ieee xplore full-text pdf

Low Power Consumption Solutionsfor Mobile Instant Messaging

Ling-San Meng, Da-Shan Shiu, Member, IEEE,

Ping-Cheng Yeh, Member, IEEE, Kuan-Chi Chen, and Hung-Yi Lo

Abstract—Instant messaging (IM) services enable real-time text and multimedia exchange and online presence awareness. Users

typically log onto instant messaging services persistently to discover available friends and also to be discovered. However, our analysis

shows that the frequency exchange of presence information incurs massive power consumption to mobile devices over cellular or

wireless local area networks. Such power consumption penalty can render persistent-instant messaging infeasible for battery-powered

mobile devices. In this paper, we propose several solutions to mitigate the power consumption problem. By reducing the network

access and keeping mobile devices in the sleep mode as much as possible, these solutions achieve significant power saving. The

power consumption of the proposed solutions is derived analytically in this paper and the proposed solutions are implemented using a

Jabber-based architecture. Actual power measurement results show that the power consumption of the proposed solutions agrees well

with our analysis, and significant power saving can be achieved on mobile handsets with our low power consumption solutions

implemented.

Index Terms—Mobile devices, power saving, presence information, system implementation, Jabber/XMPP.

Ç

1 INTRODUCTION

INSTANT messaging (IM) services, as have become arguablyone of the most popular Internet applications nowadays,

appeared as early as the introduction of the UNIX operatingsystem, where users were able to exchange short messagesusing simple commands in real time. It is not, however,until the advent of ICQ (short for “I seek you”) that IMbegan to gain wide popularity. Within only a few years,commercial IM applications such as AOL Instant Messenger(AIM), Microsoft MSN messenger (MSN), and Yahoo!messenger [1], were released one after another with millionsof registered users today [2]. Typically, IM applicationsprovide two main services: the instant message deliveryservice and the presence awareness service. The instantmessage delivery service enables real-time text messageexchange between users, while the presence awarenessservice provides the instantaneous online status of IMfriends/entities through the so-called “buddy list” [3].

Originally, IM services are designed and tailored fordesktop use only. It is commonly realized that modifica-tions to the existing IM services are necessary before IM canbe widely accepted by mobile users [4]. Issues that havebeen identified for mobile IM include enhancement forpresence awareness and security [5], [6], [7], [8], support forlocation awareness [9], and the power consumptionproblem [10]. To the best of the authors’ knowledge,however, the issue of power consumption caused byrunning IM on mobile devices has not been specifically

addressed and solved. As shown in [10] and [11], sporadicdata traffic on mobile devices leads to sever powerconsumption penalty. The occasional behavior of presenceexchange with remote IM friends/entities thus poses aproblem, which, as we show in this paper, can renderpersistent-IM infeasible for battery-powered devices.

In this paper, we propose several solutions to lower thepower consumption of mobile devices due to the presenceinformation exchange. By effectively reducing the rate ofthe presence information exchange that mobile devicesactually have to participate in, the proposed solutions arecapable of achieving great power saving at zero cost. Furtherpower saving can also be obtained on the mobile devices bycompromising a certain amount of presence update delay.The tradeoff between the presence update delay and theattainable power saving is derived analytically in this paper.The proposed solutions are then implemented on both Wi-Fiand 3G handsets using a Jabber/Extensible Messaging andPresence Protocol (XMPP)-based architecture [12], [13],based on which extensive power measurement experimentsare performed. It is observed that our analysis is accurate incapturing the amount of power saving provided by theproposed solutions. By yielding an average presence updatedelay of 30 seconds, maximum power savings of 6.5 mA and154 mA can be achieved for handsets over Wi-Fi and 3Gnetworks, respectively. Concerning the battery capacity ofmobile handsets, which typically ranges between 700-1,500 mAh, our proposed low power consumption solutionscan effectively extend the battery lifetime for mobile IM users.

The remainder of this paper is organized as follows: inSection 2, we describe the mobile IM system on which wewill explore the problem of power consumption, includingthe power saving mechanism and various presence ex-change mechanisms. In Section 3, we propose the corre-sponding solution to lower the power consumption to each

896 IEEE TRANSACTIONS ON MOBILE COMPUTING, VOL. 11, NO. 6, JUNE 2012

. The authors are with the Department of Electrical Engineering, NationalTaiwan University, Taipei 106, Taiwan.E-mail: {f95942117, r95942027, b92901121}@ntu.edu.tw,{dsshiu, pcyeh}@cc.ee.ntu.edu.tw.

Manuscript received 4 Nov. 2008; revised 19 Feb. 2011; accepted 1 Apr. 2011;published online 7 June 2011.For information on obtaining reprints of this article, please send e-mail to:[email protected], and reference IEEECS Log Number TMC-2008-11-0444.Digital Object Identifier no. 10.1109/TMC.2011.123.

1536-1233/12/$31.00 � 2012 IEEE Published by the IEEE CS, CASS, ComSoc, IES, & SPS

type of presence exchange mechanism. The tradeoffbetween the achievable power saving and the presenceupdate delay of the proposed solutions is derived in detailin Section 4. In Section 5, a Jabber/XMPP-based implemen-tation of the proposed solutions is presented along with thepower measurement results. Finally, conclusions are givenin Section 6.

2 SYSTEM DESCRIPTION

In this paper, we consider a battery-powered mobile devicewhich persistently logs onto one or multiple IM services.Since the majority of the IM traffic consists of the presenceinformation exchanges, we focus only on the powerconsumption due to the presence information exchanges.It is assumed that there is no text message exchange takingplace during the power measurement. As mentionedpreviously, the online status of remote friends changesfrom time to time. An IM client regularly exchangesmessages with remote entities, either a central server orother peer clients, to ensure its current presence informationalways being up-to-date. Such frequent network accessseverely disrupts the power saving mechanism of themobile device. Depending on the underlying protocols,the presence information exchanges affect the the powersaving mechanism in different manners. In this section, wefirst describe the power saving model of a typical mobiledevice. This model is simple, yet it captures the essence ofthe power saving operation. We then elaborate on thevarious mechanisms for exchanging presence information.

2.1 Power Saving Model

The operation of a typical mobile device can be roughlyclassified into two different modes: the active mode andthe sleep mode. In the active mode, all the communicationmodules are turned on to support data exchange at lowlatency. On the other hand, during the sleep mode, themobile device turns off the RF and other communicationmodules until it wakes up again to process data traffic. Themobile device consumes battery energy at a much higherrate in the active mode than in the sleep mode. The longerthe mobile device stays in the sleep mode, the more powerit saves.

The procedures for mobile devices to transit from theactive mode to the sleep mode generally differ inimplementations. Yet, they share similar features. Ingeneral, if a mobile remains idle for a certain periodwithout seeing any data traffic, it will enter the sleep mode.

In this paper, we assume the mobile device switches fromthe active mode to the sleep mode when a fixed amount ofwaiting time T has elapsed since the last data exchange. Anillustrative example is given in Fig. 1.

2.2 Presence Exchange Mechanisms

Most IM applications such as AIM, MSN, and Yahoo!messenger are proprietary services. As a result, they haveinteroperability problem. Though some efforts have beendone on standardizing the IM protocols [12], [14], [15], it isoften the case that different IM services implement differentpresence exchange mechanisms. While being different, IMpresence exchange mechanisms can generally be classifiedto either one or more than one of the following presenceexchange mechanisms.

2.2.1 Event-Triggered Presence Update

This presence exchange mechanism requires that a presenceupdate message being broadcast to all of the remote entitieswhen an IM client manually changes its own online status.This mechanism serves as a notification to the remotefriends of the instantaneous presence change and isimplemented in almost all IM services. As IM userstypically have a large number of remote friends, receivingevent-triggered updates comprises the dominant portion ofthe presence information exchange in IM.

2.2.2 Presence Probe

In this mechanism, an IM client initiates a probe message toa remote IM entity to request an online status update. Whenthe remote entity receives the probe message, it replies tothe sender with the requested information. For example, itis observed in our network experiments that a Skype [16]client generates probe messages to the peer clients on itsfriend list every 180 seconds. Note that for a mobile client,the power consumption induced by initiating presenceprobes can be larger than that induced by reacting topresence probes. This is due to the fact that the initiatingparty must wait for the response from the remote entity toreturn, whereas the requested party can enter the sleepmode immediately after it reacts to the probe. However,since the round-trip delay between the mobile clients isusually small as compared with T , we assume both casesconsume the same amount of battery energy in this paper.

2.2.3 Periodic Presence Update

In this case, an IM client periodically sends (receives)messages to (from) the remote entities. The message could

MENG ET AL.: LOW POWER CONSUMPTION SOLUTIONS FOR MOBILE INSTANT MESSAGING 897

Fig. 1. Transitions between the active and sleep modes on mobile devices.

be either an explicit presence update or simply a dummymessage checking the validity of the connection. Forexample, it is observed in the experiments that an MSNclient generates a packet every 29 seconds to the centralizedMSN server to report its presence.

It is worth noting that when a mobile device logs ontomultiple IM services, it must conform to each of the IMprotocols and execute the associated presence exchangemechanisms of each IM. As a result, the power consump-tion generally becomes higher in this case.

3 SOLUTIONS FOR REDUCING IM POWER

CONSUMPTION

In this section, we propose corresponding solutions to eachof the presence update mechanisms described in theprevious section. A framework for system implementationis also given.

3.1 Buffering Event-Triggered Presence Updatesfrom the Remote Entities

Recall that the power saving mechanism of a mobile devicerequires it to stay in the active mode for a fixed period of Tbefore entering the sleep mode. If the event-triggeredpresence updates are sporadic with most of the interarrivaltime intervals shorter than T , the mobile device can hardlyenter the sleep mode and thus consumes a large amount ofpower. To solve this problem, we propose to use an agent,preferably running on a nonbattery-powered machineconnected to a fixed-line network, to buffer the event-triggered presence updates. For instance, presence updatemessages can be temporarily buffered at the agent either fora predefined period or until the number of accumulatedpresence updates reaches a certain threshold. The bufferedevent-triggered presence updates are then forwarded to themobile IM client by the buffering agent. In this way, the rateof the mobile client entering the active mode can be reducedand thus achieves power saving. Note that buffering event-triggered presence information introduces update delay tothe mobile client. There is a tradeoff between the powersaving and the presence update delay. The balance betweenthe attained power saving and the resulted update delayshould be determined by the user preference.

3.2 Handling Presence Probes on Behalf of theMobile Client

For mobile devices which are required to send (receive)probe messages to (from) remote IM entities, we propose touse an agent to handle the presence probes on behalf of the

mobile device. This agent can initiate the outgoing probemessages and react to the incoming probe messageswithout disturbing the mobile device. With this solution,the mobile device seldom leaves the sleep mode to initiatepresence probes. The power consumption for initiatingprobe messages can, thus, be completely eliminated.

3.3 Periodic Presence Update on Behalf of theMobile Client

To reduce the power consumption due to the periodicpresence updates, we propose to use an agent to handle theperiodic presence updates to (from) the remote IM entitiesas required by the underlying protocol. With this solution,the duty of the periodic presence update is also completelyremoved from the mobile client.

Note that while different agents are proposed, theseagents can be integrated into one agent when multiplepresence update mechanisms are used. The agents ofdifferent IM running on a presence proxy server is shownin Fig. 2. As mentioned previously, simultaneously loggingonto multiple IM services generally causes more powerconsumption as compared with the case of single IM. Toreduce such excess power consumption, we propose toemploy an intermediate transport protocol to streamline thepresence updates from different IM services. Specifically,multi-IM protocol translators are used on the presenceproxy to translate different IM protocols into one unifyingIM protocol. The mobile client actually communicates withthe presence proxy through this IM protocol. At the mobiledevice side, a multi-IM protocol translator is used totranslate the unified-IM back to the intended IM formats.

4 POWER CONSUMPTION ANALYSIS

In this section, we first derive the power consumption for theconsidered mobile device both with and without theproposed solutions to compute the achievable power saving.The expected presence update delay experienced by themobile client after applying the proposed solutions is thenderived to characterize the tradeoff between the attainedpower saving and the resulted presence update delay.

4.1 Power Consumption for Presence InformationExchange

Typically, presence update messages contain only the useraccount and online status information. The length ofpresence update messages and the associated transmissiontime is rather small as compared with that of the regularnetwork data traffic. We assume the transmission time isnegligible so that a presence exchange triggers the modem


Fig. 2. The integrated agent for reducing IM power consumption on mobile devices.

to stay in the active mode for exactly T seconds. The powerconsumption of the mobile device is then completelydetermined by the arrival process of the presence informa-tion exchanges, which is in turn characterized by theinterarrival time distribution of the presence exchanges.

For event-triggered presence updates and presenceprobes, it is observed from our network experiments thatthe presence information exchanges are triggered in anearly random manner. The two presence exchangeprocesses are, thus, modeled as Poisson processes withrates �1;k and ð�2;k þ �2;kÞ, where �i;k and �i;k denote therates for receiving and initiating presence information dueto presence exchange mechanism i of IM protocol k,respectively. Note that we assume the IM status of themobile device considered remains unchanged for a rela-tively long period. Thus, there is nearly no outgoing event-triggered presence update and we set �1;k ¼ 0. For periodicpresence updates, the rates of receiving and initiatingpresence updates due to IM protocol k are similarly denotedas �3;k and �3;k, respectively. Table 1 summarizes thenotations. In this paper, we assume that the periodic updatemechanisms implemented in different IM protocols arenearly uncorrelated. Let tp denote the time durationbetween two consecutive presence information exchanges.The cumulative distribution function (c.d.f.) of tp can beexpressed as

FtpðtÞ ¼ 1� Prftp > tg

’ 1� exp �Xk

ð�1;k þ �2;k þ �2;kÞ t !

Yk

ð1�minf�3;kt; 1g��

1�minf�3;kt; 1gÞ;

ð1Þ

where the last equality follows since we have assumedthat different presence exchange mechanisms are nearlyuncorrelated.

Let Ta and Ts denote the respective burst time that themobile device persistently stays in the active mode and thesleep mode as depicted in Fig. 3. It is clear that Ta � T . Itcan be shown that the proportion of time staying in theactive mode for the mobile device is

fa ¼E½Ta�

E½Ta� þ E½Ts�: ð2Þ

One needs the value of fa to compute the powerconsumption incurred by presence information exchanges.Define n to be the number of presence informationexchanges within Ta; n is geometrically distributed asGEOð1� FtpðT ÞÞ. We can express Ta as

Ta ¼Xn�1

i¼1

ti þ T; ð3Þ

where ti denotes the ith interarrival time of the presenceexchanges, and ftig are independently and identicallydistributed as tp j ftp � Tg. It should be noted that in (3),we have neglected the possible paging delay introducedby the first presence exchange after waking up from thesleep mode. Such approximation is acceptable since thepaging delay is relatively small as compared withthe interarrival times ftig. Taking the expectation of (3),it can be shown that

E½Ta� ¼1

p� 1

� �E½tp j tp � T � þ T: ð4Þ

On the other hand, since Ts ¼ Tn � T is just the remainingtime for the next presence exchange event to occur giventhat a period of T has elapsed, we have

E½Ts� ¼ E½tp j tp > T � � T: ð5Þ


TABLE 1Notations

Fig. 3. A example illustrating the burst time Ta and Ts that the mobile device persistently stays in the active mode and the sleep mode, respectively.

Combining (1), (4), and (5), we can rewrite (2) as

fa ¼E½tp j tp � T �ð1� pÞ þ Tp

E½tp�

¼R T

0 1� FtpðtÞ dtR10 1� FtpðtÞ dt

:

ð6Þ

Using the notations in Table 1, the additional powerconsumption due to the presence information exchangeon the mobile device can be derived as

ðPa � PsÞ fa: ð7Þ

4.2 Attainable Power Saving of the ProposedSolutions

When a mobile client applies all of the mobile IM powersaving solutions, the only occasions it awakes are to receivethe accumulated presence updates from the buffering agent.The power consumption of the mobile device is determinedby the arrival process of the buffered update packets fromthe agent. Depending on the different buffering policies theagent adopts, the packet arrival process can be different. Inthis work, we assume that the buffering agent adopts apolicy that forwards the buffered presence updates to themobile client after Tb seconds have elapsed from receivingthe first buffered presence update. The performancecorresponding to another buffering policy where the agentforwards the buffered presence updates to the mobile clientin a periodic fashion can be found in our previous work [17].

Let t0p denote the interarrival time of two consecutivebuffered presence updates at the mobile client. Define t0 ast0 ¼ t0p � Tb to represent the time for the first presenceupdate to arrive from the latest buffer flush as depicted inFig. 4. The c.d.f. of t0p can be expressed as

Ft0pðtÞ ¼0; if t < Tb;Ft0ðt� TbÞ; otherwise:

�ð8Þ

where Ft0ðtÞ denotes the c.d.f. of t0 and is derived to be

Ft0ðtÞ ’ 1� exp �Xk

ð�1;k þ �2;kÞ t !Y

k

ð1�minf�3;kt; 1gÞ:

ð9Þ

The proportion of time staying in the active mode for themobile device after applying the proposed solutions,

denoted as f 0a, is then derived by replacing FtpðtÞ withFt0pðtÞ in (6). If no periodic presence update is received at theagent (i.e., �3;k ¼ 0; 8k), as is often the case, a closed-formexpression for f 0a can be derived for Tb < T as

f 0a ¼R TTb

exp �P

kð�1;k þ �2;kÞ ðt� TbÞ� �

dtR1Tb

exp �P

kð�1;k þ �2;kÞ ðt� TbÞ� �

dt

¼ 1�exp �

Pk �1;k þ �2;k

� �ðT � TbÞ

� �Pkð�1;k þ �2;kÞTb þ 1

:

ð10Þ

Similar derivations yield

f 0a ¼T

Tb þP

k �1;k þ �2;k

� ��1; ð11Þ

for Tb � T . The additional power consumption incurred byrunning IM on the mobile device with the proposedsolutions applied is, thus,

ðPa � PsÞ f 0a: ð12Þ

Finally, the attainable power saving Psaved of the proposedsolutions can be obtained easily from (7) and (12) as

Psaved ¼ ðPa � PsÞ � fa � f 0a� �

: ð13Þ

4.3 Tradeoff between Power Saving and MeanPresence Update Delay

The proposed solutions achieve power saving by regulatingthe arrivals of the presence updates to the mobile device.However, as stated before, this is accomplished at the costof the presence update delay. Note that the presence updatedelay here is defined as the additional time required for themobile IM client to get a particular presence informationafter applying the proposed solutions. Intuitively, to getmore power saved, one has to undergo longer delay forpresence updates, and vice versa. The tradeoff between thepower saving and the mean presence update delay canserve as the criterion for assessing the performance ofdifferent buffering policies. Thus, it is of practical interest tocharacterize such tradeoff.

It is clear that each one of the buffered presence updatesreceived at the beginning of each buffering process under-goes a delay of Tb seconds. For the rest of the presenceupdates received by the agent during Tb, we note thatthe arrival instants corresponding to a Poisson process are


Fig. 4. An example illustrating the presence information exchange with power saving solutions applied. t0 denotes the time for the first presenceupdate to arrive since the last buffer flush (Tb � T case shown only).

uniformly distributed over time. Since we have assumeddifferent periodic presence update processes are uncorre-lated, the average presence update delay experienced by themobile client can be approximated as

Tdelay ’ Tb �1

1þP

kð�1;k þ �2;k þ �3;kÞ Tb

þ Tb2�

Pkð�1;k þ �2;k þ �3;kÞ Tb

1þP

kð�1;k þ �2;k þ �3;kÞ Tb

¼ Tb �1þ 1

2

Pkð�1;k þ �2;k þ �3;kÞ Tb

1þP

kð�1;k þ �2;k þ �3;kÞ Tb:

ð14Þ

Solving for Tb from (14), we have

Tb ¼ Tdelay �1P

kð�1;k þ �2;k þ �3;kÞ

þ

ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiT 2delay þ

Xk

�1;k þ �2;k þ �3;k

!�2vuut :

ð15Þ

The tradeoff between the power saving Psaved and the meanpresence update delay Tdelay can thus be derived bysubstituting the r.h.s. of (15) for Tb in (13). For the casewhere no periodic presence update is received at the agent,the tradeoff can be further derived for Tb < T as

Psaved ¼ ðPa � PsÞ fa � 1þ

(exp

� X

k

�1;k þ �2;k

!

T � Tdelay �

ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiT 2delay þ

Xk

�1;k þ �2;k

!�2vuut !

� 1

!),

( Xk

�1;k þ �2;k

! Tdelayþ

ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiT 2delayþ

Xk

�1;k þ �2;k

!�2vuut !)!

ð16Þ

and

Psaved ¼ ðPa � PsÞ

fa �T

Tdelay þffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiT 2delay þ

Pk �1;k þ �2;k

� ��2q

0B@

1CA: ð17Þ

for Tb � T .It should be noted that while the rates of various presence

update mechanisms have to be known in advance to apply

the analysis presented above, the rates can be easily estimatedby monitoring the actual traffic of different presenceinformation exchanges. We also want to note that since mostof the IM applications today are of proprietary services,reverse engineering as well as extensive network experi-ments are often required to figure out the actual presenceexchange mechanisms implemented in different IM.

5 SYSTEM IMPLEMENTATION AND POWER

MEASUREMENT RESULTS

Here, we present our implementation of the proposedmobile IM power saving solutions. The implementationapplies a proxy-based IM architecture originating fromFig. 2. In this architecture, a mobile client logs onto IMservices through a “presence proxy” which is capable ofhandling incoming presence updates buffering, presenceprobes, and periodic presence updates for the mobileclient. The presence proxy also functions as a multi-IMprotocol translator, so that the mobile client connecting tothe presence proxy can communicate with various IMnetworks. In this work, the Jabber/XMPP protocol [12],[13] is used to realize the presence proxy-based architec-ture. Jabber (also known as XMPP) is an open IM protocolwhich is selected by the Internet Engineering Task Force(IETF) to be one of the two core protocols for instantmessaging and presence technology. Several IM services,such as Google Talk and iChat, adopt Jabber either as theprimary or the auxiliary IM and presence protocol in theirdesigns [18]. The resulting Jabber-based system implemen-tation is shown in Fig. 5. The Jabber server handles thepresence probes and the presence proxy server buffersincoming presence updates. The translation gatewaysserve as the multi-IM protocol translators between Jabberand other IM protocols. The gateways also serve as theagents to send periodic presence updates to the remote IMentities for the mobile client.


Fig. 5. The Jabber-based implementation of the proposed mobile IM power saving solutions.

TABLE 2Average Current Consumption of Various IM Services

Actual power consumption measurement on both Wi-Fi(IEEE 802.11b)1 and 3G/UMTS (3GPP release 6) PDAhandsets is performed for several IM services, includingMSN, Yahoo! messenger, AIM, and Google Talk. The IMclient in our measurement experiments is assumed to havea group of remote friends with an average rate of onlinestatus change, i.e., incoming event-triggered presenceupdates, equal to 1

15 message/second.2 In addition to theevent-triggered presence updates, it is observed that all ofthe selected IM services access the network in a periodicfashion with distinct periods. Specifically, a period of29 seconds is observed for MSN, a period of 240 secondsis observed for Yahoo! messenger, a period of 300 seconds isobserved for AIM, and a period of 30 seconds is observedfor Google Talk. Note that the reason these IM servicesshow similar behaviors in exchanging presence informationlies mainly in the fact that they are all of centralized IMarchitectures, in which an IM client always communicateswith a centralized IM server to exchange text messages andpresence information with other peer clients.

When receiving a presence update message, the PDAhandsets stay in the active mode for another timeoutperiod with much higher power consumption than that inthe sleep mode as described in Section 2. For the PDAhandsets we used, both the cellular modem and theoperating system software wake up during the timeoutperiod and contribute to the high power consumption. Asimple solution could be forcing the operating system toenter the sleep mode as soon as the reception of apresence update message. The numerical results corre-sponding to such a simple solution is included in theAppendix, which can be found on the Computer SocietyDigital Library at http://doi.ieeecomputersociety.org/10.1109/TMC.2011.123. However, one should note that

such an option to modify the timeout value of operatingsystem is generally not available on PDA handsets/smartphones. Also the 0-timeout solution results in greateruser’s experience of latency.

The results of the power consumption measurementusing “Agilent 66321D Mobile Communications DCSources” are listed in Table 2. Note that the results aregiven in units of current consumption (mA) as output bythe measuring instrument. The corresponding powerconsumption can be obtained by multiplying the currentconsumption value by 3.6 volts, which is the operatingvoltage level of the tested mobile handsets. All themeasurement results represent the average additionalcurrent consumption incurred by running a particular IMservice. For our proposed solutions, the proxy-inducedpresence update delay is targeted at 30 seconds. We can seethat the power consumption incurred by the original IMarchitecture is quite prohibitive, especially in the case of 3G.This is due to the highly inefficient implementation of themode transitions in 3G/UMTS networks.3 The waiting timefor 3G mobile handsets to enter the idle mode canpotentially exceed tens of seconds [11]. In our experiments,a timeout value T of 14 seconds is observed as shown inFig. 6. On the other hand, it is shown that significant powersaving can be achieved for all IM services once theproposed solutions are applied. Specifically, for themeasurement results over the Wi-Fi network, it is observedthat the power consumption values due to IM services areall reduced to around 1 mA. A maximum power saving of6.5 mA is achieved. As to the 3G network, the powerconsumption values are all reduced to around 55 mA, and amaximum power saving of as high as 154 mA can beachieved. Concerning the battery capacity of mobilehandsets, which typically ranges between 700-1,500 mAh,


Fig. 6. A snapshot of the power measurement results over the 3G network. A timeout value T ¼ 14 seconds is observed.

1. For experiments over the Wi-Fi network, the default power savingmode specified in IEEE 802.11 is used.

2. For the purpose of experiments, we control the update rate to be 115

message/second by manually changing the online status of the remote IMentities.

3. Mode transitions for mobile handsets in 3G/UMTS networks aretermed as the radio resource control (RRC) state transitions. The task ofRRC state transitions is controlled by the network and the correspondingtimeout value is a network-specific parameter.

our proposed low power consumption solutions caneffectively extend the battery lifetime for mobile IM users.

In Figs. 7 and 8, the analytical curves of the tradeoffbetween the attainable power saving and the meanpresence update delay are plotted along with the measure-ment results for MSN over Wi-Fi and 3G networks,respectively. The parameters for plotting the analyticalcurves are listed in Table 3. One can see that themeasurement results agree well with the analytical results.It can also be observed that our solutions are able toprovide power saving gain at zero presence update delay.This is due to the fact that the proposed solutions

completely remove the need to initiate presence probesand periodic presence updates from the mobile device.Depending on the user preference, one can determine theoperation mode of the handset by compromising betweenthe tolerable presence update delay and the desired degreeof power saving. In Figs. 9 and 10, we show the results ofthe case in which a mobile client simultaneously logs ontomultiple IM services. For a fair comparison, the total rate ofthe incoming event-triggered presence updates from multi-ple IM services is also set to 1

15 message/second. It isobserved that the proposed solutions can achieve evenhigher power saving for the case of multiple IM services.This is due to the fact that the power consumption resultedfrom the multiprotocol overhead is removed once theproposed solutions are applied.

6 CONCLUSIONS

In this paper, we have investigated the power consumptioncaused by the presence information exchange of IM onmobile devices. To analyze the severity of this problem, we


Fig. 7. The tradeoff between the attainable power saving and the meanpresence update delay for MSN over Wi-Fi networks.

Fig. 8. The tradeoff between the attainable power saving and the meanpresence update delay for MSN over 3G networks.

Fig. 9. The tradeoff between the attainable power saving and the meanpresence update delay for Multi-IM over Wi-Fi networks.

Fig. 10. The tradeoff between the attainable power saving and the meanpresence update delay for Multi-IM over 3G networks.

TABLE 3Modem Transition Delay and Operating

Power for Wi-Fi and 3G Networks

examine the various presence information exchange me-chanisms and derive the respective power consumptionimplications. We have proposed several mobile IM powersaving solutions to reduce the power consumption penalty.The proposed solutions achieve power saving by introdu-cing agents to reduce and regulate the presence informationexchange traffic. The proposed proxy-based IM architecturehas been implemented using the Jabber/XMPP protocol.Actual power measurement experiments have been con-ducted and the results coincide well with our analyticalwork. From our analysis and actual measurement, it isconcluded that

. Due to the inefficient mode transitions in 3G/UMTSnetworks, persistent-IM on 3G mobile handsetsincurs considerable power consumption.

. Significant power saving can be achieved by theproposed solutions on both Wi-Fi and 3G networks.

. The degree of offered power saving can be adjustedaccording to the user preference by trading off thepresence update delay.

The proposed solutions together with the analyzedtradeoff between the presence update delay and the powerconsumption can be applied to implement adaptive powersaving mechanism for mobile IM services by adjusting thetolerable presence update delay in real-time according tothe remaining battery life. The handset battery life can beeffectively extended with the balance between the presenceupdate delay and the attainable power saving.

REFERENCES

[1] R.B. Jennings III, E.M. Nahum, D.P. Olshefski, D. Saha, Z.-Y. Shae,and C. Waters, “A Study of Internet Instant Messaging and ChatProtocols,” IEEE Network, vol. 20, no. 4, pp. 16-21, July/Aug. 2006.

[2] Comsore Presss Release, http://www.comscore.com/press/release.asp?press=800, Apr. 2006.

[3] S.J. Vaughan-Nichols, “Presence Technology: More than JustInstant Messaging,” Computer, vol. 36, no. 10, pp. 11-13, Oct. 2003.

[4] R. Parviainen and P. Pames, “Mobile Instant Messaging,” Proc.10th Int’l Conf. Telecomm., pp. 425-430, Mar. 2003.

[5] 3GPP, “Presence Service Stage 1 (Release 5),” TS 22.141, http://www.3gpp.org/ftp/Specs, 2001.

[6] C. Faure, “Presence Service in 3G Networks,” Proc. Third Int’l Conf.3G Mobile Comm. Technologies, vol. 489, pp. 511-515, May 2002.

[7] J.W. Nah, Y.H. Cho, S.W. Kim, and J.T. Park, “Architecture forExtensible Mobile Instant Messaging and Presence Service overIMS,” Proc. Third Int’l Conf. Comm. Systems Software and Middle-ware, pp. 395-400, Jan. 2008.

[8] P. Salin, “Mobile Instant Messaging Systems—A ComparativeStudy and Implementation,” master’s thesis, Eng. Telecomm.Software and Multimedia Laboratory, Dept. of Computer Science,Helsinki Univ. of Technology, Sept. 2004.

[9] Z. NAOR, “Mobile Instant Messaging Service over CellularNetworks,” Proc. IEEE Global Telecomm. Conf. (GlobeCom),pp. 5319-5323, Nov. 2007.

[10] H. Haverinen, J. Siren, and P. Eronen, “Energy Consumption ofAlways-On Applications in WCDMA Networks,” Proc. IEEEVehicular Technology Conf. (VTC), pp. 964-968, Apr. 2007.

[11] P. Perala, A. Barbuzzi, G. Boggia, and K. Pentikousis, “Theory andPractice of RRC State Transitions in UMTS Networks,” Proc. IEEEGlobeCom Workshops, pp. 1-6, Dec. 2009.

[12] “Extensible Messaging and Presence Protocol (XMPP): Core,”IETF RFC 3920, Oct. 2004.

[13] “Extensible Messaging and Presence Protocol (XMPP): InstantMessaging and Presence,” IETF RFC 3921, Oct. 2004.

[14] “A Model for Presence and Instant Messaging,” IETF RFC 2778,Feb. 2000.

[15] “A Presence Event Package for the Session Initiation Protocol(SIP),” IETF RFC 3856, Aug. 2004.

[16] S.A. Baset and H. Schulzrinne, “An Analysis of the Skype Peer-to-Peer Internet Telephony Protocol,” Proc. IEEE INFOCOM, pp. 1-11, Apr. 2006.

[17] D. Shiu, P.-C. Yeh, L.-S. Meng, K.-C. Chen, and H.-Y. Lo, “LowPower Consumption Solutions for Instant Messaging on MobileDevices,” Proc. IEEE Vehicular Technology Conf. (VTC), pp. 2764-2768, May 2008.

[18] J.R. Smith, “Jingle: Jabber Does Multimedia,” IEEE Multimedia,vol. 14, no. 1, pp. 90-94, Jan.-Mar. 2007.

Ling-San Meng received the BS degree inelectrical engineering from National Tsing HuaUniversity in 2006. Currently, he is workingtoward the PhD degree in the Department ofElectrical Engineering and the Graduate Instituteof Communication Engineering at National Tai-wan University. His research interests includequeuing theory, cooperative communications,and cross-layer design in wireless networks.

Da-Shan Shiu received the BSEE degree fromNational Taiwan University in 1993 and the PhDdegree in electrical engineering and computersciences from the University of California,Berkeley, in 1999. From mid-1999 until mid-2004, he was with Qualcomm, Inc., where hewas involved in the research and developmentof mobile station modem (MSM) supporting dualmode WCDMA/GSM standard. In particular, helead the physical layer algorithm development

for high-speed downlink packet access (HSDPA). He joined theDepartment of Electrical Engineering and the Graduate Institute ofCommunication Engineering of National Taiwan University as anassistant professor in the fall of 2004.

Ping-Cheng Yeh received the BS degree inmathematics and the MS degree in electricalengineering from the National Taiwan Universityin 1996 and 1998, respectively, and the PhDdegree in electrical engineering and computerscience from the University of Michigan, AnnArbor, in 2005. He joined the Department ofElectrical Engineering and the Graduate Instituteof Communication Engineering at the NationalTaiwan University in August 2005. His research

interests include channel coding, coded modulation, directional anten-nas, cooperative communications, and cross-layer design in wirelessnetworks. He has received various teaching awards in the past,including the 2002 EECS Outstanding GSI Award, the 2003 Universityof Michigan Outstanding GSI Award, and the NTU Excellent TeachingAward in 2008 and 2009.

Kuan-Chi Chen received the BS degree inelectrical engineering from National TaiwanOcean University, Keelung, and the MS degreein the communication engineering from theGraduate Institute of Communication Engineer-ing, National Taiwan University, Taipei, in 2006and 2008, respectively. He joined High TechComputer Corporation, Taipei, Taiwan, in 2009.His research interests include the area of codingtheory and digital communications.

Hung-Yi Lo received the BS degree in elec-trical engineering from National Taiwan Univer-sity in 2007. Currently, he is working toward thePhD degree in the Department of ElectricalEngineering at Purdue University. His researchinterests include mobile computing and MIMOcommunications.


ieee xplore full-text pdf

Documents