Energy and QoS aware Routing in Wireless Sensor Networks

Kemal Akkaya and Mohamed Younis
Department of Computer Science and Electrical Engineering
University of Maryland, Baltimore County
Baltimore, MD 21250
kemal1 | younis@cs.umbc.edu

Abstract. Many new routing protocols have been proposed for wireless sensor networks in recent years. Almost
all of the routing protocols considered energy efficiency as the ultimate objective since energy is a very scarce
resource for sensor nodes. However, the introduction imaging sensors has posed additional challenges.
Transmission of imaging data requires both energy and QoS aware routing in order to ensure efficient usage of
the sensors and effective access to the gathered measurements. In this paper, we propose an energy-aware QoS
routing protocol for sensor networks which can also run efficiently with best-effort traffic. The protocol finds a
least-cost, delay-constrained path for real-time data in terms of link cost that captures nodes’ energy reserve,
transmission energy, error rate and other communication parameters. Moreover, the throughput for non-realtime data is maximized by adjusting the service rate for both real-time and non-real-time data at the sensor
nodes. Such adjustment of service rate is done by using two different mechanisms. Simulation results have
demonstrated the effectiveness of our approach for different metrics with respect to the baseline approach where
same link cost function is used without any service differentiation mechanism.
Keywords: Sensor networks, QoS routing, energy-aware routing, real-time traffic

1. Introduction
reveals some properties about objects located and/or
Recent advances in micro-electro-mechanical systems
(MEMS) and low power and highly integrated digital
electronics have led to the development of micro
sensors [1][2][3][4][5][5][7]. Such sensors are
generally equipped with data processing and
communication capabilities. The sensing circuitry
measures

ambient

conditions

related

to

the

environment surrounding the sensor and transforms
them into an electric signal. Processing such a signal

events happening in the vicinity of the sensor. The
sensor sends such sensed data, usually via radio
transmitter, to a command center either directly or
through a data concentration center (a gateway). The
gateway can perform fusion of the sensed data in
order to filter out erroneous data and anomalies and
to draw conclusions from the reported data over a
period of time.

The continuous decrease in the size and cost of

research. Current research on routing in wireless

sensors has motivated intensive research in the past

sensor networks mostly focused on protocols that are

few years addressing the potential of collaboration

energy aware to maximize the lifetime of the

among sensors in data gathering and processing via

network, scalable for large number of sensor nodes

an ad hoc wireless network. Networking unattended

and tolerant to sensor damage and battery exhaustion

sensor nodes is expected to have significant impact

[2][4][8][9][11][10][12]. Since the data they deal

on the efficiency of many military and civil

with is not in large amounts and flow in low rates to

applications, such as combat field surveillance,

the sink, the concepts of latency, throughput and

security and disaster management. A network of

delay were not primary concerns in most of the

sensors can be used to gather meteorological

published work on sensor networks. However, the

variables such as temperature and pressure. These

introduction of imaging sensors has posed additional

measurements can be used in preparing forecasts or

challenges

detecting harsh natural phenomena. In disaster

Transmission of imaging data requires careful

management situations such as earthquakes, sensor

handling in order to ensure that end-to-end delay is

networks can be used to selectively map the affected

within acceptable range. Such performance metrics

regions directing emergency response units to

are usually referred to as quality of service (QoS) of

survivors. In military situations, sensor networks can

the communication network. Therefore, collecting

be used in surveillance missions and can be used to

sensed imaging data requires both energy and QoS

detect moving targets, chemical gases, or presence of

aware routing in order to ensure efficient usage of the

micro-agents.

sensors and effective access to the gathered

However, sensor nodes are constrained in energy

for

routing

in

sensor

networks.

measurements.

supply and bandwidth. Such constraints combined

QoS protocols in sensor networks have several

with a typical deployment of large number of sensor

applications including real time target tracking in

nodes have necessitated energy-awareness at the

battle environments, emergent event triggering in

layers of networking protocol stack including

monitoring applications etc. Consider the following

network layer. Routing of sensor data has been one of

scenario: In a battle environment it is crucial to

the challenging areas in wireless sensor network

locate, detect and identify a target. In order to identify

a target. After and maximizes the throughput for best effort traffic is locating and detecting the target without the need of found. is defined for each link. Sensor Network Architecture Energy-aware QoS routing in sensor networks will ensure guaranteed bandwidth (or delay) through A set of sensors is spread throughout an area of interest to detect and possibly track events/targets in the duratio n of a connection as well as providing the this area. In that case. Our proposed protocol Sensor nodes Gateway Node extends the routing approach in [12] and considers Command Node only end-to-end delay. The sensors are battery-operated with use of the most energy efficient path. Alternative paths with bigger costs are tried until one. it is a sensor network architecture that we consider and battle environment. this requires a real-time data summarize the related work. However. with minimum possible delay. The availability of imaging sensors is of particular interest due to the present an energy-aware QoS routing mechanism for wireless sensor networks. we with limited data processing engines. In this paper. we can turn on those sensors to get overhead to the sensors. In section 2. 1 : Three-tier sensor network architecture . for instance an image of the target periodically for In the balance of this section we describe the sending to the base station or gateway. The cost function which captures remaining and Command Node transmission energy and error rate. guarantee the reliable delivery of the real-time data. which meets the end-to-end delay requirement Command Node Fig. no previous research has addressed QoS routing in sensor networks. we should deal networks and describe our approach. The protocol looks for a Command Node delay-constrained path with the least possible cost. we should employ imaging sensors. Section 3 with real-time data. Our protocol does not introduce any extra imaging sensors. a service Finally we conclude the paper in section 4 and outline differentiation mechanism is needed in order to our future research. To the best of diverse capabilities and types and are empowered our knowledge. 1. which requires certain bandwidth includes simulations and evaluations of the protocol. we analyze exchange between sensors and controller in order to the complexity of the QoS routing problem in sensor take the proper actions. Since.1.

[13]. For sensor nodes and interacting with command nodes. coordinating communication among technical issues untraditional and challenging. using the receiver can be independently turned on and off and approaches presented in [1][14]. rescue vehicles and evacuation helicopters are organization. only focus on the energy-aware and QoS routing of The architecture is depicted in Fig 1. while paramedics. the naturally resource nodes to achieve a mission. All the sensors are assumed to be considered for typical networks. can be stationary or mobile. In addition both the radio transmitter and to be active in signal processing. e. It . A gateway node command node communication protocol.quality of service constraints associated with data is worth noting that most of these capabilities are generated by such sensors. The mission for these available on some of the advanced sensors. the transmission power can be programmed for a required range. While physical proximity of sensors. While The sensor is assumed to be capable of operating the gateway will take charge of sensor organization in an active mode or a low-power stand-by mode. Command nodes The gateway node is assumed to know its location.g. one or multiple command nodes. fusing data collected by constrained sensor-based environment makes these sensor nodes. e.g. It is also assumed that the sensor can act as a relay to forward data from another sensor. In a disaster management e. In this paper. we assume knowledge of which sensors need and off. support of is a less energy-constrained node deployed in the QoS traffic generated by imaging sensors. network management. a factor that has not been nodes as stationary.g. The based on the mission and available energy in each sensing and processing circuits can be powered on sensor. via the use of GPS. etc. example energy efficiency has to be a core objective We are considering both the gateway and sensor of the system design. gateway to examples of mobile command nodes. the sensors is dynamically changing to serve the need of Acoustic Ballistic Module from SenTech Inc. fire interesting issues such as mission-oriented sensor trucks. coordination centers are typical The described system’s architecture raises many stationary command nodes. we within the communication range of the gateway node. sensor data among the communicating nodes. The gateway is many of these issues are studied in the context of responsible for organizing the activities at sensor wireless networking research. environment.

Using routes found cases where real-time or multimedia data are through the network core. Related Work protocols consider the imprecise state information In traditional best-effort routing throughput and delay while determining the routes [20][21]. Lin [23] and Zhu et al. multiple paths from sink to sensors are formed. CEDAR is are the main concerns. By using created wireless environments and limited resources.2. Both protocols routing can build a QoS route from a source to destination approaches address unconstrained traffic. for TDMA-based ad-hoc networks. several new protocols have One of these paths is selected according to the energy been proposed for QoS routing in wireless ad-hoc resources and achievable QoS on each path. will cost too much in terms of resource usage to bandwidth and delay jitter are needed. bandwidth. However. Such reconstruct the core. such as trees. if any node in the core is broken. some performance found. The been proposed for routing QoS constrained real-time SAR protocol creates trees routed from one-hop multimedia networks neighbor of the sink by taking the QoS metric. they cannot be directly applied energy resource on each path and the priority level of to wireless networks due to inherent characteristics of each packet into consideration. However. While contemporary best-effort reservation in a connection-oriented communication Another protocol for wireless networks that in order to meet the QoS requirements for each includes some notion of QoS in its routing decisions individual connection. in some determining the paths [22]. Therefore. requirement.1. In our networks taking the dynamic nature (due to mobility approach. the [15][16][17][18][19] . but maximize the throughput for best data in Some wire-based of these proposed . which uses the idea of certain performance in throughput or delay will be core nodes (dominating set) of the network while ensured throughout the connection. While many mechanisms have is the Sequential Assignment Routing (SAR) [4]. it guarantees in certain metrics such as delay. The bandwidth calculation routing is usually performed through resource is done hop-by-hop in a distributed fashion. There is no guarantee that a another QoS aware protocol. we not only select a path from a list of of the nodes) of the network into account candidate paths that meet the end-to-end delay [20][21][22][23][24]. QoS with reserved bandwidth. [24] have guarantees can be achieved by employing special proposed QoS routing protocols specifically designed mechanisms known as QoS routing protocols. a QoS path can be easily involved in communication.

none of them considers any QoS or (PCPO) problems. This is because most of the crit ical applications with the QoS parameters. which are proved to be NP- service differentiation mechanism in order to handle complete [25]. Although they are well such as battlefield surveillance have to receive for suited to mobile ad hoc networks. in this case we have both each sensor node and maintaining the multiple paths real-time and non-real-time traffic coexisting in the from each node to the sink. Therefore it is important to prevent the real- will be an over-kill for the systems where nodes are time traffic from consuming the bulk of network not mobile and have limited resources. Such 2. protocols proposed specifically for wireless sensor The described QoS routing problem is very networks are designed according to the needs of similar to typical path constrained path optimization sensor networks. End-to- . applications of sensor networks. challenges posed by imaging sensors and real-time which meets the end-to-end delay path constraint. in our case there is an extra goal. such as bandwidth and leave non-real-time data starving and bandwidth and energy. the SAR approach end delay requirements are associated only with the suffers the overhead of maintaining the node states at real-time data. require sensor’s involvement in route setup. We are trying to find least-cost path. Our protocol does not network. In addition. function for each link and used a K least cost path algorithm to find a set of candidate routes.effort traffic as well. Energy-aware QoS Routing Our aim is to find an optimal path to the gateway in routes are checked against the end-to-end constraints and the one that provides maximum throughput is terms of energy consumption and error rate while meeting the end-to-end delay requirements. but need to this section are based on the mobility of the nodes maximize the throughput for non-real time traffic as and none of them consider energy awareness along well. Our proposed However. Note that. which makes the problem more complex. the emerging instance acoustic data regularly in order not to miss complexity from mobility in such routing algorithms targets. which is approach tackles these challenges into account so that basically to maximize the throughput of non-real-time both the system lifetime will be maximized and QoS traffic. On the other hand. routing thus incurring large amount of delay. Our approach is based on associating a cost requirements are met. We not only should find paths that meet the Most of the QoS routing algorithms discussed in requirements for real-time traffic.

There is also a scheduler. The model we employ is G inspired from class-based queuing [26].picked. Before explaining the details of proposed Queuing model on a particular node algorithm. Real-time packet Non-real-time packet Sensing only node Relaying node 2. both classes can borrow bandwidth from each other when one of the two types of traffic is non-existent or under the limit. First we calculate the candidate paths without considering the end-toend delay. the appropriate queue. we introduce the queuing model. We use different queues for the two different types of traffic. 3. Fig. As indicated in Figure 3. Our approach is based on a two-step strategy incorporating both link-based costs and end-to-end constraints. Moreover. which checks the type of the incoming packet and sends it to real-time data on sensor node i. Therefore we should first find a list of candidate least-cost paths and then select one that meets the end-to-end delay requirement. this r-value is also used to calculate the service rate of real-time and non-real-time traffic on that bwNRT = (1 − ri )bwij bwRT = ri bwij Fig. is actually an initial value set by the gateway and represents the amount of bandwidth to be dedicated both to the real-time and non-real-time traffic on a particular outgoing link. What we do is simply calculate costs for . On each node there is a classifier.1 Queuing Model Classifier The queuing model is specifically designed for the G Gateway node case of coexistence of real-time and non-real-time Scheduler traffic in each sensor node. 2. Bandwidth sharing and service rates for a sensor node we cannot calculate the end-to-end delay for a particular path without knowing the r-value. with ri µ and (1 − ri ) µ being (normal) respectively the service rate for real-time and non- traffic whose packets are labeled accordingly. The bandwidth ratio r. 2. we have real-time traffic and non-real-time particular node. which determines the order of packets to be RT serviceratei = ri µ i NRTservicerate i = (1 − ri ) µ transmitted from the queues according to the bandwidth ratio “r” of each type of traffic on that j link. Queuing model in a particular sensor node Basically. Since the queuing delay depends on this r-value. The model is depicted in Fig.

Cost factors delay for a particular path P. The more energy the node contains. Once we obtain these candidate transmission paths. where l c 0 is a weighting constant and the parameter l . which favors nodes with more requirement. We define the following better it is for routing. • cost function for a link between nodes i and j: cos t ij = 2 ∑ CF k ( ) ( ) = c 0 × dist ij + c1 × f energy j + k =0 l function of distance between nodes i and j and buffer size on node j (i. cost function.2 Calculation of link costs CF1 (Energy Stock)= c1 × f (energy j ) . and typically equals of Dijkstra's algorithm to find an ascending set of to 2.3 Estimation of end-to-end delay for a path In order to find a QoS path for sending real-time data to the gateway. we further check them to identify those that proportional to the distance raised to some power meet our end-to-end QoS requirements by trying to l. we introduce the are defined as follows: notation below: • CF0 (Communication Cost)= c 0 × (dist ij ) . The closer a node to the destination. thus will be avoided. • CF2 (Error rate)= c 2 × f (eij ) where f is a f (eij ) is the function for finding the error rate on the link between i and j. it’s not part of the cost function. 2.e. • f (energy j ) is the function for finding current residual energy of node j. c 2 × f (eij ) where. particular link separately except the end-to-end delay energy usage rate). the less its find an optimal r-value that will also maximize the cost factor CF0 and more attractive it is for throughput for non-real-time traffic. This We consider the factors for the cost function on each factor reflects the remaining battery lifetime (i.each particular link and then use an extended version depends on the environment. The links with high error rate will increase the • dist ij is the distance between the nodes i and j. • 2.e. dist ij / buffer _ size j ). Before explaining the computation of the Hence. end-to-end delay requirement should be met. power. the all the links on that path). which is directly routing. which should be for the whole path (i. energy.e. This factor reflects the cost of the wireless least cost paths.

Total realtime data rate by pi nodes will be p i λRT and total end delay will be: Tend −end = ∑ ∀i∈Path 1 qi ri µ − pi λRT − ∑ p j λ(RTj ) j =1 [2] . TQRT on a node i is: (i ) QRT = traffic 1 ri µ − λ(RTi) [1] We make an approximation to simplify the end-to- TE : End-to-end queuing delay for a particular path P (ignoring propagation delay) end queuing delay by assuming the incoming traffic to Tend −end Nodes and non-real-time queues are stochastically independent. total (i ) queuing delay (including the service time). total end-toWe also assume that all the imaging sensors have the same real-time data generation rate λ RT . Thus.λ RT : Real-time data generation rate for imaging sensors ri µ : Service rate for real-time data on sensor real-time data rate by q i nodes will be added recursively for each relaying only node. Hence. Since we ignore the propagatio n delay. the end-to-end Trequired : End-to-end delay requirement for all paths m real-time : End-to-end delay for a particular path P : The number of nodes on path P : The set of all the sensing nodes that are part of path P queuing delay for a particular path is: TE = ∑Q (i ) RT = ∀i∈Path 1 = ∑ (i ) ∀i∈Path ri µ −λRT ∀i∈Path ∑ 1 qi ri µ − pi λRT − ∑ p j λ(RTj) j =1 We assume that the propagation delay is negligible. Then total real-time data load on a sensor node is: node i (1 − ri )µ : Service rate for non-real-time data on sensor node i pi : The number of sensing neighbors (data generators) of node i on path P qi λ(iRT) = p i λRT + : The number of relaying neighbors (data forwarders) of node i on path P λ(iRT) : Real-time data rate on sensor node i (i ) QRT : Queuing delay on a node i for real-time qi ∑p λ j j =1 ( j) RT The average waiting time including the service time in the queue in M/M/1 model is stated as W = 1 µ−λ where µ is the link transmission rate or service rate and ? is the packet arrival rate [27].

Moreover. an r-value is calculated on the current path . We divide Trequired into m equal time slots. finding the optimal finding an r-value which will satisfy the last hop r-values for each link as far as the queuing delay is node’s delay since the last node will be getting the concerned. 4. we propose the algorithm shown in Fig. The Tend− end = ∑ ∀i∈Path 1 qi rµ − pi λ RT − ∑ p j λ(RTj ) j =1 Then the problem is stated as an optimization problem as follows:   Max  ∑ ((1 − r ) µ )   ∀i∈ Path  subject to : Tend −end ≤ Trequired and 0 ≤ r < 1 algorithm calculates the cost for each link. the distribution of these other nodes before the last node will already be r-values to each node is not an easy task because each satisfied with that r-value and will use the same value should be unicasted to the proper sensor node value. for each node the least cost path to the gateway is found by running Dijkstra’s shortest path algorithm in line 2. the problem to solve. For each sensor node that has imaging capability. we follow The calculation of r for the last hop node m is as an approach. rather than broadcasting it to all the sensors. we define each follows: Trequired m m = QRT = 1 rµ − λRT ( pm + r-value to be same on each link so that the optimization problem will be simple and this unique r= m µTrequired + λRT (p + ∑ p ) µ m ∀k∈Nodesk ∑p ) k ∀k∈Nodes [3] r-value can be easily broadcasted to all the sensors by By considering the optimization problem the gateway.2. As a consequence. which meets the constraints and formula will be stated as: maximizes the throughput for non-real-time data. Then. Between lines 5-15. Therefore. will be very difficult optimization actual longest queuing delay. might bring a lot of overhead.2. for simplification we consider to-end delay for a particular path. line 1 of Fig. above. which will eliminate the overhead and complexity of the problem. appropriate rvalues are calculated for paths from imaging sensors to the gateway. based on the cost function defined in section 3. Basically. 4. to If we let all r-values be same for every link then the find a least-cost path.4 Single-r Mechanism In order to find r-value from the above inequality of While we generate a formula for calculating the end- Tend −end ≤ Trequired . which where m is the number of nodes on a particular path.

e. Since.1)) then 13 break. the use it to derive their r-value. Interested reader is referred to [28] for further information. The least-cost path is the whole network (line 17). 4.1 Calculate algorithm can suffer loops during execution. the r-values are stored in a every node on the path proportional to arrival rate of list. only node- Tend −end ( pik ) = Trequired 12 if (r is in range [0. K nodes on the path. 14 if no appropriate r is found 15 Reject the connection 16 end 17 Find max r from the list efficiency. The gateway then broadcasts the shortest paths). the connection request of that node nodes for better resource allocation. The gateway calculates a delay factor “∆d” the end-to-end delay requirement for all the paths by dividing the value of the end-to-end delay “d” by established from imaging sensors to the gateway. That r-value will satisfy picked. ∀i . we modified an extended version of value of “∆d” to all nodes on the path so that they can Dijkstra’s algorithm given in [28]. K values for each node by assuming a unique r-value different least-cost paths are tried in order to find a for each node. Since. extended Since the single -r mechanism is just an Dijsktra algorithm for K-shortest path is run in order approximation to the optimal solution of allocating r- to find alternative paths with bigger costs (line 9). j ∈ V 2 Find least cost path for each node by using Dijkstra 3 for each imaging sensor node i do 4 begin 5 Compute r from Tend −end ( pi ) = Trequired (as above) modified the algorithm in such a way that each time a 6 7 8 disjoint paths are considered during the process. 2. . In order to find to the gateway is rejected. we extended the model so that it will proper r-value between 0 and 1 (lines 10-13).1)) then Add r to a list corresponding to node i else 9 Find K least cost paths 10 for each 11 k∈K (P ) to the gateway eliminates loops and ensures simplicity and K i do Recompute r from new path is searched for a particular node.5 Multi-r Mechanism Fig. If there allow different r-values to be assigned to sensor is no such r-value. we cos t ij . each node’s r-value is calculated The algorithm might generate different r-values by setting maximum allowable queuing delay for for different paths. the maximum of them is selected to be used for real-time traffic to that node. the accumulative arrival rates of real-time traffic at all In order to find the K least cost paths (i. This if (r is in range [0. If that value is not between 0 and 1. Pseudo code for the proposed algorithm (line 5). This might also help finding a proper rvalue easily since that node-disjoint path will not inherit the congestion in the former path. different r-values.

∆d = ri µ − λ RT algorithm. This metric also shows how 1 qi   *  pi + ∑ p j  j =1   qi   1 λ ri = + RT *  pi + ∑ p j  ∆d * µ µ  j =1  efficient is the algorithm in energy consumption. is desirable. However. The name network partitioning [29] with the capacity set to 2Mbps. 3. the network within the cluster is said A free space propagation channel model is assumed to be partitioned. simulation independently.1. out of energy.  ∀i∈Path  sensitive. The gateway position is determined • Time to first node to die: When the first node runs randomly within the boundaries of deployment area. Most energy aware routing algorithms try Then the problem will be to maximize the total to minimize the consumed energy. the throughput on each particular path: applications that deal with real-time data is delay   Max ∑ (1 − ri )  where 0 ≤ ri < 1 . This time and non-real-time traffic will be considered section describes the performance metrics. which maximizes the lifetime of the network. Experimental Results • Network Throughput: Defined as the total number of data packets received at the gateway divided by The effectiveness of the energy-aware QoS routing the simulation time. and experimental results. environment. • Average delay per packet: Defined as the average time a packet takes from a sensor node to the [5] gateway. 3. Packet lengths are 10 Kbit for data packets and 2 Kbit for routing .2 Environment Setup 3.Then ∆d will be calculated as follows: ∆d = reflects the fact that some routes become invalid and cluster-wide rerouting may be immanent. so this metric is important in our case. Trequired qi    p + ∑  i ∑ p j  * λRT ∀ i∈ Path  j =1  [4] • Average lifetime of a node: This gives a good measure of the network lifetime. A routing Each sensor node i will calculate its r-value ri by using ∆d as follows: From [1]. Performance Metrics In the experiments we have considered a network of We have used the following metrics to capture the 100 randomly placed nodes in a 1000×1000 meter performance of our QoS routing approach: square area. The throughput for both real- approach is validated through simulation.

which are the closest nodes to the target. Acoustic Ballistic Module from SenTech Inc.and refresh packets. Each sensor node is capable of taking the image of a target data packet is time-stamped when it is generated to allow the calculation of average delay per packet. The buffers for real-rime be 0. the imaging sensor set may change with the movement of the target. which have imaging capability is greater than usually turned off but can be turned on according to the normal rate. This si used to make the simulator more data and normal data have default size of 20 packets realistic and to simulate the deviation of the gateway [30]. The set of generated at a constant rate of 1 packet/sec. The default value is 6 packets/sec. sensing nodes is chosen to be the nodes on the convex For a node in the sensing state. packets are hull of sensors in the deployment area. energy level reaches 0. For the term CF1 in the cost We assume that the network is tasked with a target- function. we used the linear discharge curve of the tracking mission in the experiment. The sensing circuitry of other nodes are nodes. Since value is consistent with the specifications of the targets are assumed to come from outside the area. Each node is assumed to have an part of the cost. [13]. Therefore. During simulation. the sensing circuitry of all boundary nodes is always The real-time packet generation rate (λ RT ) for the turned on.01. In addition. We also assume that each service rate ( µ ) of 20 packets/sec is assumed. each packet has an energy field that is updated during the packet transmission to calculate the average energy per packet since our cost function defined for each link is using remaining energy as to identify it clearly and can turn on its imaging capability on demand. The initial set of alkaline battery [31]. This sensing nodes changes as the target moves. A packet drop probability is taken to initial energy of 5 joules. A node is considered non-functional if its energy model from the actual energy model of nodes. a small subset of current active nodes. . A the target movement. are selected to turn on their imaging capability.

we have not differentiated between packets and have used only one queue in each sensor node. In this case. That is. This target remains active until it leaves the deployment region.5 imaging sensors get activated. The default end-to-end 320 310 300 290 280 270 260 250 240 230 Baseline Single-r Multi-r 2 4 6 8 10 RT Data rate Fig. Different parameters are have compared this approach with our single -r and such as buffer size.The packet-generation rate for imaging sensors is As a baseline approach. We obtained by the simulation.e. It is assumed that only one target is active at a time. which 3. When we compare the average delay defined earlier in this section. Targets are uniformly from the range 4 meters/s to 6 meters/s and Baseline Single-r Multi-r 0. a new target is generated.5 1 0 delay requirement for a QoS path is taken to be 0. packet drop probability and real- multi-r mechanisms by looking at the average delay time data generation rates are considered in order to per packet. Dijkstra) are generated when imaging sensors are employed. Therefore.3 Performance Results accommodates all kinds of packets. Targets are assumed to start at a random the initial target position in order for the target to cross the convex hull region. no In this section. we present some performance results bandwidth sharing on any path is performed. per real-time packets generated in our model with the Performance comparison of three different protocols average delay per packet generated in single queue . 5. 2.8 2 sec [32]. Time characterized by having a constant speed chosen a constant direction chosen uniformly depending on 4 6 RT Data rate 8 10 Fig. average lifetime of a node and time to capture the effects on the performance metrics first node to die. The r-value is Time initially assumed to be 0 but it is recalculated as 1. Average delay per packet with different real-time data rates position outside the convex hull.5 These packets are labeled as real-time packets and 2 treated differently in sensor nodes. hence more packets function with same routing algorithm (i. we have used the same cost bigger than the normal sensors. Average lifetime of a node with different real-time data rates without doing any service differentiation. 6.

Hence. the energy usage of the single -r mechanism is slightly less than the others. However. Since the same cost function is used for all protocols. The results are depicted in figures 8 and 9. Furthermore. we ran simulation for different values of real-time packet data rates. it gets more difficult to satisfy increasing number of QoS paths. such decrease is very less. In figures 6 and 7. This is more efficient than the single -r case in which a unique r-value is imposed by the gateway for all the nodes. 7. sometimes an r-value for the whole 150 network cannot be found. This decreases the throughput 0 2 4 6 8 RT Data Rate 10 Fig. While the number of real-time packets increase. This is not the case for the baseline protocol.model. First. 5). causing the rejection of 100 Baseline Single-r Multi-r 50 some connections. Such increase incurs a little more energy consumption in the sensor nodes. Furthermore. For the single -r 250 200 Time mechanism. The average lifetime of a node and the time for first node to die decreases when real-time data increases. multi-r mechanism performs better than single -r mechanism as expected. Time for first node to die with different real-time data rates mechanisms have less average delay (See Fig. Effect of real-time data rate on throughput and delay In order to study the performance of the algorithm for different real-time data rates. we have looked at the non-real-time data throughput. causing the nodes to sense and transmit more packets. we observed that both multi-r and single -r However. On the other hand. the lifetime of the nodes and the time for first node to die are very close to each other as confirmed by figures 6 and 7. This is due to the priority given to real-time packets when transmitting to the gateway. in all cases the average delay per packet increases for higher rates and real-time data causes more queuing delay at each sensor node. every particular node adjusts its r-value based on the resources it has. we have looked at the energy usage of the protocols. On the other hand. for the multi-r mechanism. Because. it is easier to find an r-value for a particular node. the efficiency in the usage of resources for multi-r mechanism causes an increase in the throughput especially for non-realtime data as seen in figure 8. . this can cause rejection of paths or packet drops for non-real-time data causing throughput for such data to decrease. This can be explained 300 by looking at the throughput. hence fewer packets are relayed. becoming constant after a certain point (See figure 8).

Multi-r 0. The network r-value goes down while the end-toend delay requirement gets looser.4 increases with the rate since packets incur more 0. 2 requirement and monitored how this change affects 1.5 1 the network r-value. Non-real-time packet delay for different real-time data rates 10. In multi-r mechanism. The results are depicted in figure 0.5 queuing delay on the nodes. the nodes will be able to meet the end- multi-r mechanism is bigger than the single -r to-end delay requirement with a smaller r-value as . leading non-real-time 1 0. It is interesting to note that the average 1 1.5 3 stringent conditions. 0. 11.4 1 since the resources are handled more efficiently. 9 shows the effect of real-time data rate on 0.92 0.5 Single-r Multi-r 0 2 4 6 8 RT Date Rate 10 Fig.84 real-time data for the sake of real-time data.2 queuing delay and share the same amount of 0 bandwidth.3 mechanism. 9. Network r-value with different end-toend delay values mechanism has greater throughput than the single -r average delay per non-real-time packet.9 0.94 causing the throughput for non-real-time data ( (1 − r ) µ ) to stay greater than 0. the increase in the Throughput 2. Hence.5 Single-r Multi-r 0 2 4 6 8 RT Data Rate 10 Fig.2 End-to-End Delay Fig.5 throughput of non-real-time packets cause extra 2 1.8 2 4 6 RT Data Rate 8 10 Fig.96 0.5 Time packets to have more end-to-end delay. we varied the end-to-end delay 2. Since the delay is We restricted r-value to be strictly less than 1. Non-real-time data throughput for different real-time data rates Effect of end-to-end delay requirement and real-time date generation rate on r-values In order to see how the algorithm behaves under 3. 0. the r-value 0. 10. The delay 0.6 0.8 r-value Fig. 8.6 1. Network r-value with different real-time data rates packet delay for non-real-time packets in the case of not too strict.88 algorithm does not sacrifice the throughput for non- 0.86 0.

8 delay per packet increases with the buffer size since 0. we varied the buffer size to see if this has any effect on the performance of the algorithm.04 Packet Drop Probability 0.2 small number of hops and thus incurring less delay. The average 1 Time 0.03 0. The average delay per packet decreases buffers.4 0. This means that the packets that 1 0. more bandwidth will be required for real-time 280 Single-r Multi-r 270 packets.expected from equation 3.02 0.4 the throughput increases. This will cause the r-value to increase so that 260 0.02 0.8 0.05.04 Packet Drop Probability when there is enough space in the buffers. Average delay per RT packets for different packet drop probabilities increase the number of packets arriving to the gateway.01 to 0.6 0. Average delay per RT packets for different buffer size . Average lifetime of a node for different packet drop probabilities each node can serve more real-time packets (See figure 11).4 1. 12. Packets are not dropped 0.2 results are shown in figures 14 and 15. the probability that it will be arrive to the gateway are most probable to take a Time dropped increases. not all packets reach their destination and thus the Effect of packet drop probability on delay and node energy is conserved. the queuing model we employed uses buffers from 0. The average node lifetime increases since 10 15 Buffer Size 20 25 Fig.2 Single-r Multi-r 0 0.05 Fig.2 traverse increases. while 350 340 330 we congest the network with more real-time data packets by increasing the real-time data generation Time 320 310 300 290 rate.05 Fig.03 0.6 0. the throughput decreases due to lost packets. The results are depicted in figures in each node and there is a limit on the size of those 12 and 13. 14.01 0.01 0. we varied the probability of packet drop Since. This will 0. Single-r Multi-r 0 5 As expected. This can be explained by noting that as the number of hops the packet 1. On the other hand. 13. average lifetime of a node Effect of buffer size on delay and average lifetime of To study the effect of packet drop probability on a node performance. The packets from far nodes will be also able with the increasing probability. The 1.

Single -r mechanism sets a network wide rpackets arriving to the gateway will also increase the value for every sensor node. Average lifetime of a node for different buffer size allows the throughput for normal data not to diminish by utilizing that service rate on each node. we presented a new energy-aware QoS has provided better end-to-end delay for real-time routing protocol for sensor networks. a nodes causing an increase in the average delay per class-based queuing model is employed. throughput and average delay. A ratio r is defined as an initial 400 350 value set by the gateway and is used to calculate the 300 Time 250 amount of bandwidth to be dedicated to the real-time 200 150 and non-real-time traffic on a particular outgoing 100 Single-r Multi-r 50 0 5 10 15 Buffer Size 20 link.g. The protocol packets with a slight increase in energy usage. 15. which eventually increases the multi-r. Conclusion that use the same link cost. The multi-r mechanism In this paper. The effectiveness of the It is worth noting that. calculate their own r-value. The queuing non-real-time packets. The increasing number of presented. The selected queuing model for the protocol 25 Fig.real-time traffic. In the multi-r energy consumption by increasing the number of mechanism. e. the gateway broadcasts the necessary transmission and reception costs. to reach the gateway. It has finds QoS paths for real-time data with certain end- also increased the throughput for non-real-time data to-end delay requirements. 13. Simulation results have shown that our because of the more efficient adjustment of the packet protocol consistently performs well with respect to service rates on sensor nodes as depicted in figures QoS metrics. comparison to a baseline non-QoS aware protocol 4. for both delay and average protocol for both mechanisms is validated by lifetime metrics. namely single -r and mean more delay. 14 and 15. More packets from far nodes Two different mechanisms. multi-r mechanism performs better simulation. In order to support both packets. which has extended the queuing delay on the best effort and real-time traffic at the same time. in 12. for setting that service rate on each node are average delay per packet. model allows service sharing for real-time and non- . therefore decreasing information to the sensor nodes in order for them to the average lifetime of a node.

Seattle. "Low Power Wireless Sensor by Genetic Algorithms. March 2002. 16-27. Vol. Conference on Systems Through Artificial Neural Networks. Computer Networks. of the ACM. MASCOTS2002. Shah and J. October 2002. “Energy-Efficient Scalable Coordination in Sensor Networks. “Wireless [12] M. Bangalore.. Min et al. Seattle. we plan on addressing issues related to the Dust. 33. Protocols for Wireless [3] G. Washington.. Vol. No 5. J. March 422. Rabaey. MA. Rabaey et al. Akyildiz et al. "PicoRadio supports ad hoc ultra low power wireless networking. FL. WA. H. "Energy Aware Routing for Low Energy Ad Hoc Sensor Networks". July 2000. pp. pp. August 1999. 5. ASME Press. in the Proc. Seattle.. 7. May 2000. Orlando. 1998. Intanagonwiwat et al. propagation of that information through the network WA. [2] D. pp. Dagli. “Next Century Challenges: [11] W. October 2000. Katz et al.While our proposed protocol fits a fixed gateway [7] R. of IEEE Wireless Communications and survey”. the frequent information dissemination in wireless sensor update of the position of the gateway and the networks.com/ [14] A. Pottie and W. 2002.. Vol.” Communications of Cluster-Based Sensor Networks”. Boston. C.” in the Proc..). of MobiCom’99. Jamalabad. in the Proc. et al. 38.. relocation and mobility of the gateway under QoS [8] W. Kaiser.sentech-acoustic.” IEEE Personal Communications. [13] "Data sheet for the Acoustic Ballistic Module". http://www. SenTech Inc. Heinzelman et al. 393- Networking Conference. India . TX. Buczak and V.. "Protocols for self-organization of a wireless sensor network. January 2000.” in the Proc. “Adaptive protocols for traffic as a future work. VLSI Design. [6] J. In such cases. pp. 42-48.M. of MobiCom’99. References [9] C. Fort Worth. August 2000.” in Hawaii International August 1999. Estrin et al.H. 259-264. “Energy-Aware Routing in integrated network sensors. We plan to extend to model in order to handle the overhead of mobility and topology changes. F. New York. “Mobile Networking for Smart model. “Wireless sensor networks: a Proc. in the Proc. [4] K. [10] R. .” in Communication the Proc. J. Vol. Conference on System Sciences. of Int..." Intelligent Engineering Networks". of MobiCom’99. Sohrabi et al.. may excessively drain the energy of nodes. 43. 51 – 58. Younis et al.. in the [1] I. No. "Directed diffusion: A scalable and robust communication paradigm for sensor networks". Microsensor Networks. January 2001. pp. August 1999. (eds. Heinzelman et al. of MobiCom'00. Vol. 8. "Self- organization of a Heterogeneous Sensor Network [5] R." IEEE Computer.

Networks. Zhu and M. qosr-framework-06. Corson.” Research Report. S. August 1999.” IETF Internet draft draft-ietf-manet-cedar-spec-00. May 1999..” IEEE Communications Magazine.3-20. Vol. Crowcraft..R. “QoS-based routing in networks with inaccurate information: Theory and algorithms. Gerla et al. Feng et al. [22] R." IEEE Delay-Constrained Journal on Selected Area of Communications.1 pp. C. Quality-of-Service Routing in ad-hoc Networks. 1997. in the Proc. NY. draft-ietf- [20] R.M. “Propagation Measurements and Models for Wireless Communications [21] S. NJ. No. 75-83. Dallas. “Real-time tracking of moving objects with an active camera.” in Proc. Andresen et al.” RealTime Imaging Journal. [29] J.txt. “QoS routing for Service Constraints Integrated Communication mobile ad hoc networks. Japan.V. Orda. Querin and A.. [17] Q.” in the Proc. Jacobson. 33. of IEEE INFOCOM.. ReServation Protocol. June 1998. “Performance Evaluation of for Supporting Resource Reservation. 8. [31] M. of MobiCom’98. August 1995.. 17. INFOCOM. June 2002. TX.. of IEEE Networks. 4 pp. Vol. April 2002. 3 No. Ernesto et al. Sept [26] S. 2002. "QoS-based Routing [25] G. 1998." in the Proc. [19] E. “QoS and fairness constrained 1993.txt.” on wireless cellular and ad hoc networks. [16] Z. July 2000. Lagrange Relaxation.365-386." IEEE Network . of ICC'2002.. Ma and P. "Routing Subject to Quality of [24] C. of IEEE/ACM FOCUS'99. Vol.. Floyd and V. “On Demand QoS routing in Multihop Mobile Networks.. . 4. "Quality-of-Service routing with Performance Guarantees. Danilidis et al. Woo et al. Lee et al.” [18] L. No. No.Workshop on Quality of Service. [23] C. Algorithms Based on Linear and Nonlinear Sept 1996. Chen and K. [27] D. Crowley et al. “A framework for QoS based routing in the Internet. “Core extraction distributed ad hoc routing specification.” in the Proc. 1995. Channels. IEEE Routing New York. February 1998. Aug. "RSVP: A New Resource Least-Cost Selected areas in Communications. Lin. “Power aware routing in mobile ad hoc networks”. Zhang et al. of the 4 th Int. 1998. “Link Sharing and Resource Management Models IEEE/ACM for Packet Transactions on Networking. IEEE INFOCOM’97. May 1997. 1." IEEE Network . “Distributed Journal convex optimization of resource allocation for Proc. “The K shortest paths problem.” in the Proc... Julian et al. Vol. pp.” in the [28] Q. New Brunswick. “Wireless.Sivakumar et al. 1998 [32] K.” Internet-draft. Wang and J.[15] W. January 1995 [30] M. July/Aug. Mobile Ad-Hoc Network Routing. CISUC. Steenkiste.” IEICE Transactions on Communications. Nahrstedt.