Computer Networks 55 (2011) 225–240

Contents lists available at ScienceDirect

Computer Networks

journal homepage: www.elsevier .com/ locate/comnet

An event-aware MAC scheduling for energy efficient aggregationin wireless sensor networks

Donggook Kim ⇑, Jaesub Kim 1, Kyu Ho ParkComputer Engineering Research Laboratory, Department of Electrical Engineering and Computer Science,Korea Advanced Institute of Science and Technology (KAIST), Republic of Korea

a r t i c l e i n f o a b s t r a c t

Article history:Received 12 October 2009Received in revised form 31 August 2010Accepted 31 August 2010Available online 8 September 2010Responsible Editor: T. Melodia

Keywords:Wireless sensor networksData aggregationEnergy efficiencyEvent triggered application

1389-1286/$ - see front matter � 2010 Elsevier B.Vdoi:10.1016/j.comnet.2010.08.010

⇑ Corresponding author. Tel.: +82 42 350 5425; faE-mail addresses: dgkim@core.kaist.ac.kr (D. Ki

ac.kr (J. Kim), kpark@ee.kaist.ac.kr (K.H. Park).1 Present address: SK Broadband Co. Ltd., 267 Namd

Seoul 100-711, Republic of Korea.

The data aggregation scheme of wireless sensor networks (WSNs) can reduce the numberof transmitted packets by aggregating multiple packets in one packet; thus, it can reduceenergy consumption at the same time. However, the energy consumption of idle listeningin WSNs remains dominant in the total energy consumption for WSNs. Therefore, the dutycycle MAC protocols have been used to reduce idle listening. However, if aggregation isused on duty cycle MAC protocols, it has very low performance caused by the latency ofthe sleep-wake scheduling.

The goal of the present work is to design energy efficient event-aware MAC schedulingthat can be used by a data aggregation technique. In order to aggregate the packets spa-tially and temporally, two corresponding mechanisms are proposed: the event-awareand energy-aware routing (EE routing) protocol at the routing layer, and the aggregationscheduling MAC (A-MAC) protocol at the MAC layer. The EE routing protocol chooses thebest path for better aggregation and for energy balance in WSNs. The A-MAC protocoladjusts schedules to aggregate more packets while maintaining a low latency.

The performances of the proposed protocols are evaluated through ns2 simulations. Theresults indicate that the proposed protocols reduce energy consumption by 80–90% andhave balanced energy consumption while maintaining similar latencies and aggregationrates compared with previous aggregation protocols.

� 2010 Elsevier B.V. All rights reserved.

1. Introduction

Wireless sensor networks (WSNs) are used widely forsensing various events, such as forest fires, intruders,animal movements, gas leaks, and so on. These events mayoccur infrequently; thus, the sensor nodes should use as lit-tle energy as possible when in an idle state, because mostsensor nodes operate with batteries, which are often diffi-cult to replace or recharge. Once an event occurs, however,multiple nodes near the event send data simultaneously,

. All rights reserved.

x: +82 42 350 3410.m), jskim@core.kaist.

aemoon-ro Joong-gu,

leading to competition between and collisions of data pack-ets. Gupta and Kumar [1] showed that the throughput pernode is Oð 1

ffiffi

np Þ, even under optimal circumstances for n nodes.

As a result, the bursty traffic causes a long latency and highenergy consumption. However, Pottie and Kaiser [2] re-ported that the energy consumption for executing 3000instructions is equivalent to sending 1 bit 100 m by radio;thus, it would be better to reduce the packet size using theCPU resources. Because nodes near the event have depen-dent data, the number of data packets decreases with thedata aggregation scheme, thus reducing the latency and en-ergy consumption of WSNs.

Data aggregation requires cooperation between theapplication layer and the network layer. The applicationlayer aggregates data packets by sharing data packetheaders, compressing data, processing data according to

226 D. Kim et al. / Computer Networks 55 (2011) 225–240

functions (min, max, average, count, event position, etc.),or removing duplications. The network layer assists inthe collection of nearby packets and waits for the arrivalof additional packets using spatial locality and temporallocality, respectively. The focus of this paper is the networklayer that helps the data aggregation of the applicationlayer. Data can be collected at a node with the assistanceof the network layer, and the collected data can be aggre-gated using the data aggregation function of the applica-tion layer. The existing data aggregation protocols in thenetwork layer can be classified [3] into two approaches:structured approaches and structure-free approaches.

Structured approaches [4–12] are suitable for staticdata gathering applications where fixed nodes send dataperiodically, because packets can be easily aggregatedwhen nodes send packets to parent nodes along pre-constructed structures and the parent nodes await thepackets of the child nodes for a fixed time. However, in dy-namic event triggered applications where the traffic pat-terns are time varying, structure-based approaches arenot efficient. Because a parent node does not knowwhether its child nodes have data, it cannot determinethe length of the wait time. A short wait time leads to few-er aggregations and a long wait time leads to a higher la-tency. Furthermore, packets are not aggregated wellbecause the traffic patterns change but the structure isbased on the previous traffic pattern. Therefore, somestructured approaches [4,5] reconstruct the structure forcurrent event distributions, but the overhead of the struc-ture reconstruction and maintenance could be larger thanthe benefit of data aggregation.

To reduce the overhead of structure maintenance, struc-ture-free approaches have recently been proposed forevent triggered applications [13,3,14]. One structure-freeapproach [3] uses a data-aware anycast and random waitapproach, hereafter called the DR protocol. To increasethe aggregation rate, a node attempts to route its packetto another node that has a packet. Because the sender nodedoes not know whether neighbor nodes have packets, theDR protocol uses anycast, which does not specify a destina-tion when sending packets. Multiple receiver nodes com-pete with others, and the ‘winning’ node is chosen as a

Table 1Power consumption of the Mica2 radio (CC1000) and the TelosB radio (CC2420) [

Hardware Communication chip Power consumption

Transmit (mW)

Mica2 [16] CC1000 [17] 31.2

TelosB [18] CC2420 [19] 52.2

Table 2Comparison of the energy consumption without data aggregation (500 packets tranduring 100 s). TX: transmit, IDLE: idle listen.

H/W Energy w/o agg. (J)

TX IDLE Total

Mica2 1.1 1.4 2.5TelosB 2.0 3.5 5.5

receiver node. To collect additional packets, nodes usingthe DR protocol wait for a random amount of time beforebeginning transmission. However, all sensor nodes in theDR protocol must operate on an always awake networkto aggregate or forward data.

Wireless interfaces of sensor nodes consume largeamounts of energy when they are awake although theyare not active, which is called idle listening. Table 1shows the power consumption of various radio statesin CC1000 [17] and CC2420 [19] RF transceivers, whichare widely used in sensor devices. Although the transmit-ting power has the largest consumption, the idle listen-ing power is also large and is similar to the receivingpower.

Table 2 shows the energy consumption of a sender nodewith and without data aggregation in a simple environ-ment. In this test, the sender transmits 500 packets withno aggregation and 50 packets with 90% aggregation over100 s. The energy consumption of the data transmissiondecreases when aggregation is used, but the energy con-sumption of idle listening increases because the idle listen-ing time increases with the reduced number of datatransmissions. As a result, the total energy consumptiondoes not significantly decrease with aggregation. In WSNs,the energy consumption during idle listening is consider-able. Therefore, to achieve energy efficiency in WSNs, idlelistening must be reduced in addition to increasing theaggregation rate.

Several previous works on reducing idle listening havebeen reported. The sensor MAC (S-MAC) [20,21] protocolreduces idle listening by using periodic listen and sleep cy-cles, though it also reduces the bandwidth and causes de-lays at each hop. Thus, the S-MAC protocol has a highlatency when there are many packets to send or whenthe hop-count to the destination increases.

To reduce the latency in multi-hop forwarding, severalmulti-hop scheduling protocols such as the look-aheadschedule MAC (LAS-MAC) [22,23] protocol and the rout-ing-enhanced MAC (R-MAC) [24] protocol have been pro-posed, which reserve packet forwarding schedules acrossmultiple nodes in the listen period and forward data inthe sleep period by following the reserved schedules.

15].

Receive (mW) Listen (mW) Sleep (lW)

22.2 22.2 3

56.4 56.4 3

smission during 100 s) and with data aggregation (50 packets transmission

Energy w/agg. (J)

TX IDLE Total

0.1 (�1.0) 2.1 (+0.7) 2.2 (�0.3)0.2 (�1.8) 5.4 (+1.9) 5.6 (+0.1)

D. Kim et al. / Computer Networks 55 (2011) 225–240 227

However, these protocols do not consider data aggrega-tion nor do they construct an appropriate path for dataaggregation. These protocols reserve multiple data sched-ules independently and send each data piece as soon aspossible; hence, data aggregation does not occur well.Therefore, both data aggregation and the idle listeningproblem must be considered in order to reduce the totalenergy consumption.

To solve the above challenges, an event-aware andenergy-aware routing (EE routing) protocol and an aggre-gation-MAC (A-MAC) protocol is proposed and referred toas the EA protocol hereafter. The goal in this study is todesign energy efficient event-aware MAC scheduling thatcould be used by a data aggregation technique. The EErouting protocol selects the path for better aggrega-tion and energy balance in WSNs. The A-MAC protocol al-lows forwarding schedules across multiple nodes withthe help of the EE routing protocol in the listen periodand sends data according to the reserved schedules inthe sleep period. In the listen period, the nodes thatoverhear the schedules of neighbors opportunisticallyjoin the schedules of neighbors for aggregation (schedulejoin mechanism). The A-MAC protocol creates space be-tween schedules for the potential joining of neighbors(schedule margin mechanism) and rearranges the sched-ules to increase the aggregation rate (schedule shiftmechanism).

This work contributes the following to the aggregationprotocol in WSNs.

1. An analysis of the importance of idle listening in theaggregation protocol.

2. A proposal for methods of generating an aggregationstructure when events occur.

3. A proposal for new multi-hop scheduling approaches toincrease the aggregation rate by adjusting the transmis-sion order.

4. Validation of the operation by means of an extensivesimulation.

The remainder of this paper is structured as follows. Wedescribe the background and related work in Section 2.Section 3 presents the EE routing protocol and the A-MACprotocol. In Section 4, we compare the performance of theEA protocol to that of the DR protocol via a simulation. Fi-nally, Section 5 concludes the paper.

` ` `

` ` `

` ` `

SinkB

C D E

F G H

` ` `

` ` `

` ` `

SinkB

C D E

F G HEvent

` : nodes that have a

: data packet

: pre-constructed ro

AA

Fig. 1. A structure-ba

2. Related work

2.1. Structure-based approaches

Fig. 1 shows the aggregation procedure of a structure-based approach. Structured approaches [4–12] constructa structure that connects all sensor nodes, and fixed nodesperiodically send data to a sink node along the structure instatic data gathering applications. For event triggeredapplications where the traffic pattern changes continu-ously, structured approaches are not appropriate becausestructure-based approaches have a low aggregation ratedue to the data only being sent along pre-constructedstructure, a high latency due to waiting for all data fromthe child nodes, and poor energy balance as a result ofusing a fixed structure in event triggered applications.The overhead of structure reconstruction and maintenancecould be larger than the benefit of data aggregation if thestructure is reconstructed to make a new structure forthe current event distribution.

2.2. Structure-free approaches

In an effort to reduce the overhead of the structure con-struction and maintenance, structure-free approaches havebeen introduced [13,3,14]. Because there is no structure,nodes make two decisions before sending data locally:where to send the data and when to send the data. TheDR protocol [3] uses data-aware anycast and random wait:Fig. 2 shows the aggregation procedure of the DR protocol.

Because the sender node does not know whether theneighbor node has a packet, the DR protocol uses anycast.The sender node sends a Request-To-Send (RTS) packetwith no destination; a receiver node sends a Clear-To-Send(CTS) packet. The receiver nodes wait according to theirpriority before sending the CTS packet. Each node has a dif-ferent priority class according to its position and whetherthe node has data. Class A nodes are nodes that are closerto the sink node than the sender node and that have data toaggregate. Class B nodes are nodes that have data butwhich are further from the sink node. Class C nodes arenodes that have no data but are closer to the sink node.Class A nodes send a CTS packet earlier than class B nodes,and class B nodes send a CTS packet earlier than class Cnodes. When a node hears the CTS packet from othernodes, it cancels its CTS transmission. To avoid collisions

` ` `

` ` `

` ` `

SinkB

C D E

F G H

` ` `

` ` `

` ` `

SinkB

C D E

F G H

packet to send

utes

AA

sed approach.

` ` `

`

` `

SinkB

C D E

F G H

` `

`Event

` ` `

` `

` ` `

SinkB

C D E

F G H

`

` ` `

` `

` ` `

SinkB

C D E

F G H

`

` ` `

` `

` ` `

Sink

C D E

F G H

`

B

` : nodes that have a packet to send

: data packet

: control packet (RTS or CTS)

A A A A

Fig. 2. The DR protocol (structure-free approach).

228 D. Kim et al. / Computer Networks 55 (2011) 225–240

with other nodes in the same class, nodes in the same classtransmit CTS packets after a random delay during the dura-tion of the class.

However, to aggregate or to forward data, the DRprotocol must be operated in an always awake networkbecause receiver nodes should send CTS packets when-ever sender nodes send RTS packets. Thus, nodes in aDR protocol system consume large amounts of energyduring idle listening. Even though the DR protocol re-duces the energy consumption of the data transmissionvia data aggregation, the total energy reduction is not sig-nificant because the energy consumption of idle listeningremains dominant.

The I-FrameComm protocol [14] performs aggregationon the B-MAC protocol [25]. The listen times of sensornodes are not synchronized, so a sender transmits a trailof identical packets for the receiver to catch during itslisten time. When an immediate neighbor node with apacket to aggregate overhears the packet transmission ofthe sender, it sends an interrupt packet to the sender andsends data to the sender for aggregation. The I-Frame-Comm protocol also aggregates data opportunistically,but it has a high latency in a multi-hop network becausethe sender must wait for the receiver nodes to wake upat each hop.

2.3. Approaches for reducing idle listening

Because the power consumption of idle listening ismuch higher than that of the sleeping state, several MACprotocols have been developed to reduce idle listening.The sensor MAC (S-MAC) [20,21] protocol reduces idle lis-tening using periodic listen and sleep cycles. Nodes ex-change synchronization packets, so they wake up andsleep at the same time in a time-synchronized manner.The S-MAC protocol sends control packets and data pack-ets only during the listen period and sleeps during thesleep period. When the listen period is over, the node cansend data in the next listen period. Thus, the S-MAC proto-col has a high latency because it should wait for the nextlisten period when there are many packets to send or whenthe hop-count to the destination is high. Therefore, if theDR protocol is operated on the sleep-wake schedule aswith the S-MAC protocol, it will have low energy consump-tion but a high latency.

Multi-hop scheduling protocols, such as the look-aheadschedule MAC (LAS-MAC) protocol [22,23] and the routing-enhanced MAC (R-MAC) protocol, are also based on peri-odic listen and sleep cycles as in the S-MAC protocol, butthey have lower latency than the S-MAC protocol becausethey send control packets more hops during the listenperiod.

The proposed protocol follows the approach of the mul-ti-hop scheduling protocols, which use a periodic listenand sleep strategies to reduce energy consumption, and re-serve schedules in the listen period to decrease latency.However, they are not designed for aggregation. Fig. 3shows the aggregation procedure of the LAS-MAC protocol.The LAS-MAC protocol selects a next hop without consider-ation of the event distribution. The LAS-MAC protocol doesnot wait to receive other data, but rather makes schedulesto send the packet as soon as possible.

2.4. Summary

Table 3 shows a comparison of the EA protocol with pre-vious aggregation protocols. Most of previous works arequery-based and have a static structure. The DR protocolis an event-based structure-free protocol, but it cannotoperate on the sleep-wake scheduling.

The EA protocol is the first event-based structure-freeprotocol that is operated on the sleep-wake scheduling.The event-based structure-free protocol that is operatedon the sleep-wake scheduling cannot be achieved by sim-ply combining the multi-hop scheduling protocol and theDR protocol. The DR protocol has the time overhead forwaiting for the CTS packet and for waiting for the packetof neighboring nodes, so it consumes a great deal of timeto transmit a control packet at each hop. Thus, the DR pro-tocol cannot be operated well with the multi-hop schedul-ing protocol and has high latency. Moreover, the multi-hopscheduling protocols are not appropriate for aggregationbecause they only make schedules for the packetforwarding.

Therefore, we designed the EA protocol to utilize theadvantage of the multi-hop scheduling protocol. The EAprotocol selects a routing path in the direction that eventsmay occur and reserves schedules for aggregating the datapackets of neighboring nodes by using event prediction.Neighboring nodes can join the schedules after hearing

` ` `

`

` `

S

C D E

F G H

` `

`Event

` ` `

`

` `

SinkB

C D E

F G H

` `

`

` ` `

`

` `

SinkB

C D E

F G H

` `

`

` ` `

`

` ` `

SinkB

C D E

F G H

` `

SLEEP

` : nodes that have a packet to send

: data packet

: control packet (RTS or CTS)

A B A A A

Fig. 3. The LAS-MAC protocol (multi-hop scheduling approach).

Table 3Comparison of previous aggregation protocols.

Protocol Year Cause Duty-cycled Structure

TAG [6] 2002 Query-based Yes StaticSynopsis-Diffusion [11] 2004 Query-based Yes StaticRideSharing [12] 2006 Query-based Yes StaticDR [13] 2007 Event-based No DynamicEA – Event-based Yes Dynamic

D. Kim et al. / Computer Networks 55 (2011) 225–240 229

the neighboring schedules. Furthermore, the schedules canbe modified dynamically when the event prediction iswrong.

3. Event-aware and energy-aware routing protocol andaggregation-MAC protocol

3.1. Overview

In this paper, an EA protocol is proposed: it is a hybrid ofthe event-aware and energy-aware routing (EE routing)protocol and the aggregation-MAC (A-MAC) protocol. Thereare two decisions for data aggregation:

� Where to send the data for spatial locality– The EE routing protocol: Nodes decide the next hop

that is likely to have data to be aggregated andretains a large amount of energy.

� When to send the data for temporal locality– Schedule margin of the A-MAC protocol: When a node

makes a schedule, it creates an empty space insidethe schedule. The node predicts the number of joinsof neighbor nodes and uses it for the size of theempty space.

– Schedule join of the A-MAC protocol: When the nodeoverhears schedules from neighbor nodes, it can joinan empty space of ongoing schedules. This mecha-nism is called schedule join.

– Schedule shift of the A-MAC protocol: When there isinsufficient space to accept the join request of neigh-bor nodes with incorrect predictions, the node shiftsthe schedules to accept the join request of neighbornodes.

In the listen period, the A-MAC protocol reserves sched-ules and the multi-hop path is determined by the EE rout-ing protocol. When the nodes make schedules, they setempty space, called a schedule margin, inside the schedulesfor the possible joins of neighbors. Neighbor nodes thatoverhear the control packets try to join those schedules.The A-MAC protocol rearranges the schedules using theschedule shift to increase the aggregation rate if needed.In the sleep period, only those nodes involved wake up,send, receive, and aggregate the data packets accordingto the schedule. The detailed operation will be explainedin the following sections.

3.2. Event-aware and energy-aware routing (EE routing)protocol

The event-aware and energy-aware routing (EE routing)protocol is used to select paths for aggregation and forenergy balance. It is assumed that the nodes know theircoordinates from a global positioning system (GPS) orother localization protocols [26,27]. There are two primarypurposes of the EE routing protocol: to increase the aggre-gation rate and to balance the energy consumption distri-bution. To increase the probability of data aggregation,the EE routing protocol chooses the next hop that is mostlikely to have data. In order to find the nodes that are likelyto have data to send, a node traces the event rate informa-tion of its neighbor nodes. The event rate is the averagenumber of event occurrences in the unit time. In order tobalance the energy consumption distribution, the EE rout-ing protocol chooses the next hop that has a larger energybecause a node that aggregates the data of its neighbornodes consumes more energy to gather and aggregate

NodeRouting path

A

Mini-map

KF

230 D. Kim et al. / Computer Networks 55 (2011) 225–240

the data packets. To achieve these purposes, the EE routingprotocol operates as follows:

1. Using a neighbor table, a node finds candidate nodesthat are closer to the sink node than itself.

2. Among the closer nodes, it selects candidate nodes thathave a higher event rate than the average event rate ofthe candidate nodes.

3. Among the nodes with above average event rates, thenode selects the next hop nodes that have higherremaining energies than the average remaining energyof the candidate nodes.

4. If there are two or more candidate nodes, the nodeselects the node nearest to the sink node.

With the EE routing protocol, a node selects the next hopthat has higher probability to have a packet to send and ahigher remaining energy. In order to trace the event rate ofits neighbor nodes, the node sends a control packet with aflag when it has a packet to send. The other nodes checkthe control packet during the listen period and accumulatethe number of event occurrences in their neighbor tables.The event occurrence value of the neighbor table is up-dated using the exponentially weighted moving average(EWMA). Then, the event rate of a sensor node i (ERi) is rep-resented as:

ERi ¼ ERi � ð1� aÞ þ ERcurrenti � a; ð1Þ

where ERcurrenti is the observed event rate in the currentlisten period. A weighting factor, a, is chosen by the useraccording to the event characteristics such as the eventspeed. The remaining energies of the neighbor nodes areacquired from the synchronization packets of neighborswith additional remaining energy information. Packetsare forwarded to a sink node when there are nodes closerto the sink node than the sender node. When there areholes in a network and the sender does not have a nodethat is closer to the sink node, it avoids the holes usingthe perimeter mode forwarding from the Greedy PerimeterStateless Routing (GPSR) protocol [28].

A B C

R N

List

en p

erio

dSl

eep

perio

dN

time

S R N

S R

A-RTS

A-RTS

S : send

R : receive

N : nav

ScheduleDATA

ACK

ACK

DATA

A-RTS(A)

A-RTS

B

C

E

D

L

M

O

N

G

H

J

I(B) (B)

(C)

(B) (B) (B)

(B) (C) (C)

(D)

(E) (F)

(G)

1

2

3

4

5

6

Sch

edul

e sl

ot n

umbe

r

Fig. 4. Schedule setup process. A-RTS: the Aggregation-Request-To-Sendpacket, DATA: the data packet, ACK: the acknowledgment packet.

3.3. Schedule setup of the aggregation-MAC (A-MAC) protocol

Each node maintains its schedule table, which main-tains the reserved operations for the sleep period. Forsimplicity, it is assumed that applications process dataaccording to functions such as MIN, MAX, AVERAGE, COUNT,and EVENT POSITION, in which many packets can be aggre-gated into one packet. The sleep period is divided intoidentically sized schedule slots. The schedule slot is usedinstead of the schedule time in order to decrease the over-head of the control packet by reducing the packet size. Thenode forwards the data packet after a Short Inter-FrameSpace (SIFS), which is a sufficiently long time to processthe packet, to aggregate the packet, and to switch radiomodes (from a receiving state to a transmitting state).The schedule slot size is the sum of the data transmissiontime, acknowledgment transmission time, and SIFS aftereach process.

There are three types of schedules: S (send), R (receive),and N (network allocation vector). In the S schedule, thenode wakes up, sends a packet, receives an acknowledg-ment, and sleeps. In the R schedule, the node wakes up, re-ceives a packet, sends an acknowledgment, and sleeps. Inthe N schedule, the node sleeps to avoid collisions withthe neighboring S and R schedules.

The schedules are established by forwarding an aggre-gation RTS (A-RTS) packet. Because the wireless transmis-sion has broadcast characteristics, the A-RTS packet ofnodes in a multi-hop transmission can be simultaneouslyused as a CTS packet and an RTS packet in order to reducethe redundant header fields of the CTS and RTS packets.The A-RTS packet contains a final destination address, aschedule slot number, a previous node address, and a pre-vious schedule slot number in addition to the original RTSpacket. The final destination address is used to find thenext forward hop using the EE routing protocol. Becausethe MAC protocols do not know the next hop node forthe final destination address, the EE routing protocolreceives the final destination address from the MAC layerand returns the next forward hop to the MAC layer whenthe node receives the A-RTS packet. The schedule slotnumber is used to reserve the new S schedule of the cur-rent node, to reserve the new R schedule of the target node,and to make the N schedules of the sender node’s neigh-boring nodes. The previous node address and the previousschedule slot number are used to confirm the S schedule ofthe previous hop node, to confirm the R schedule of thecurrent node, and to make the N schedules of the receivernode’s neighboring nodes.

Fig. 4 shows the process of the schedule setup in the lis-ten period and the data transmission in the sleep period.The node schedule tables according to the process aredepicted in Table 4. This process completes the following:in the listen period, nodes exchange the A-RTS packets and

Table 4Schedule tables of nodes. Schedule is hoperation, schedule slot number,target nodei. S is send, R is receive, and N is Network Allocaion Vector (NAV)operations. [ , , ] means estimated schedule with no confirmation, whichturns to the confirmed schedule after receiving the acknowledgmentpacket.

Node (A) A-RTS (B) A-RTS (C) A-RTS

Node A [S, 1, B] hS, 1, Bi, hN, 2, Bi hS, 1, Bi, hN, 2, BiNode B [R, 1, A] hR, 1, Ai, [S, 2, C] hR, 1, Ai, hS, 2, CiNode C hN, 1, Bi, [R, 2, B] hN, 1, Bi, hR, 2, Bi

List

en p

erio

dSl

eep

perio

d

A B C

S R N

S R

A-RTS

G

N

N

schedulemargin

time

NodeRouting path

S : send

R : receive

N : nav

Schedule

A-RTSA-RTS

A-RTS

A

B

C

E

D

K

L

M

O

N

F

G

H

J

I

Schedule margin

Schedule margin1

2

3

4

5

6

Sch

edul

e sl

ot n

umbe

r

(A)(B) (B)

(C)

(A)(B) (B) (B)

(B) (C) (C)

Fig. 5. Schedule setup with schedule margin.

D. Kim et al. / Computer Networks 55 (2011) 225–240 231

reserve schedules in their schedule tables, as shown in pro-cedures from (A) to (C) in Fig. 4 and Table 4.

(A) Node A sends the A-RTS packet. Node B receives theA-RTS packet from node A.1. Node A finds that the next hop is node B with the

final destination node E from the EE routing pro-tocol. Node A finds that Slot 1 is empty in itsschedule table and reserves an estimated S sche-dule in Slot 1. It sends an A-RTS packet, whichcontains the information: the sender node isnode A, the target node is node B, the final desti-nation node is node E, and the schedule slotnumber is 1.

2. When node B receives the A-RTS packet fromnode A, it reserves an estimated R schedule inSlot 1 if there are no prior schedules in Slot 1.The estimated schedule is kept until node Bsends the A-RTS packet as a confirmationpacket.

(B) Node B sends the A-RTS packet. Node A and node Creceive the A-RTS packet from node B.1. Node B finds that the next hop is node C with the

final destination node E from the EE routing pro-tocol. Node B finds that Slot 2 is empty in itsschedule table and reserves an estimated Sschedule in Slot 2. It sends the A-RTS packet withthe information that the sender node is node B,the target node is node C, the final destinationnode is node E, the schedule slot number is 2,the previous node address is node A, and the pre-vious schedule slot number is 1. Node B changesthe estimated R schedule to the confirmed Rschedule in Slot 1 in its schedule table if the A-RTS packet is transmitted without a problem.

2. When node A receives the A-RTS packet fromnode B, it changes the estimated S schedule tothe confirmed S schedule in Slot 1, because theboth nodes agree on this schedule. Node A makesthe N schedule in Slot 2 in its schedule table, inorder to avoid collisions during node B’stransmission.

3. When node C receives the A-RTS packet fromnode B, it reserves an estimated R schedule inSlot 2 and makes the N schedule in Slot 1 in itsschedule table, in order to avoid collisions duringthe reception of the data from node B.

(C) Node C sends the A-RTS packet. Node B receives theA-RTS packet from node C.

1. Node C sends the A-RTS packet with the informa-tion that the sender node is node C, the previousnode address is node B, and the previous sche-dule slot number is 2. Node C changes the esti-mated R schedule to the confirmed R schedulein Slot 2 in its schedule table.

2. When node B receives the A-RTS packet fromnode C, it changes the estimated S schedule tothe confirmed S schedule in Slot 2 in its scheduletable.

This procedure continues until the listen period ends orthe A-RTS packet reaches its destination. In the sleep per-iod, all nodes sleep and behave according to theirreserved schedules, as shown in procedures from (D) to(G) in Fig. 4.

(D) Node A sends a data packet to node B at the time ofSlot 1.

(E) Node B sends an acknowledgment packet to node A.Node A then sleeps.

(F) Node B sends a data packet to node C at the time ofSlot 2.

(G) Node C sends an acknowledgment packet to node B.Node B and node C then go to sleep.

After the nodes operate according to the reservedschedules, they go to sleep until the next listen period.

3.4. Schedule margin

The A-MAC protocol makes space, called schedule mar-gins, between schedules to accept the possible joins ofneighbor nodes with minimal schedule modification. Asshown in Fig. 5, node B makes its S schedule in Slot 3 afterthe schedule margin of Slot 2 from the R schedule in Slot 1.The schedule margin is the duration during which the pre-dicted nodes send data. By using the collected event rateinformation again, which is used in the EE routing protocol,

List

en p

erio

dSl

eep

perio

dtim

e

S : send

R : receive

N : nav

Schedule

A B C

A-RTS

S R N

S R

A-RTS

G

N

N

NodeRouting path

A-RTSA-RTS

A

B

C

E

D

K

L

M

O

N

F

G

H

J

I

Mini-map

1

2

3

4

5

6

Sch

edul

e sl

ot n

umbe

r

(A)

(B) (B)

(C)

(A)(B) (B) (B)

(B) (C) (C)

A-RTS

A-RTSA-RTS

List

en p

erio

dep

per

iod

S : send

Schedule

A B C

N

S R N

S R

G

S

N

N

R

Schedule join

N

A-RTS

A

B

C

E

D

K

L

M

O

N

F

G

H

J

I

Schedule join

Schedule join

1

2

3

slo

t num

ber

(E) (E)

(D)

(E) (E) (E)(D)

(D)

232 D. Kim et al. / Computer Networks 55 (2011) 225–240

the node can predict the number of neighbors that mayhave data to send. From the event rate information of theneighbor table of node B, node B predicts that there isone node, node G, which may join node B. Therefore, nodeB sets a schedule margin of one slot.

The expected number of neighbors that will send data iscalculated from the number of neighbor nodes that have anevent rate over the threshold. In the proposed protocol, thethreshold value is the average event rate of all neighbornodes. The administrator can adjust the threshold valueaccording to the characteristics of the application. Thesmaller the threshold is, the higher the schedule marginis; thus, the protocol will have a high aggregation rate,but will also have a high average latency. The larger thethreshold value is, the smaller the schedule margin is, andthe aggregation rate will be lower. However, a largerthreshold value is preferred because the position of theschedule can be adjusted with the schedule shift mecha-nism, which will be explained in Section 3.6.

3.5. Schedule join

When node B exchanges the A-RTS packet to makeschedules, node G will overhear the A-RTS packet of nodeB and will know the schedules of the neighboring node B.As can be seen in Fig. 6, node G can join the empty slot cre-ated by the schedule margin of node B with an A-RTS packetwith the information that the sender node is node G, thetarget node is node B, and the schedule slot number is 2.The A-RTS packet does not need to be forwarded to the fi-nal destination, node E, further than node C, because thepacket of node G will be aggregated with that of node B,and node B has already reserved the schedule for the finaldestination, node E.

3.6. Schedule shift

When there are more schedule joins than expected, anode loses the chance to aggregate more data packets.

A-RTS

List

en p

erio

dSl

eep

perio

d

A B C

N

S R N

S R

G

S

N

N

R

Schedule join

time

NodeRouting path

S : send

R : receive

N : nav

Scheduleschedulemargin N

A-RTSA-RTS

A

B

C

E

D

K

L

M

O

N

F

G

H

J

I

Schedule join

Schedule join

A-RTS

1

2

3

4

5

6

Sch

edul

e sl

ot n

umbe

r

(D) (D)

(E) (E)

(D) (E) (E) (E)

Fig. 6. Schedule join.

Thus the schedule shift mechanism that shifts the schedulein order to aggregate more packets is suggested for imple-mentation in this process. In Fig. 7(a), node B makes the Sschedule for node C at Slot 2 with the assumption thatthere is no join of neighboring nodes. In Fig. 7(b), node Breceives the schedule join of node G and reserves the Rschedule from node G in Slot 3. If node B does not modifyits schedules, node B will receive a data packet from nodeA, send it to node C, and then receive another data packet

Sle

time

R : receive

N : nav

4

5

6

Sch

edul

e

A-RTS

List

en p

erio

dSl

eep

perio

dtim

e

S : send

R : receive

N : nav

Schedule

A B C

N

S R N

S R

G

S

N

N

R

S R

Schedule shift

A-RTS

Routing path

N

A-RTS

N

A

B

C

E

D

K

L

M

O

N

F

G

H

J

I

Schedule shift

Schedule shift

1

2

3

4

5

6

Sch

edul

e sl

ot n

umbe

r

(E) (E)

(F)

(E) (E) (E)

(E) (F) (F)

Fig. 7. An example of schedule shift.

D. Kim et al. / Computer Networks 55 (2011) 225–240 233

from node G later; thus, it cannot aggregate the two datapackets.

Therefore, if node B can send an additional A-RTS packetthat updates its schedule in the listen period, it sends an A-RTS packet that is used to shift the schedule in order tosend the data packets after aggregation of the two packets.Node B sends an A-RTS packet with the information thatthe sender node is node B, the target node is node C, andthe schedule slot number is 4. When node C receives theA-RTS packet from node B, it already has the R schedulefrom node B at Slot 2. So, node C cancels the previous Rschedule for Slot 2 and makes a new R schedule in Slot 4.As shown in Fig. 7(c), with the schedule shift mechanism,node B moves the S schedule from Slot 2 to Slot 4, and nodeC moves the R schedule from Slot 2 to Slot 4, thus sendingthe data packet after receiving all other data packets dur-ing the sleeping time. Therefore, although the number ofschedule joins exceeds the expected number, the scheduleshift mechanism adjusts the previous schedules with alow overhead, so the A-MAC protocol achieves a higheraggregation rate.

3.7. Summary

Fig. 8 shows the data aggregation procedure of the EAprotocol. As shown in Fig. 8(a), the EA protocol reservesthe schedule with control packets along the multi-hoppath with the assistance of the EE routing protocol. The pathis selected for aggregating more packets and for energybalance. The nodes make a schedule with a schedule mar-gin, so neighbor nodes can join to the schedule. Node Doverhears the A-RTS packet of node E and joins the sched-ules, as shown in Fig. 8(b). In the sleep period, the nodessend data packets according to the reserved aggregationschedule as shown in Fig. 8(c). As shown in Fig. 8(d), dataaggregation occurs in node E with data transmissionaccording to the schedules.

3.8. Discussion

The event rate is updated when the neighboring sensornodes that detect the event send the A-RTS packet via pig-gybacking whether the event occurs or not. Thus, the event

` ` `

`

` `

Sink

C D E

F G H

` `

`Event

` ` `

`

` `

Sink

C D E

F G H

` `

`

B B

` : nodes that have a

: data packet

: control packet (A-

A A

Fig. 8. The summary o

rate is not increased for the intermediate nodes, which donot detect the event but only forward the A-RTS packet.The event rate is updated only when the event is detectedwithout regard for the duplicate-sensitive aggregation orthe duplicate-insensitive aggregation. This is because anode can select as the next hop node any among the nodesthat have duplicate data packets. Whether the data packetis duplicated or not is judged not when the routing pathand the schedule are made using the A-RTS packets, butwhen the actual data packets are transmitted and col-lected. Therefore, the A-RTS packet is transmitted to thedirection in which more events can be collected by usingonly the event rate. The goal of the transmission of A-RTSpackets is to collect the events that will be transferred tothe same sink.

4. Performance evaluation

4.1. Simulation setup

The performance of the EA protocol was simulated usingthe ns2 network simulator (version 2.33) [29] and wascompared with that of the DR protocol, because the DRprotocol has a lower latency and higher aggregation ratiothan the structured approach [3]. In this simulation, allprotocols aggregate every packet of events. The networkis a 900 m by 900 m square region with a grid topology,which consists of 49 nodes separated by 140 m. The sinknode is located at one corner of the grid network. The cen-ter of the event moves according to the random way pointmodel. Unless otherwise stated, the event moving speed is30 m/s for 1900 s with a pause time of 0 s, the radius of theevent is 200 m; and the event generation interval is 10 s. Ifnodes are inside the event radius, they generate data pack-ets with a 50 bytes payload periodically according to theevent interval and send data packets to the sink node.The duty cycle is set to 10%; the listen period is 591 ms;the synchronization period is 55.2 ms; and the sleep periodis 5.231 s. A weighting factor, a, of 0.5 is used because thelocation where the event occurs continuously changes.

The simulation included a sensor node with the TwoRay Ground radio propagation model in the air, a singleomni-directional antenna, and 20 kbps wireless interface

` ` `

`

` `

Sink

C D E

F G H

` `

`SLEEP

` ` `

`

` ` `

SinkB B

C D E

F G H

` `

SLEEP

packet to send

RTS)

AA

f the EA protocol.

0

5

10

15

20

25

30

35

40

45

50

0 20 40 60 80 100

Aver

age

late

ncy

(s)

Event intervals (s)

DR on SMAC

LAS-MAC

DR

EA

Fig. 9. Average latency.

234 D. Kim et al. / Computer Networks 55 (2011) 225–240

with a 250 m transmission range and a 550 m carrier sens-ing range, which works in a similar manner to the 914 MHzLucent WaveLAN direct sequence spread spectrum (DSSS)radio interface. To evaluate the energy consumption, theenergy model used had Ptransmission = 24 mW, Preceive =13 mW, and Pidle = 13 mW, where P represents power;these parameters are sourced from Model TR1000 of RFMonolithics Inc. [30].

The sizes and transmission times of packets used in thesimulations are presented in Table 5. The nodes sense car-riers during a Distributed Inter-Frame Space (DIFS) and arandom time lower than the contention window (CW) asin the IEEE 802.11 protocol. In the simulation, DIFS is10 ms, SIFS is 5 ms and the CW is 64 ms. The schedule slottime is the duration of the data packet and the acknowl-edgment packet including the SIFS after each packet, whichis 64 ms.

To compare these protocols, the three metrics listed be-low are used:

� Average latency: The average delay of all received pack-ets at the sink.� Aggregation rate: The ratio of aggregated packets to total

received packets at the sink. The aggregation rate willbe higher if more packets are aggregated.� Power consumption: The average power consumption of

all nodes.

The performance of the DR protocol depends on themaximum random wait value. The larger the maximumrandom wait value is, the higher the average latency is.However, the smaller the maximum random wait valueis, the lower the aggregation rate is. A maximum randomwait value of 2 s was used because the DR protocol sug-gests 2 s as a default value for the random wait time [3].

100

4.2. Simulation results

In this section, the results for different event intervalsare shown. The event interval varies from 2 s to 100 s. Toreduce the total energy consumption, the DR protocolcan be operated on a periodic listen and sleep scheme,hereafter called the DR-on-SMAC protocol. The EA protocolis also compared with the LAS-MAC protocol and the DR-on-SMAC protocol. The size of the LAS-RTS packet of theLAS-MAC protocol is 14 bytes. The LAS-MAC protocol usesthe GPSR routing protocol [28] as its routing protocol.

40

60

80

gatio

n ra

te (%

)

DR on SMAC

EA

4.2.1. Average latencyIn Fig. 9, the EA protocol has the lowest latency among

the protocols compared. The DR protocol and the LAS-MAC protocol has two times higher latency than the EA

Table 5The size and the transmission latency of packets.

Packet Size (bytes) Transmission time (ms)

RTS/CTS/ACK/SYNC 10 11.0A-RTS 16 15.8DATA 50 43.0

protocol, and the DR-on-SMAC protocol has seven timeshigher latency than the EA protocol. For the DR protocol,the data-aware anycast mechanism takes an inefficientlonger routing path to the sink node to aggregate morepackets and the random wait mechanism increases the de-lay to aggregate possible packets. The DR-on-SMAC proto-col has a much higher latency than the DR protocol due tothe latency of the sleep-wake scheduling. The DR protocoland the DR-on-SMAC protocol attempt to aggregate thedata packets of neighbor nodes even though the data pack-et is located in a backwards direction; thus, they havehigher average latency as the event interval decreases be-cause another data packet is generated at neighbor nodesbefore they send the data packet to the sink node. How-ever, the EA protocol has a relatively low and constant aver-age latency because the EA protocol reserves schedules tothe sink node first, aggregates packets of neighboringnodes opportunistically in the listen time, and sends dataaccording to schedules in the sleep time. The LAS-MACprotocol has a higher average latency than the EA protocolbecause more time is needed to transfer data packets withlower aggregation.

0

20

0 20 40 60 80 100

Agg

re

Event intervals (s)

DR

LAS-MAC

Fig. 10. Aggregation rate.

0

20

40

60

80

100

120

Aver

age

late

ncy

(s)

Event radius (m)

DR on SMACDREA

0

10

20

30

40

50

60

70

80

90

100

Agg

regr

atio

n ra

te (%

)

Event radius (m)

DR on SMACEADR

0 100 200 300 400 500 600 700 800

0 100 200 300 400 500 600 700 800

D. Kim et al. / Computer Networks 55 (2011) 225–240 235

4.2.2. Aggregation rateIn Fig. 10, the LAS-MAC protocol has the lowest aggre-

gation rate because it forwards data to the sink node with-out considering aggregation. As the event intervaldecreases, the number of packets to aggregate in the samenode also increases, so the aggregation rate of the LAS-MAC protocol increases. The DR-on-SMAC protocol hasthe highest aggregation rate because the node does not for-ward data packets to the sink node, but rather forwardsdata packets to neighboring nodes that have data packets,even if it is located away from the sink node. Therefore,even though the DR-on-SMAC protocol has the highestaggregation rate, it has the longest average latency. TheEA protocol has a similar aggregation rate to that of theDR protocol because the EA protocol chooses a path basedon the event generation rate of neighboring nodes andreorders schedules.

4.2.3. Power consumptionThe power consumption of the EA protocol, the LAS-

MAC protocol, and the DR-on-SMAC protocol is eight timeslower than that of the DR protocol, as depicted in Fig. 11.The DR protocol consumes a large amount of energy be-cause it is always awake. The EA protocol, the LAS-MACprotocol, and the DR-on-SMAC protocol use a much lowerenergy, because they reduce idle listening by periodic lis-ten and sleep cycles.

In summary, the EA protocol has a lower latency thanthe DR protocol and a similar aggregation rate to the DRprotocol, while it consumes a much lower amount of en-ergy than the DR protocol. Although the EA protocol hasan overhead in obtaining the neighbor information, theoverhead of the EA protocol is similar to that of theS-MAC protocol and the LAS-MAC protocol. Most of the ob-tained information is used to synchronize nodes to sleepand wake at the same time in a time-synchronized man-ner. The additional event count information occupies 1bit in the packet size and about 1 byte in the memory ofthe sensor node. Another overhead is that the A-RTS packetsize is larger than the control packet of other protocols be-cause it includes more entries, such as the final destinationaddress and the previous node address. However, by using

0

0.002

0.004

0.006

0.008

0.01

0.012

0.014

0 20 40 60 80 100

Pow

er c

onsu

mpt

ion

(W)

Event intervals (s)

DR

DR on SMAC

LAS-MAC

EA

Fig. 11. Power consumption.

the A-RTS packet for the RTS and CTS packets simulta-neously, the overhead of the larger size control packetcould be decreased. Furthermore, a schedule slot (1 byte)is used instead of a schedule time (2 bytes) by dividingthe synchronized sleep time to equal sized slots. Therefore,the EA protocol including the overhead has a lower latencyand lower energy consumption than other protocols.

0

0.002

0.004

0.006

0.008

0.01

0.012

0.014

0 100 200 300 400 500 600 700 800

Pow

er c

onsu

mpt

ion

(W)

Event radius (m)

DRDR on SMACEA

Fig. 12. Results for different event sizes.

236 D. Kim et al. / Computer Networks 55 (2011) 225–240

4.2.4. Effect of event radiusIn this simulation, the performance of the EA protocol,

DR protocol and DR-on-SMAC protocol were evaluated byvarying the event radius from 100 m to 900 m. The averagelatency with the varying event radius is depicted inFig. 12(a). As the event radius enlarges, the number ofpackets in the network also increases. The DR-on-SMAC

0

2

4

6

8

10

12

Aver

age

late

ncy

(s)

Event speed (m/s)

DR EA

82

83

84

85

86

87

88

89

Agg

rega

tion

rate

(%)

Event speed (m/s)

DR EA

0

0.002

0.004

0.006

0.008

0.01

0.012

0.014

0 10 20 30 40 50 60

0 10 20 30 40 50 60

0 10 20 30 40 50 60

Pow

er c

onsu

mpt

ion

(W)

Event speed (m/s)

DR EA

Fig. 13. Results for different event moving speeds.

protocol collects the data packets in the network, althoughthe movement direction is away from the sink node. Thus,the average latency of the DR-on-SMAC protocol increasesas the number of packets in the network increases. The DRprotocol does not have a sleeping time, thus it can collect

0

1

2

3

4

5

6

7

Aver

age

late

ncy

(s)

Event intervals (s)

E routingshortestEE routing

50

55

60

65

70

75

80

85

90

95

100

Agg

rega

tion

rate

(%)

Event intervals (s)

EE routingshortestE routing

0.00114

0.00116

0.00118

0.0012

0.00122

0.00124

0.00126

0.00128

0 10 20 30 40 50 60

0 10 20 30 40 50 60

0 10 20 30 40 50 60

Pow

er c

onsu

mpt

ion

(W)

Event intervals (s)

E routingshortestEE routing

Fig. 14. The EA protocol with various routing protocols. EE routing: eventand energy-aware routing, E routing: energy-aware routing, shortest:shortest path routing.

D. Kim et al. / Computer Networks 55 (2011) 225–240 237

data packets from neighboring nodes and send aggregateddata packets to the sink node before other packets are gen-erated. The EA protocol has a relatively constant average la-tency with the increasing the event radius, because mostpackets are aggregated and arrive at the sink node simulta-neously. From these results, it is concluded that the EAprotocol operates well with a large event radius and main-tains a low overhead.

For the aggregation rate, when the event radius is small,there are fewer nodes with packets; thus, the nodes are notable to aggregate the packets of neighboring nodes andmaintain a low aggregation rate. The EA protocol has simi-lar aggregation rate as the DR protocol and the DR-on-SMAC protocol, as shown in Fig. 12(b). Because the EAprotocol aggregates the data packets of neighboring nodesat one node, it has a higher aggregation rate. As shown inFig. 12(c), the energy consumption of the EA protocol andthe DR-on-SMAC protocol are much lower than that ofthe DR protocol.

Fig. 16. Node distribution of a realistic scenario.

4.2.5. Effect of event speedIn this simulation, the performance of the EA protocol

and the DR protocol were evaluated with varying eventspeeds from 0 m/s to 60 m/s. With a higher event speed,the prediction of the event position can be incorrect. How-ever, it can be seen that the results remain steady at differ-ent speeds. As both protocols do not create a structure,mobility has little impact on the performance. InFig. 13(a), the EA protocol has a lower average latency thanthe DR protocol. The aggregation rate of the EA protocol is

0 140 840280 420 560 700

0140

840700560420280

x (m)

y (m

)

0 140 840280 420 560 700

0140

840700560420280

x (m)

y (m

)

Low energy consumption

Fig. 15. Energy consumption distribution of the EA protocol w

similar to that of the DR protocol, as seen in Fig. 13(b).The energy consumption of the EA protocol is much lowerthan that of the DR protocol, as illustrated in Fig. 13(c).

4.2.6. Effect of the EE routing protocolIn this simulation, the performance of the EA protocol

is evaluated while varying the routing protocols. Theevent speed is 0 m/s to emphasize the effect of energybalance. The shortest path routing protocol, the energy-aware routing protocol, and the EE routing protocol arecompared.

Fig. 14(a) shows the average latency with differentevent intervals. The EA protocol with the EE routing protocolhas the lowest average latency because it chooses the pathto aggregate more packets and the average latencydecreases with data aggregation.

0 140 840280 420 560 700

0140

840700560420280

x (m)

y (m

)

0 140 840280 420 560 700

0140

840700560420280

x (m)

y (m

)

High energy consumption Event

ith various routing protocols and of the DR protocol.

0

2

4

6

8

10

12

14

Ave

rage

late

ncy

(s)

Event speed (m/s)

DR EA

0 2 4 6 8 10 12 14 16

238 D. Kim et al. / Computer Networks 55 (2011) 225–240

Fig. 14(b) shows the aggregation rate with differentevent intervals. The EA protocol with the EE routing protocolhas the highest aggregation rate. Thus, the EA protocol withthe EE routing protocol selects a more efficient path foraggregation.

In Fig. 14(c), the shortest path routing has the lowestenergy consumption because it chooses the shortest path;thus, minimal nodes are used for forwarding the data pack-ets. However, the shortest path routing has an unbalancedenergy consumption, which is shown in next section. Theenergy-aware routing (E routing) has the largest energyconsumption because it chooses the next hop node thathas the largest energy; thus, the maximum number ofnodes is used. EE routing has a smaller energy consumptionthan E routing because it chooses the next hop that has apacket to aggregate, thus reducing the number of packettransmissions by aggregating the data packets.

0

10

20

30

40

50

60

70

80

90

100

Agg

rega

tion

rate

(%)

Event speed (m/s)

DR EA

0.004

0.006

0.008

0.01

0.012

0.014

0 2 4 6 8 10 12 14 16

wer

con

sum

ptio

n (W

)

DR EA

4.2.7. Energy consumption distribution of the ee routingprotocol

The energy consumption distributions of the EA protocolwith varying routing protocols are compared. Fig. 15shows the event distribution and the consumed energy ofeach node in the topology. The event interval is 40 s. Theshortest path routing protocol consumes more energy inthe static path, as depicted in Fig. 15(a). The EE routingprotocol balances the energy consumption distributionacross nodes and has the lowest energy consumption, ascan be seen in Fig. 15(b). The energy-aware routing proto-col also balances the energy consumption distribution, butit does not select the path in which the node aggregates thepackets of neighbor nodes. Therefore, the energy-awarerouting protocol causes more traffic with a lower aggrega-tion rate. Thus, the energy-aware routing protocol has ahigher average energy consumption than the EE routingprotocol, and it consumes more energy near the sink nodeswith a higher number of packets, as depicted in Fig. 15(c).

As shown in Fig. 15(d), the DR protocol consumed themost energy with idle listening; thus, every node has ahigh energy consumption. Hence, by using the A-MAC pro-tocol, the EA protocol reduces the energy consumption ofidle listening. Also, by selecting an efficient path for aggre-gation and energy balance, the EA protocol with the EE-routing protocol has the lowest and most balanced energyconsumption.

0

0.002

0 2 4 6 8 10 12 14 16

Po

Event speed (m/s)

Fig. 17. Results with varying event speed in realistic scenario.

4.2.8. Realistic scenarioIn this simulation the performance of the EA protocol

and the DR protocol are evaluated in a realistic scenario.Fig. 16 shows the node distribution of a realistic scenario.The total number of nodes is 120 and they are uniformrandomly distributed in 900 m by 900 m square area. Theevent radius is 200 m with varying speeds from 3 m/s to15 m/s.

Fig. 17(a) shows that the average latency of the EA pro-tocol is lower than the DR protocol in the realistic scenario.In Fig. 17(b), the EA protocol has a higher aggregation ratethan the DR protocol. The power consumption of the EAprotocol is much lower than the DR protocol. In the realisticscenario presented, the EA protocol maintains a lower

latency and higher aggregation rate than the DR protocolwhile having a lower power consumption.

The EA protocol consumes 10–20% of the energy con-sumption of the DR protocol, but the EA protocol has equiv-alent or better performance (latency, aggregation rate)than the DR protocol. The EA protocol reduces the idle

D. Kim et al. / Computer Networks 55 (2011) 225–240 239

listening time by using the sleep-wake scheduling and usessmart scheduling mechanisms (EE-routing, schedule margin,schedule join, schedule shift) in order to collect many pack-ets and aggregate them during a short listening period;thus, the EA protocol has similar performance to the DRprotocol although the EA protocol consumes much lessenergy. If only the sleep-wake scheduling is used, theenergy consumption will be decreased, but the latency willincrease and the aggregation rate will decrease.

5. Conclusion

In this paper, an EA protocol, which consists of theevent-aware and energy-aware routing (EE routing) proto-col and the aggregation MAC (A-MAC) protocol, was pro-posed. It achieves energy efficiencies by reducing idlelistening with periodic listen and sleep cycles. In order toaggregate more packets and balance the energy consump-tion, the EE routing protocol was suggested. In order to re-duce the increase of latency due to the sleep-wakescheduling, the A-MAC protocol uses the listening time formulti-hop scheduling for aggregation. In the listen period,the nodes that overhear the schedule of neighboring nodesopportunistically join the schedules of the neighboringnodes for aggregation. The schedule margin and scheduleshift mechanisms were suggested to provide more chancesfor the neighboring nodes to join and to make aggregationschedules in order to maximize the aggregation rate. As aresult, the EA protocol has a similar latency and aggregationrate to the DR protocol, but more importantly, has up tonine times lower energy consumption compared with thatof the DR protocol.

References

[1] P. Gupta, P.R. Kumar, The capacity of wireless networks, IEEETransactions on Information Theory 46 (2) (2000) 388–404.

[2] G.J. Pottie, W.J. Kaiser, Wireless integrated network sensors,Communications of the ACM 43 (5) (2000) 51–58. http://doi.acm.org/10.1145/332833.332838.

[3] K.W. Fan, S. Liu, P. Sinha, Structure-fraggregation in sensor networks,IEEE Transactions on Mobile Computing 6 (8) (2007) 929–942.http://doi.ieeecomputersociety.org/10.1109/TMC.2007.1011.

[4] C. Intanagonwiwat, R. Govindan, D. Estrin, Directed diffusion: ascalable and robust communication paradigm for sensor networks,in: International Conference on Mobile Computing and Networking(MOBICOM ’00), ACM, 2000, pp. 56–67.

[5] W. Zhang, G. Cao, DCTC: dynamic convoy tree-based collaborationfor target tracking in sensor networks, IEEE Transactions on WirelessCommunication 3 (5) (2004) 1689–1701.

[6] S. Madden, M.J. Franklin, J.M. Hellerstein, W. Hong, TAG: a tinyaggregation service for ad-hoc sensor networks, ACM SIGOPSOperating System Review 36 (SI) (2002) 131–146. http://doi.acm.org/10.1145/844128.844142.

[7] W. Heinzelman, A. Chandrakasan, H. Balakrishnan, Energy-efficientcommunication protocol for wireless microsensor networks, in:Proceedings of the 33th Hawaii International Conference on SystemSciences (HICSS ’00), IEEE, 2000, pp. 3005–3014. http://www-mtl.mit. edu/wendi/papers/hicss00.ps.

[8] W. Heinzelman, A. Chandrakasan, H. Balakrishnan, An application-specific protocol architecture for wireless microsensor networks,IEEE Transactions on Wireless Communications 1 (4) (2002) 660–670, doi:10.1109/TWC.2002.804190.

[9] A. Manjeshwar, D.P. Agrawal, TEEN: a routing protocol for enhancedefficiency in wireless sensor networks, in: Proceedings of the 15thInternational Symposium on Parallel and Distributed Processing(IPDPS ’01), IEEE, 2001, pp. 2009–2015.

[10] A. Manjeshwar, D.P. Agrawal, APTEEN: a hybrid protocol for efficientrouting and comprehensive information retrieval in wireless sensornetworks, in: Proceedings of the 16th International Symposium onParallel and Distributed Processing (IPDPS ’02), IEEE, 2002, pp. 195–202.

[11] S. Nath, P.B. Gibbons, S. Seshan, Z.R. Anderson, Synopsis diffusion forrobust aggregation in sensor networks, in: Proceeding of the 2ndinternational Conference on Embedded Networked Sensor Systems(SenSys ’04), ACM, 2004, pp. 250–263.

[12] S. Gobriel, S. Khattab, D. Mosse, J. Brustoloni, R. Melhem,Ridesharing: fault tolerant aggregation in sensor networks usingcorrective actions, in: Communications Society Conference onSensor, Mesh and Ad Hoc Communications and Networks (SECON’06), IEEE, 2006, pp. 595–604.

[13] K.W. Fan, S. Liu, P. Sinha, On the potential of structure-free dataaggregation in sensor networks, in: Proceedings of the 25thInternational Conference on Computer Communications.(INFOCOM ’06), IEEE, 2006, pp. 1–12. http://dx.doi.org/10.1109/INFOCOM.2006.192.

[14] J. Benson, T. O’Donovan, U. Roedig, C.J. Sreenan, Opportunisticaggregation over duty cycled communications in wireless sensornetworks, in: Internationcal Conference on InformationProcessing in Sensor Networks. (IPSN ’08), ACM/IEEE, 2008, pp.307–318.

[15] W. Ye, F. Silva, J. Heidemann, Ultra-low duty cycle mac withscheduled channel polling, in: Proceedings of the 4th InternationalConference on Embedded Networked Sensor Systems (SenSys ’06),ACM, 2006, pp. 321–334. http://doi.acm.org/10.1145/1182807.1182839.

[16] Crossbow Technology Inc. <http://www.xbow.com/products/Product_pdf_files/Wireless_pdf/MICA2_Datasheet.pdf>.

[17] Chipcon Products from Texas Instruments, CC1000 – SingleChip Very Low Power RF Transceiver. <http://www.ti.com/lit/gpn/cc1000>.

[18] Crossbow Technology Inc. <http://www.xbow.com/Products/Product_pdf_files/Wireless_pdf/TelosB_Datasheet.pdf>.

[19] Chipcon Products from Texas Instruments, CC2420 – 2.4 GHz IEEE802.15.4/ZigBee-ready RF Transceiver. <http://www.ti.com/lit/gpn/cc2420>.

[20] W. Ye, J. Heidemann, D. Estrin, An energy-efficient mac protocol forwireless sensor networks, in: Proceedings of the 21th InternationalConference on Computer Communications. (INFOCOM ’02), IEEE,2002, pp. 1567–1576.

[21] W. Ye, J. Heidemann, D. Estrin, Medium access control withcoordinated adaptive sleeping for wireless sensor networks, IEEE/ACM Transactions on Networking 12 (3) (2004) 493–506.

[22] J. Kim, K. Park, D. Park, An energy-efficient scheduling MAC protocolfor wireless sensor networks, in: Proceedings of the 4th AnnualConsumer Communications and Networking Conference (CCNC ’07),IEEE, 2007, pp. 655–659, doi:10.1109/CCNC.2007.134.

[23] J. Kim, K.H. Park, An energy-efficient, transport-controlled macprotocol for wireless sensor networks, Computer Networks 53 (11)(2009) 1879–1902. http://dx.doi.org/10.1016/j.comnet.2009.03.002.

[24] S. Du, A.K. Saha, D.B. Johnson, Rmac: A routing-enhanced duty-cyclemac protocol for wireless sensor networks, in: Proceeding of the26th International Conference on Computer Communications(INFOCOM ’07), IEEE, 2007, pp. 1478–1486, doi:10.1109/INFCOM.2007.174.

[25] J. Polastre, J. Hill, D. Culler, Versatile low power media access forwireless sensor networks, in: Proceedings of the 2nd InternationalConference on Embedded Networked Sensor Systems (SenSys ’04),ACM, 2004, pp. 95–107. http://doi.acm.org/10.1145/1031495.1031508.

[26] N. Bulusu, J. Heidemann, D. Estrin, GPS-less low-cost outdoorlocalization for very small devices, IEEE Personal Communications7 (5) (2000) 28–34, doi:10.1109/98.878533.

[27] D. Moore, J. Leonard, D. Rus, S. Teller, Robust distributed networklocalization with noisy range measurements, in: Proceedings of the2nd International Conference on Embedded Networked SensorSystems (SenSys ’04), ACM, 2004, pp. 50–61. http://doi.acm.org/10.1145/1031495.1031502.

[28] B. Karp, GPSR: Greedy perimeter stateless routing for wirelessnetworks, in: Proceedings of the 6th Annual ACM/IEEE InternationalConference on Mobile Computing and Networking (MOBICOM ’00),ACM, 2000, pp. 243–254.

[29] K. Fall, K. Varadhan, ns Notes and Documentation, The VINT Project.<http://www.isi.edu/nsnam/ns/doc/>.

[30] Ash transceiver TR1000 data sheet. <http://www.rfm.com>.

240 D. Kim et al. / Computer Networks 55 (2011) 225–240

Donggook Kim received the B.S. degree inElectrical Engineering from Korea AdvancedInstitute of Science and Technology (KAIST) in2000 and the M.S. degree in Electrical Engi-neering from KAIST in 2002. He is currently aPh.D. student at KAIST. His research interestsinclude sensor network, multichannel net-work, localization system, and middleware.

Jaesub Kim received the B.S. degree in Elec-

trical Engineering from Kyungpook NationalUniversity in 2000, the M.S. degree in Elec-trical Engineering from Korea AdvancedInstitute of Science and Technology (KAIST) in2002, and the Ph.D. degree in Electrical Engi-neering from KAIST in 2009. His researchinterests include sensor network, internetserver systems, and operating systems.

Kyu Ho Park received the B.S. degree inElectronics Engineering from Seoul NationalUniversity, Korea in 1973, the M.S. degree inElectrical Engineering from the KoreaAdvanced Institute of Science and Technology(KAIST) in 1975, and the Dr.Ing. degree inelectrical engineering from the University deParis XI, France in 1983. He has been a Pro-fessor of the Division of EECS, KAIST since1983. He was a president of Korea Institute ofNext Generation Computing in 2005–2006.His research interests include computer

architectures, file systems, storage systems, ubiquitous computing, andparallel processing. He is a member of KISS, KITE, Korea Institute of NextGeneration Computing, IEEE and ACM.