You are on page 1of 8

2328 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 8, NO.

5, MAY 2009

Performance Evaluation of Efficient and Reliable
Routing Protocols for Fixed-Power Sensor Networks
Peter Kok Keong Loh, Hsu Wen Jing, and Yi Pan

Abstract—Fixed-power wireless sensor networks are prevalent energy limitations still limit sustainability of operations in
and cost-effective. However, they face mote failures, RF interfer- WSNs and new, low energy devices are still experimental
ence from environmental noise and energy constraints. Routing [2], [15], [21]. Several routing protocols in fixed-power, multi-
protocols for such networks must overcome these problems to
achieve reliability, energy efficiency and scalability in message hop WSNs use shortest-path routing [3], [8]. Since operation
delivery. Achievement of these requirements, however, poses is often over long unattended periods, the protocol must be
conflicting demands. In this paper, we propose an efficient and energy efficient. The environment is also unpredictable and
reliable routing protocol (EAR) that achieves reliable and scal- often disrupts operation. As such, routing protocols must
able performance with minimal compromise of energy efficiency. ensure that the WSN can reconfigure [6], [13], be energy
The routing design of EAR is based on four parameters -
expected path length and a weighted combination of distance efficient and resilient to failures [14], [21]. These non-trivial
traversed, energy levels and link transmission success history, to requirements pose conflicting demands on protocol design [5],
dynamically determine and maintain the best routes. Simulation [10], [12], [16], [26],[ 30]–[31]. With these issues in mind,
experiments of EAR with four existing protocols demonstrate we propose an efficient and reliable routing protocol (EAR)
that a design based on a combination of routing parameters that routes messages to one or more hubs for data-aggregation
exhibits collectively better performance than protocols based on
just hop-count and energy or those using flooding. applications. EAR takes into account the expected path length
and a weighted combination of distance traversed, energy level
Index Terms—Energy efficiency, hop count, reliable routing, and past performance of an RF link for its routing decisions.
routing protocol, wireless sensor network.
Control overheads in EAR are low.
The rest of this paper is organised as follows. Section II
I. I NTRODUCTION surveys some reliable and efficient routing protocols for multi-

W IRELESS sensor networks (WSNs) open up new ap- hop WSNs. Section III highlights our motivation. Section IV
plication areas such as tactical surveillance, intelligent describes the detailed design of EAR. Simulation results and
environmental and structural monitoring and target tracking some open problems are presented and discussed in Section
[18]. In a WSN, large numbers of tiny nodes (sensor motes) V. Section VI concludes this paper followed by the references
may be deployed in an ad hoc manner. These nodes automati- and acknowledgements.
cally configure a topology by communicating and coordinating
with each other [13]. II. R ELATED W ORK
Nodes assume the roles of both sensing device and router.
Here we discuss four recently proposed routing protocols
Messages are relayed to other nodes or to a hub in a multi-
for reliable and efficient many-to-one routing in multi-hop
hop fashion. Multi-hop routing in an energy-constrained WSN WSNs: Dynamic Source Routing (DSR) [19-20], Gradient-
has been shown to give rise to significant gains in network
Based Routing (GBR) [26], Gradient Broadcast (GRAB) [23],
performance [27]. With more nodes, the area being monitored
[28] and Adhoc On-Demand Vector routing (AODV) [11],
can be increased or with the same area, the increase in node
[24]. These routing protocols are similar in that they use
density gives more precise and timely data and also provides
neighbourhood information to route packets to the hub. In
a degree of operational reliability. Power-controlled networks
DSR, nodes dynamically discover a complete route across
have nodes with variable RF power transceivers that provide
multiple hops to the required destination in the network. To
greater routing performance at the expense of higher power
do this, each packet header contains the complete, ordered list
consumption and costs [8], [17], [25]. Fixed-power networks of traversed nodes (including the source node). The significant
have cheaper motes with fixed-power RF transceivers but may
information overheads here are partially offset by not requiring
be more prone to communication disruptions [2], [14]. Despite
periodic route updates and the monitoring of only routes in
advances in Micro-Electro Mechanical Systems technology, use. If an intermediate node is not the destination or it does
Manuscript received October 12, 2006; revised January 12, 2007; accepted not have any route to the destination in its route cache, it will
January 14, 2007. The associate editor coordinating the review of this paper initiate a route discovery process via request broadcasts to its
and approving it for publication was Y. B. Lin. neighbors. If available, the complete route to the destination is
P. K. K. Loh and W. J. Hsu are with Nanyang Technological University,
School of Computer Engineering, Nanyang Avenue, Singapore 639798 (e- found and returned to the initiator. Otherwise, the neighbour
mail: {askkloh, hsu}@ntu.edu.sg). appends its address to the route record and re-broadcasts the
Y. Pan is with Georgia State University, Department of Computer Science, route request to its neighbors. In noisy environments, route
34 Peachtree Street, Suite 1450, Atlanta, GA 30302-4110 USA (e-mail:
pan@cs.gsu.edu). discovery incurs large overheads. When routes become invalid,
Digital Object Identifier 10.1109/TWC.2009.060772 DSR adapts by sending a route error packet to the source
1536-1276/09$25.00 
c 2009 IEEE

An optimal route need not be the change in gradient must be propagated as far as needed to shortest and a more stable RF link may be more efficient keep other gradients consistent. A new packet’s header is The procedural design of EAR may be divided into three initialised with the source node’s cost to the hub and assigns phases: setup. increasing overheads. The onus is to prolong the operational same gradient. only the record traversed nodes but only the traversed node count.: PERFORMANCE EVALUATION OF EFFICIENT AND RELIABLE ROUTING PROTOCOLS FOR FIXED-POWER SENSOR NETWORKS 2329 node. nodes include their own costs in the packet header but do not need to designate next hop nodes as flooding is used. multi-round discovery may be of its neighbour is the gradient of that link. a portion and a sequence number generated by the destination that of nodes will receive a RREP packet before they have begun indicates whether this information is current or dated. This is repeated until the source nodes using that link are keeps only one route entry for a neighbour that has a route to reached. This enables faster propagation of entry also records the addresses of active neighbors through routes and saves on the amount of control packets generated in which packets for the specified destination can be routed. neighbours with lower costs may continue routing packets. With multiple neighbours having links with the operating environment. No additional route management overheads dynamically is expensive. A node may store more than one route to the Whenever a link in use is no longer valid. Each their initialisation process. and data dissemination. A node’s cost is the minimum energy overhead to route a packet to IV. For each route entry in the route table. there the hub even though that neighbour could have multiple routes is reduced recovery overhead as the route request does not to the hub. reply has either a greater destination sequence number (more In GBR. the node Nodes do not propagate the ADV packet received. hub it skips the initialisation process. In AODV. A node link. The challenge then is to develop a optimal routes. Otherwise. hop-counts to the hub are computed for each current). Thus. or the same sequence number with a smaller hop node and the difference between a node’s hop count and that count. A best route is stored. When a packet has enough credit. it lowers its gradient to not account for dynamically varying RF link conditions in discourage other nodes from routing packets through it. it will store the only one entry for each destination. broadcast a RREP packet if it has a route to a hub. The hub has a cost of 0 and the value increases with node distance from the hub. path to the hub. Duplicate packets that are routed to enhance significant overhead. The performance of the routing protocol also has transmission varies from one node pair to another. when a node receives a RREQ packet. it will store the route to the hub in its routing table. packet III. thus established from nodes to the hub and all messages will flow in the direction of the greatest gradient. Only (ADV) packet indicating that it wants to receive RPT packets. By introducing random Each table entry records the next hop to that destination delay for each node to begin initialisation process. even on to scale with network size. Unlike DSR. This mechanism provides good reliability at the expense of high A. Intermediate are detailed in the following sections. lifetime of the WSN while tolerating mote failures and RF This can spread traffic over time but in WSNs where RF disruptions. AODV incurs route in its routing table. while minimizing compromise. Setup Phase energy consumption. The selection of best routes is described node that receives route replies. D ESIGN OF EAR the hub. which stops using the route. A route in the routing table is indexed using the next hop of that link immediately notifies the active neighbors of the node’s ID . the on-demand route discovery process is similar it will ignore the RREQ packet. As a result. It is also offset by the increased packet collisions. When it has at least one route to the similar information overheads as DSR with lower reliability. it will broadcast a Route Reply (RREP) width of the broadcast mesh with the latter being widest at packet. latency and increased packet collisions When a hub is powered on. When a node receives a RREP packet. routing is based on cost and credit.LOH et al. In noisy environments. Although this process incurs and reliable. the RF link with the next highest increased latency. In GRAB. Route recovery is then initiated. A When a neighbouring node around the hub receives this ADV packet’s credit is consumed at each intermediate node on the packet. Gradients are needed to establish a route. it broadcasts an Advertisement especially in a noisy and/or high traffic volume WSN. If delivery success incur additional energy consumption and a link fails or is disrupted. especially in a noisy. additional routing information is required to routing protocol that can meet these conflicting requirements identify better performing links. A node starts the of that packet may travel in multiple paths in a mesh from the initialisation process by broadcasting a Route Request (RREQ) source to the hub. This the operating environment.that is the ID of the neighbouring node. However. Nodes do not propagate RREQ to DSR except that each node maintains a routing table with packets. DSR and propagates the reply back towards the source only if the maintains multiple route entries in its routing table. Keeping track of the entire message route gradient is chosen. For better reliability. . updates its routing table entry next. duplicate copies before starting an initialisation process. one is randomly chosen to route the message. When a will forward the packet else it will be dropped. route selection. When a hub receives delivery depends greatly on the fault locations relative to the a RREQ packet. the setup phase. M OTIVATION FOR C URRENT W ORK routing is similar to distance-based techniques utilised by several existing protocols. the upstream node hub. By assigning an node is powered on. it delays for a random interval of time appropriate amount of credit to each packet. this increased in probability of packet asking for a route to a hub. These a credit value to the packet before broadcasting it. When a node detects that its energy Routing based on hop-count and node energy levels does level has decreased to 50% or less. it is done only once for each node. it will the source. disruptive are incurred. Similarly.

a variable Buf U tilLvl is used at each node to a better link.7 and WT = 0. Conversely. D ment. A metric. The route selected need not be the shortest blacklisted and omitted from the table. the route with the highest routing table. As RPT packets flow through these links. Selecting the shortest route at every a hub at the given time. BLT hreshold . Since the routing table contains only intermediate node will lead the packet to a hub and no node one entry to the next hop node. may be determined node in the route. PT rises (or drops) when threshold value. WE = 0. then it will be route selected. This allows for adaptiveness.0. from the routing table may be re-admitted again only after a period of time. To prevent potential deadlocks takes on a value from 0 to 100 and a higher value indicates from occurring. PE starts < Bmax (max size of buffer).0 to 1. Since each algorithm selects the shortest route in the routing table for node stores and maintains the best available RF links in its forwarding. Let h against the incoming route and the shorter route is admitted be the number of hops traversed by a packet. link replacement is initiated. In an actual environ.0 since the PE – energy level of the next hop node (0. Some RF links are affected by temporary external disruption and should be given the chance to be re- The first field is the expected number of hops the packet will admitted. a temporary loop may be possible. NO. This worst route is then compared Proof: We recall that ExpP athLen = N H × α. the best route is the shortest as it incurs the lowest ls=48 latency and consumes the least energy. ignores blacklisted links and targets the link with the lowest Proposition 1: EAR is deadlock-free in a connected wire- LinkScore to be replaced. but ExpP athLen is bounded by the network diameter. its size scales slowly with is revisited. WE and WT . The search packets to be routed. on energy conservation in path selection.2330 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS. the neighbouring Sensor nodes generate RPT packets at periodic intervals node with the highest LinkScore may have the same hop or sleep.0 to 100. MAY 2009 B. Route Selection Phase F Ideally. It is defined as: a sliding window that keeps track of the last N attempts to ExpP athLen = N H × α. If there is a tie in LinkScore. N H route packets on a specified link. When the next node receives the packet.3 could be the expected number. packets travel on the best route from a node to LinkScore is selected.0). An arbitrary value is initially assigned to PT as store the current utilisation level of the packet output buffer. where is an assigned weight such that 0.3 By assigning α >> 0. the probability of successful transmission with ExpP athLen. PT – transmission N umHopT raversed is the distance a packet has traversed success rate (0. a packet may favour a route with may be chosen allowing routing decisions to focus more better performing links rather than just the shortest route (Fig. it empirically but their sum must equal 1. WE minimum number of hops to reach the hub is at least 1. During this period of instability. Case 2: ExpP athLen > h: the data dissemination algo- rithm selects the route with the highest LinkScore in the routing table for forwarding. into the routing table. network size and multiple hubs. a deadlock can never . The algorithm is as follows: is higher.0 < α ≤ 1. The packet is forwarded to the next PT (0.0 to 100. in a low will increment N umHopT raversed by one and compare it noise environment. giving higher emphasis to the selection of high topology. the entry with the longest length is chosen to be replaced. When there is more than one entry less sensor network. α is defined as LinkScore = (PE × WE + PT × WT ). VOL. there must be changes in the network chosen instead. with the same LinkScore. If the number of hops that a packet has traversed exceeds noisy environment. LinkScore.0) and WT – assigned weight for and is initialised as 0.0 < α ≤ 1. If Buf U tilLvl is greater than at 100 and drops as a node consumes its energy resources. then Case 1: ExpP athLen ≤ h: the data dissemination the route with the higher LinkScore is admitted. is defined where BLT hreshold subsequent packet transmissions succeed (or fail). the packet will be relayed on the shortest route LinkScore is used when there are two links of different to the hub. Although fields in its header: ExpP athLen and N umHopT raversed. Links that are omitted Fig. in a 1). 8. In EAR. BLT hreshold .7 and WE = 0. In this scenario. Weights. WT = 0. For example. In this case. If a link fails to relay all is the number of hops from this node to the hub for the packets in the last N consecutive attempts.0). C. where 0. When a new link is received and the and there will always be at least one buffer space for transit routing table is full. less desirable ones will start to exhibit high packet loss rate and are eventually blacklisted C and omitted from the routing table. In the case of a tie. 5. An RPT packet count to the hub as the sender node and this may result contains information of interest to network users and has two in the packet being re-transmitted to the sender.0. LinkScore then the shortest route to the hub. Data Dissemination the shorter route is chosen.0). If there is a tie in route length. A the link performance is unknown.0 to 1. waiting for some event to happen. the packet will take reliability paths over energy conservation. – assigned weight for PE (0. This buffer control mechanism ensures that new routes with the same hub distance competing to be admitted packets will not be injected when the buffer is almost full to the routing table. the performance of an RF link varies with physical dis- tance and the terrain between nodes and should be accounted A B E G for in routing decisions. 1. shortest routes are initially ls=75 ls=94 Hub admitted into the routing table based on hop-count. Illustration of forwarding based on LinkScore metric. The mechanism uses have to traverse before it reaches the hub.

RTS/CTS packet compared to sending an explicit control shaking mechanism at the MAC layer. RTS and CTS packets have to be received bits per second. Nodes in blue have received updated route K be the size of the route information in bits. When α is set to 1. routing information between nodes is costly in terms of energy Proof: In a non-ideal environment. D. Node B may not be on the shortest route in 0 0 50 100 150 200 250 300 350 400 the routing table and in the worst case. 2. the packet will be forwarded on the shortest route and the packet will nodes upon hearing a RTS or CTS packet that is not addressed need a maximum of D hops to reach the hub. thus breaking the loop. we have: and processed by all nodes as part of the collision avoidance • Energy cost of piggybacking = K units. When 30 GRAB ExpP athLen > N umHopT raversed. . 4. and As an example. EAR uses the hand. Instead of explicit control packets. a packet eventually arrives at a hub after a energy consumption as the packet size increases. S-MAC uses the same (SCT S /B) + ((K + W )/B) + (SACK /B) seconds four-way handshaking mechanism as IEEE 802. Hence. the packet will be Number of nodes relayed along the longest route at every intermediate node and the packet will traverse N H hops until ExpP athLen = Fig. maximum of N H + D hops. EAR can use S-MAC (Sensor MAC) [17] • Latency cost of explicit control = (SRT S /B) + that has energy saving mechanisms. W be the size information from either node X’s RTS or node Y’s CTS of the routing control packet in bits and the bandwidth be B packet or both. Then. where D is the distance of the no extra packets need be generated and additional costs are furthest node from the hub in hops. In the worst case. negligible compared to the cost incurred in relaying explicit route information (control) packets. the packet is for.11 to achieve reliable link-to-link transmission. will eventually be exited.LOH et al. 3. Latency cost of mechanism employed by the MAC protocol [29]. where D is the distance of the AODV 60 furthest node from the hub in terms of hops. W + SACK ) units. be full. Fig. Periodically exchanging Corollary 1: The routing algorithm of EAR is livelock-free. By mation onto existing RTS and CTS packets incurs additional Theorem 2. One of the energy-saving where SRT S −RT S packet size in bits. to them will go into sleep mode. loops formed are consumption and bandwidth usage. 90 % of packets delivered successfully Proposition 2: The distance a packet travels before reaching 80 70 a hub is bounded by N H + D. 50 DSR EAR Proof: Consider a packet that is generated at node 40 GBR A. However. 2 illustrates and we assume the energy cost to transmit 1 bit is 1 unit. N umHopT raversed. SCT S −CT S packet mechanisms known as Overhearing Avoidance specifies that size in bits and SACK − ACK packet size in bits. 20 warded to its neighbour node B which has the highest 10 LinkScore. ExpP athLen = N H.: PERFORMANCE EVALUATION OF EFFICIENT AND RELIABLE ROUTING PROTOCOLS FOR FIXED-POWER SENSOR NETWORKS 2331 Packet Delivery Ratio (PDR) with 10% Source Nodes 1 2 1 2 100 90 % of packets delivered successfully (1) Node 1 sends RTS (2) Node 2 sends CTS 80 70 AODV 60 DSR Legend 50 EAR Nodes without latest GBR 40 route info 1 2 GRAB Nodes with latest 30 route info 20 (3) Node 1 sends data packet 10 0 Fig. Route Update Cost analysis of route update: We show the energy Sensor nodes continually update “best” routes in the routing and latency costs in piggybacking route information on the table. The energy consumed is proportional to packet size piggybacked onto both RTS and CTS packets. Packet Delivery Ratio results (with 50% active sources). the packet will travel on the shortest 100 route to hub. 0 50 100 150 200 250 300 350 400 Number of nodes occur since BLT hreshold < Bmax and buffers will never Fig. Now. ising RTS-CTS handshaking instead of separate DATA-ACK • Energy cost of explicit control = (SRT S + SCT S + K + would result in more current route information for a node. Packet Delivery Ratio (PDR) with 50% Source Nodes If h ≥ ExpP athLen. the packet will have been routed to the furthest node from the hub. as in case 1. A packet will be forwarded on the shortest route if the packet output buffer exceeds BLT hreshold . ExpP athLen is bounded by the network diameter. Let this scenario. Piggybacking route infor- temporary and by Theorem 1. util- piggybacking = K/B seconds. Route information is packet. Packet Delivery Ratio results (with 10% active sources). Piggybacking on RTS/CTS packet.

Besides path length. Packet Latency results (with 50% active sources). 3–8. [13]. Packet latency measures the average time it takes to route V. [23] A. Packet Latency results (with 10% active sources). 3. It is expressed GloMoSim [1] was used to emulate a WSN with crossbow as  MICA2 motes [9] for data aggregation. nodes generate data packets that are routed to the hub located Total number of data packets delivered in the centre of the WSN. To ensure consistency. Also. 8. Fig. delivered. expressed as Multiple paths are kept in the routing table giving DSR a Total number of data packets succesfully delivered good degree of reliability. VOL. shadowing and multipath losses. 5. 4). The noise factor specifies metrics are recorded for 10% and 50% active source nodes the probability that packets to be received by that node are with a specified fault model: 50% of randomly selected nodes corrupted or lost. per delivered data packet. Energy Consumption results (with 10% active sources). We simulated measures the amount of control packet overhead of a routing network sizes from 50 to 400 nodes with 10% and 50% active protocol. the same model is applied to all routing protocols tested. GBR considers a node’s . This metric also indirectly that is compounded with noise and node failures. Total number of data packets delivered The routing protocols are subjected to a series of tests to We calculate energy expended in transmission and reception evaluate their performances in a realistic WSN environment by the nodes’ RF transceivers. 8. Fig.5 20 10 0 0 0 50 100 150 200 250 300 350 400 0 50 100 150 200 250 300 350 400 Number of nodes Number of nodes Fig. 5. DSR exhibits moderately high PDR × 100% Total number of data packets sent (Figs. The following standard metrics [10]. 7. Energy Consumption results (with 50% active sources).5 100 90 Energy expended per packet (mJ) 2 80 Latency per packet (s) 70 AODV AODV 1. In the simulation. source nodes. It is expressed as we used a simulation model that precludes effects such as  Energy expended by each node fading. To focus on performance variations Energy Consumption: This measures the energy expended due to design differences among the routing protocols tested. S IMULATION a data packet from the source node to the hub. Energy Consumption with 50% Source Nodes Packet Latency with 50% Source Nodes 250 8 Energy expended per packet (mJ) 7 200 6 Latency per packet (s) AODV AODV 150 5 DSR DSR EAR 4 EAR 100 GBR GBR 3 GRAB GRAB 2 50 1 0 0 0 50 100 150 200 250 300 350 400 0 50 100 150 200 250 300 350 400 Number of nodes Number of nodes Fig. 6.2332 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS.5 60 DSR DSR EAR 50 EAR GBR 40 GBR 1 GRAB GRAB 30 0. MAY 2009 Packet Latency with 10% Source Nodes Energy Consumption with 10% Source Nodes 2. Performance Evaluation were used: Packet Delivery Ratio: DSR operates in promiscuous mode Packet delivery ratio (PDR) measures the percentage such that nodes can obtain the latest routing information of data packets generated by nodes that are successfully and packets are routed on valid paths with high probability. Results were ranging from 10% to 50% to disrupt RF communications. 50% of randomly selected nodes fail fail and surviving nodes are exposed to a random noise factor at random times within the simulation duration. averaged over 30 runs each with a different seed and are shown in Figs. Every node except the hub takes on a random Fault Tolerance and Scalability: Values for the above noise factor between 10% and 50%. . all Individual data packet latency . NO.

As an improvement over AODV.023 449.406 331. Less control packets are needed compared DSR exhibits large packet latencies because its route discovery to AODV especially for route discovery. 6). itself. GRAB is less robust (Figs. 138. appreciably higher energy consumption levels of both AODV Hence.139 Spreading of network traffic reduces the chance of collisions.3s build up the cost field. especially manifested in larger networks above 150 cies. which incurs additional energy overheads during routing fewer routes are discovered with the consequence that the compared to data packets of routing protocols that carry only number of packets delivered is less (Figs. The improvement over DSR and GRAB is evident with larger networks and higher traffic levels (Fig. despite the piggy-backing. A packet broadcasted by a source node may consistency. 6). EAR achieved is proportional to network size. A single route discovery in AODV reveals intensive.479 235. average. 50% active). overheads.777 111. Besides expected path length. This is due to the higher control overhead compared permits a more gradual and uniform rate of energy consump- to GBR. With 50% GRAB incurs high levels of energy consumption as it floods source nodes.126 114.768 824.814 300. 4). packet transmission slows down. AODV (Fig. Every intermediate node tries to extract and energy overheads show an improvement over AODV’s for record information before forwarding a reply. Hence. energy and link performance history. 6). AODV 7. however. within the same time. Although route the resulting trends. In GBR. this is good tion over the network. GBR exhibits the lowest packet latencies of all protocols (Figs. There is no need for regular updates of routing provide for link-to-link transmission and acknowledgement. them from the routing table. This reduces total energy maintains only one routing table entry for each destination. However. 3. GRAB 61.473 683. 7. nodes (Table 1).578 183. Number of control packets than the other protocols. As network size increases. 3. With 50% active source nodes. More Energy Expended Per Packet (mJ) for 10% active sources 50 100 150 200 250 300 350 400 active sources contribute to a higher rate of packet collisions. GBR’s selection of the next hop node does not Energy Expended Per Packet (mJ) for 50% active sources 50 100 150 200 250 300 350 400 account for the stability of RF links that may vary over time.102 20. this process is initiated only once by also be lost for good during transmission as broadcast does not each node. the packet GBR and EAR exhibit the lowest energy overheads (Fig. 6). 3.6s (10% active).066 packets in a mesh to the hub. of EAR are competitive with that of GBR. 8).741 112. 4). With larger networks from about produced by GRAB fluctuates significantly across simulations 200 nodes upwards. AODV does not record traversed nodes. each node only maintains a single routing table entry for Energy Consumption: Route discovery in AODV is energy- each destination. it 5.426 293. 3.: PERFORMANCE EVALUATION OF EFFICIENT AND RELIABLE ROUTING PROTOCOLS FOR FIXED-POWER SENSOR NETWORKS 2333 TABLE I energy level to distribute packet traffic more uniformly among E NERGY C ONSUMED P ER PACKET FOR AODV AND GRAB all nodes. tables and hence. 3. The gradient change provement over GBR with larger networks may be attributed . of gradient changes to neighbouring nodes consume energy. GRAB’s takes more time. 4).362 314. However. the additional redundant routes available and therefore it shows widely fluctuation performance. The data packet carries pointers to the full route in less information than in DSR.559 215.041 206. when a node’s energy reserves fall below 50%. Hence. while route discovery in DSR yields more information and GRAB relative to the other protocols are evident from for delivery. DSR makes use of promiscuous mode to discovery process in AODV is similar to DSR.705 186.48 214. Energy overheads the node reduces its gradient with respect to its neighbours.394 175. Values recorded are: 36. energy level goes below the 50% threshold. 4).239 However. The additional energy consumed With both 10% and 50% active source nodes.05 seconds higher than GBR on route cache to reduce route discovery costs (Fig. the process is done only once for each node – when its 5.697 182.777 585. With an operating environment about 97% PDR on average compared to 82% for GRAB and that is plagued with dynamically occurring link disruptions 70% for GBR. Large to packets appreciably increase packet delivery success over error bars have also been observed in an independent study [3]. The result is exacerbated energy Packet Latency: GRAB and DSR incur the highest laten. 7. consumption as less re-transmissions and control packets are Unlike DSR. the consumption (200 nodes.4 seconds higher than GBR on average (Figs. 5. The same thing larger network sizes (Fig. consuming more energy. DSR uses a EAR’s packet latency is 0. 6). 94. route discovery broadcasts in reducing route update latencies. each node constantly update route information.713 131. and levels of GRAB are at least an order of magnitude higher 318.966 272.358 510. Energy is also expended to graph (Fig. traversed. 4).484 367. DSR can lead to significant energy consumption especially in this is more evident (Figs.233 GRAB has the second highest PDR on average as it floods GRAB 69. 50% active). The performance is consistent Route discovery packets will accordingly flood the network even with larger networks up to 400 nodes (Figs. latency is 0. The slight im- discouraging further routing through it. larger networks. DSR. 8). EAR’s routing and node failures.9s (300 nodes. As in AODV. As GBR directly monitors each node’s energy reserves. Despite the large error bars. Additional packet loss is contributed by collisions as flooding generates many is progressively propagated to neighbouring nodes to maintain redundant packets.033 104. latency is reduced significantly due to lower Although the route discovery process in AODV is similar to control overheads. off the packets in a mesh to the hub. 8).843 148. the happens when a data packet is routed from node to node. however. Excepting AODV. GRAB floods packets in a mesh to the hub. EAR blacklists The source will keep sending the route discovery packet but RF links with consistently poor performance and discards will not receive a definite route response from the destination.59 100. it may be very difficult to establish a full decisions are based on a weighted combination of distance route from source to the destination at a given point in time.474 256. neighbourhood information. Though the progressive propagation compromise resulting in appreciably higher PDR (Figs. 50% active). packet latencies increase exponentially. 4).9s (400 nodes. With network sizes of 350 nodes upwards. especially on networks with 200 nodes upwards (Figs. 5. thus needed.LOH et al. AODV 29.

Commun. R. neighbouring nodes to the hub route more packets than other [12] G. Like AODV and DSR. Goldsmith and S. “Networking issues in sensor networks. “A performance comparison of data GRAB depends very much on the location of node failures dissemination protocols for wireless sensor networks. L. pp. Takai. packet latency. Simulation results have shown that it performs com. Chua. vol. Its performance is largely [7] A. Estrin. exhibits consistently higher PDR than GBR over all network [9] CrossBowMICA2 mote specs [Online]. no. 14. E. “Ad hoc on-demand distance In the proposed network setup. current route information clustered WSNs.xbow. EAR sensor networks. One possible solution to this problem could be to [13] D. J. Oct. and energy consumption. “Performance evaluation of routing protocols for ad hoc there’s a higher chance of avoiding it as the mesh is wider. By and investigating the security aspects of EAR. We would like to thank the Nanyang Technological Univer- Fault Tolerance and Scalability: AODV and DSR utilise sity and Georgia State University for supporting our research on-demand route discovery based on flooding the network with efforts." B. Chang and L. Estrin. 4. G. “Fault-tolerant that is used up as packets get closer to the hub. Su. E. also scale well with network size. In networks with a single. Parallel and Distributed Computing. pp. 1. centrally-located hub. . nodes relay all data packets vector (AODV) routing. N.. As the cost analysis in Section IV-D shows. vol." IEEE/ACM Trans.. 272–285. Akyildiz. This paper has proposed a viable routing protocol. Feb. EAR. Yi Pan’s work was also supported by the 111 project route requests and are more suitable for a small. http://bullseye. Yu. Johnson. 16–29. data aggregation in fixed-power WSNs. L. Perkins. Apr.-H. 799–814. Girod. R. and M. R. and T. Royer. Das.. 3. delivery ratio. of China under the grant No." in Proc. C. W. no. Speech Signal Processing (ICASSP). “Physical layer driven protocol and algorithm design for energy-efficient wireless sensor networks. Networking. Hu. Fault tolerance exhibited by [3] T. Bokareva." IEEE/ACM Trans. VOL. Mobile Computing Networking. Cho. S. wireless networks. no. Sankarasubramaniam. and link disruptions. and E. pp. Estrin and J. EAR exhibits performance close to [8] J. and F. J." Internet draft. 2006. Pottie. “Instrumenting the world with wireless sensor networks. 272–287. is disseminated quickly with low energy costs. E. link disruptions and node failure rates draft-ietf-manet-dsr-09. 2004. Bourkerche. 12.-H. International Conf.to moderate. Available: sizes. GRAB is more suitable for small to moderate-sized [1] A. no. Silva. IEEE Globecom. the energy cost of a single ACKNOWLEDGEMENT route update is appreciably lower when piggybacking than using explicit control packets. Pazzi. Ye. Mobile Computing. R EFERENCES however. Tassiulas. “Self-configuring disruptions. Cerpa. 2001. “Performance comparison of two on-demand routing protocols for ad hoc networks. Zhao. overheads incurred and UCLA. 2003 (work in progress). Packet latency wireless sensor network routing protocols for the supervision of context- and energy consumption exhibited by GBR and EAR scale aware physical environments. “A survey packet collisions encountered rise disproportionately leading on sensor networks. 111-2-14. no. Bulusu. Open Issues IEEE Personal Commun. Computer Science Dept. Maltz. Heidemann. B. J. R. Tran. A. energy level and RF link performance fading channels. costs associated with route repair recommendations that have significantly improved this paper. Estrin. With scalable network simulation environment. no. Feb. Intanagonwiwat. F. “Wireless sensor networks: a bridge to the implemented with variable-powered motes. Y. Marina. and M. VI. vol. Y. The mesh-broadcast feature of GRAB may increase the robustness of data delivery in the presence of faulty nodes and noisy channels. and D. A. “GloMoSim: a networks with a small proportion of active source nodes. 85–89. 2002. R." IETF Internet draft.txt. faster rate. Parallel Distributed Computing. pp. 27–40. 7th Annual Intl. 9. Estrin." J. vol. physical world. based on expected path length and a weighted combination [17] A. C. We also thank the editor’s and anonymous reviewers’ node and link failures. 2004. Chandrakasan. 45. 102–114. especially in piggybacking on RTS/CTS packets. pp. E. Bajaj. [19] Y. vol. Bagrodia. [6] N. Small. 2004. Aug. N. due to its one-off approach in monitoring energy levels and its and A. 2003. “The dynamic sumption while operating in a noisy wireless environment source routing protocol for mobile ad hoc networks. NO. no. “ASCENT: adaptive self-configurable sensor adopt a clustering approach wherein cluster node members networks topologies. scalability and energy con. Deepak." Wireless Sensor Networks. Govindan. larger networks and heavy traffic. Jha. pp. Commun. Sinha.2334 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS. We acknowledge sized network with minimal noise and RF disruptions. vol. Apr.com/Products/productdetails. pp. draft-ietf-manet-aodv- to the hub. Royer. As Long Say Huan’s help in the development and testing of network size scales upwards and in the presence of appreciable EAR. 586–599. vol. Cayirci. D. and M. C. Heidemann. Jetcheva. A." IEEE Trans. 609–619. 7. 8. where network traffic. pp. 3. nodes. pp." IEEE Trans. E. Srivastava. Bulusu. B. 8." IEEE Trans. and D. 1. W. “A new networking model for biological petitively against existing routing protocols in terms of packet applications of ad hoc sensor networks. M." in Proc. When these occur nearer the source node. Gerla. Networking. GBR latency and energy consumption figures are localization systems: design and experimental evaluation. and S. no. 2004." in Proc. C. scale exponentially. [2] I. D." technical report 990027. 2006.aspx?sid=192 [10] S. for “Directed diffusion for wireless sensor networking. [4] A. to poor performance scalability. use of shortest routes whenever possible. [5] A. Elson. For packet latency Conf. 13. May 2001. [11] S." Mobile Networks Applications." J. pp. vol. [16] D. history." ACM Trans. 4. These nodes will drain their energy-reserve at a much 64. pp. 2–16. essentially independent of the number of active sources and Embedded Computing Systems. 2004. By also accounting for link performance history. Das. Cerpa. pp. 11." IEEE/ACM Trans. 2001. [15] D. J. “Variable-rate variable-power M-QAM for of distance traveled. Feb. C ONCLUSION Acoustics. Perkins. Bourkerche. [18] Z. July 2004. MAY 2009 in part to EAR dynamically accounting for variable link are high. G.. vol. K. M. pp. 1. 1997. R. vol. 3. Wang. 5. 24–60. pp. M. 1999. Routes are managed Networking. Haas and T. J. Araujo. 1–4. route only to the cluster heads and cluster heads may be [14] D. 1–21. Ickes. 1218–1230. “Maximum lifetime routing in wireless GBR. Future work includes studying the open issues performance and forwarding packets only on “best” routes. The width of the mesh is controlled by the amount of credit 333–342. 2003. and R. no. Shih. pp. 2004. 1. N. Min. and W. 4. Estrin and A. well with network size in the presence of node faults and link 66.txt.

73-91." ACM Wireless has published a book Anatomy of HCTS. He is a co-inventor of works: proactive versus reactive.txt. Srivastava. Zhang. B. Ye. Charles E. USA. Nov. pers on parallel and distributed processing.11-1997. His pioneering work on computing using reconfigurable ACM Int’l Workshop Performance Evaluation Wirelesss Ad Hoc. 2002 (work in progress).. Pan’s July 2003. Pister. Dr. JSPS. IISF. ed. He has authored or co-authored many pa- [22] LAN MAN Standards Committee of the IEEE Computer Society. 1999. A. Pan has published more than 80 Networks. vol. He is also a registered Professional Engineer conferences and universities worldwide. works. in 1982 and 1984. Zhao and G." in Proc. Lu. and was a Visiting Scientist at the IBM T. and S. V. and M. G. He has served as a reviewer/panelist for many research foundations/agencies Nanyang Technological University of Singapore. “Energy efficiency of large-scale wireless net- in books. ICC. In addition. J.LOH et al." ACM Trans. and data delivery protocol for large scale sensor networks. D. theses. 2. journal and conference papers. Areas Commun. 2005. Zhong. Embedded degree in computer science from the University of Computing Systems. Mohan. 431-447. . the Australian Research Council.) and the BEng from the National over 50 invited talks. His recent research has been supported by NSF. Das. He has organized and the Parallel Processing Lab. Who’s Who in America. in 1991. vol. the as INFOCOM.S. JSPS. Addison. 2. Peter K. Zhang. no. Chap. Raghunathan. H. conferences (including IPDPS. Ye and J. projects. System for Mega Vessels. and a senior member of IEEE. Katz. and the IBM Palo Alto Scientific Center.. a Yamacraw Distinguished Speaker (2002). such as the U. wireless sensor networks. Zhong. pp. 2003. Pan has served as an editor- systems. 84–91. He has previously in-chief or editorial board member for eight journals including three IEEE held positions as Head of Software Cluster Labs Transactions and as a guest editor for seven special issues. no. Toumpis and A. Currently. AFRL. MIT Alliance (SMA) Computer Science Program [21] J. 2001. [25] V. “Topology knowledge range control for lifetime He has also co-edited 13 books (including proceedings) and contributed maximization in sensor networks with data aggregation. respectively. Meng. research interests include parallel and distributed [28] F. Select. and L." in Proc. pp. 5. “Ad hoc on demand Advanced Information Systems. 2. Belding-Royer. USA." Ad Hoc puter Engineering. AFOSR. 1100–1112. and his Ph. Lu. “Gradient broadcast: a robust Director of the Maritime Research Center. Speaker (2000–2002). Broch. no. Rodoplu and T. Who’s Who in Computational Science and Engineering. 2005. S. pp. K. National Science Foundation. aodv-12. J. Dr. He is listed in Men of Achievement. “DSR: the dynamic source Wen-Jing Hsu is presently with the School of Com- routing protocol for mulit-hop wireless ad hoc networks. 736–748. journal papers with 28 papers published in various [29] W. M. NTU and data delivery protocol for large scale sensor networks. M. a Shell Oil Colloquium Speaker (2002). 11. he has published over 90 papers in refereed networks. and J. Pan is an IEEE Distinguished (Singapore). IISF and the Mellon Foundation. Ye. He also held positions on the faculty of Michigan State University. Chartered Engineer (UK). vol." IEEE Trans. Aug. parallel and distributed and the Hong Kong Research Grants Council." IEEE J. 285–298. H. AFOSR. Goldsmith. degrees Aug. Areas Commun. ICPP. fessor of the School of Computer Engineering. Dr. and K.. Johnson. and Ubiquitous Networks." Wireless Sensor Networks. no. he has initiated and com- mobile networking for smart dust. vol. Dr. Select. he is the chair [27] S. Nanyang Technological Univer- Networking.Engg. his research results have been cited by more than 100 researchers worldwide [31] Q. ICDCS. no. Awards (with two awarded). and a Senior Member of IEEE. three U.." IEEE J. the Natural Sciences and He has authored and co-authored several papers in Engineering Research Council of Canada. Loh is currently an Associate Pro. “Capacity regions for wireless ad hoc and a full professor in the Department of Com- networks. IEEE Std 802. B. 285–298. He is presently Deputy [23] F. Heidemann. 2005. in computer engineering from Tsinghua University. R. vol." ACM Wireless bioinformatics. Who’s Who in American Education. sity. patents (pending) and five provisional patents. Perkins. ment for energy-aware communication systems. At NTU. among “Wireless LAN medium access control (MAC) and physical layer (PHY) them four papers were nominated for Best Paper specification. [26] C. 5th Annual ACM/IEEE pleted several large scale externally funded research International Conf. He was previously Director of the Center of [24] C. INFOCOM. pp. pp.a High-capacity Container Handling Networks. and Deputy Director of Financial Engineering distance vector (AODV) routing.Sc. 8. Who’s Who in the Midwest. Watson Research Center. pp. Dr. at NTU. “Next century challenges: since 2001.: PERFORMANCE EVALUATION OF EFFICIENT AND RELIABLE ROUTING PROTOCOLS FOR FIXED-POWER SENSOR NETWORKS 2335 [20] D. including keynote speeches and colloquium talks at University of Singapore. S. E. GLOBECOM. and M. Pittsburgh. 1999.D. He obtained his several international conferences and workshops and has also served as a Ph. wireless networks. pp. 2nd several book chapters. NSFC. 139–172. Pan has delivered MSc (Elect. G.D. vol. pp. Zhao and L. no. puter Science at Georgia State University. and GLOBECOM). S. 5. pp. and computer gaming. and Who’s Who of Asian Americans. Kahn. Wireless Commun. and ICPP. 3. Yorktown Heights. “Minimum energy mobile wireless net. “Power manage- China. pp. 1333-1344. many awards from agencies such as NSF. He is also a Faculty Fellow of the Singapore- Wesley." IETF Internet draft. 11. 4." New York.Eng. J. 271–278. Schurgers. “Medium access control in wireless sensor IEEE journals. Yi Pan received his B. 17. IPDPS. 2005. optical networks. Tong. USA. and L. 2. [30] Q. Mobile Computing Networking. NIH. Maltz. “Gradient broadcast: a robust computing. and has received 23.Eng. Sensor optical buses has inspired extensive subsequent work by many researchers.) from Manchester Univeristy. E. Perkins.S. from Nanyang Technological University. the program committee member for several major international conferences such MSc (Comp. 2004. draft-ietf-manet. and the states of Georgia and Ohio.