ESRT: Event-To-Sink Reliable Transport in Wireless Sensor ∗ Networks

8/14/2019 ESRT: Event-To-Sink Reliable Transport in Wireless Sensor Networks

1/12

ESRT: Event-to-Sink Reliable Transport in Wireless SensorNetworks

Yogesh Sankarasubramaniam zgr B. Akan Ian F. AkyildizBroadband & Wireless Networking LaboratorySchool of Electrical & Computer Engineering

Georgia Institute of Technology

{yogi,akan,ian}@ece.gatech.edu

ABSTRACT

Wireless sensor networks (WSN) are event based systemsthat rely on the collective effort of several microsensor nodes.Reliable event detection at the sink is based on collective in-formation provided by source nodes and not on any individ-

ual report. Hence, conventional end-to-end reliability defini-tions and solutions are inapplicable in the WSN regime andwould only lead to a waste of scarce sensor resources. How-ever, the absence of reliable transport altogether can seri-ously impair event detection. Hence, the WSN paradigm ne-cessitates a collective event-to-sink reliability notion ratherthan the traditional end-to-end notion. To the b est of ourknowledge, reliable transport in WSN has not been studiedfrom this perspective before.

In order to address this need, a new reliable transportscheme for WSN, the event-to-sink reliable transport (ESRT)protocol, is presented in this paper. ESRT is a novel trans-port solution developed to achieve reliable event detectionin WSN with minimum energy expenditure. It includes acongestion control component that serves the dual purposeof achieving reliability and conserving energy. Importantly,the algorithms of ESRT mainly run on the sink, with min-imal functionality required at resource constrained sensornodes. ESRT protocol operation is determined by the cur-rent network state based on the reliability achieved and con-gestion condition in the network. If the event-to-sink reli-ability is lower than required, ESRT adjusts the reportingfrequency of source nodes aggressively in order to reach thetarget reliability level as soon as possible. If the reliability ishigher than required, then ESRT reduces the reporting fre-quency conservatively in order to conserve energy while stillmaintaining reliability. This self-configuring nature of ESRTmakes it robust to random, dynamic topology in WSN. An-

This work is supported by the National Science Foundationunder contract ECS-0225497.

Permission to make digital or hard copies of all or part of this work forpersonal or classroom use is granted without fee provided that copies arenot made or distributed for profit or commercial advantage and that copiesbear this notice and the full citation on the first page. To copy otherwise, torepublish, to post on servers or to redistribute to lists, requires prior specificpermission and/or a fee.MobiHoc03, June 13, 2003, Annapolis, Maryland, USA.Copyright 2003 ACM 1-58113-684-6/03/0006 ...$5.00.

alytical performance evaluation and simulation results showthat ESRT converges to the desired reliability with mini-mum energy expenditure, starting from any initial networkstate.

Categories and Subject DescriptorsC.2 [Computer-Communication Networks]: NetworkProtocols, Wireless Communications

General Terms

Algorithms, Design, Reliability, Performance

Keywords

Wireless Sensor Networks, Reliable Transport Protocols, Event-to-Sink Reliability, Congestion Control, Energy Conserva-tion

1. INTRODUCTIONThe Wireless Sensor Network (WSN) is an event driven

paradigm that relies on the collective effort of numerousmicrosensor nodes. This has several advantages over tra-ditional sensing including greater accuracy, larger coveragearea and extraction of localized features. In order to real-ize these potential gains, it is imperative that desired eventfeatures are reliably communicated to the sink.

To accomplish this, a reliable transport mechanism is re-quired in addition to robust modulation and media access,link error control and fault tolerant routing. The function-alities and design of a suitable transport solution for WSNare the main issues addressed in this paper.

The need for a transport layer for data delivery in WSNwas questioned in a recent work [11] under the premise thatdata flows from source to sink are generally loss tolerant.While the need for end-to-end reliability may not exist dueto the sheer amount of correlated data flows, an event inthe sensor field needs to be tracked with a certain accuracyat the sink. Hence, unlike traditional communication net-works, the sensor network paradigm necessitates an event-to-sink reliability notion at the transport layer. This is atruly novel aspect of our work and is the main theme of theproposed Event-To-Sink Reliable Transport (ESRT) proto-col for WSN. Such a notion of collective identification ofdata flows from the event to the sink is illustrated in Fig. 1.

177


2/12

SinkEvent radius

Figure 1: Typical sensor network topology withevent and sink. The sink is only interested in col-lective information of sensor nodes within the eventradius and not in their individual data.

Our work is also motivated by the results in [10], whichemphasize the need for congestion control in WSN. It wasshown in [10] that exceeding network capacity can be detri-mental to the observed goodput. However, the authorsstopped short of providing a solution to this problem.

ESRT is a novel transport solution that seeks to achieve

reliable event detection with minimum energy expenditureand congestion resolution. It has been tailored to match theunique requirements of WSN. Some of its salient featuresare

1. Self-configuration - Reliable event detection must beestablished and maintained in the face of dynamic topol-ogy in WSN. Topology dynamics can result from eitherthe failure or temporary power-down of energy con-strained sensor nodes. Spatial variation of events andrandom node deployment only exacerbate the aboveproblem. ESRT is self-configuring and achieves flexi-bility under dynamic topologies by self-adjusting theoperating point (see Section 4).

2. Energy awareness - Although the primary goal of ESRTis reliable event detection, it aims to accomplish thiswith minimum possible energy expenditure. For in-stance, if reliability levels at the sink are found to be inexcess of that required, the source nodes can conserveenergy by reducing their reporting rate (see Section 4).

3. Congestion Control - Packet loss due to congestion canimpair event detection at the sink even when enoughinformation is sent out by the sources. Hence, con-gestion control is an important component for reli-able event detection in WSN. An important featureof ESRT is that congestion control is also used to re-duce energy consumption. Correlated data flows areloss tolerant to the extent that event features are re-

liably communicated to the sink. Due to this uniquecharacteristic of WSN, required event detection accu-racy may be attained even in the presence of packetloss due to network congestion. In such cases however,a suitable congestion control mechanism can help con-serve energy while maintaining desired accuracy levelsat the sink. This is done by conservatively reducingthe reporting rate. Details of such a mechanism arepresented in Section 4.

4. Collective identification - In typical WSN applications,the sink is only interested in the collective information

provided by numerous sensor nodes and not in their in-dividual reports. In accordance with this, ESRT doesnot require individual node IDs for operation. Thisis also in tune with our proposed event-to-sink modelrather than the traditional end-to-end model. Moreimportantly, this can ease implementation costs andreduce overhead.

5. Biased Implementation - The algorithms of ESRT mainlyrun on the sink with minimum functionalities requiredat sensor nodes. This helps conserve limited sensorresources and shifts the burden to the high-poweredsink. Such a graceful transfer of complexity is possibleonly due to the event-to-sink reliability notion.

We emphasize that ESRT has been designed for use intypical WSN applications involving event detection and sig-nal estimation/tracking, and not for guaranteed end-to-enddata delivery services. Our work is motivated by the factthat the sink is only interested in reliable detection of eventfeatures from the collective information provided by numer-ous sensor nodes and not in their individual reports. This

notion of event-to-sink reliability distinguishes ESRT fromother existing transport layer models that focus on end-to-end reliability. To the best of our knowledge, reliable trans-port in WSN has not been studied from this perspectivebefore.

The remainder of the paper is organized as follows. InSection 2, we present a review of related work in trans-port protocols, both in WSN and other communication net-works, and point out their inadequacies. We formally definethe transport problem in WSN in Section 3 and identifyfive characteristic reliability regions. These regions deter-mine the appropriate actions taken by ESRT. The operationof ESRT is described in detail in Section 4 and a pseudo-algorithm is also presented. ESRT performance analysis andsimulation results are presented in Section 5. Finally, thepaper is concluded in Section 6.

2. RELATED WORKDespite the considerable amount of research on several

aspects of sensor networking, the problems of reliable trans-port and congestion control are yet to be efficiently studiedand addressed. The urgent need for congestion control ispointed out within the discussion of infrastructure tradeoffsfor WSN in [10]. However, the authors do not propose anysolution for the problem they identify.

In another recent work [11], the PSFQ (Pump Slowly,Fetch Quickly) mechanism is proposed for reliable retasking/reprogramming in WSN. PSFQ is based on slowly injectingpackets into the network, but performing aggressive hop-by-hop recovery in case of packet loss. The pump operation inPSFQ simply performs controlled flooding and requires eachintermediate node to create and maintain a data cache tobe used for local loss recovery and in-sequence data deliv-ery. Although this is an important transport layer solutionfor WSN, it is applicable only for strict sensor-to-sensor re-liablity and for purposes of control and management in thereverse direction from the sink to sensor nodes. Event de-tection/tracking in the forward direction does not requireguaranteed end-to-end data delivery as in PSFQ. Individualdata flows are correlated and loss tolerant to the extent that

178


3/12

desired event features are collectively and reliably informedto the sink. Hence, the use of PSFQ for the forward direc-tion can lead to a waste of valuable resources. In addition tothis, PSFQ does not address packet loss due to congestion.

In contrast, ESRT is based on an event-to-sink reliabilitymodel and provides reliable event detection without any in-termediate caching requirements. ESRT also seeks to achieve

the required event detection accuracy using minimum energyexpenditure and has a congestion control component.

A novel transmission control scheme for use at the MAClayer in WSN is proposed in [12] with the main objective ofper-node fair bandwidth share. Energy efficiency is main-tained by controlling the rate at which MAC layer injectspackets into the channel. Although such an approach cancontrol the transmission rate of a sensor node, it neitherconsiders congestion control nor addresses reliable event de-tection. For similar reasons, the use of other MAC protocolslike the IEEE 802.11 DCF or S-MAC [13] that provide someform of hop reliability is inadequate for reliable event detec-tion in WSN.

Next, we briefly examine transport solutions in other wire-

less networks and point out their inadequacies when appliedto WSN. These studies mainly focus on reliable data trans-port following end-to-end TCP semantics and are proposedto address the challenges posed by wireless link errors andmobility [1]. The primary reason for their inapplicability inWSN is their notion of end-to-end reliability. Furthermore,all these protocols bring considerable memory requirementsto buffer transmitted packets until they are ACKed by thereceiver. In contrast, sensor nodes have limited bufferingspace (


4/12

200 sensor nodes were randomly positioned in a 100x100sensor field. Node parameters such as radio range and IFQ(buffer) length were carefully chosen to mirror typical sensormote values [5]. One of these nodes was chosen as the sinkto which all source data was sent. Event centers (Xev, Yev)were randomly chosen and all sensor nodes within the eventradius behave as sources for that event. In order to com-

municate source data to the sink, we employed a simpleCSMA/CA based MAC protocol and Dynamic Source Rout-ing (DSR) [4]. The impact of using other routing protocolson the achieved goodput behavior with reporting period wasshown to be insignificant in [10]. Hence, it is reasonable toassume that the r vs. f behavior and ESRT performanceare insensitive to the underlying routing protocol.

Table 1: NS-2 simulation parametersArea of sensor field 100x100 m2

Numb er of sensor no des 200Radio range of a sensor node 40 m

Packet length 30 bytesIFQ length 65 packets

Transmit Power 0.660 WReceive Power 0.395 W

Decision interval () 10 sec

The results of our study are shown in Fig. 2 for number ofsource nodes n = 41, 52, 62. Note that each of these curveswas obtained by varying the reporting rate f for a certainevent center (Xev, Yev) and corresponding number of sendersn. These values are tabulated in Table 2. The event radiuswas fixed throughout at 30m.

101

100

101

102

0

1000

2000

3000

4000

5000

6000

7000

Reporting frequency (f)

Reliability(r):Numberofreceivedpackets

n=41n=52n=62

Figure 2: The effect of varying the reporting rate, f,of source nodes on the event reliability, r, observedat the sink. The number of source nodes is denotedby n.

We make the following observations from Fig. 2

1. The reliability, r, shows a linear increase (note the logscale) with source reporting rate, f, until a certainf = fmax, beyond which the reliability drops. This isbecause the network is unable to handle the increased

injection of data packets and packets are dropped dueto congestion.

2. Such an initial increase and subsequent decrease in re-liability is observed regardless of the number of sourcenodes, n.

3. fmax decreases with increasing n, i.e., congestion oc-curs at lower reporting frequencies with greater num-ber of sources.

4. For f > fmax, the behavior is rather wavy and notsmooth. An intuitive explanation for such a behavioris as follows. The number of received packets, whichis our reliability, r, is the difference between the totalnumber of source data packets, s, and the number ofpackets dropped by the network, d. While s simplyscales linearly with f, the relationship between d andf is non-linear. In some cases, the difference s d isseen to increase eventhough the network is congested.The important point to note however, is that this wavybehavior always stays well below the maximum relia-

bility at f = fmax

5. The drop in reliability due to network congestion ismore significant with increasing n.

Table 2: Event centers for the three curves withn=41,52,62 in Fig. 2

Number of Event Centersource nodes (Xev,Yev)

41 (88.2,62.8)52 (32.6,79.3)62 (39.2,58.1)

Fig. 3 shows a similar trend between r and f with furtherincrease in n (n = 81, 90, 101). As before, we tabulate theevent centers in Table 3. The event radius was fixed at 40mfor this set of experiments.

The wavy behavior for f > fmax observed in Fig. 2 per-sists in Fig. 3, but appears rather subdued because of muchsteeper drops due to congestion (see observation 5 earlier).All the other trends observed earlier are confirmed in Fig.3.

Table 3: Event centers for the three curves withn=81,90,101 in Fig. 3

Number of Event Centersource nodes (Xev,Yev)

81 (32.6,79.3)90 (61.1,31.5)

101 (60.0,63.6)

3.3 Characteristic RegionsA general trend of initial reliability, r, increase with re-

porting frequency, f, and subsequent decrease due to con-gestion loss is evident from our preliminary studies in Fig.

180


5/12

101

100

101

102

103

0

1000

2000

3000

4000

5000

6000

7000


n=81n=90n=101

Reliability(r):Num

berofreceivedpackets

Figure 3: The effect of varying the reporting rate, f,of source nodes on the event reliability, r, observedat the sink. The number of source nodes is denotedby n.

2 and Fig. 3. This confirms the urgent need for an event-to-sink reliable transport solution with a congestion controlmechanism in WSN. We now take a closer look at the rvs. f characteristics and identify five characteristic regions.As will be seen shortly, these regions are important for theoperation of ESRT.

Consider a representative curve from Fig. 3 for n = 81senders. This is replicated for convenience in Fig. 4. Allour subsequent discussions use this particular case for illus-tration. However, it was verified that the r vs. f behaviorshows the general trend of initial increase and subsequentdecrease due to congestion regardless of the parameter val-

ues. This is indeed observed in Figs. 2 and 3 for varyingvalues ofn. Hence, our discussions and results in this paperapply to a general r vs. f behavior in WSN with any set ofparameter values, with the specific case (n = 81) used onlyfor illustration purposes.

Let the desired reliability as laid down by the applicationbe R. Hence, a normalized measure of reliability is = r

R.

As before, i denotes the normalized reliability at the endof decision interval i.

Our aim is to operate as close to = 1 as possible, whileutilizing minimum network resources (f close to f in Fig.4). We call this the optimal operating point, marked as P1in Fig. 4. For practical purposes, we define a tolerance zoneof width 2 around P1, as shown in Fig. 4. Here, is aprotocol parameter. The suitable choice of and its impacton ESRT protocol operation is dealt with in Section 5.3.

Note that the = 1 line intersects the reliability curve attwo distinct points P1 and P2 in Fig. 4. Though the event isreliably detected at P2, the network is congested and somesource data packets are lost. Event reliability is achievedonly because the high reporting frequency of source nodescompensates for this congestion loss. However, this is awaste of limited energy reserves and hence is not the optimaloperating point. Similar reasoning holds for > 1 + .

From Fig. 4, we identify five characteristic regions (boundedby dotted lines) using the following decision b oundaries

(NC,LR) : f < fmax and < 1 (No Congestion,Low Reliability)

(NC,HR) : f fmax and > 1 + (No Congestion,High Reliability)

(C,HR) : f > fmax and > 1 (Congestion, High

Reliability) (C,LR) : f > fmax and 1 (Congestion, Low Reli-

ability)

OOR : f < fmax and 1 1 + (OptimalOperating Region)

As seen earlier, the sink derives a reliability indicator i atthe end of decision interval i. Coupled with a congestiondetection mechanism (to determine f>< fmax), this can helpthe sink determine in which of the above regions the net-work currently resides. Hence, these characteristic regionsidentify the state of the network. Let Si denote the networkstate variable at the end of decision interval i. Then,

Si

(NC,LR),(NC,HR),(C,HR),(C,LR),OOR

The operation of ESRT is closely tied to the current net-work state Si. The ESRT protocol state model and transi-tions are shown in Fig. 5. We now proceed to discuss thespecifics of ESRT and its operation in each of these statesin detail.

4. ESRT: EVENT-TO-SINK RELIABLE

TRANSPORT PROTOCOLESRT is a novel solution that is proposed to address the

transport problem in WSN. The primary motive of ESRTis to achieve and maintain operation in state OOR. Hence,the aim is to configure the reporting frequency f to achievethe desired event detection accuracy with minimum energy

expenditure. To help accomplish this, ESRT uses a con-gestion control mechanism that serves the dual purpose ofreliable detection and energy conservation.

Recall that the r vs. f characteristic shown in Fig. 4can change with dynamic topology resulting from either thefailure or temporary power-down of sensor nodes. Hence,an efficient transport protocol should keep track of the re-liability observed at the sink and accordingly configure theoperating point. Ifi is within the desired reliability limits(1 i 1 + ) and no congestion notification alert isreceived, then state OOR has been reached and the sinkinforms source nodes to maintain the current reporting fre-quency fi. Here, we make the reasonable assumption thatthe sink is powerful enough to reach all source nodes bybroadcast.

In general, the network can reside in any one of the fivestates Si


.Depending on the current state Si, ESRT calculates an up-dated reporting frequency fi+1, which is then broadcast tothe source nodes. For example, if Si

(NC,LR),(C,LR)

,the observed reliability levels are inadequate to detect thedesired event features. In such a case, ESRT aggressivelyupdates the reporting frequency to reliably track the eventas soon as possible.

This self-configuring nature of ESRT helps it adapt todynamic topology and random deployment, both typical of

181


6/12

101

100

101

102

103

0

0.2

0.4

0.6

0.8

1

1.2

1.4

1.6

1.8


Normalizedreliability()

Required reliability

(NC,LR)

(NC,H

R)

(C,H

R)

(C,LR)

Optimal operating point P1

= (1,f*)

P2

fmax

1+

1

OOR

Figure 4: The five characteristic regions in the normalized reliability, , vs. reporting frequency, f, behavior

WSN. Another important feature of ESRT is its inclinationto conserve scarce energy resources when reliability levelsexceed those required for event detection. This is the casewhen Si

(NC,HR),(C,HR)

. The motivation to re-duce the reporting frequency in this case comes from energyconservation. However, our primary motive of reliable eventdetection must not be compromised. Hence, ESRT takesa conservative approach in this case and decreases f in acontrolled manner.

The algorithms of ESRT mainly run on the sink, withminimal functionality at the source nodes. More precisely,sensor nodes only need the following two additional func-tionalities

Sensor nodes must listen to the sink broadcast at theend of each decision interval and update their report-ing rates

Sensor nodes must deploy a simple and overhead-freelocal congestion detection support mechanism

While the former is an implementation issue and is notwithin the scope of this work, the details of a congestiondetection mechanism are provided in Section 4.2. Such agraceful transfer of complexity from sensor nodes to the sink

node reduces management costs and saves on valuable sen-sor resources. Further simplifying implementation is the factthat ESRT works on the collective identification principleand does not require unique source IDs.

In the following subsection, we discuss the operation ofESRT in each network state and also present a pseudo-algorithm for its implementation.

4.1 ESRT Protocol OperationESRT identifies the current state Si from

Reliability indicator i computed by the sink for deci-sion interval i

A congestion detection mechanism,

using the decision boundaries defined in Section 3.3. De-pending on the current state Si, and the values of fi and i,ESRT then calculates the updated reporting frequency fi+1to be broadcast to the source nodes. At the end of the nextdecision interval, the sink derives a new reliability indica-tor i+1 corresponding to the updated reporting frequencyfi+1 of source nodes. In conjunction with any congestionreports, ESRT then determines the new network state Si+1.This process is repeated until the optimal operating region

182


7/12

(NC,LR) (NC,HR)(C,LR)

(C,HR)

f 1;

f 1;

max

f < f =1;

f > fmax >=1;

f > fmax >1;

max

f < f ;1


8/12

and f fmax. This is because source nodes reportmore frequently than required. The most importantconsequence of this condition is excessive energy con-sumption by sensor nodes. Therefore the reporting fre-quency should be reduced in order to conserve energy.However, this reduction must be performed cautiouslyso that the event-to-sink reliability is always main-

tained. Hence, the sink reduces reporting frequencyf in a controlled manner with half the slope, as op-posed to the aggressive approach in the previous case.Intuitively, we are striking a balance here between sav-ing the maximum amount of energy and losing reliableevent detection. Thus the updated reporting frequencycan be expressed as

fi+1 =fi2

1 +1

i (2)

It is shown in Section 5 that such an update policyreduces the energy consumption in the network anddoes not compromise on event reliability.

3. (C,HR) (Congestion, High Reliability) : In this state,the reliability is higher than required, and congestionis experienced, i.e., > 1 and f > fmax. This is dueto the unique feature of WSN where required event de-tection reliability can be attained even when some ofthe source data packets are lost. In this case ESRT de-creases the reporting frequency in order to avoid con-gestion and conserve energy in sensor nodes. As be-fore, this decrease should be performed carefully suchthat the event-to-sink reliability is always maintained.However, the network operating in state (C,HR) isfarther from the optimal operating point than in state(NC,HR). Therefore, we need to take a more aggres-sive approach so as to relieve congestion and enterstate (NC,HR) as soon as possible. This is achievedby emulating the linear behavior of state (NC,HR)with the use of multiplicative decrease as follows

fi+1 =fii

(3)

It can be shown that such a multiplicative decreaseachieves all objectives (see Section 5).

4. (C,LR) (Congestion, Low Reliability) : In this statethe observed reliability is inadequate and congestionis experienced, i.e., 1 and f > fmax. This isthe worst possible state since reliability is low, conges-tion is experienced and energy is wasted. ThereforeESRT reduces reporting frequency aggressively in or-

der to bring the network to state OOR as soon aspossible. Note that reliability is a non-linear functionof reporting frequency in state (C,LR) as shown inFig. 4. Hence in order to assure sufficient decrease inthe reporting frequency, it is exponentially decreasedand the new frequency is expressed by

fi+1 = f(i/k)i (4)

where k denotes the number of successive decision in-tervals for which the network has remained in state(C,LR) including the current decision interval, i.e.,

k = 1;ESRT()

If (CONGESTION)If ( < 1)/* State=(C,LR) *//* Decrease Reporting Frequency

Aggressively */

f = f/k

;k = k + 1;else if ( > 1)/* State=(C,HR) *//* Decrease Reporting Frequency

to Relieve Congestion; No Compromise onReliability */

k = 1;f = f/;

end;else if (NO CONGESTION)

k = 1;If ( < 1 )/* State=(NC,LR) *//* Increase Reporting Frequency

Aggressively */f = f/;

else if ( > 1 + )/* State=(NC,HR) *//* Decrease Reporting Frequency

Cautiously */

f = f2

1 + 1

;

end;else if (1 1 + )/* Optimal Operating Region *//* Hold Reporting Frequency */

f = f;end;

end;

Figure 6: Algorithm of the ESRT protocol opera-tion.

k 1. The aim is to decrease f with greater aggres-sion if a state transition is not detected. Such a policyalso ensures convergence for = 1 in state (C,LR).

5. OOR (Optimal Operating Region) : In this state, thenetwork is operating within tolerance of the optimalpoint, where the required reliability is attained withminimum energy expenditure. Hence, the reportingfrequency of source nodes is left unchanged for thenext decision interval.

fi+1 = fi (5)

The entire ESRT protocol operation is summarized in thepseudo-algorithm given in Fig. 6

4.2 Congestion DetectionIn order to determine the current network state Si in

ESRT, the sink must be able to detect congestion in thenetwork. However the conventional ACK/NACK-based de-tection methods for end-to-end congestion control purposescannot be applied here. The reason once again lies in thenotion of event-to-sink reliability rather than end-to-end re-liability. Only the sink, and not any of the sensor nodes, can

184


9/12

determine the reliability indicator i and act accordingly.Moreover, end-to-end retransmissions and ACK/NACK over-heads are a waste of limited sensor resources. Hence, ESRTuses a congestion detection mechanism based on local bufferlevel monitoring in sensor nodes. Any sensor node whoserouting buffer overflows due to excessive incoming packetsis said to be congested and it informs the sink of the same.

The details of this mechanism are as follows.In our event-to-sink model, the traffic generated during

each reporting period, i.e., 1/f, mainly depends on the re-porting frequency f and the number of source nodes n. Thereporting frequency f does not change within one reportingperiod since it is controlled periodically by the sink at theend of each decision interval with period of > 1/f. As-suming n does not significantly change within one reportingperiod, the traffic generated during the next reporting pe-riod will have negligible variation. Therefore the amount ofincoming traffic to any sensor node in consecutive reportingintervals is assumed to stay constant. This, in turn, signifiesthat the increment in the buffer fullness level at the end ofeach reporting interval is expected to be constant.

bk1

b

bk

f

B

Figure 7: An illustration of buffer level monitoringin sensor nodes.

Let bk and bk1 be the buffer fullness levels at the end of

kth

and (k 1)th

reporting intervals respectively and B bethe buffer size as in Fig. 7. For a given sensor node, let bbe the buffer length increment observed at the end of lastreporting period, i.e.,

b = bk bk1 (6)

Thus if the sum of current buffer level at the end of kth

reporting interval and the last experienced buffer length in-crement exceeds the buffer size, i.e., bk + b > B, the sensornode infers that it is going to experience congestion in thenext reporting interval. Hence it sets the CN (CongestionNotification) bit in the header of the packets it transmitsas shown in Fig. 8. This notifies the sink for the upcom-ing congestion condition to be experienced in next reportinginterval.

Event

IDFECPayloadCN

(1 bit)

Time

StampDestination

Figure 8: A typical data packet with congestion no-tification field, which is marked to alert the sink forcongestion.

Hence if the sink receives packets whose CN bit is marked,then it infers that congestion is experienced in the last de-cision interval. In conjunction with the reliability indicator

i, the sink can now determine the current network stateSi at the end of decision interval i and act according to therules in Section 4.1.

5. ESRT PERFORMANCEIn this section, we present both analytical and simula-

tion results on the performance of ESRT protocol. Our re-sults show that ESRT converges to state OOR starting fromany of the other four initial network states Si

(NC,LR),(NC,HR),(C,HR),(C,LR) . ESRT is self-configuring inthis sense and can hence perform efficiently under random,dynamic topology frequently encountered in WSN applica-tions.

The convergence times presented in this section are de-rived under the assumption that the r vs.f characteristicdoes not change appreciably within this duration. They canhence be interpreted as achievable lower bounds.

5.1 Analytical ResultsWe first present some analytical results on ESRT perfor-

mance depending on the initial network state S0. Recallthat ESRT aims to reach state OOR starting from any ini-tial state S0.

Lemma 1. Starting fromS0=(NC,HR), and with linearreliability () behavior when the network is not congested,the network state remains unchanged until ESRT convergesto state OOR.

Proof. The linear reliability () behavior for f < fmaxcan be expressed as f = , where denotes the slope.ESRT conservatively decrements f as follows (equation (2))

fi+1 =fi

2

1 +1

i (7)

Hence,

i+1 =1 + i

2(8)

Since fi+1 < fi from (7), it follows that Si

(NC,HR),(NC,LR),OOR , i 0 until ESRT converges. If possi-ble, let Si+1=(NC,LR) when Si=(NC,HR) for some i 0before ESRT converges. Then,

i+1 =1 + i

2< 1 (9)

This implies that i < 12, but i > 1+ since Si=(NC,HR).

Hence, Si=(NC,LR) for any i 0 until ESRT converges.In conjunction with our earlier inference, we conclude thatSi=(NC,HR) i 0, until ESRT converges to state OOR.

Lemma 2. Starting fromS0=(NC,HR), and with linearreliability () behavior when the network is not congested,ESRT converges to state OOR inlog2

01 time units,

where is the duration of the decision interval.

Proof. To establish the convergence time, we proceed asfollows. Let the jth decision interval be the first one whereSj=OOR. It follows from Lemma 1 that j is the least index

185


10/12

such that j < 1 + . Using equation (8),

j =j1+1

2< 1 +

j1 =j2+1

2< 1 + 2

...1 =

0+12

< 1 + 2j1

(10)

Hence, j > log2 01

and the result follows. Note thatthis represents the time required to reach state OOR inorder to conserve maximum energy. Our primary objectiveof reliable event detection is maintained all along by virtueof the conservative decrease (equation (7)).

Lemma 3. With linear reliability () behavior when thenetwork is not congested, the network state transitionSi=(C,HR)Si+1=(NC,LR) is not possible for any i 0.

Proof. The linear reliability () behavior for f < fmaxcan be expressed as f = , where denotes the slope. Itis seen from the r vs. f characteristics in Figs. 2, 3, and 4,that for every f > fmax in state (C,HR), there exists one

f

< fmax (in linear region) such that (f) = (f

).The proof now proceeds by contradiction. Let us assumethat Si+1=(NC,LR) when Si=(C,HR), for some i 0.From the state definitions in Section 3.3 and update policyin Section 4.1, it follows that

fi(1 )

i>

fii

(11)

Hence, a necessary condition is

fi >fi

1 > fi, (12)

but this is not true since fi > fmax > f

i . This completes theproof. In accordance with this result, there is no transitionfrom state (C,HR) to (NC,LR) in the state diagram shown

in Fig. 5. This achieves our objective of relieving congestionand reducing energy consumption while not compromisingon the event reliability (see Section 4.1).

In order to determine the convergence times of the ESRTprotocol starting from S0

(C,HR),(C,LR)

, the non-linear r vs. f behavior needs to be tracked analytically.However, this is beyond our present scope. Hence, we studythe convergence in these two cases using simulations.

5.2 Simulation ResultsIn order to study the convergence of ESRT using simula-

tions, we once again developed an evaluation environmentusing ns-2 [9]. Our convergence results are shown in Figs.9 through 12 for initial network states S0=(NC,LR),(NC,HR),(C,HR), and (C,LR), respectively. The correspond-ing trace values (fi, i) and states are listed within each fig-ure. The energy conservation property of ESRT for S0=(NC,HR) is illustrated in Fig. 13. For all our simulation resultspresented here, number of senders n = 81 and tolerance = 5%. The event radius was fixed at 40m. Other simu-lation parameters are the same as those listed in Table 1 inSection 3.2.

It is seen from Fig. 9 that the ESRT protocol for S0=(NC,LR) converges in a total of two decision intervals (2=20s).This is expected from the aggressive multiplicative policy

0 5 10 15 20 25 30 35 40 45 500

0.2

0.4

0.6

0.8

1

1.2

1.4

1.6

1.8

2

Time (s)

Normalizedreliability(

)

= 10 s

f0

= 0.1

0

= 0.0203

f1

= 4.938

1

= 1.0048

S0

= (NC,LR)

S1

= OOR

Figure 9: The ESRT protocol trace forS0=(NC,LR). Convergence is attained in a to-tal of two decision intervals. The trace values andstates are also shown in the figure.

0 10 20 30 40 50 601

1.1

1.2

1.3

1.4

1.5

1.6

1.7

1.8

1.9

2


Time (s)

= 10 s

f0

= 8.000

0

= 1.6168

f1

= 6.474

1

= 1.3158

f2

= 5.697

2

= 1.1548

f3

= 5.316

3

= 1.0733

f4

= 5.134

4

= 1.0403

S0

= (NC,HR)

S3

= (NC,HR)

S2

= (NC,HR)

S1

= (NC,HR)

S4

= OOR

Figure 10: The ESRT proto col trace forS0=(NC,HR). Convergence is attained in a total offive decision intervals. The trace values and statesare also shown in the figure.

employed. Lemmas 1, 2 and 3 in Section 5.1 can be verifiedfrom the trace values (fi, i) and states listed within Figs.10 and 11.

5.3 Suitable Choice ofFor practical purposes, ESRT uses a tolerance zone of

around the optimal operating point P1 in Fig. 4. If at theend of decision interval i, the reliability i is within [1-,1+]and if no congestion is detected in the network, then the net-work is in state OOR. The event is deemed to be reliablydetected at the sink and the reporting frequency remains un-changed. Greater proximity to the optimal operating pointcan hence be achieved with small . However, as seen fromLemma 2 in Section 5.1, smaller the , greater the conver-gence time. Hence, a good choice of is one that balances thetolerance and convergence requirements. For example, a 1%

186


11/12

Table 4: Summary of ESRT protocol operation in each of the five states

Network State (Si) Description ESRT Action

(NC,LR) No Congestion, Low Reliability Multiplicatively increase fAchieve required reliability as soon as p ossible

(NC,HR) No Congestion, High Reliability Decrease f conservativelyCautiously reduce energy consumption so as not compromise on reliability

(C,HR) Congestion, High Reliability Decrease f aggressively to state (NC,HR) to relieve congestionThen follow action in (NC,HR)

(C,LR) Congestion, Low/equal Reliability Decrease f exponentiallyRelieve congestion as soon as possible

OOR Optimal Operating Region f remains unchanged

0 10 20 30 40 50 60 701

1.05

1.1

1.15

1.2

1.25

1.3

1.35

1.4

1.45

1.5

Time (s)


= 10 s

f0

= 10.000

0

= 1.1365

f1

= 8.799

1

= 1.3478

f2

= 6.529

2

= 1.3218

f3

= 5.734

3

= 1.1630

f4

= 5.332

4

= 1.0828

f5

= 5.128

5

= 1.0418

S0

= (C,HR)

S4

= (NC,HR)

S3

= (NC,HR)

S1

= (C,HR)

S2

= (NC,HR)

S5

= OOR

Figure 11: The ESRT protocol trace for S0=(C,HR).Convergence is attained in a total of six decisionintervals in this case. The trace values and statesare also shown in the figure.

0 5 10 15 20 25 30 35 40 45 500

0.2

0.4

0.6

0.8

1

Time (s)

Normalizedreliability(

)

= 10 s

f0

= 400.000

0

= 0.5465

f1

= 0.038

1

= 0.0203

f2

= 1.868

2

= 0.3848

f3

= 4.858

3

= 0.9923

S

1

= (NC,LR)

S3

= OOR

S2

= (NC,LR)

S0

= (C,LR)

Figure 12: The ESRT protocol trace for S0=(C,LR).Convergence is attained in a total of four decisionintervals in this case. The trace values and statesare also shown in the figure.

tolerance requirement can offset the convergence time by asmuch as 7 time units when S0=(NC,HR). Note however

0 5 10 15 20 25 30 35 40 45 5022

24

26

28

30

32

34

36

Time (s)

Averagepowerconsumption(J)

Figure 13: The average power consumption of sensornodes in each decision interval for S0=(NC,HR).

that reliable event detection is maintained all along (Lemma2 in Section 5.1) due to the conservative decrease.

6. CONCLUSIONThe notion of event-to-sink reliability is necessary for re-

liable transport of event features in WSN. This is due to thefact that the sink is only interested in the collective informa-tion of a number of source nodes and not in individual sensorreports. This is also the reason why traditional end-to-endreliability notions and transport solutions are inappropriatefor WSN. Based on such a collective reliability notion, a newreliable transport scheme for WSN, the event-sink reliabletransport (ESRT) protocol, is presented in this paper.

ESRT is a novel transport solution developed to achievereliable event detection with minimum energy expenditureand congestion resolution functionality. To the best of ourknowledge, this is the first study of reliable transport inWSN from the event-to-sink perspective.

ESRT has been tailored to meet the unique requirementsof WSN. Its congestion control component serves the dualpurpose of achieving reliability and conserving energy. Thealgorithms of ESRT mainly run on the sink and requireminimal functionality at resource constrained sensor nodes.The primary objective of ESRT is to configure the net-work as close as possible to the optimal operating point,where the required reliability is achieved with minimum en-

187


12/12

ergy consumption and without network congestion. Thus,ESRT protocol operation is determined by the current net-work state based on the reliability achieved and the conges-tion condition. In this regard, five possible network statesSi


wereidentified and ESRT operation in each of these states wasdiscussed in detail in Section 4.1. The main ideas are sum-

marized in Table 4.Analytical performance evaluation and simulation results

show that ESRT converges to state OOR regardless of theinitial network state S0. This self-configuring aspect ofESRT is valuable under random, dynamic topology frequentlyencountered in WSN applications. Future work includes ex-tending ESRT to address multiple concurrent events in thesensor field and investigating other possible reliability met-rics.

7. REFERENCES[1] H. Balakrishnan, V. N. Padmanabhan, S. Seshan, R.

H. Katz, A Comparison of Mechanisms for ImprovingTCP Performance over Wireless Links, IEEE/ACM

Trans. Networking, Vol. 5, No. 6, pp. 756-769,December 1997.

[2] S. Floyd, V. Jacobson, C. Liu, S. Macanne, L. Zhang,A Reliable Multicast Framework for LightweightSessions and Application Level Framing, IEEE/ACMTrans. Networking, Vol. 5, No. 6, pp.784-803, Dec.1997.

[3] C. Intanagonwiwat, R. Govindan, and D. Estrin,Directed Diffusion: A Scalable and RobustCommunication Paradigm for Sensor Networks, InProc. ACM MOBICOM 00, Boston, Massachussetts,August 2002.

[4] D. Johnson, D. Maltz, Y. Hu, and J. Jetcheva, TheDynamic Source Routing Protocol for Mobile Ad Hoc

Networks (DSR), http://www.ietf.org/internet-

drafts/draft-ietf-manet-dsr-07.txt, IETF MANETworking group, Internet draft, February 2002.

[5] MICA Motes and Sensors,\http://www.xbow.com/Products/Wireless Sensor

Networks.htm

[6] S. D Servetto, and G. Barrenechea, ConstrainedRandom Walks on Random Graphs: RoutingAlgorithms for Large Scale Wireless SensorNetworks, In Proc. WSNA 2002, September 2002,Atlanta, GA, USA.

[7] E. Shih, S. Cho, N. Ickes, R. Min, A. Sinha, A. Wang,and A. Chandrakasan, Physical Layer DrivenProtocol and Algorithm Design for Energy-EfficientWireless Sensor Networks, In Proc. ACMMOBICOM01, pp. 272-286, Rome, Italy, July 2001.

[8] K. Sohrabi, B. Manriquez and G. Pottie,Near-Ground Wideband Channel Measurements, InProc. IEEE VTC99, New York, 1999.

[9] The Network Simulator - ns-2,\http://www.isi.edu/nsnam/ns/index.html

[10] S. Tilak, N. B Abu-Ghazaleh and W. Heinzelman,

Infrastructure Tradeoffs for Sensor Networks, InProc. WSNA 2002, September 2002, Atlanta, GA,USA.

[11] C. Y. Wan, A. T. Campbell and L. Krishnamurthy,PSFQ: A Reliable Transport Protocol for WirelessSensor Networks, In Proc. WSNA 2002, September2002, Atlanta, GA, USA.

[12] A. Woo, D. E. Culler, A Transmission ControlScheme for Media Access in Sensor Networks, InProc. ACM MOBICOM 2001, pp.221-235, Rome, Italy2001.

[13] W. Ye, J. Heidemann and D. Estrin, AnEnergy-Efficient MAC Protocol for Wireless SensorNetworks, In Proc. INFOCOM02, New York, USA,

June, 2002.

188

Documents

ESRT: Event-To-Sink Reliable Transport in Wireless Sensor ∗ Networks