Computer Networks xxx (2011) xxx–xxx
Contents lists available at ScienceDirect
Computer Networks
journal homepage: www.elsevier .com/ locate/comnet
SDRCS: A service-differentiated real-time communication schemefor event sensing in wireless sensor networks q
Yuyan Xue, Byrav Ramamurthy, Mehmet C. Vuran ⇑Department of Computer Science and Engineering, University of Nebraska, Lincoln, NE 68588, United States
a r t i c l e i n f o
Article history:Received 5 October 2010Received in revised form 20 June 2011Accepted 21 June 2011Available online xxxx
Keywords:Service-differentiated real-timecommunicationPrioritized MACReceiver-contention-based forwardingDistributed event sensingWireless sensor networks
1389-1286/$ - see front matter � 2011 Elsevier B.Vdoi:10.1016/j.comnet.2011.06.018
q An earlier version of this work was presented at tworkshop [1].⇑ Corresponding author.
E-mail addresses: [email protected] (Y. Xue), bRamamurthy), [email protected] (M.C. Vuran).
Please cite this article in press as: Y. Xue et al.,less sensor networks, Comput. Netw. (2011), d
a b s t r a c t
Real-time communication is crucial for wireless sensor networks (WSNs) to accomplishcollaborative event sensing tasks with specific timing constraints. In this work, a service-differentiated real-time communication scheme (SDRCS) is developed to provide softreal-time guarantees for event-based traffic in WSNs. SDRCS features a cross-layer packetforwarding design to integrate the real-time routing functionality with a novel prioritizedmedium access control scheme. Based on this design, SDRCS performs distributed packettraversal speed estimation for traffic classification and admission control. SDRCS also per-forms prioritized packet forwarding so that the routing decisions are locally performed formaximized packet traversal speed. SDRCS requires no extra hardware for localization,transmission power adaptation or multi-channel transmission. It also adapts well tonetwork dynamics, such as channel quality and communication voids. Performance evalu-ations show that SDRCS significantly improves the on-time delivery ratio and service-dif-ferentiation granularity for mixed priority traffic flows in unsynchronized WSNs, comparedwith currently used communication schemes. SDRCS also provides higher end-to-endthroughput in terms of supporting higher source data rates with tight end-to-end latencyrequirements.
� 2011 Elsevier B.V. All rights reserved.
1. Introduction
Wireless Sensor Networks (WSNs) have emerged as anew generation of distributed embedded systems that pro-vide observations on the physical world at low cost andwith high accuracy. Most current WSN applications, suchas battlefield surveillance [2], industrial production control[3], and structural monitoring [4], pose various kinds ofreal-time constraints in response to the physical world[5–8,1]. In a typical real-time WSN application, as shownin Fig. 1, a number of sensor nodes are deployed to coverthe sensing field. Predefined events can be detected bythe nearby sensor nodes. The collected event information
. All rights reserved.
he IEEE SenseApp ’08
SDRCS: A service-differenoi:10.1016/j.comnet.201
must be sent to a sink by a certain deadline so that theproper event response can be performed in a timely man-ner. Depending on the urgency of the event, the data pack-ets associated with different events can be assigneddifferent end-to-end deadline requirements. Only thepackets that are delivered to the sink before the deadlineare deemed useful. Mission-critical applications call fornew service-differentiated real-time communication pro-tocols designed for WSNs.
Providing end-to-end real-time guarantees in WSNs isextremely challenging, compared with the situation in tra-ditional networks such as wireless local area networks(WLANs). First, WSNs utilize multi-hop communicationover lossy channels. The dynamic network and channelconditions make firm real-time guarantees (e.g., a guaran-teed packet reception rate and end-to-end transmissiondelay for a specific data rate) almost impossible. Second,the event-based traffic in WSNs may exhibit highly diversereal-time constraints [5] depending on the event locations
tiated real-time communication scheme for event sensing in wire-1.06.018
Fig. 1. A service-differentiated real-time application in event-basedWSNs.
2 Y. Xue et al. / Computer Networks xxx (2011) xxx–xxx
and urgencies. As a result, traditional flow-based trafficclassification methods (e.g., dividing QoS traffic into data,voice, video, and control categories) may not be able toprovide sufficient service-differentiation capability for theevent traffic and ensure prioritized transmission. Third,the limited resources available to a sensor node restrictthe design space of feasible WSN communication scheme.For example, location awareness and network synchroni-zation may not be realistic assumptions for low-cost wire-less sensor nodes. More importantly, the use of duty-cycleoperation is essential for event-based applications toprolong the network lifetime.
Supporting service-differentiated real-time communi-cation in WSNs is a cross-layer task. First, an efficient pri-oritized medium access control (MAC) mechanism isrequired to provide service differentiation so that the pack-ets with tighter deadline requirements are given higherpriority to access the channel and are delivered to thedestination earlier than others. Some existing real-timecommunication protocols [9] use non-prioritized MACdesigns, such as B-MAC, with multiple priority queues toimplement intra-node traffic prioritization. In this case, ifa sender has multiple outgoing packets in queue, the pack-et with the tightest deadline requirements is scheduledfirst for transmission. However, a non-prioritized MACwith a priority queue cannot resolve inter-node traffic pri-oritization. When a number of senders within the samecontention area try to send the packets with differentdeadline requirements, such MAC schemes cannot priori-tize the transmission attempts of different senders, whichleads to inter-node priority mismatches. Another type ofprioritized MAC design has been proposed in [10,11]; thisscheme uses a dynamic inter frame space (IFS) and back-off window (BW) extension-based CSMA/CA MAC schemeto resolve inter-node traffic prioritization. However, IFS/BW extension based MAC schemes may experience severebandwidth under-utilization in multi-hop WSNs with fine-grained traffic classification. The limitations of theseschemes are discussed in detail in Section 2.
In addition to medium access, routing is a major chal-lenge for real-time communication provisioning in WSNs.To support high sustainable throughput with tight end-
Please cite this article in press as: Y. Xue et al., SDRCS: A service-differenless sensor networks, Comput. Netw. (2011), doi:10.1016/j.comnet.201
to-end deadline constraints, both the end-to-end hop countand the per-hop transmission delay must be considered indetermining the routing metric so that the overall end-to-end communication delay can be minimized. To this end,most existing real-time routing schemes [10–12] use ta-ble-based geographical forwarding techniques. Each sensornode must maintain a routing table with location informa-tion of all the neighboring nodes and the average pairwisetransmission delay. Based on this table, forwarding deci-sions are made to obtain the maximum forwarding speed.Table-based geographical forwarding techniques requireeach sensor node to have global localization capabilityand maintain the accurate one-hop connectivity informa-tion. Under dynamic wireless channel conditions, thistranslates into the frequent exchange of control packets,even when no event traffic exists on the network. The sig-nificant control overhead introduced by table-based geo-graphic routing inevitably deteriorates the WSN lifetimein event-based, real-time applications.
Traffic admission control is another crucial componentused to improve the bandwidth utilization and energy effi-ciency of a real-time communication scheme. By estimat-ing the schedulability of packet transmissions, a properadmission control policy can be applied to the outgoingtraffic in a per-hop manner. As a result, a packet transmis-sion that is unlikely to meet the required latency con-straints should be rejected at an early stage of the end-to-end transmission. However, most existing WSN real-time communication schemes do not consider admissioncontrol or simply drop packets only when the end-to-endtransmission deadline is missed.
In this work, a novel service-differentiated real-timecommunication scheme (SDRCS) is proposed to providesoft, real-time guarantees for event-based traffic in WSNsusing a cross-layer design. Compared with existing real-time communication schemes, the main contributions ofthis work are as follows.
1.1. Cross-layer real-time forwarding
SDRCS uses a dynamic forwarding technique to inte-grate routing functionality with a CSMA/CA-based priori-tized MAC scheme. In this way, a receiver contentionprocess is performed at each hop of packet forwardingbased on the proposed real-time forwarding metric. Neigh-boring nodes with better forwarding distance, lower trafficload, and higher channel quality, i.e., those satisfying thereal-time requirements, receive a higher priority to for-ward packets. No routing tables or neighboring node infor-mation need to be maintained or periodically exchangedfor end-to-end communication; hence, the control over-head is mitigated. Since the forwarding decision is madeon-demand, the SDRCS adapts well to network dynamics.More importantly, the fully distributed and on-demandforwarding design makes SDRCS suitable for duty-cycledWSNs.
1.2. Efficient prioritized MAC design
To provide better service-differentiation capability fordiverse end-to-end deadline requirements in WSN applica-
tiated real-time communication scheme for event sensing in wire-1.06.018
Y. Xue et al. / Computer Networks xxx (2011) xxx–xxx 3
tions, a novel polling contention period-based prioritizedMAC is proposed in SDRCS, as an alternative to traditionalIFS/BW extension-based MAC schemes. The proposed MACdesign helps decrease the average IFS and BW sizes whenthe number of traffic priority categories is large, thusimproving the overall bandwidth utilization of the end-to-end communication when four or more traffic prioritycategories are supported in the network.
1.3. Light-weight packet schedulability estimation
SDRCS includes a light-weight packet schedulabilityestimation mechanism utilizing a received signal strength(RSS)-based sensor node grouping technique and a uniformpolling contention period design for traffic within any pri-ority category. Based on packet schedulability estimation,proper admission control and early missed-deadline pack-et-dropping policies are designed to prevent unschedula-ble packets from being injected into the network anddegrading bandwidth utilization.
The rest of the paper is organized as follows: In Section 2,the existing solutions for real-time communication in WSNsare discussed. An overview of SDRCS is provided in Section 3and the design details and protocol operations are de-scribed. The results of extensive simulation evaluationsare presented in Section 4 to evaluate the performance ofSDRCS, and compare it with two existing protocols, RAP[11] and MMSpeed [10], using two important metrics: aver-age end-to-end latency and on-time delivery rate. The paperis concluded in Section 5.
2. Related work
In this section, we discuss existing real-time communi-cation schemes for wireless sensor networks. We point outthe limitations of the existing solutions and the motivationfor our cross-layer SDRCS design.
2.1. MAC layer solutions
A prevalent approach for achieving prioritized MAC inWSNs was recently developed [10,11], based on the IEEE802.11e or 802.11 EDCA [13] standards. These MACschemes are designed to dynamically adapt the inter framespace (IFS) and/or back-off window (BW) length accordingto different priority classes. A larger IFS is used to transmitpackets with lower priority levels. When the number ofpackets with the same priority level increases, a largerBW is used to resolve the collision. These prioritized MACapproaches are generally referred to as Dynamic IFS/BWExtension-based approaches.
However, IFS/BW extension-based MAC designs do notscale well when the number of supporting priority levelsincreases under diverse end-to-end deadline requirementsin WSN applications. In this case, the MAC design attemptsto prioritize the medium access by increasing the IFS andBW sizes for low-priority traffic. Therefore, the averageIFS and BW sizes allocated in end-to-end communicationand the probability of priority reversion [14] dramaticallyincrease, resulting in significantly degraded bandwidth uti-
Please cite this article in press as: Y. Xue et al., SDRCS: A service-differenless sensor networks, Comput. Netw. (2011), doi:10.1016/j.comnet.201
lization. Mainly for this reason, the revised version of802.11 EDCA [13] limits the supported number of prioritylevels to four. However, supporting fewer priority levels inthe network results in more traffic being classified into thesame priority level. Such a situation leads not only to a de-graded service differentiation ability but also an increasedcollision possibility in the medium access contention pro-cess. Accordingly, a higher collision possibility introducesa larger average BW size. Thus, the average communicationthroughput deteriorates. Based on the above observations,providing fine service differentiation capability and limitedIFS/BW extension is vital in efficient real-time MAC designfor WSN applications.
2.2. Routing layer solutions
The majority of existing real-time communicationschemes [9–12] adopt traditional table-based pro-activerouting approaches with different real-time routing met-rics. In table-based pro-active routing, each sensor nodemaintains a routing table listing all its neighboring nodes.Based on specific real-time routing metrics, one neighbor-ing node that satisfies the application-specific deadlinerequirement is selected as the next hop to complete packetforwarding. RAP [11] uses a greedy geographic forwardingmetric, in which any outgoing packets are routed to theneighboring node with the shortest distance to the recei-ver. A major limitation of the RAP design is that greedygeographic forwarding does not consider local networkconditions, such as load-balance, congestion level, andchannel quality. Therefore, the RAP routing decision leadsto unpredictable per-hop transmission delay in dynamicWSN environments, which affect not only communicationthroughput but also packet traversal speed estimation.
RPAR [9], SPEED [12], and MMSpeed [10] improve thereal-time routing metric by considering both the geo-graphic information and the average pairwise transmissiondelay of neighboring nodes. The pairwise transmission de-lay is usually affected by the local contention level, conges-tion level, and channel quality. Using the location anddelay information, the sender can evaluate the spped ofpacket progress achieved by a neighboring node and thusmake a forwarding decision to minimize end-to-endlatency.
Table-based real-time routing techniques encountercommon limitations in WSNs. First, to maintain the accu-racy of the information listed in the routing table in dy-namic WSNs, a number of control mesages must beexchanged periodically. This introduces significant controloverhead, especially for event-based WSN applications.Second, table-based routing techniques are not suitablefor duty-cycle design, which is vital for energy conserva-tion in WSNs. In an unsynchronized WSN, the sensor nodeswith duty cycle design randomly go into sleep mode to de-crease their energy consumption. In this case, table-basedrouting techniques cannot properly identify the activenext-hop candidate.
In contrast to table-based forwarding techniques, recei-ver-contention-based dynamic forwarding techniqueshave been proposed in recent studies [15–19]. Here, rout-ing functionality is combined with a CSMA/CA-based
tiated real-time communication scheme for event sensing in wire-1.06.018
4 Y. Xue et al. / Computer Networks xxx (2011) xxx–xxx
MAC design so that an adaptive receiver contention is per-formed at each hop. Sensor nodes with better forwardingdistances than others, lower traffic loads, higher channelquality or higher residual energy levels receive higher pri-ority to respond to the RTS packet with a CTS packet andthus, become the next hop. No routing tables or neighbor-ing node information must be maintained or periodicallyexchanged. Since the forwarding decision is madeon-demand, these schemes can easily adapt to a distrib-uted duty-cycle design.
The existing dynamic forwarding techniques motivatesthe SDRCS design by allowing for an efficient cross-layercommunication approach. However, since the existingdynamic forwarding approaches do not consider softreal-time provisioning in forwarding decision, new for-warding metrics based on prioritized MAC operations mustbe designed so that the application-specific deadlinerequirements can be enforced in end-to-end packetforwarding.
2.3. Other solutions
In addition to the aforementioned link- and network-layer solutions, physical- and transport-layer protocolshave recently been developed to address energy conserva-tion and reliable communication in delay-constrainedWSN applications. In [20], it is pointed out that eventdetection probability and detection latency are functionsof the duty cycles of the sensor nodes. Based on this obser-vation, a distributed algorithm is proposed to regulate theprobability of sensor nodes being active, such that an eventthat occurs anywhere in the network can be detected bythe sink within a maximum detection latency and a mini-mum detection probability. In [21], it is shown that theend-to-end communication reliability and latencyachieved in event-based WSNs can be regulated throughtransport-layer rate control. By observing the averageend-to-end communication delay and the on-time deliveryrate at the sink, proper control mechanisms for the event
AdmissionControl /
Deadline-misspacket drop
Prioritized Queuing
RSS-basedGrouping
(0) Group ID contains the end-to-end hop cou
(1) Incoming packet with an end-to-end deadl
(2) Admitted packet with a per-hop deadline a
(3) Prioritized packet with a priority level ass
(4) The highest-priority packet within the con
(5) The packet is forwarded to a next-hop who
(2) (3)
)0(
SDRCS
(1)
Fig. 2. SDRCS components and th
Please cite this article in press as: Y. Xue et al., SDRCS: A service-differenless sensor networks, Comput. Netw. (2011), doi:10.1016/j.comnet.201
data rate are applied to the sensor nodes within the eventarea so that application-specific event transport reliabilityor latency requirements can be met at the sink. Since thesedesigns are independent of the MAC and network layeroperations, the proposed SDRCS operations can be comple-mented with these solutions.
3. SDRCS: a service-differentiated real-timecommunication scheme
In this section, the details of the SDRCS are given interms of five components. The protocol operations exe-cuted in each component are described. The relationshipamong different components and the process by which areal-time packet is scheduled and forwarded in the SDRCSdesign are shown in Fig. 2. The void-avoidance capability ofSDRCS is discussed at the end of the section.
3.1. Assumptions
We consider a static WSN with homogeneous sensornodes, and a single sink (see Fig. 1). The nodes communi-cate through multihop wireless links, using a single chan-nel and fixed transmission power. The sensor nodes areunsynchronized devices without location awareness. Thesensor nodes are capable of measuring the received signalstrength for each received packet. The above assumptionsreflect the current hardware configurations of wirelesssensor nodes [22].
We consider a mission-critical event sensing applica-tion [5], where the predefined events are detected by thenearby sensor nodes and the event information should beconverge-casted [23] to the sink. According to the urgencyof each event, data packets can be assigned different end-to-end deadline requirements. Only the packets deliveredto the sink before the deadline are deemed useful. We alsoassume the networks to be connected, where at least oneend-to-end forwarding path exists.
Real-timeMAC
Dynamic Forwarding
nt information
ine assignment
ssignment
ignment
tention area obtains the channel
wins through the receiver-contention
(4)
)0(
(5)
e packet forwarding steps.
tiated real-time communication scheme for event sensing in wire-1.06.018
Y. Xue et al. / Computer Networks xxx (2011) xxx–xxx 5
3.2. RSS-based sensor node grouping
Many existing real-time communication protocols forWSNs assume precise location awareness at each sensornode [10,12], which requires GPS equipment or complexlocalization schemes. In the absence of such precise loca-tion awareness, we design a Received Signal Strength(RSS) based sensor node grouping method to roughly stripthe sensing field into layers, as shown in Fig. 3. The layerinformation can be used to estimate the hop-distancefrom the node to the sink, which enables the estimationof packet traversal speed in the packet forwarding pro-cess. The accuracy of the hop distances resulting fromthe grouping can be controlled by grouping granularity,defined as GRA. The basic grouping operations are givenbelow:
� Step 1: The sink initializes a Grouping Message broadcastwith its group ID, where G_ID = 0.� Step 2: Each sensor node, that receives a grouping mes-
sage with received signal strength RSS higher than apre-defined threshold RSSth and does not have a groupID, is assigned a group ID G_ID = G_IDr + 1, whereG_IDr is the group ID value contained in the receivedgrouping message. It then sets its back-off window asBW = [G_ID⁄slot, (G_ID + 1)⁄slot], and broadcasts agrouping message, which contains its own group ID,once.� Step 3: Each sensor node, that receives a grouping mes-
sage with received signal strength RSS lower than RSSth
and does not have a group ID, is assigned a temporalgroup ID as G_IDtemp = G_IDr + GRA. It then sets a timerthat expires in GRA * Broadcast period. If a groupingmessage is received with received signal strength RSShigher than RSSth before the timer expires and verifiesthat G_IDtemp > G_IDr + 1, a sensor node will assign itsgroup ID as G_ID = G_IDr + 1. It then cancels the timer,sets its back-off window as BW = [G_ID⁄slot,(G_ID + 1)⁄slot], and broadcasts a grouping message,which contains its own group ID, once.
Fig. 3. Received signal strength (RSS) based sensor node groupingexample with GRA = 2.
Please cite this article in press as: Y. Xue et al., SDRCS: A service-differenless sensor networks, Comput. Netw. (2011), doi:10.1016/j.comnet.201
� Step 4: Each sensor node that has a G_IDtemp is assigneda group ID as G_ID = G_IDtemp when the timer expires. Itthen sets its back-off window as BW = [G_ID⁄slot,(G_ID + 1)⁄slot], and broadcasts a grouping message,which contains its own group ID, once.� End: Each sensor node can broadcast the grouping mes-
sage at most once based on Steps 2–4. Hence, thegrouping process ends when all the nodes finish theirbroadcast.
The performance of the proposed grouping process canbe analyzed using log-normal shadow fading channel mod-el [24]. First, the RSS value obtained by a receiver at a dis-tance R from the transmitter is given by
RSSðRÞ ¼ Pt � PLðR0Þ � 10glog10RR0
� �þ Xr; ð1Þ
where Pt is the transmit power in dBm, PL(R0) is the pathloss at a reference distance R0 in dBm, g is the path lossexponent, and Xr is the shadow fading component, whereXr � N(0,r). With RSS(R) = RSSth, the expected transmissionrange E[R] of a broadcast message is given by
E½R� ¼ R0 � 10Pt�PLðR0 Þ�RSSth
10g � E 10Xr10g
h ið2Þ
¼ R0 � 10Pt�PLðR0 Þ�RSSth
10g � er
10g�ln10: ð3Þ
We also define the maximum transmission range E[Rmax]based on the noise power floor Pn, where
E½Rmax� ¼ R0 � 10Pt�PLðR0 Þ�Pn
10g � er
10g�ln10: ð4Þ
The grouping granularity, GRA, is then defined as
GRA ¼ E½Rmax�E½R� ¼ 10
RSSth�Pn10g : ð5Þ
By properly increasing RSSth, the grouping granularity isincreased as more layers are assigned to the network. Thisresults in finer end-to-end hop-distance awareness at thesensor nodes. The design of the back-off window, BW, en-sures that the sensor nodes with higher group IDs cannotinterrupt the Grouping Message broadcast from a lower-group node; hence, the grouping process can propagatefrom the sink in a layered manner.
The node grouping should be done at the post-deploy-ment stage. After the RSS-based grouping process, the sen-sor nodes can be mapped into strip-style groups, usingE[R]/GRA as the average width of the strip. The density ofthe WSN affects the grouping structure. With increasingnode density, the result of this grouping would approachperfectly circular strips if the channel fading and noisecomponents in the network are homogeneous [25]. Thegroup ID can be used to estimate the hop distance fromthe node to the sink and the packet forwarding can beguided towards the sink without precise locationinformation.
The main difference between using an RSS-basedgrouping technique and a traditional geographic localiza-tion technique in the design of real-time communicationschemes lies in the definition of the node-to-sink distance.In a traditional geographic forwarding approach, the end-to-end distance is defined as the Euclidean distance,
tiated real-time communication scheme for event sensing in wire-1.06.018
6 Y. Xue et al. / Computer Networks xxx (2011) xxx–xxx
whereas in SDRCS, the distance is defined as the end-to-end hop count. As stated earlier, in a densely deployedWSN with homogeneous channel fading and noise compo-nents, the group ID can serve as a good indicator of node-to-sink geographic distance. However, in a sparselydeployed WSN or a WSN with dynamic channel fadingand noise components, the group ID is more of an end-to-end hop count estimation, which may not be linearly re-lated to geographic distance. Therefore, the proposed RSS-based grouping technique can help improve the accuracyof end-to-end hop-distance estimation in real WSNdeployment, while avoiding the use of the expensive pre-cise localization schemes or devices as in [10,11].
3.3. Per-hop deadline based prioritized queueing policy
In WSNs, an application-specific real-time requirementis usually presented as an end-to-end deadline, which indi-cates the maximum packet traversal time from the senderto the receiver [26]. However, in a multi-hop network, theend-to-end deadline is not the only criterion for determin-ing the urgency of packet delivery. The end-to-end hopcount also affects the packet delivery schedule. For exam-ple, if there are two schedulable packets with the sameend-to-end deadline requirements competing for the chan-nel, the one with a higher end-to-end hop count should bescheduled first. If we assume that each sensor node is ableto predict the end-to-end hop-count to the sink, the end-to-end deadline requirement can be broken down into aper-hop deadline requirement, LReq
hop, where
LReqhop ¼
Le2e
HCe2e: ð6Þ
Le2e is an application-specific parameter which reflects therequired end-to-end delay for packet delivery. HCe2e is thepredicted hop-count value based on the G_ID of the sender,the GRA value and the forwarding strategy, which is dis-cussed in Section 3.5. LReq
hop reflects the per-hop traversalspeed required to achieve the end-to-end real-time guar-antees in a contention-based WSN. It can be used as anaccurate indicator for packet delivery priority classification[26].
We use FIFO priority queues for packet scheduling at anode, as shown in Fig. 4. Since the prioritized MAC can onlyprovide differentiated service for a limited number of pri-ority classes, the per-hop deadline requirements are fur-
Packet Priority Level Assignment
FIFO Priority QueuesApplication
Specific Deadline
PTx = 1
PTx = 2
PTx = N
Polling Contention Period Based
Real-Time MAC
Fig. 4. Per-hop deadline based priority queues at each sensor node forintra-node real-time traffic classification.
Please cite this article in press as: Y. Xue et al., SDRCS: A service-differenless sensor networks, Comput. Netw. (2011), doi:10.1016/j.comnet.201
ther mapped into N priority levels, where N is thenumber of the priority queues allocated at each sensornode.
In this paper, we use the following packet priority levelassignment policy:
PTx ¼ minLReq
hop
LMinhop
$ %;N
!; ð7Þ
where PTx is the assigned packet priority level in the queueand LMin
hop is the minimum time required for one-hop packetforwarding, which depends on the MAC operationsadopted in the dynamic forwarding design. The LMin
hop valuefor SDRCS is given in Section 3.5. Since Early Deadline First(EDF) has been proven as the most efficient scheduling pol-icy for channel access in wireless networks [27], the packetin a higher priority queue is scheduled earlier fortransmission.
Note that the above priority level assignment policyworks well when the application-specific LReq
hop is uniformly
distributed within its design space LMinhop ;N � LMin
hop
h i. For dif-
ferent real-time applications with different LReqhop design
spaces and distributions, different priority level assign-ment policies can be used such that the incoming packetswith various LReq
hop values can be classified properly into N
priority classes and placed into an associated priorityqueue for transmission [10].
3.4. Polling contention period based real-time MAC
To better support the diverse end-to-end deadlinerequirements in WSN applications, we designed a pollingcontention period based real-time MAC to support priori-tized channel access.
As mentioned in Section 2, Dynamic IFS/BW Extensionis used by most existing real-time communicationschemes for prioritized MAC support in WSNs. Such ap-proaches employ extended arbitrary inter frame space(AIFS) and back-off window (BW) size adaptation for prior-itized medium access contention. For a packet with prior-ity level i, according to the IEEE 802.11 EDCA [13], theAIFS and BW values are derived as follows:
AIFSi ¼ SIFSþ i � SLOT TIME; ð8ÞBWi ¼ ðBW1 þ 1Þ � i� 1; ð9Þ
where SIFS is the short inter frame space for controllingpacket transmission contention. In the dynamic IFS/BWextension based MAC design, a higher number of prioritylevels supported in the network results in higher averageAIFS and back-off window values and a lower averagethroughput.
In SDRCS, a fixed number of polling slots are used forprioritized packet transmission contention instead of vari-able inter frame space and back-off window sizes. This de-sign was inspired by the bus access control mechanismsused in computer systems. The basic MAC operationadopted by SDRCS is shown in Fig. 5.
For any packet transmission, a sender first senses themedium. If the medium is idle, the sender waits for the
tiated real-time communication scheme for event sensing in wire-1.06.018
Fig. 5. Polling period-based transmission contention in SDRCS Real-timeMAC.
Table 1Polling-slot design for maximum priority level = 7.
Priority level Slot 1 Slot 2 Slot 3
1 Active Active Active2 Active Active Inactive3 Active Inactive Active4 Active Inactive Inactive5 Inactive Active Active6 Inactive Active Inactive7 Inactive Inactive Active
Y. Xue et al. / Computer Networks xxx (2011) xxx–xxx 7
AIFS period of time and senses the medium again. If themedium remains idle, the sender attempts to initiate pack-et transmission by sending out an RTS packet. Since severalnodes within the interference range may have been wait-ing for this chance to transmit, all these nodes enter thepolling period to compete for the transmission of an RTSpacket based on the priority level associated with the out-going packet. The entire polling period consists of dlog2Nepolling slots for contention entities with N priority levels.For example, if 7 priority levels are supported in SDRCS,3 polling slots are required for medium access contentionamong all possible competitors within the interferencearea. According to Table 1, any sensor node with an outgo-ing packet at priority level i will transmit a burst signal inits active polling slots and remain silent in its inactive poll-ing slots. Any node that senses a burst in its inactive poll-ing slots will be suppressed in the following transmissionperiod. In this manner, only the nodes with the highest pri-ority level among all competitors can survive the pollingperiod. Note that more than one node may survive thepolling contention period because they have the outgoingpackets with the same priority or they are located in a hid-den-terminal scenario. Therefore, an extra back-off periodis used after the polling period to handle the possible col-lision (Exponential Back-off state at the bottom-left inFig. 5). Since much fewer number of competing nodes
Please cite this article in press as: Y. Xue et al., SDRCS: A service-differenless sensor networks, Comput. Netw. (2011), doi:10.1016/j.comnet.201
can enter the back-off period after the polling contentionperiod, the BW can be set to a much smaller size comparedto that used in dynamic IFS/BW extension based prioritizedMAC design. Assuming each priority level has the sameamount of traffic load, the proposed polling-contentionmechanism provides better overall throughput when thenumber of priority levels satisfies N P 4.
3.5. Receiver contention based dynamic forwarding
Motivated by the cross-layer forwarding design dis-cussed in Section 2, we developed a receiver contention-based dynamic forwarding for converge-cast packet rout-ing. This approach is combined with the RTS/CTS exchangeperiod of the proposed real-time MAC design. In the real-time MAC design in Section 3.4, if a sender i wins in a poll-ing contention period and gains access to the medium afterthe exponential back-off period, it will initiate an RTSbroadcast containing its own group ID, Gi. All the neighbor-ing nodes that overhear this RTS message enter the recei-ver contention period, in which only sensor nodes withgroup IDs equal to or lower than Gi, become the qualifiednext-hop candidates. Therefore, the packet can only be for-warded towards the sink and gain a non-negative packettraversal speed. The unqualified nodes enter the NAV (Net-work Allocation Vector) period. Each qualified next-hopcandidate is required to evaluate its capability for maxi-mizing the packet traversal speed for this transmission.This capability is classified into M priority levels for recei-ver contention.
Based on the aforementioned forwarding process andthe grouping mechanism described in Section 3.2, thegrouping granularity GRA gives the maximum number ofgroups that a packet can traverse within one hop. LMin
hop givesthe minimum time for one-hop packet transmission.Therefore, the maximum packet traversal speed achievableby a forwarding decision without queuing delay is given by
Speedmax ¼GRA
LMinhop
: ð10Þ
Regarding a specific next-hop candidate j, the average tra-versal speed Speedj by forwarding the packets to j is de-rived from its average pairwise packet transmission timetAvg
i;j , the queue length LQ, the average per-packet queuingdelay tAvg
Q , and its group ID Gj, where
Speedj ¼Gi � Gj
tAvgi;j þ LQ � tAvg
Q
: ð11Þ
In (11), the packet progresses by a forwarding decisionbased on the group ID difference between the sender andreceiver, i.e., Gi � Gj. The per-hop packet delivery delayconsists of two parts, the packet transmission delay andpacket queuing delay. The packet transmission delay, tAvg
i;j ,is calculated using a Exponentially Weighted MovingAverage (EWMA) algorithm [28]:
tAvgi;j ¼ ati;j þ ð1� aÞtAvg
i;j ; ð12Þ
where a is the moving average coefficient, 0 < a < 1, whichrepresents the degree of weighting decrease. A higher adiscounts the older delay value faster, and makes the re-
tiated real-time communication scheme for event sensing in wire-1.06.018
8 Y. Xue et al. / Computer Networks xxx (2011) xxx–xxx
cent delay value more important in determining the aver-age delay. With low traffic rates and highly unreliabletime-varying channel conditions, a larger moving averagecoefficient is generally used. In (12), ti,j is the instantaneouspacket transmission time, which is measured as the timebetween an RTS transmission and the corresponding ACKis receipt. If the packet is dropped because it exceeds themaximum retransmission time, NRe_Trans, then we have,
ti;j ¼ LMinhop � NRe Trans:
In (12), tAvgi;j is a good indicator of the link quality (packet
error rate) of a potential receiver. A higher tAvgi;j than LMin
hop
indicates the possible retransmission for a packet deliverybetween node i and j.
The average per-packet queuing delay, tAvgQ , reflects the
local contention level for a particular next-hop candidateand is again calculated using an EWMA:
tAvgQ ¼ btQ þ ð1� bÞtAvg
Q ;
where b is the moving average coefficient, 0 < b < 1, and tQ
is the instantaneous per-packet queuing delay, which ismeasured as the time between two consecutive packetsdequeued from a priority queue. A larger queuing lengthand per-packet queuing delay indicate a lower packet tra-versal speed for a given forwarding decision.
Based on (10) and (11), the contention priority for anext-hop candidate j is given as
PRxj ¼ min M �
Speedj �MSpeedmax
� �;M
� �: ð13Þ
The above receiver priority assignment guarantees thefollowing:
� The next-hop candidate with the lowest group IDreceives the highest priority for transmitting its CTSpacket.� For multiple next-hop candidates with the same group
ID, the one with a better channel quality and lower traf-fic load gets a higher priority for transmitting its CTSpacket.� The sensor node with the maximum the packet tra-
versal speed is assigned the highest priority.� A packet is forwarded only to achieve non-negative tra-
versal speed.
Upon receiving the RTS broadcast and evaluating its for-warding priority based on (13), each receiver candidatecompetes to reply with a CTS packet based on its deter-mined forwarding priority. The same prioritized MACmechanism is used for receiver contention, as describedin Section 3.4. The nodes with the highest forwarding pri-ority among all candidates capture the channel throughthe dlog2Me polling period. After an extra back-off periodBWCTS, the winning receiver notifies the sender by a CTS
AIFS PollingRTS BWRTS
SIFS
Sender
Receiver Pollin
CTS
Fig. 6. A complete prioritized packet transmission contenti
Please cite this article in press as: Y. Xue et al., SDRCS: A service-differenless sensor networks, Comput. Netw. (2011), doi:10.1016/j.comnet.201
packet with its node ID. Accordingly, the sender will uni-cast the data packet to the winner, wait for an acknowledg-ment and finish the one-hop packet forwarding.
A complete prioritized packet transmission contentionand receiver contention period for real-time MAC is shownin Fig. 6. According to our receiver-contention based dy-namic forwarding operation, two important parametersare determined for (6) and (10). The minimum per-hop la-tency for the packet transmission with any priority levelassignment, Lmin
hop , is given as
Lminhop ¼ AIFSþ tRTS
Polling þ12
BWRTS;min þ tRTS þ SIFSþ tData
þ SIFSþ tCTSPolling þ
12
BWCTS;min þ tCTS þ SIFSþ tACK ; ð14Þ
where AIFS and SIFS are arbitrary and short IFSs, tRTSPolling and
tCTSPolling are the fixed times of the polling periods for RTS and
CTS packets, BWRTS,min and BWCTS,min are the minimumback-off window values, and tRTS, tCTS,tData and tACK are theRTS, CTS, Data and ACK packet transmission times, respec-tively. The end-to-end hop count estimation for sender i isgiven by:
HCe2e ¼Gi
AvgðGiFwÞ
; ð15Þ
where AvgðGiFwÞ is the moving average of the number of
groups a packet can traverse within a single hop transmis-sion from node i and 0 6 AvgðGi
FwÞ 6 GRA. The receiver’sgroup ID is obtained by i for each transmission by piggy-backing on the CTS packets. The initial value of AvgðGi
FwÞis set to GRA. The AvgðGi
FwÞ value depends on the local net-work density, the channel quality and the network conges-tion level at node i.
3.6. Admission control and early deadline-miss packet drop
Admission control is important in real-time provision-ing for WSNs. A well-designed admission control policycan prevent unschedulable traffic from entering the net-work, thus improving bandwidth utilization and energyefficiency. According to the receiver-contention based dy-namic forwarding design, as explained in Section 3.5, theminimum end-to-end deadline requirement for a packetinitiated at sender i can be derived from (10) as
Lmine2e ¼
Gi � Lminhop
GRA: ð16Þ
Using (15) and (16) in (6) and (7), any schedulable end-to-end requirement can be mapped to a priority level i, where1 6 i 6 N. Therefore, a simple admission control policy canbe adopted at the sender, where packets with priority lev-els greater than N are not admitted to the network.
At the relay-node, SDRCS employs an early-deadline-miss (EDM) drop policy for relaying nodes. For any relaying
CTS
DATA
ACK
SIFS
gCTS BWCTS SIFS
on and receiver contention period for real-time MAC.
tiated real-time communication scheme for event sensing in wire-1.06.018
Fig. 7. An example topology with communication voids.
Y. Xue et al. / Computer Networks xxx (2011) xxx–xxx 9
node k, the cumulative packet transmission time is re-corded as tA and the remaining deadline for a packet, Lr,is calculated as:
Lr ¼ Le2e � tA:
From (6) and (15), the updated per-hop deadline is calcu-lated at each relay node as
LReqhop ¼
Lr
Gk=AvgðGkFwÞ
: ð17Þ
If the updated LReqhop is mapped into a priority level greater
than N, the packet will be dropped because it is unlikelyto be delivered to the sink on time based on the end-to-end hop count estimation at node k. In contrast to thepacket drop policies adopted by [10] or [12], which dependon periodically updated per-hop delay information storedin the neighbor list, this early drop policy better adaptsto the dynamic channel and load conditions encounteredin WSNs, thus avoiding false packet drops due to outdatedper-hop pairwise delay information.
3.7. Void avoidance
SDRCS relies on a greedy forwarding strategy at everyhop to transmit a data packet to a locally optimal next-hop node ensuring positive progress toward the sink. How-ever, this may not always be possible. For example, in a sit-uation where all the neighboring nodes of a sender areassociated with higher groupIDs, the sender will fail to lo-cate a qualified next-hop node that has a positive progresstoward the sink. This undesirable phenomenon is usuallycalled a communication void [29]. The presence of commu-nication voids is a challenging problem for any greedy for-warding approach. Although a dense deployment ofwireless nodes reduces the likelihood of the occurrenceof a void in the network, it is still possible for some packetsto encounter voids that are caused by the presence of deadnodes or the boundaries of a wireless network. These pack-ets must be discarded only when a single greedy-forward-ing strategy is used, even though a topologically valid pathto the destination node may still exist. Thus, it is impera-tive to provide an effective and efficient void-handling ap-proach in SDRCS.
MMSpeed [10] and SPEED [12] use passive participationto deal with communication voids. The idea of passive par-ticipation was introduced in [15], and it exploits a self-healing property of the network topology itself. Once anode identifies itself as a void node, it simply discardsthe data packet and keeps itself from forwarding any sub-sequent data packets toward the destination. The nodemay periodically check whether it can locate a neighboringnode guaranteeing positive progress to participate in pack-et forwarding at a later time. This simple strategy has a re-verse-propagation effect, which eventually informs otherintermediate nodes to explore other possible paths in thenetwork, such that nodes leading to a broken route canbe avoided on routing paths. However, passive participa-tion is not always effective. For instance, as shown inFig. 7, source node S wants to deliver a sequence of datapackets to sink D. The first data packet is greedily for-
Please cite this article in press as: Y. Xue et al., SDRCS: A service-differenless sensor networks, Comput. Netw. (2011), doi:10.1016/j.comnet.201
warded to node V at the first hop. However, node V cannotcontinue to greedily forward the data packet. Node V dropsthe data packet and will not participate in forwarding sub-sequent data packets for destination D. It appears to S thatnode V no longer exists in the topology. However, no othernode in its neighborhood capable of positive progress canhelp forward the subsequent data packets. Thus, data pack-ets must be discarded, even hough a topologically validpath does exist from S to D: S � V � A � B � C � E � D. Itwas argued in [30] that passive participation is not effec-tive in a randomly deployed wireless network with lowdensity.
In SDRCS, a communication void is handled inherentlyby grouping ID assignments and the design of forwardingmetrics. First, note that RSS-based grouping is a limitedbroadcast process initiated at the sink, as described inSection 3.2. Any node can be reached by the broadcastgrouping message and assigned a group ID while thenetwork is connected. In addition, any node with a groupID assignment must be able to reach the sink through thereversed broadcast path, if symmetric links are assumedbetween each pair of connected nodes. Then, consideringthe receiver-contention based dynamic forwarding opera-tion described in Section 3.5, packet forwarding fails tofind a next hop only if one of the following conditions istrue:
� There is no node within the transmission range of thesender.� Any node within the transmission range has a higher
group ID than the sender.
However, RSS-based grouping design prevents eithercondition from occurring in SDRCS operation. First, sincethe network is assumed to be connected, there must beat least one node within the transmission range of the sen-der. Second, any sender must have at least one neighboringnode with a lower group ID, from which the grouping mes-sage is received. As a result, the SDRCS operation guaran-tees that a packet can always be forwarded from thesender to a node with lower group ID and finally reachthe sink, whose group ID equals 0. In Section 4.4, wediscuss how grouping results adapt to the communicationvoids.
tiated real-time communication scheme for event sensing in wire-1.06.018
Table 2Simulation parameters.
Sensing field dimensions (500 � 500) mSink location (25,25)Number of sensor nodes 100Node placement Random uniformPacket length 128 bytesRadio bandwidth 250 kbpsChannel model log-normal shadow fadingPath loss exponent 4Shadow fading variance 6Transmission power 1 dBmNoise power floor �95 dBmMaximum transmission range 125 mReference distance 0.3 mMoving average coefficient 0.5
10 Y. Xue et al. / Computer Networks xxx (2011) xxx–xxx
4. Performance evaluation
The performance of our real-time communicationscheme, SDRCS, was analyzed in GlomoSim [31] usingthe simulation parameters shown in Table 2. We chose alog-normal shadow fading channel model [24] to reflectthe channel dynamics in real WSN deployments andimplemented the model in the simulator. The node relatedparameters were also carefully chosen to reflect typical Mi-caZ node capabilities [22]. We explored extensive simula-tion scenarios for SDRCS and compared its performancewith the existing service differentiated real-time commu-nication schemes, RAP [11] and MMSpeed [10]. The MACoperation parameters for dynamic IFS/BW extension basedprioritized MAC design (used by RAP and MMSpeed) andreal-time MAC (used by SDRCS) are listed in Table 3. ForRAP and MMSpeed, AIFS[i], the BWMin
RTS [i] and BWMaxRTS [i] val-
ues are defined based on the simulation settings in [10]and (8), where i is the data packet transmission priority le-vel. For SDRCS, the Lmin
hop value was derived according to(14).
Two important end-to-end metrics were measured forreal-time performance evaluation in our simulations:
� End-to-End On-Time Packet Delivery Rate: The ratio ofthe number of unique packets received at the sink withend-to-end latency less than or equal to the end-to-enddeadline requirement, to the total number of packets
Table 3Dynamic IFS/BW extension based prioritized MAC and real-time MACparameters.
MMSpeed & RAP SDRCS
Retransmission Limit 7 7Number of Priority Classes 7 7SIFS 10 ls 10 lsTime Slot 20 ls 20 lsAIFS[1] 30 ls 80 ls
BWMinRTS [1] 15 Slots 10 Slots
BWMaxRTS [1] 255 Slots 200 Slots
BWCTS N/A 4 Slots
Lminper�hop
N/A 2200 ls
GRA N/A 2
Please cite this article in press as: Y. Xue et al., SDRCS: A service-differenless sensor networks, Comput. Netw. (2011), doi:10.1016/j.comnet.201
sent by the source node. This metric reveals the end-to-end real-time capacity of the network achieved bya given communication scheme.� Average End-to-End Packet Transmission Latency: Aver-
age end-to-end transmission time for all on-time deliv-ered packets. The packets that are dropped enroute dueto missed deadlines are not included in the averageend-to-end latency calculation. This metric gauges theservice-differentiation capability of a given communi-cation scheme.
4.1. RSS-based grouping with varying grouping granularity
In this simulation scenario, we examined the perfor-mance of the RSS-based grouping scheme. In Fig. 8(a) and(b), the RSS-based geographic grouping results withGRA = 1 and GRA = 2 are shown, respectively, on a samplenetwork topology generated for the simulation. It can beobserved that in the sample network topology with homo-geneous channel fading and noise components, RSS-basedgeographic grouping can properly divide the sensing fieldinto near circular strips. Since the node distribution, andthe channel fading and noise components are homoge-neous in the network, the node group ID assignment showsa nearly perfect linear relation with the group-to-sinkdistance.
Next, we examined how the grouping granularity GRAaffects the end-to-end performance of the SDRCS. Two setsof simulations were performed with different source nodesand end-to-end deadline requirements. Using the networktopology shown in Fig. 8, we chose the source nodes lo-cated in the left-bottom corner to maximize the possibleend-to-end hop count. In the first simulation set, one con-stant-bit-rate (CBR) event data flow CBR1 is generatedfrom a node located at (459,411), with end-to-end dead-line requirement LReq
e2e ¼ 30 ms. In the second set, anotherevent data flow CBR2 is generated from a node located at(402,451), with LReq
e2e ¼ 60 ms. Accordingly, PriorityTxCBR1 ¼ 2
and PriorityTxCBR2 ¼ 5. For each set, 2000 real-time packets
were sent from the source node. The simulation was con-ducted 10 times with different random seeds, and the aver-age value is shown.
In Fig. 9(a), the average on-time delivery rates of CBR1and CBR2 are shown for GRA values ranging from 1 to 4.It can be observed that increasing the grouping granularity,GRA, to some extent helps improve the end-to-end real-time performance at both priority levels. In this case, forGRA = 2 and node degree of 15, up to a 30% improvementin the end-to-end on-time delivery rate is achieved withboth traffic flows. However, increasing GRA without con-sidering the network density leads to groups with unevennode distributions or empty groups. In these cases, thepacket traversal speed cannot be estimated properly andthe end-to-end real-time performance is degraded. In addi-tion, a larger GRA introduces higher control overhead (dueto more temporary group ID updates) and longer groupingtimes (due to the larger back-off window).
To further investigate the relationship between net-work density and optimal GRA value, we conducted simu-lations with varying node degree values (15, 22 and 30)and show the resulting average on-time delivery rate of
tiated real-time communication scheme for event sensing in wire-1.06.018
Fig. 8. A sample simulation network topology with a node degree of 15 for grouping granularities of GRA = 1 and GRA = 2. The numbers shown next to eachnode are the resulting group IDs.
0
20
40
60
80
100
10 100
On-
time
Del
iver
y R
ate
(%)
Source Rate (Pkt/s)
GRA=1(60ms)GRA=2(60ms)GRA=3(60ms)GRA=4(60ms)GRA=1(30ms)GRA=2(30ms)GRA=3(30ms)GRA=4(30ms)
40
50
60
70
80
90
100
1 2 3 4 5 6
On-
time
Del
iver
y R
ate
(%)
Grouping granularity (GRA)
Density=30(60ms)Density=30(30ms)Density=22(60ms)Density=22(30ms)Density=15(60ms)Density=15(30ms)
Fig. 9. On time delivery rate of CBR1 and CBR2 traffic.
Y. Xue et al. / Computer Networks xxx (2011) xxx–xxx 11
CBR1 and CBR2 in Fig. 9(b). The simulation results confirmthat a larger GRA value can achieve better localizationinformation for higher network density. For node degreesof 15, 22 and 30, the best GRA values are found to be 2,3.5, and 5, respectively. The optimal GRAs in a WSN mayvary with time due to node failures. Such situations requirea network regrouping based on a different GRA. We will ad-dress dynamic GRA adjustment strategies in a future work.
4.2. Performance comparison
In this simulation scenario, we compare the real-timeperformance of SDRCS with the existing service-differenti-
Please cite this article in press as: Y. Xue et al., SDRCS: A service-differenless sensor networks, Comput. Netw. (2011), doi:10.1016/j.comnet.201
ated real-time communication schemes, RAP [11] andMMSpeed [10]. To fully test the protocol performance,we randomly generate 10 network topologies with nodedegrees of 15. For each network topology, we chose threenodes located in the lower-right corner as the event datasources for maximizing the possible end-to-end hop count.
In this scenario, three constant bit rate (CBR) event dataflows CBR1, CBR2, and CBR3, each with different end-to-end deadline requirements are generated simultaneouslyat the source nodes. The end-to-end deadline requirementsof the three flows are 30 ms, 60 ms, 90 ms, respectively.According to the sample packet priority assignment policygiven in (7), PTx
CBR1 ¼ 2 PTxCBR2 ¼ 4, and PTx
CBR3 ¼ 5. Note that,the PTx values listed here are the initial packet priority levelassignment at the sender. The priority level of each packetis updated at each hop according to (7) and (17). For eachCBR flow, 2,000 packets are generated and sent to the sink.
The average end-to-end transmission latencies andaverage on-time delivery rates achieved by SDRCS andRAP are shown in Fig. 10. In Fig. 10(a), it can be observedthat both RAP and SDRCS provide service differentiationfor traffic flows with different end-to-end deadlinerequirement in terms of different average end-to-endtransmission latencies, because both designs provide prior-itized queuing and MAC support. However, RAP always re-sults in a higher end-to-end latency. In our simulatedenvironment setting, a log-normal shadow fading channelmodel is used to reflect the channel dynamics in real WSNdeployments. The channel quality enroute affects the end-to-end delay in terms of per-hop retransmission. Thetradeoff between larger per-hop forwarding distance andshorter per-hop latency plays an important role in deter-mining the end-to-end real-time performance. Since RAPassumes a perfect channel model, it simply chooses thenext-hop to maximize the per-hop forwarding distancebut fails to consider this tradeoff in its forwarding metricdesign. As a result, the average end-to-end delay issignificantly higher for RAP thatn for SDRCS. Moreover,beacuse RAP does not have a dynamic packet traversalspeed-estimation strategy in its protocol design, only abaseline missed-deadline packet drop policy is used in
tiated real-time communication scheme for event sensing in wire-1.06.018
100
80
60
40
20
100 50 20 10 5 2
110
90
70
50
30Aver
age
End-
to-E
nd D
elay
(s)
Source Rate (Pkt/s)
SDRCS(30ms)SDRCS(60ms)SDRCS(90ms)
RAP(30ms)RAP(60ms)RAP(90ms)
100
80
60
40
20
100 50 20 10 5 2
90
70
50
30
On-
time
Del
iver
y R
ate
(%)
Source Rate (Pkt/s)
SDRCS(30ms)SDRCS(60ms)SDRCS(90ms)
RAP(30ms)RAP(60ms)RAP(90ms)
Fig. 10. End-to-end real-time performance of SDRCS and RAP [11].
100
80
60
40
20
100 50 20 10 5 2
90
70
50
30
Aver
age
End-
to-E
nd D
elay
(ms)
Source Rate (Pkt/s)
MMSpeed(30ms)MMSpeed(60ms)MMSpeed(90ms)
SDRCS(30ms)SDRCS(60ms)SDRCS(90ms)
100
80
60
40
20
100 50 20 10 5 2
90
70
50
30
On-
time
Del
iver
y R
ate
(%)
Source Rate (Pkt/s)
MMSpeed(30ms)MMSpeed(60ms)MMSpeed(90ms)
SDRCS(30ms)SDRCS(60ms)SDRCS(90ms)
Fig. 11. End-to-end real-time performance of SDRCS and MMSpeed [10].
12 Y. Xue et al. / Computer Networks xxx (2011) xxx–xxx
end-to-end transmission. Therefore, unnecessary packetforwarding cannot be eliminated to improve bandwidthutilization. The long transmission latency and the lowbandwidth utilization also affect RAP performance in termsof the on-time delivery rate. Fig. 9(a) showing the on-timedelivery rates, reveals that SDRCS provides up to 40% high-er average on-time delivery rate for 2,000 packet transmis-sions and maintains steady on-time delivery rates formuch higher source rates without congestion.
In Fig. 11, the comparison of the real-time performanceof SDRCS and MMSpeed are shown, in terms of averageend-to-end transmission latency and the average on-timedelivery rate. In Fig. 11(a), it can be observed that bothschemes provide a relatively low average end-to-end la-tency in achieving soft real-time guarantees. This resultindicates that, in contrast to purely geographic forwarding,utilization of both geographic information and channelquality is necessary delay-sensitive communication. InFig. 11(b), it can be observed that, compared withMMSpeed, SDRCS improves the on-time delivery rate fortraffic of any priority level by approximately 20% whenthe event source rate is less than 5 Pkt/s. When the eventsource rate reaches 5 Pkt/s, MMSpeed starts to experiencenetwork congestion with a significantly decreased on-timedelivery rate. In contrast, SDRCS maintains a steady on-time delivery rate for high event source rates up to20 Pkt/s. The better real-time performance of SDRCS canbe attributed to the following factors.
First, SDRCS provides better overall per-hop transmissionlatency for traffic with low priority levels. According to the
Please cite this article in press as: Y. Xue et al., SDRCS: A service-differenless sensor networks, Comput. Netw. (2011), doi:10.1016/j.comnet.201
MAC operations described in Section 3.4, the average per-hop latency for SDRCS is 2200 l s regardless of the prioritylevels associated with the traffic. According to (8), this va-lue is approximately the same as the minimum per-hop la-tency for a dynamic IFS/BW extension-based approachwith a priority level equal to 2. However, for MMSpeed,traffic with a lower priority level experiences longer IFSand back-off times in each packet transmission attempt,which results in significantly increased transmission de-lays. Therefore, the dynamic IFS/BW extension basedMAC design adopted by MMSpeed leads to lower transmis-sion throughput and a lower on-time delivery rate, espe-cially for low priority-level traffic. Accordingly, as shownin Fig. 11(b), traffic with looser end-to-end deadlinerequirements experiences congestion earlier in MMSpeed;whereas in SDRCS, traffic of all priority levels of suffersfrom congestion at approximately the same source rate.In addition, SDRCS achieves a better on-time delivery rateat the same packet source rate.
4.3. Energy efficiency analysis
In this section, we discuss the energy efficiency of theSDRCS in terms of packet forwarding, packet duplication,and packet dropping. In Fig. 12, the average number of datapacket forwarded per unique end-to-end packet delivery(forwardings per delivery; FPD) is shown for SDRCS andMMSpeed under varying end-to-end deadline require-ments and source rates. FPD reflects the energy efficiencyof the protocol in terms of average end-to-end hop count,
tiated real-time communication scheme for event sensing in wire-1.06.018
10
20
30
40
50
60
70
80
10 20 30 40 50 60 70
Aver
age
FPD
Source Rate (Pkt/s)
SDRCS(30ms)SDRCS(60ms)
MMSpeed(30ms)MMSpeed(60ms)
Fig. 12. Energy efficiency performance of SDRCS and MMSpeed, in termsof the average number of packets forwarded for each unique end-to-endon-time packet delivery.
40
20
50 20 10 5 2
50
30
Dup
licat
ed P
acke
t Tra
nsm
issi
on R
ate
(%)
Source Rate (Pkt/s)
SDRCS(Single)SDRCS(Mixed)
MMSpeed(Single)MMSpeed(Mixed)
Fig. 13. Percentage of duplicated packets received at the sink for all on-time delivered packets with single or mixed priority CBR traffic flows(SDRCS vs. MMSpeed).
Y. Xue et al. / Computer Networks xxx (2011) xxx–xxx 13
average retransmission time, percentage of dropped pack-ets and duplicated packet rate. When the data rate is low,and the end-to-end deadline requirement is loose, mostpackets reach the sink on time. Therefore, FPD capturesthe product of the average end-to-end hop count and theaverage number of per-hop retransmissions introducedby channel fading, which accounts for the majority of en-ergy consumption in WSNs. The simulation results inFig. 12 show that SDRCS always results in a lower FPD thanMMSpeed under similar deadline and source rate require-ments, thus leading to better energy efficiency.
When the data rate is high and the end-to-end deadlinerequirement is tight, the network contention and conges-tion levels are increased, resulting in higher packet errorrates and longer per-hop transmission delays. Accordingly,the percentage of unschedulable packets or duplicatedpacket transmissions significantly increase. As a result,large amount of energy is wasted on the delivery of thesepackets. Under such circumstances, the energy efficiencyof the system depends on properly designed admissioncontrol and packet-dropping policies.
Compared with MMSpeed, SDRCS adopts an improvedpolicy based on a more accurate and adaptive packet tra-versal speed estimation method. MMSpeed uses table-based forwarding, where the neighbor list and the averagetransmission time between a pair of neighboring nodes areperiodically exchanged. The packet traversal speedachieved by a certain neighbor is determined by the senderbased on this periodically exchanged information. For anend-to-end packet delivery process, if any intermediatenode cannot find a neighbor satisfying the required packettraversal speed, the packet is dropped. Such a rigid packet-dropping policy requires an accurate end-to-end hop-count estimation and frequent neighbor information ex-change. Under dynamic network conditions, this drop pol-icy may experience a large percentage of false missed-deadline packet drops.
In contrast, with SDRCS, both the packet traversal speedand the required per-hop deadline are estimated based onthe information instantaneously updated at the receiver,which helps improve the accuracy of the schedulabilityestimation. In Fig. 11(a), it can be observed that the aver-age end-to-end latency for SDRCS increases with increas-ing source rate until it approaches the end-to-enddeadline requirement. This indicates that the EDM droppolicy correctly estimates most packet drops. Therefore,
Please cite this article in press as: Y. Xue et al., SDRCS: A service-differenless sensor networks, Comput. Netw. (2011), doi:10.1016/j.comnet.201
the average end-to-end transmission latency for each on-time packet delivery increases with an increasing networkqueuing delay. However, for MMSpeed, the average end-to-end latency does not increase proportionally withincreasing source rate, even when the network is fully con-gested. This situation indicates that MMSpeed’s tight droppolicy results in a large portion of schedulable packetsbeing dropped because the end-to-end transmission la-tency is close to the deadline requirement.
The energy efficiency of a communication protocol isalso affected by the number of duplicate packets gener-ated. To analyze this metric, two kinds of traffic flows wereset up: a single flow sent from one lower-right node withan end-to-end deadline of 30 ms and a mixed flow sentfrom two lower-right nodes with end-to-end deadlines of30 ms and 60 ms. The duplicated packet transmission ratesof the SDRCS and MMSpeed are shown in Fig. 13, whereSDRCS reduces duplicate packet transmissions in the net-work by more than 70% in comparison with MMSpeed.This difference is mainly due to MMSpeed’s probabilisticper-hop multicast mechanism, where the number of mul-ticast receivers in each hop is determined by an end-to-end link error rate estimation. However, due to channeldynamics, such a link error rate estimation is not accurate.Improper per-hop multicasting results in duplicated pack-ets being delivered to the sink and greatly reduces thebandwidth utilization. In contrast, SDRCS results in a muchbetter bandwidth utilization and a higher end-to-end on-time delivery rate.
Based on above analysis, we conclude that SDRCS yieldshigher energy efficiency for both low-data-rate, loose-deadline and high-data-rate, tight-deadline conditions.
4.4. Void avoidance performance
In this section, we discuss the void-avoidance perfor-mance of SDRCS and compare its end-to-end real-time per-formance with MMSpeed. A network topology wasmanually generated by removing 30 nodes from the net-work to create two communication voids located in thelower-left and top-right corner as shown in Fig. 14 withthe resulting RSS-based group formation. As described inSection 3.7, each node can obtain a group ID assignmentin the sample network topology because the network re-mains connected. The group ID assignment adapts to the
tiated real-time communication scheme for event sensing in wire-1.06.018
Fig. 14. A sample network topology with two communication voids andthe RSS-based group formation results with GRA = 2. The source nodes aremarked with solid squares.
14 Y. Xue et al. / Computer Networks xxx (2011) xxx–xxx
network void and reflects the end-to-end hop-count esti-mation instead of a distance estimation.
We repeated the simulation scenario described in Sec-tion 4.2 using three source nodes marked with solidsquares in Fig. 14. The average end-to-end transmission la-tency and on-time delivery rate are observed for bothSDRCS and MMSpeed. The simulation results are shownin Fig. 15. Compared with the real-time performance based
100
80
60
40
20
100 50 20 10 5 2
90
70
50
30
On-
time
Del
iver
y R
ate
(%)
Source Rate (Pkt/s)
MMSpeed(30ms)MMSpeed(60ms)MMSpeed(90ms)
SDRCS(30ms)SDRCS(60ms)SDRCS(90ms)
100
80
60
40
20
100 50 20 10 5 2
90
70
50
30
Aver
age
End-
to-E
nd D
elay
(ms)
Source Rate (Pkt/s)
MMSpeed(30ms)MMSpeed(60ms)MMSpeed(90ms)
SDRCS(30ms)SDRCS(60ms)SDRCS(90ms)
Fig. 15. End-to-end real-time performance of SDRCS and MMSpeed basedon the sample network topology with communication voids.
Please cite this article in press as: Y. Xue et al., SDRCS: A service-differenless sensor networks, Comput. Netw. (2011), doi:10.1016/j.comnet.201
on a void-free topology, as shown in Fig. 11, SDRCS did notsuffer from significant service degradation, whereasMMSpeed suffered severe degradation.
These results are attributed to two majors factors. First,SDRCS guarantees a route from the sender to the sink, irre-spective of the route taken by the dynamic forwarding pro-cess. Therefore, a communication void can only slightlyincrease the end-to-end hop count according to the group-ing results. Second, the network capacity bottleneck is lo-cated around the sink, where all the packets converge.Therefore, even though fewer parallel forwarding pathscan be taken from the sender to the sink, the end-to-endthroughput cannot be affected dramatically.
For MMSpeed, the on-time delivery rates dropped by10% for traffic flows at all priority levels. Since MMSpeedrelies on negative participation for void avoidance, someof the forwarding decisions may lead to broken routes. Asa result, relaying nodes drop packets and stop participatingin further transmissions. Since it takes time for the nega-tive participation to propagate in the reverse direction tothe upstream nodes and trigger new route exploration, anumber of packets are dropped due to communicationvoids. This situation becomes severe when the total num-ber of packet transmissions is small.
5. Conclusions
In this paper, we developed a service-differentiatedreal-time communication scheme (SDRCS) for multi-hopcommunication in WSNs. Using RSS-based grouping,SDRCS enables end-to-end hop-distance awareness forsensor nodes with a low control overhead. The hop-dis-tance estimation accuracy can be controlled by adjustingthe grouping granularity parameter to meet various appli-cation requirements. Along with this grouping approach, achannel-aware dynamic forwarding approach is utilized,with a polling contention-period based prioritized MACfor inter-node traffic differentiation. Compared with thecommonly adopted dynamic IFS/BW extension basedMAC approaches, the developed MAC design features bet-ter service differentiation capability with better bandwidthutilization when the number of priority levels in the net-work is greater than 4. By including a receiver contentionprocess in the MAC operation, the forwarding decision islocally performed to maximize packet traversal speed.Based on this MAC operation, we also designed a per-hopdeadline based prioritized queueing policy for intra-nodetraffic differentiation.
SDRCS requires no extra hardware for localization,transmission power adaptation, or multiple channel trans-mission support. It also adapts to network dynamics, suchas varying channel quality, local congestion and communi-cation voids. Our analysis showed that SDRCS achieves abetter on-time delivery ratio and a higher throughput, withbetter energy efficiency, than existing approaches such asRAP [11] or MMSpeed [10]. The dynamic forwarding tech-nique eliminates the neighbor exchange packets found inrecent approaches. This difference is especially beneficialin high density WSNs, where each node has a high degree.Consequently, SDRCS provides a complete design for real-time communication in WSNs.
tiated real-time communication scheme for event sensing in wire-1.06.018
Y. Xue et al. / Computer Networks xxx (2011) xxx–xxx 15
References
[1] Y. Xue, B. Ramamurthy, M.C. Vuran, A service-differentiated real-time communication scheme for wireless sensor networks, in:Proceedings of SenseApp ’08, Montreal, Canada, 2008.
[2] S. Oh, L. Schenato, P. Chen, S. Sastry, Tracking and coordination ofmultiple agents using sensor networks: system design, algorithmsand experiments, Proceedings of IEEE 95 (1) (2007) 234–254.
[3] J.R. Moyne, D.M. Tilbury, The emergence of industrial controlnetworks for manufacturing control, diagnostics, and safety data,Proceedings of IEEE 95 (1) (2007) 29–47.
[4] S.N. Pakzad, G.L. Fenves, S. Kim, D.E. Culler, Design andimplementation of scalable wireless sensor network for structuralmonitoring, Journal of Infrastructure Engineering 14 (1) (2008) 89–101.
[5] T. He, V.P., T. Yan, L. Luo, L. Gu, G. Zhou, Achieving real-time targettracking using wireless sensor networks, in: Proceedings of the IEEERTAS’06, San Jose, California, 2006, pp. 37–48.
[6] I.F. Akyildiz, T. Melodia, K.R. Chowdhury, A survey on wirelessmultimedia sensor networks, Computer Networks 51 (4) (2007)921–960.
[7] Y. Wang, M.C. Vuran, S. Goddard, Cross-layer analysis of the end-to-end delay distribution in wireless sensor networks, in: Proceedingsof IEEE Real-Time Systems Symposium (RTSS ’09), Washington, D.C.,2009, pp. 138–147.
[8] Y. Wang, M.C. Vuran, S. Goddard, Analysis of event detection delay inwireless sensor networks, in: Proceedings of IEEE InternationalConference on Computer Communications (INFOCOM ’11), in press.
[9] O. Chipara, Z. He, G. Xing, Q. Chen, X. Wang, C. Lu, J. Stankovic, T.Abdelzaher, Real-time power-aware routing in sensor networks, in:Proceedings of 14th IEEE International Workshop on Quality ofService: IWQoS 2006, New Haven, CT, 2006, pp. 83–92.
[10] E. Felemban, C. Lee, E. Ekici, MMSPEED: multipath multi-speedprotocol for QOS guarantee of reliability and timeliness in wirelesssensor networks, IEEE Transactions on Mobile Computing 5 (6)(2006) 738–754.
[11] C. Lu, B. Blum, T. Abdelzaher, J. Stankovic, T. He, RAP: A real-timecommunication architecture for large-scale wireless sensor networks,in: Proceedings of IEEE Real-Time and Embedded Technology andApplications Symposium (RTAS 2003), Washington, DC, 2003.
[12] T. He, J. Stankovic, C. Lu, T. Abdelzaher, A spatiotemporal communi-cation protocol for wireless sensor networks, IEEE Transactions onParallel and Distributed Systems 16 (2005) 995–1006.
[13] IEEE802.11WG, Draft supplement to IEEE standard 802.11-1999:Medium access control (MAC) enhancements for quality of service(QoS).
[14] I. Aad, C. Castelluccia, Differentiation mechanisms for IEEE 802.11,in: Proceedings of IEEE INFOCOM 2001, vol. 1, Anchorage, AK, 2001,pp. 209–218.
[15] M. Zorzi, R. Rao, Geographic random forwarding (GeRaF) for ad hocand sensor networks: multihop performance, IEEE Transactions onMobile Computing 2 (4) (2003) 337–348.
[16] M.C. Vuran, I.F. Akyildiz, XLP: a cross layer protocol for efficientcommunication in wireless sensor networks, IEEE Transactions onMobile Computing 9 (11) (2010) 1578–1591.
[17] P. Casari, M. Nati, C. Petrioli, et al., Efficient non-planar routingaround dead ends in sparse topologies using random forwarding, in:Proceedings of IEEE ICC’07, Glasgow, UK, 2007.
[18] L. Galluccio, A. Leonardi, G. Morabito, S. Palazzo, A MAC/routingcross-layer approach to geographic forwarding in wireless sensornetworks, Ad Hoc Networks 5 (6) (2007) 872–884.
[19] Y. Xue, M.C. Vuran, B. Ramamurthy, Cost-efficiency of Anycast-basedforwarding in duty-cycled WSNs with lossy channel, in: IEEEConference on Sensor, Mesh and Ad Hoc Communications andNetworks (SECON), Boston, MA, 2010, pp. 1–9.
[20] Y. Zhu, N.L.M., Probabilistic approach to provisioning guaranteed qosfor distributed event detection, in: Proceedings of the IEEEINFOCOM’08, Phoenix, AZ, 2008, pp. 592–600.
[21] V.C. Gungor, O.B. Akan, I.F. Akyildiz, A real-time and reliabletransport (RT) 2 protocol for wireless sensor and actor networks,IEEE/ACM Transaction Networks 16 (2) (2008) 359–370.
[22] MicaZ, MicaZ wireless module, <http://www.cmt-gmbh.de/MICAz.pdf>.
[23] H. Zhang, A. Arora, Y. Choi, et al., Reliable bursty convergecast inwireless sensor networks, Computer Communication 30 (13) (2007)2560–2576.
[24] M. Zuniga, B. Krishnamachari, An analysis of unreliability andasymmetry in low-power wireless links, ACM Transaction onSensor Networks 3 (2) (2007) 7–31.
Please cite this article in press as: Y. Xue et al., SDRCS: A service-differenless sensor networks, Comput. Netw. (2011), doi:10.1016/j.comnet.201
[25] S. Vural, E. Ekici, Hop-distance based addressing and routing fordense sensor networks without location information, Ad HocNetworks 5 (4) (2007) 486–503.
[26] T.F. Abdelzaher, S. Prabh, R. Kiran, On real-time capacity limits ofmultihop wireless sensor networks, in: Proceedings of IEEE RTSS2004, Lisbon, Portugal, 2004.
[27] Y. Tian, A. Dogan, F. Özgüner, An urgency-based prioritized MAClayer protocol for real-time traffic in ad-hoc wireless networks, in:Proceedings of the 17th International Symposium on Parallel andDistributed Processing, Washington, USA, 2003.
[28] D.G. Kleinbaum, L.L. Kupper, K.E. Muller (Eds.), Applied RegressionAnalysis and Other Multivariable Methods, PWS Publishing Co.,Boston, MA, USA, 1988.
[29] D. Chen, P.K. Varshney, A survey of void handling techniques forgeographic routing in wireless networks, IEEE CommunicationsSurveys and Tutorials 9 (2007) 50–67.
[30] D. Chen, J. Deng, P.K. Varshney, On the forwardingarea of contention-based geographic forwarding for ad hoc and sensor networks, in:Proceedings of IEEE SECON 2005, Santa Clara, CA, 2005.
[31] X. Zeng, R. Bagrodia, M. Gerla, GloMoSim: A library for parallelsimulation of large-scale wireless networks, in: Proceedings ofWorkshop on Parallel and Distributed Simulation, Alberta, Canada,1998, pp. 154–161.
Yuyan Xue received the BS and MS degrees incontrol engineering from Nankai University,China, in 2002 and 2005, respectively. Shereceived the Ph.D. degree in computer engi-neering from University of Nebraska, Lincoln,under the supervision of Dr. Byrav Rama-murthy and Dr. Mehmet C. Vuran, in 2010.Her research interests include real-time andsecure communication, cross-layer design andanalysis in wireless sensor networks.
Byrav Ramamurthy is currently an AssociateProfessor in the Department of ComputerScience and Engineering at the University ofNebraska-Lincoln (UNL). He is the author ofthe book ‘‘Design of Optical WDM Networks –LAN, MAN and WAN Architectures’’ and a co-author of the book ‘‘Secure Group Communi-cations over Data Networks’’ published byKluwer Academic Publishers/Springer in 2000and 2004 respectively. He serves as the Chairof the IEEE Communication Society’s OpticalNetworking Technical Committee (ONTC). He
serves as the IEEE INFOCOM 2011 TPC Co-Chair. His research areasinclude optical and wireless networks, peer-to-peer networks for multi-media streaming, network security and telecommunications. His research
work is supported by the US National Science Foundation, US Departmentof Energy, US Department of Agriculture, AT& T Corporation, AgilentTech., Ciena, HP and OPNET Inc.Mehmet C. Vuran received his B.Sc. degree inelectrical and electronics engineering fromBilkent University, Ankara, Turkey, in 2002.He received his M.S. and Ph.D. degrees inelectrical and computer engineering fromBroadband Wireless Networking Laboratory,School of Electrical and Computer Engineer-ing, Georgia Institute of Technology, Atlanta,GA, under the supervision of Prof. Ian F.Akyildiz in 2004 and 2007, respectively. Cur-rently, he is an Assistant Professor in theDepartment of Computer Science and Engi-
neering at the University of Nebraska-Lincoln and director of Cyber-Physical Networking Laboratory. Dr. Vuran received the NSF CAREERaward in 2010. He has received numerous academic honors, including the
tiated real-time communication scheme for event sensing in wire-1.06.018
16 Y. Xue et al. / Computer Networks xxx (2011) xxx–xxx
2010 Maude Hammond Fling Faculty Interdisciplinary Research Fellow-ship from the UNL Research Council and the 2007 ECE Graduate ResearchAssistant Excellence Award from Georgia Tech’s School of Electrical andComputer Engineering. He is an associate editor of Computer NetworksJournal and Journal of Sensors. He is a member of the Institute of Elec-
Please cite this article in press as: Y. Xue et al., SDRCS: A service-differenless sensor networks, Comput. Netw. (2011), doi:10.1016/j.comnet.201
trical and Electronics Engineers (IEEE) and the IEEE CommunicationSociety. His current research interests include cross-layer design andanalysis, wireless sensor networks, underground sensor networks, cog-nitive radio networks, and deep space communication networks.
tiated real-time communication scheme for event sensing in wire-1.06.018
Top Related