In this paper, since we consider the energy saving efficiency of a mobile host compared to legacy schemes under the assumption that the mobile host does not move, we establish an AP, which acts as a Home-AP, to initiate the paging procedure, presented in Section 4. The VoIP system emulator has a simple protocol for call control. With this protocol, we can build an experimental environment to test a paging procedure. Fig. 6 shows the procedure for call setup. Upon receiving the Call-SetupRequest transmitted by user in wired network, the mobile user replies with a Call-Setup-Response after successful paging. After the simple processing for a new call, the user in wired network begins to transmit VoIP packets. When completing the communication, the user in wired network transmits a Call-Terminate-Request the mobile user replies with a Call-Terminate-Response, and then, it enters the IM.

Table 6: BIOS configuration for the mobile host BIOS function Processor speed System suspend timer Automatic turnoff timer for LCD Automatic turnoff timer for HDD Select hibernate by timer

Configuration Fixed max Disabled Disabled Disabled Disabled

Table 7: Battery information Design capacity Last full capacity Battery technology Design voltage Design capacity warning Design capacity low Battery type

61920 mWh Dependent on each time for charging Rechargeable 14400 mV 3096 mWh 619 mWh LION

for our experiments.

5.2 Figure 6: Call control procedure Table 5: WNIC parameters in each mode IM Tx-Power Signal level Tx-Power Signal level

0 dBm -95 dBm

Link Quality Noise level AM 18 dBm Link Quality -35.3 dBm Noise level

0/94 -95 dBm 59.6/94 -95 dBm

For VoIP traffic, we assume G.711 supporting 64 kbps Pulse Coded Modulation (PCM) as the voice codec. This codec generates 160-byte (= 64 kbits/s · 20 ms) voice data every 20 ms. Typically, RTP over UDP is utilized as the transfer protocol for the VoIP data. For this reason, we assume that RTP protocol is used for the VoIP system emulator. Along with the given parameters, we can calculate the payload size for the VoIP system emulator as follows: 160-byte DATA + 12-byte RTP header = 172 bytes. For more accurate measurements, we configure the Linux OS and testbed. In the Linux OS, Advanced Configuration and Energy Interface specification (ACPI) [18] daemon takes a role of listening and dispatching ACPI events from the kernel. By using such events, the daemons for power management can provide system-wide power saving schemes. So, we decide to turn off the ACPI daemon in order to accurately determine the energy consumed by the WNIC operations. In order to exclude possible factors to disturb our measurements, we reconfigure power management functions defined in BIOS as shown in Table 6, and additionally, turn off LCD panel of the testbed. During our experiments, as the operating mode of a WNIC changes, the WNIC parameters related to energy consumption varies over time. The measured WNIC parameters in each mode are shown in Table 5. ACPI provides a very useful means to monitor the battery status. The “/proc/acpi/battery/BAT0/info” file provided by ACPI shows basic information for the battery, including the remaining energy. We use a script, which reads the file every two minutes, and saves it into the hard disk periodically. Table 7 shows the information about the battery used

Measurement Results

Observation on paging operation: Utilizing the VoIP system emulator explained above, we perform an experiment in order to observe the paging operation. For this experiment, we prepare a scenario that a mobile host spends five minutes in the IM and one minute in the AM repeatedly. Fig. 7 shows the frames captured by a commercial tool, called airopeek [21]. In the figure, (1) and (2) represent Probe-Request and Probe-Response frames for scanning, respectively. (3) and (4) indicate the message exchanges to authenticate each other. After the authentication, the mobile host attempts to associate with the Home-AP as shown in (5) and (6). Finally, the mobile host enters the IM by transmitting a Disassociation-Request with the PM bit set to ‘1’ as shown by (7). After the mobile host spends five minutes staying in the IM, the mobile host is paged by receiving the paging beacon (8). Immediately after receiving the paging beacon, the mobile host wakes up to exchange ReassociationRequest and Reassociation-Response frames with the HomeAP (9), (10). After a paging success, the VoIP session lasts for one minute. In the figure, the procedure, in which the mobile host enters the IM and wakes up, repeats three times. Fig. 8 and Fig. 9 show the variation of the energy consumption rate as time goes. The x-axis represents the time and the y-axis represents the remaining energy in the battery of the mobile host, respectively. The slope of each line represents the energy consumption rate, i.e., power consumption. The slopes of the measured energy consumption rates are alsmost linear. So, we can regard it as power consumption dependent on each mode. Fig. 8 shows the energy consumption rate while the VoIP system emulator establishes and terminates calls according to the predefined call pattern. In this measurement, the call pattern for energy consumption measurement is different from that used for the experiment to capture 802.11 frames shown in Fig. 7. The initial mode of the WNIC is the IM. After twenty minutes, the mode changes into the AM because the mobile host gets reassociated with the HomeAP due to the success of the paging. After receiving VoIP traffic for twenty minutes, the WNIC returns to the IM. For this reason, in Fig. 8, the energy consumption rate is restored accordingly. In the figure, we observe that the the energy consumption rate changes at the moments of 20, 40,

Table 8: Energy consumption rate of the mobile host and the WNIC Operation mode HD alone IM, Tsleep = 1000 ms (PW N IC IM ) IM, Tsleep = 100 ms (PW N IC IM ) AM w/o traffic (PW N IC AM ) AM w/ traffic (PW N IC AM tr )

Figure 7: Captured IEEE 802.11 WLAN frames 36000 Paging scenario

34000

Remaining energy (mWh)

32000

30000

28000

26000

24000

22000 0

20

40

80

90

Time (min)

Figure 8: Remaining energy variation over time 40000 HD alone AM w/o traffic AM w/ traffic IM, Tsleep = 1000 ms IM, Tsleep = 100 ms

35000

Remaining energy (mWh)

30000

25000

20000

15000

10000

5000

0 100

200 Time (min)

300 320 340 360 380

Figure 9: Energy consumption measurement of the mobile host

mobile host 6775 mW 7231 mW 7310 mW 7990 mW 8230 mW

WNIC 456 mW 535 mW 1215 mW 1455 mW

80, and 90 minutes. We can easily imagine that the VoIP traffic is transferred during two time intervals, i.e., from 20 to 40, and from 80 to 90 minutes. Energy consumption measurement: We obtain the energy consumption rate of a WNIC in IM, PSM and AM, respectively, because our major interest is to get the energy consumption rate of the WNIC itself and to apply the obtained values to another platform to estimate the standby time of our target device. As discussed in Section 4, since the energy consumption of a WNIC in the PSM is almost identical to that of the IM with sleep duration (Tsleep ) set to 100 ms as long as there is no traffic, we measure the energy consumption rate of the IM with Tsleep = 100 ms instead of the PSM. In case of the AM, we measure the energy consumption rate of a WNIC when the WNIC is activated with/without VoIP traffic. The VoIP traffic is transmitted by the user agent in the VoIP system emulator. In addition, we measure the energy consumption rate when only the HD, i.e., without a WNIC, is activated. Based on these scenarios, we observe the energy consumption rate of a mobile host. For every measurement, we measure the remaining energy of a mobile host every two minutes until the battery of the mobile host is drained of the remaining energy. Fig. 9 shows the measurement results dependent on each mode. The fully charged energy varies each time after the charging is finished. In order to make the fair comparison, we measure the remaining energy until the remaining energy difference between fully charged energy and remaining energy reaches 40000 mAh. For this reason, in Fig. 9, the maximum values are adjusted to 40000 mAh while the minimum values are zero. Fig. 9 shows the energy consumption rate of the IM with sleep duration (Tsleep ) set to 1000 ms is not much different from that of the IM with (Tsleep ) set to 100 ms. The reason is because the energy consumption rate of a LG-IBM X40 is much large compared with that of WNIC. With these experimental results, we can obtain the energy consumption rate of a WNIC when each of IM, PSM, and AM is employed, respectively, by subtracting the energy consumption rate of a mobile host without WNIC from that of the mobile host with WNIC in each mode. The obtained results are summarized in Table 8. We observe that the energy consumption rate of the mobile host in the IM is significantly smaller than that in the AM. The value of Tsleep in the IM and the existence of the traffic in the AM makes some difference too while it is less influential than the operation mode itself. We also observe that the the HD, i.e., LG-IBM X40 laptop, itself consumes remarkable energy compared to the WNIC. It is due partly to the fact that we disable the ACPI daemon for accurate measurements for the WNIC energy consumption. Standby time estimation: It would be commonly un-

Table 9: Power consumption of the PDA (HP3715) Active mode (PHD 207 mW

active )

Standby mode (PHD 5.6 mW

standby )

Table 10: Estimated standby time of the PDA (HP3715) in each mode IM 692

PSM 430

PDA in active PDA in standby

AM 224

derstood that a user would carry portable devices, i.e., cellular phone or PDA, for her/his wireless communication rather than holding a laptop at hand. Under such an assumption, we estimate the standby time extension when the IM is applied to a PDA. For our estimation, we choose HP3715 [19] as a portable device and measure the energy consumption of it. When there is no traffic, the PDA must be in the standby mode, in which the PDA turns off almost every component inside the PDA. While doing the measurement for both active mode and standby mode, the LCD panel is turned off and no application runs. The measurement results of the energy consumption rate of the PDA are summarized in Table 9. For the AM, both the WNIC and the HD should be activated all the time, even when traffic does not exist, thus incurring a significant energy consumption. In the case of the PSM, a WNIC can save its energy when there is no traffic. However, a mobile host in the PSM still requires the HD to remain activated since the IP packets should be managed since VoWLAN is operated upper IP layer. For this reason, when the PSM is employed, the energy consumption required by the HD should be considered. The IM provides a benefit entirely distinguishable from the PSM and the AM. When the IM is employed, the HD can be in standby mode because the IM enable the IP layer to be paged without IP paging. For this reason, when utilizing the IM, we can consider the energy required for the HD is that of the HD for standby mode. Considering these features about each operation of AM, PSM, and IM, we can estimate the standby time in each mode. The amount of the energy consumption rate for each mode is obtained as follows:

90

Remaining energy (%)

WNIC mode Standby time (minutes)

100

80

70

60

50 0

1000

2000

3000 Time (min)

4000

5000

6000

Figure 10: Energy consumption measurement of the PDA

IM PSM AM w/o traffic AM w/ traffic

5000

PP SM = PW N IC PIM = PW N IC

+ PHD active = 1422 mW P SM + PHD active = 742 mW

AM

IM

+ PHD

standby

= 461.6 mW

Each of PAM , PP SM , and PIM indicates the energy consumed by the PDA in AM, PSM, and IM, respectively. Considering the obtained energy consumption rate values and the battery capacity of HP3715, we estimate the standby time when each of IM, PSM, and AM is employed for the PDA, respectively. As shown in Fig. 11, if the portable device employs the IM, the standby time is extended remarkably. The standby time we have estimated are summarized in Table 10.

6.

CONCLUSION

In this paper, we propose an Idle Mode operation in the 802.11 WLANs. The proposed protocol can be easily applied to already-deployed products by just updating their firmwares or device drivers. In order to prove the above statement, we have implemented our proposed schemes based

Remaining energy (mWh)

4000

PAM = PW N IC

3000

2000

1000

0 0

100

200

300 400 Time (min)

500

600

Figure 11: Estimated energy consumption of the PDA (HP3715)

700

on the device driver source code distributed under the madwifi project. The results obtained through experiments demonstrate that our proposed IM operation outperforms legacy schemes with respect to the energy consumption. As a result, it enables a longer standby time of the 802.11-equipped mobile hosts.

7.

REFERENCES

[1] M. Liebsh and X. P´erez-Costa, “Utilization of the IEEE 802.11 Power Save Mode with IP Paging,” in Proc. IEEE ICC’05, 2005. [2] A. Mishra, M. Shin, and W. A. Arbaugh, “Context Caching using Neighbor Graphs for Fast Handoffs in a Wireless Network,” in Proc. IEEE INFOCOM’04, 2004. [3] IEEE Computer Society LAN MAN Standards Committee, IEEE Standard for Information Technology: Part 11: Wireless LAN Medium Access Control (MAC) and Physical Layer (PHY) Specifications, 1999. [4] IEEE 802.11k/D2.0, Part 11: Wireless LAN Medium Access Control (MAC) and Physical Layer (PHY) Specifications: Radio Resource Measurement: Draft Amendment 7 to IEEE 802.11 Standard-1999 Edition, Draft 2.0, 2004. [5] IEEE 802.11f, Part 11: IEEE Trial-Use Recommended Practice for Multi-Vendor Access Point Interoperability via an Inter-Access Point Protocol Across Distribution Systems Supporting IEEE 802.11 Operation, 2003. [6] C. Perkins, “IP Mobility Support,” IETF, RFC 2002, Oct. 1996. [7] X. Zhang, J. Gomez, G. Castellanos, and A. T. Campbell, “P-MIP: Paging Extensions for Mobile IP,” ACM Mobile Networks and Applications, July 2002.

[8] R. Ramjee, L. Li, T. L. Porta, and S. Kasera, “IP Paging Service for Mobile Hosts,” in Proc. ACM MobiCom’01, 2001. [9] E. Shih, P. Bahl, and M. J. Sinclair, “Wake on Wireless: An Event Driven Energy Saving Strategy for Battery Operated Devices,” in Proc. ACM MobiCom’02, 2002. [10] IEEE 802.11i, Part 11: Wireless Medium Access Control (MAC) and Physical layer (PHY) Specifications: Amendment 6: Medium Access Control (MAC) Security Enhancements, 2004. [11] J. Chen, M. Jiang, and Y. Liu, “Wireless LAN security and IEEE 802.11i,” IEEE Wireless Communications, Feb. 2005. [12] M. Anand, E. B. Nightingale, and J. Flinn “Self-Tuning Wireless Network Power Management,” in Proc. ACM MobiCom’03, 2003. [13] http://www.nespot.com/web/eng/index.html. [14] http://www.hpl.hp.com/personal/ Jean Tourrilhes/Linux/Tools.html. [15] http://hostap.epitest.fi. [16] http://www.netfilter.org. [17] http://www.madwifi.org. [18] http://www.acpi.info. [19] http://www.shopping.hp.com/shopping/pdf/ fa281a.pdf. [20] H. Yokota, A. Idoue, T. Hasegawa, and T. Kato, “Link Layer Assisted Mobile IP Fast Handoff Method over Wireless LAN Networks,” in Proc. ACM MobiCom’02, 2002. [21] http://www.airopeek.com.