184

IEEE TRANSACTIONS ON DEPENDABLE AND SECURE COMPUTING,

VOL. 9,

NO. 2,

MARCH/APRIL 2012

Design and Implementation of TARF: A Trust-Aware Routing Framework for WSNs

Guoxing Zhan, Weisong Shi, Senior Member, IEEE, and Julia Deng
AbstractThe multihop routing in wireless sensor networks (WSNs) offers little protection against identity deception through replaying routing information. An adversary can exploit this defect to launch various harmful or even devastating attacks against the routing protocols, including sinkhole attacks, wormhole attacks, and Sybil attacks. The situation is further aggravated by mobile and harsh network conditions. Traditional cryptographic techniques or efforts at developing trust-aware routing protocols do not effectively address this severe problem. To secure the WSNs against adversaries misdirecting the multihop routing, we have designed and implemented TARF, a robust trust-aware routing framework for dynamic WSNs. Without tight time synchronization or known geographic information, TARF provides trustworthy and energy-efficient route. Most importantly, TARF proves effective against those harmful attacks developed out of identity deception; the resilience of TARF is verified through extensive evaluation with both simulation and empirical experiments on large-scale WSNs under various scenarios including mobile and RF-shielding network conditions. Further, we have implemented a low-overhead TARF module in TinyOS; as demonstrated, this implementation can be incorporated into existing routing protocols with the least effort. Based on TARF, we also demonstrated a proof-of-concept mobile target detection application that functions well against an antidetection mechanism. Index TermsWireless sensor networks, routing protocols, security.

1 INTRODUCTION

original headers, are replayed without any modification. IRELESSSENSOR networks (WSNs) [2] are ideal candidates for applications to report detected events of Even if this malicious node cannot directly overhear the interest, such as military surveillance and forest fire valid nodes wireless transmission, it can collude with other monitoring. A WSN comprises battery-powered sensor malicious nodes to receive those routing packets and replay http://ieeexploreprojects.blogspot.com nodes with extremely limited processing capabilities. With them somewhere far away from the original valid node, a narrow radio communication range, a sensor node which is known as a wormhole attack [5]. Since a node in a wirelessly sends messages to a base station via a multihop WSN usually relies solely on the packets received to know path. However, the multihop routing of WSNs often about the senders identity, replaying routing packets allows becomes the target of malicious attacks. An attacker may the malicious node to forge the identity of this valid node. tamper nodes physically, create traffic collision with After stealing that valid identity, this malicious node is able to misdirect the network traffic. For instance, it may seemingly valid transmission, drop or misdirect messages drop packets received, forward packets to another node in routes, or jam the communication channel by creating not supposed to be in the routing path, or even form a radio interference [3]. This paper focuses on the kind of transmission loop through which packets are passed among attacks in which adversaries misdirect network traffic by a few malicious nodes infinitely. It is often difficult to know identity deception through replaying routing information. whether a node forwards received packets correctly even Based on identity deception, the adversary is capable of with overhearing techniques [4]. Sinkhole attacks are another launching harmful and hard-to-detect attacks against kind of attacks that can be launched after stealing a valid routing, such as selective forwarding, wormhole attacks, identity. In a sinkhole attack, a malicious node may claim sinkhole attacks and Sybil attacks [4]. itself to be a base station through replaying all the packets As a harmful and easy-to-implement type of attack, a from a real base station [6]. Such a fake base station could malicious node simply replays all the outgoing routing lure more than half the traffic, creating a black hole. This packets from a valid node to forge the latter nodes identity; same technique can be employed to conduct another strong the malicious node then uses this forged identity to form of attackSybil attack [7]: through replaying the participate in the network routing, thus disrupting the routing information of multiple legitimate nodes, an attacker network traffic. Those routing packets, including their may present multiple identities to the network. A valid node, if compromised, can also launch all these attacks. The harm of such malicious attacks based on the . G. Zhan and W. Shi are with the Department of Computer Science, Wayne technique of replaying routing information is further State University, 5057 Woodward Avenue, Detroit, MI 48202. aggravated by the introduction of mobility into WSNs and E-mail: {gxzhan, weisong}@wayne.edu. . J. Deng is with the Intelligent Automation, Inc., 15400 Calhoun Drive, the hostile network condition. Though mobility is introSuite 400, Rockville, MD 20855. E-mail: hdeng@i-a-i.com. duced into WSNs for efficient data collection and various Manuscript received 16 Nov. 2010; revised 3 Sept. 2011; accepted 12 Sept. applications [8], [9], [10], [11], it greatly increases the chance 2011; published online 10 Nov. 2011. of interaction between the honest nodes and the attackers. For information on obtaining reprints of this article, please send e-mail to: tdsc@computer.org, and reference IEEECS Log Number TDSC-2010-11-0216. Additionally, a poor network connection causes much Digital Object Identifier no. 10.1109/TDSC.2011.58. difficulty in distinguishing between an attacker and a
1545-5971/12/$31.00 2012 IEEE Published by the IEEE Computer Society

ZHAN ET AL.: DESIGN AND IMPLEMENTATION OF TARF: A TRUST-AWARE ROUTING FRAMEWORK FOR WSNS

185

honest node with transient failure. Without proper protection, WSNs with existing routing protocols can be completely devastated under certain circumstances. In an emergent sensing application through WSNs, saving the network from being devastated becomes crucial to the success of the application. Unfortunately, most existing routing protocols for WSNs either assume the honesty of nodes and focus on energy efficiency [12], or attempt to exclude unauthorized participation by encrypting data and authenticating packets. Examples of these encryption and authentication schemes Fig. 1. Multihop routing for data collection of a WSN. for WSNs include TinySec [13], Spins [14], TinyPK [15], and TinyECC [16]. Admittedly, it is important to consider which is not achieved by previous security protocols. Even efficient energy use for battery-powered sensor nodes and under strong attacks such as sinkhole attacks, wormhole the robustness of routing under topological changes as well attacks as well as Sybil attacks, and hostile mobile network as common faults in a wild environment. However, it is also condition, TARF demonstrates steady improvement in critical to incorporate security as one of the most important network performance. The effectiveness of TARF is verified goals; meanwhile, even with perfect encryption and authen- through extensive evaluation with simulation and empirical tication, by replaying routing information, a malicious node experiments on large-scale WSNs. Finally, we have implecan still participate in the network using another valid mented a ready-to-use TARF module with low overhead, nodes identity. The gossiping-based routing protocols offer which as demonstrated can be integrated into existing certain protection against attackers by selecting random routing protocols with ease; the demonstration of a proofneighbors to forward packets [17], but at a price of of-concept mobile target detection program indicates the considerable overhead in propagation time and energy use. potential of TARF in WSN applications. In addition to the cryptographic methods, trust and We start by stating the design considerations of TARF in reputation management has been employed in generic ad Section 2. Then, we elaborate the design of TARF in Section hoc networks and WSNs to secure routing protocols. 3, including the routing procedure as well as the EnergyBasically, a system of trust and reputation management Watcher and TrustManager components. In Section 4, we assigns each node a trust value according to its past present the simulation results of TARF against various http://ieeexploreprojects.blogspot.com performance in routing. Then, such trust values are used to attacks through replaying routing information in static, help decide a secure and efficient route. However, the mobile and RF-shielding conditions. Section 5 further proposed trust and reputation management systems for presents the implementation of TARF, empirical evaluation generic ad hoc networks target only relatively powerful at a large sensor network and a resilient proof-of-concept mobile target detection application based on TARF. Finally, hardware platforms such as laptops and smartphones [18], we discuss the related work in Section 6 and conclude this [19], [20], [21]. Those systems cannot be applied to WSNs paper in Section 7. due to the excessive overhead for resource-constrained sensor nodes powered by batteries. As far as WSNs are concerned, secure routing solutions based on trust and 2 DESIGN CONSIDERATIONS reputation management rarely address the identity decep- Before elaborating the detailed design of TARF, we would tion through replaying routing information [22], [23]. The like to clarify a few design considerations first, including countermeasures proposed so far strongly depends on certain assumptions in Section 2.1 and the goals in Section 2.3. either tight time synchronization or known geographic information while their effectiveness against attacks ex- 2.1 Assumptions ploiting the replay of routing information has not been We target secure routing for data collection tasks, which are examined yet [4]. one of the most fundamental functions of WSNs. In a data At this point, to protect WSNs from the harmful attacks collection task, a sensor node sends its sampled data to a exploiting the replay of routing information, we have remote base station with the aid of other intermediate nodes, designed and implemented a robust trust-aware routing as shown in Fig. 1. Though there could be more than one base framework, TARF, to secure routing solutions in wireless station, our routing approach is not affected by the number sensor networks. Based on the unique characteristics of of base stations; to simplify our discussion, we assume that resource-constrained WSNs, the design of TARF centers on there is only one base station. An adversary may forge the trustworthiness and energy efficiency. Though TARF can be identity of any legal node through replaying that nodes developed into a complete and independent routing proto- outgoing routing packets and spoofing the acknowledgment col, the purpose is to allow existing routing protocols to packets, even remotely through a wormhole. incorporate our implementation of TARF with the least effort Additionally, to merely simplify the introduction of and thus producing a secure and efficient fully functional TARF, we assume no data aggregation is involved. Noneprotocol. Unlike other security measures, TARF requires theless, our approach can still be applied to cluster-based neither tight time synchronization nor known geographic WSNs with static clusters, where data are aggregated by information. Most importantly, TARF proves resilient under clusters before being relayed [24]. Cluster-based WSNs various attacks exploiting the replay of routing information, allows for the great savings of energy and bandwidth

186

IEEE TRANSACTIONS ON DEPENDABLE AND SECURE COMPUTING,

VOL. 9,

NO. 2,

MARCH/APRIL 2012

2.3 Goals TARF mainly guards a WSN against the attacks misdirecting the multihop routing, especially those based on identity theft through replaying the routing information. This paper does not address the denial-of-service (DoS) [3] attacks, where an attacker intends to damage the network by exhausting its resource. For instance, we do not address the DoS attack of congesting the network by replaying numerous packets or physically jamming the network. TARF aims to achieve the following desirable properties: High throughput. Throughput is defined as the ratio of the number of all data packets delivered to the base station to the number of all sampled data packets. In our evaluation, throughput at a moment is computed over the period from the beginning time (0) until that particular moment. Note that single-hop retransmission may happen, and that duplicate packets are considered as one packet as 2.2 Authentication Requirements far as throughput is concerned. Throughput reflects how Though a specific application may determine whether data efficiently the network is collecting and delivering data. encryption is needed, TARF requires that the packets are Here, we regard high throughput as one of our most properly authenticated, especially the broadcast packets important goals. from the base station. The broadcast from the base station is Energy efficiency. Data transmission accounts for a asymmetrically authenticated so as to guarantee that an major portion of the energy consumption. We evaluate http://ieeexploreprojects.blogspot.com adversary is not able to manipulate or forge a broadcast energy efficiency by the average energy cost to successfully message from the base station at will. Importantly, with deliver a unit-sized data packet from a source node to the authenticated broadcast, even with the existence of attack- base station. Note that link-level retransmission should be ers, TARF may use TrustManager (Section 3.4) and the given enough attention when considering energy cost since received broadcast packets about delivery information each retransmission causes a noticeable increase in energy (Section 3.2.1) to choose trustworthy path by circumventing consumption. If every node in a WSN consumes approxicompromised nodes. Without being able to physically mately the same energy to transmit a unit-sized data packet, capturing the base station, it is generally very difficult for we can use another metric hop-per-delivery to evaluate the adversary to manipulate the base station broadcast energy efficiency. Under that assumption, the energy packets which are asymmetrically authenticated. The consumption depends on the number of hops, i.e., the asymmetric authentication of those broadcast packets from number of one-hop transmissions occurring. To evaluate the base station is crucial to any successful secure routing how efficiently energy is used, we can measure the average protocol. It can be achieved through existing asymmetri- hops that each delivery of a data packet takes, abbreviated cally authenticated broadcast schemes that may require as hop-per-delivery. loose time synchronization. As an example, TESLA [14] Scalability and adaptability. TARF should work well achieves asymmetric authenticated broadcast through a with WSNs of large magnitude under highly dynamic symmetric cryptographic algorithm and a loose delay contexts. We will evaluate the scalability and adaptability of schedule to disclose the keys from a key chain. Other TARF through experiments with large-scale WSNs and examples of asymmetric authenticated broadcast schemes under mobile and hash network conditions. requiring either loose or no time synchronization are found Here, we do not include other aspects such as latency, in [25], [26]. load balance, or fairness. Low latency, balanced network Considering the great computation cost incurred by a load, and good fairness requirements can be enforced in strong asymmetric authentication scheme and the difficulty specific routing protocols incorporating TARF. in key management, a regular packet other than a base station broadcast packet may only be moderately authenticated through existing symmetric schemes with a limited 3 DESIGN OF TARF set of keys, such as the message authentication code TARF secures the multihop routing in WSNs against provided by TinySec [13]. It is possible that an adversary intruders misdirecting the multihop routing by evaluating physically captures a nonbase legal node and reveals its key the trustworthiness of neighboring nodes. It identifies such for the symmetric authentication [27]. With that key, the intruders by their low trustworthiness and routes data adversary can forge the identity of that nonbase legal node through paths circumventing those intruders to achieve

through aggregating data from children nodes and performing routing and transmission for children nodes. In a cluster-based WSN, the cluster headers themselves form a subnetwork; after certain data reach a cluster header, the aggregated data will be routed to a base station only through such a subnetwork consisting of the cluster headers. Our framework can then be applied to this subnetwork to achieve secure routing for cluster-based WSNs. TARF may run on cluster headers only and the cluster headers communicate with their children nodes directly since a static cluster has known relationship between a cluster header and its children nodes, though any link-level security features may be further employed. Finally, we assume a data packet has at least the following fields: the sender id, the sender sequence number, the next-hop node id (the receiver in this one-hop transmission), the source id (the node that initiates the data), and the sources sequence number. We insist that the source nodes information should be included for the following reasons because that allows the base station to track whether a data packet is delivered. It would cause too much overhead to transmit all the one-hop information to the base station. Also, we assume the routing packet is sequenced.

and joins the network legally. However, when the adversary uses its fake identity to falsely attract a great amount of traffic, after receiving broadcast packets about delivery information, other legal nodes that directly or indirectly forwards packets through it will start to select a more trustworthy path through TrustManager (Section 3.4).

ZHAN ET AL.: DESIGN AND IMPLEMENTATION OF TARF: A TRUST-AWARE ROUTING FRAMEWORK FOR WSNS

187

satisfactory throughput. TARF is also energy efficient, highly scalable, and well adaptable. Before introducing the detailed design, we first introduce several necessary notions here. Neighbor. For a node N, a neighbor (neighboring node) of N is a node that is reachable from N with one-hop wireless transmission. Trust level. For a node N, the trust level of a neighbor is a decimal number in [0, 1], representing Ns opinion of that neighbors level of trustworthiness. Specifically, the trust level of the neighbor is Ns estimation of the probability that this neighbor correctly delivers data received to the base station. That trust level is denoted as T in this paper. Energy cost. For a node N, the energy cost of a neighbor is the average energy cost to successfully deliver a unitsized data packet with this neighbor as its next-hop node, from N to the base station. That energy cost is denoted as E in this paper.

Fig. 2. Each node selects a next-hop node based on its neighborhood table, and broadcast its energy cost within its neighborhood. To maintain this neighborhood table, EnergyWatcher and TrustManager on the node keep track of related events (on the left) to record the energy cost and the trust level values of its neighbors.

3.1 Overview For a TARF-enabled node N to route a data packet to the base station, N only needs to decide to which neighboring node it should forward the data packet considering both the trustworthiness and the energy efficiency. Once the data 3.2 Routing Procedure packet is forwarded to that next-hop node, the remaining TARF, as with many other routing protocols, runs as a task to deliver the data to the base station is fully delegated periodic service. The length of that period determines how to it, and N is totally unaware of what routing decision its frequently routing information is exchanged and updated. next-hop node makes. N maintains a neighborhood table At the beginning of each period, the base station broadcasts a with trust level values and energy cost values for certain message about data delivery during last period to the whole known neighbors. It is sometimes necessary to delete some neighbors entries to keep the table size acceptable. The network consisting of a few contiguous packets (one packet http://ieeexploreprojects.blogspot.com technique of maintaining a neighborhood table of a may not hold all the information). Each such packet has a moderate size is demonstrated by Woo et al. [28]; TARF field to indicate how many packets are remaining to complete the broadcast of the current message. The complemay employ the same technique. In TARF, in addition to data packet transmission, there tion of the base station broadcast triggers the exchange of are two types of routing information that need to be energy report in this new period. Whenever a node receives exchanged: broadcast messages from the base station about such a broadcast message from the base station, it knows data delivery and energy cost report messages from each that the most recent period has ended and a new period has node. Neither message needs acknowledgment. A broad- just started. No tight time synchronization is required for a cast message from the base station is flooded to the whole node to keep track of the beginning or ending of a period. network. The freshness of a broadcast message is checked During each period, the EnergyWatcher on a node monitors through its field of source sequence number. The other type energy consumption of one-hop transmission to its neighof exchanged routing information is the energy cost report bors and processes energy cost reports from those neighbors message from each node, which is broadcast to only its to maintain energy cost entries in its neighborhood table; its neighbors once. Any node receiving such an energy cost TrustManager also keeps track of network loops and processes broadcast messages from the base station about report message will not forward it. For each node N in a WSN, to maintain such a data delivery to maintain trust level entries in its neighborneighborhood table with trust level values and energy cost hood table. To maintain the stability of its routing path, a node may values for certain known neighbors, two components, EnergyWatcher and TrustManager, run on the node (Fig. 2). retain the same next-hop node until the next fresh broadcast EnergyWatcher is responsible for recording the energy cost message from the base station occurs. Meanwhile, to reduce for each known neighbor, based on Ns observation of one- traffic, its energy cost report could be configured to not hop transmission to reach its neighbors and the energy cost occur again until the next fresh broadcast message from the report from those neighbors. A compromised node may base station. If a node does not change its next-hop node falsely report an extremely low energy cost to lure its selection until the next broadcast message from the base neighbors into selecting this compromised node as their station, that guarantees all paths to be loop-free, as can be next-hop node; however, these TARF-enabled neighbors deducted from the procedure of next-hop node selection. eventually abandon that compromised next-hop node based However, as noted in our experiments, that would lead to on its low trustworthiness as tracked by TrustManager. slow improvement in routing paths. Therefore, we allow a TrustManager is responsible for tracking trust level values of node to change its next-hop selection in a period when its neighbors based on network loop discovery and broadcast current next-hop node performs the task of receiving and messages from the base station about data delivery. Once N delivering data poorly.

is able to decide its next-hop neighbor according to its neighborhood table, it sends out its energy report message: it broadcasts to all its neighbors its energy cost to deliver a packet from the node to the base station. The energy cost is computed as in Section 3.3 by EnergyWatcher. Such an energy cost report also serves as the input of its receivers EnergyWatcher.

188

IEEE TRANSACTIONS ON DEPENDABLE AND SECURE COMPUTING,

VOL. 9,

NO. 2,

MARCH/APRIL 2012

Next, we introduce the structure and exchange of routing information as well as how nodes make routing decisions in TARF.

3.2.1 Structure and Exchange of Routing Information A broadcast message from the base station fits into at most a fixed small number of packets. Such a message consists of 3.2.2 Route Selection some pairs of <node id of a source node, an undelivered Now, we introduce how TARF decides routes in a WSN. sequence interval [a, b] with a significant length>, <node id Each node N relies on its neighborhood table to select an of a source node, minimal sequence number received in last optimal route, considering both energy consumption and period, maximum sequence number received in last reliability. TARF makes good efforts in excluding those period>, as well as several node id intervals of those nodes that misdirect traffic by exploiting the replay of without any delivery record in last period. To reduce routing information. overhead to an acceptable amount, our implementation For a node N to select a route for delivering data to the selects only a limited number of such pairs to broadcast base station, N will select an optimal next-hop node from (Section 5.1) and proved effective (Sections 5.3, 5.4). Roughly, the effectiveness can be explained as follows: the its neighbors based on trust level and energy cost and fact that an attacker attracts a great deal of traffic from forward the data to the chosen next-hop node immediately. many nodes often gets revealed by at least several of those The neighbors with trust levels below a certain threshold nodes being deceived with a high likelihood. The undeliv- will be excluded from being considered as candidates. ered sequence interval [a, b] is explained as follows: the Among the remaining known neighbors, N will select its base station searches the source sequence numbers received next-hop node through evaluating each neighbor b based on in last period, identifies which source sequence numbers for a tradeoff between T and ENb , with E and T being bs Nb Nb Nb TNb the source node with this id are missing, and chooses energy cost and trust level value in the neighborhood table, certain significant interval [a, b] of missing source sequence numbers as an undelivered sequence interval. For example, respectively, (see Sections 3.3, 3.4). Basically, ENb reflects the base station may have all the source sequence numbers the energy cost of delivering a packet to the base station for the source node 2 as {109, 110, 111, 150, 151} in last from N assuming that all the nodes in the route are honest; period. Then, [112, 149] is an undelivered sequence interval; T1 approximately reflects the number of the needed Nb [109, 151] is also recorded as the sequence boundary of attempts to send a packet from N to the base station via delivered packets. Since the base station is usually multiple hops before such an attempt succeeds, considering http://ieeexploreprojects.blogspot.com connected to a powerful platform such as a desktop, a the trust level of b. Thus, ENb combines the trustworthiness T program can be developed on that powerful platform to and energy cost. However,Nbthe metric ENb suffers from the TNb assist in recording all the source sequence numbers and fact that an adversary may falsely reports extremely low finding undelivered sequence intervals. Accordingly, each node in the network stores a table of energy cost to attract traffic and thus resulting in a low ENb <node id of a source node, a forwarded sequence interval value of TNb even with a low TNb . Therefore, TARF prefers [a, b] with a significant length> about last period. The data nodes with significantly higher trust values; this preference packets with the source node and the sequence numbers of trustworthiness effectively protects the network from an falling in this forwarded sequence interval [a, b] have adversary who forges the identity of an attractive node already been forwarded by this node. When the node such as a base station. For deciding the next-hop node, a Nb receives a broadcast message about data delivery, its specific tradeoff between TNb and ENb is demonstrated in T TrustManager will be able to identify which data packets Fig. 5 (see Section 5.2). forwarded by this node are not delivered to the base station. Observe that in an ideal misbehavior-free environment, Considering the overhead to store such a table, old entries all nodes are absolutely faithful, and each node will choose will be deleted once the table is full. a neighbor through which the routing path is optimized in Once a fresh broadcast message from the base station is terms of energy; thus, an energy-driven route is achieved. received, a node immediately invalidates all the existing energy cost entries: it is ready to receive a new energy 3.3 EnergyWatcher report from its neighbors and choose its new next-hop node Here, we describe how a node Ns EnergyWatcher computes afterward. Also, it is going to select a node either after a the energy cost ENb for its neighbor b in Ns neighborhood timeout is reached or after it has received an energy cost table and how N decides its own energy cost EN . Before report from some highly trusted candidates with acceptable going further, we will clarify some notations. ENb menenergy cost. A node immediately broadcasts its energy cost tioned is the average energy cost of successfully delivering a to its neighbors only after it has selected a new next-hop unit-sized data packet from N to the base station, with b as node. That energy cost is computed by its EnergyWatcher Ns next-hop node being responsible for the remaining (see Section 3.3). A natural question is which node starts route. Here, one-hop retransmission may occur until the reporting its energy cost first. For that, note that when the acknowledgment is received or the number of retransmisbase station is sending a broadcast message, a side effect is sions reaches a certain threshold. The cost caused by onethat its neighbors receiving that message will also regard hop retransmissions should be included when computing this as an energy report: the base station needs 0 amount of ENb . Suppose N decides that A should be its next-hop node energy to reach itself. As long as the original base station is after comparing energy cost and trust level. Then, Ns

faithful, it will be viewed as a trustworthy candidate by TrustManager on the neighbors of the base station. Therefore, those neighbors will be the first nodes to decide their next-hop node, which is the base station; they will start reporting their energy cost once that decision is made.

ZHAN ET AL.: DESIGN AND IMPLEMENTATION OF TARF: A TRUST-AWARE ROUTING FRAMEWORK FOR WSNS

189

3.4 TrustManager A node Ns TrustManager decides the trust level of each neighbor based on the following events: discovery of network loops, and broadcast from the base station about data delivery. For each neighbor b of N, TNb denotes the trust level of b in Ns neighborhood table. At the beginning, ENb EN!b Eb : each neighbor is given a neutral trust level 0.5. After any of Since each known neighbor b of N is supposed to broadcast those events occurs, the relevant neighbors trust levels are its own energy cost Eb to N, to compute ENb , N still needs updated. to know the value EN!b , i.e., the average energy cost of Note that many existing routing protocols have their successfully delivering a data packet from N to its neighbor own mechanisms to detect routing loops and to react b with one hop. For that, assuming that the endings (being accordingly [31], [32], [28]. In that case, when integrating acknowledged or not) of one-hop transmissions from N to b TARF into those protocols with antiloop mechanisms, are independent with the same probability psucc of being TrustManager may solely depend on the broadcast from acknowledged, we first compute the average number of the base station to decide the trust level; we adopted such a one-hop sendings needed before the acknowledgment is policy when implementing TARF later (see Section 5). If received as follows: antiloop mechanisms are both enforced in the TARF component and the routing protocol that integrates TARF, 1 X 1 i psucc 1 psucc i1 : then the resulting hybrid protocol may overly react toward psucc i1 the discovery of loops. Though sophisticated loop-discovDenote Eunit as the energy cost for node N to send a unit- ery methods exist in the currently developed protocols, they sized data packet once regardless of whether it is received often rely on the comparison of specific routing cost to reject routes likely leading to loops [32]. To minimize the effort to or not. Then, we have integrate TARF and the existing protocol and to reduce the Eunit overhead, when an existing routing protocol does not Eb : ENb psucc provide any antiloop mechanism, we adopt the following The remaining job for computing ENb is to get the probability mechanism to detect routing loops. To detect loops, the psucc that a one-hop transmission is acknowledged. Con- TrustManager on N reuses the table of <node id of a source sidering the variable wireless connection among wireless node, a forwarded sequence interval [a, b] with a significant http://ieeexploreprojects.blogspot.com sensor nodes, we do not use the simplistic averaging method length> (see Section 3.2) in last period. If N finds that a to compute psucc . Instead, after each transmission from N to received data packet is already in that record table, not only b, Ns EnergyWatcher will update psucc based on whether that will the packet be discarded, but the TrustManager on N transmission is acknowledged or not with a weighted also degrades its next-hop nodes trust level. If that nextaveraging technique. We use a binary variable Ack to record hop node is b, then Told Nb is the latest trust level value of b. the result of current transmission: 1 if an acknowledgment is We use a binary variable Loop to record the result of loop received; otherwise, 0. Given Ack and the last probability discovery: 0 if a loop is received; 1 otherwise. As in the value of an acknowledged transmission pold succ , an intuitive update of energy cost, the new trust level of b is way is to use a simply weighted average of Ack and pold succ 8 > 1 wdegrade Told Nb wdegrade Loop; as the value of pnew succ . That is what is essentially adopted in > < if Loop 0: the aging mechanism [29]. However, that method used Tnew Nb > 1 wupgrade Told Nb wupgrade Loop; against sleeper attacks still suffers periodic attacks [30]. To > : if Loop 1: solve this problem, we update the psucc value using two different weights as in our previous work [30], a relatively Once a loop has been detected by N for a few times so big wdegrade 2 0; 1 and a relatively small wupgrade 2 0; 1 as that the trust level of the next-hop node is too low, N will follows: change its next-hop selection, thus that loop is broken. 8 Though N cannot tell which node should be held > 1 wdegrade pold succ wdegrade Ack; > < responsible for the occurrence of a loop, degrading its if Ack 0: pnew succ next-hop nodes trust level gradually leads to the breaking > 1 wupgrade pold succ wupgrade Ack; > : of the loop. On the other hand, to detect the traffic if Ack :1: misdirection by nodes exploiting the replay of routing The two parameters wdegrade and wupgrade allow flexible information, TrustManager on N compares Ns stored table application requirements. wdegrade and wupgrade represent the of <node id of a source node, forwarded sequence interval extent to which upgraded and degraded performance are [a, b] with a significant length> recorded in last period with rewarded and penalized, respectively. If any fault and compromise is very likely to be associated with a high risk, the broadcast messages from the base station about data wdegrade should be assigned a relatively high value to delivery. It computes the ratio of the number of successfully penalize fault and compromise relatively heavily; if a few delivered packets which are forwarded by this node to the positive transactions cant constitute evidence of good number of those forwarded data packets, denoted as connectivity which requires many more positive transac- DeliveryRatio. Then, Ns TrustManager updates its nexttions, then wupgrade should be assigned a relatively low value. hop node bs trust level as follows:
energy cost is EN ENA . Denote EN!b as the average energy cost of successfully delivering a data packet from N to its neighbor b with one hop. Note that the retransmission cost needs to be considered. With the above notations, it is straightforward to establish the following relation:

190

IEEE TRANSACTIONS ON DEPENDABLE AND SECURE COMPUTING,

VOL. 9,

NO. 2,

MARCH/APRIL 2012

Tnew

Nb

8 > 1 wdegrade Told Nb > > w > > degrade DeliveryRatio; > > > < if DeliveryRatio < Told Nb : > > 1 wupgrade Told Nb > > > > wupgrade DeliveryRatio; > > : if DeliveryRatio > Told Nb :

Fig. 3. An example to illustrate how TrustManager works. 3.5 Analysis on EnergyWatcher and TrustManager Now that a node N relies on its EnergyWatcher and TrustManager to select an optimal neighbor as its next-hop attacker node as its next-hop node. The attacker drops every node, we would like to clarify a few important points on the packet received and thus any data packet passing node A will not arrive at the base station. After a while, node A design of EnergyWatcher and TrustManager. First, as described in Section 3.1, the energy cost report discovers that the data packets it forwarded did not get is the only information that a node is to passively receive delivered. The TrustManager on node A starts to degrade and take as fact. It appears that such acceptance of the trust level of its current next-hop node B although node energy cost report could be a pitfall when an attacker or a B is absolutely honest. Once that trust level becomes too compromised node forges false report of its energy cost. low, node A decides to select node C as its new next-hop Note that the main interest of an attacker is to prevent data node. In this way, node A identifies a better and successful delivery rather than to trick a data packet into a less route (A - C - D - base). In spite of the sacrifice of node Bs efficient route, considering the effort it takes to launch an trust level, the network performs better. Further, concerning attack. As far as an attack aiming at preventing data the stability of routing path, once a valid node identifies a delivery is concerned, TARF well mitigates the effect of this trustworthy honest neighbor as its next-hop node, it tends pitfall through the operation of TrustManager. Note that the to keep that next-hop selection without considering other TrustManager on one node does not take any recommenda- seemingly attractive nodes such as a fake base station. That tion from the TrustManager on another node. If an attacker tendency is caused by both the preference to maintain stable forges false energy report to form a false route, such routes and the preference to highly trustable nodes. Finally, we would like to stress that TARF is designed to intention will be defeated by TrustManager: when the guard a WSN against the attacks misdirecting the multihop TrustManager on one node finds out the many delivery routing, especially those based on identity theft through failures from the broadcast messages of the base station, it http://ieeexploreprojects.blogspot.com information. Other types of attacks replaying the routing degrades the trust level of its current next-hop node; when such as the denial-of-service [3] attacks are out of the that trust level goes below certain threshold, it causes the discussion of this paper. For instance, we do not address the node to switch to a more promising next-hop node. attacks of injecting into the network a number of data packets Second, TrustManager identities the low trustworthiness containing false sensing data but authenticated (possibly of various attackers misdirecting the multihop routing, through hacking). That type of attacks aim to exhaust the especially those exploiting the replay of routing informa- network resource instead of misdirecting the routing. tion. It is noteworthy that TrustManager does not distin- However, if the attacker intends to periodically inject a few guish whether an error or an attack occurs to the next-hop routing packets to cause wrong route, such attacks can still node or other succeeding nodes in the route. It seems unfair be defended by TARF through TrustManager. that TrustManager downgrades the trust level of an honest next-hop node while the attack occurs somewhere after that next-hop node in the route. Contrary to that belief, 4 SIMULATION TrustManager significantly improves data delivery ratio in We have developed a reconfigurable emulator of wireless the existence of attack attempts of preventing data delivery. sensor networks on a 2D plane with Matlab to test TARF. First of all, it is often difficult to identify an attacker who We have conducted extensive simulation experiments; participates in the network using an id stolen from however, due to the page limit, interested readers may another legal node. For example, it is extremely difficult to refer to our technical report [33] and the conference version detect a few attackers colluding to launch a combined of paper [1] for detailed simulation settings and experiwormhole and sinkhole attack [4]. Additionally, despite the mental results. In our experiments, initially, 35 nodes are certain inevitable unfairness involved, TrustManager en- randomly distributed within a 300 300 rectangular area, courages a node to choose another route when its current with unreliable wireless transmission. All the nodes have route frequently fails to deliver data to the base station. the same power level and the same maximal transmission Though only those legal neighboring nodes of an attacker range of 100 m. Each node samples six times in every might have correctly identified the adversary, our evalua- period; the timing gap between every two consecutive tion results indicate that the strategy of switching to a new samplings of the same node is equivalent. We simulate the route without identifying the attacker actually significantly sensor network in 1,440 consecutive periods. Regarding the network topology, we set up three types improves the network performance, even with the existence of wormhole and sinkhole attacks. Fig. 3 gives an example to of network topologies. The first type is the static-location illustrate this point. In this example, nodes A, B, C, and D case under which all nodes stand still. The second type is a are all honest nodes and not compromised. Node A has customized group-motion-with-noise case based on Refernode B as its current next-hop node while node B has an ence Point Group Mobility (RPGM) model that mimics the

ZHAN ET AL.: DESIGN AND IMPLEMENTATION OF TARF: A TRUST-AWARE ROUTING FRAMEWORK FOR WSNS

191

behavior of a set of nodes moving in one or more groups [34], [35]. The last type of dynamic network incorporated in the experiments is the addition of scattered RF-shielded areas to the aforementioned group-motion-with-noise case. The performance of TARF is compared to that of a link connectivity-based routing protocol adapted from what is proposed by Woo et al. [28]. We denote the link connectivity-based routing protocol as Link connectivity. With the Link-connectivity protocol, each node selects its next-hop node among its neighborhood table according to an link estimator based on exponentially weighted moving average (EWMA). The simulation results show, in the presence of misbehaviors, the throughput in TARF is often much higher than that in Link connectivity; the hop-perdelivery in the Link-connectivity protocol is generally at least comparable to that in TARF. Under a misbehavior-free environment, the simulation Fig. 4. TrustManager component. results show that TARF and Link connectivity have comparable performance when there is no adversary. Both feature. As we worked on the first implementation, we protocols are also evaluated under three common types of noted that the existing protocols provide many nice attacks: 1) a certain node forges the identity of the based features, such as the analysis of link quality, the loop station by replaying broadcast messages, also known as detection and the routing decision mainly considering the the sinkhole attack; 2) a set of nodes colludes to form a communication cost. Instead of providing those features, forwarding loop; and 3) a set of nodes drops received data our implementation focuses on the trust evaluation based packets. These experiments were conducted in the static on the base broadcast of the data delivery, and such trust case, the group-motion-with-noise case, and the addition of information can be easily reused by other protocols. Finally, RF-shielded areas to the group-motion-with-noise case instead of using TinySec [13] exclusively for encryption and separately. Generally, under these common attacks, TARF authentication as in the first implementation on TinyOS 1.x, produces a substantial improvement over Link connectivity this re-implementation let the developers decide which in terms of data collection and energy efficiency. Further, encryption or authentication techniques to employ; the we have evaluated TARF under http://ieeexploreprojects.blogspot.com more severe attacks: encryption and authentication techniques of TARF may be multiple moving fake bases and multiple Sybil attackers. different than that of the existing protocol. As before, the experiments are conducted under all the 5.1 TrustManager Implementation Details three types of network topology. Under these two types of most severe attacks which almost devastates the Link- The TrustManager component in TARF is wrapped into an connectivity protocol, TARF succeeds in achieving a steady independent TinyOS configuration named TrustManaimprovement over the Link-connectivity protocol. Finally, we gerC. TrustManagerC uses a dedicated logic channel for have conducted certain experiments to explore the choice of communication and runs as a periodic service with a the period length and the trust updating scheme. Our configurable period, thus not interfering with the applicaexperiments reveal that a shorter period or a faster trust tion code. Though it is possible to implement TARF with a updating scheme may not necessarily benefit TARF. period always synchronized with the routing protocols period, that would cause much intrusion into the source code of the routing protocol. The current TrustManagerC 5 IMPLEMENTATION AND EMPIRICAL EVALUATION uses a period of 30 seconds; for specific applications, by In order to evaluate TARF in a real-world setting, we modifying a certain header file, the period length may be implemented the TrustManager component on TinyOS 2.x, reconfigured to reflect the sensing frequency, the energy which can be integrated into the existing routing protocols efficiency, and trustworthiness requirement. TrustManafor WSNs with the least effort. Originally, we had gerC provides two interfaces (see Fig. 4), TrustControl implemented TARF as a self-contained routing protocol and Record, which are implemented in other modules. The [1] on TinyOS 1.x before this second implementation. TrustControl interface provides the commands to enable However, we decided to redesign the implementation and disable the trust evaluation, while the Record interface considering the following factors. First, the first implemenprovides the commands for a root, i.e., a base station, to add tation only supports TinyOS 1.x, which was replaced by TinyOS 2.x; the porting procedure from TinyOS 1.x to delivered message record, for a nonroot node to add TinyOS 2.x tends to frustrate the developers. Second, rather forwarded message record, and for a node to retrieve the than developing a self-contained routing protocol, the trust level of any neighboring node. The implementation on second implementation only provides a TrustManager a root node differs from that on a nonroot node: a root node component that can be easily incorporated into the existing stores the information of messages received (delivered) protocols for routing decisions. The detection of routing during the current period into a record table and broadcast loops and the corresponding reaction are excluded from the delivery failure record; a nonroot node stores the informaimplementation of TrustManager since many existing pro- tion of forwarded messages during the current period also tocols, such as Collection Tree Protocol [32] and the link in a record table and compute the trust of its neighbors connectivity-based protocol [28], already provide that based on that and the broadcast information. Noting that

192

IEEE TRANSACTIONS ON DEPENDABLE AND SECURE COMPUTING,

VOL. 9,

NO. 2,

MARCH/APRIL 2012

much implementation overhead for a root can always be transferred to a more powerful device connected to the root, it is reasonable to assume that the root would have great capability of processing and storage. A root broadcasts two types of delivery failure record: at most three packets of significant undelivered intervals for individual origins and at most two packets of the ids of the origins without any record in the current period. For each origin, at most three significant undelivered intervals are broadcast. For a nonroot node, considering the processing and memory usage overhead, the record table keeps the forwarded message intervals for up to 20 source nodes, with up to five nonoverlapped intervals for each individual origin. Our later experiments verify that such size limit of the table on a nonroot node produces a resilient TARF with moderate overhead. The record table on a node keeps adding entries for new origins until it is full. With our current implementation, a valid trust value is an integer between 0 and 100, and any node is assigned an initial trust value of 50. The weigh parameters are: wupgrade 0:1, wdegrade 0:3. The trust table of a nonroot node keeps the trust level for up to 10 neighbors. Considering that an attacker may present multiple fake ids, the implementation evicts entries with a trust level close to the initial trust of any node. Such eviction policy is to ensure that the trust table remembers those neighbors with high trust and low trust; any other neighbor not in this table is deemed to have the initial trust value of 50.

Fig. 5. Routing decision incorporating trust management.

for an adversary to misguide other nodes into a wrong routing path by forging the identity of an attractive node such as a root; on the other hand, forwarding data packets to 5.2 Incorporation of TARF into Existing Protocols a candidate with a low trust level would result in many To demonstrate how this TARF implementation can be unsuccessful link-level transmission attempts, thus leading http://ieeexploreprojects.blogspot.com integrated into the exiting protocols with the least effort, we to much retransmission and a potential waste of energy. incorporated TARF into a collection tree routing protocol When the network throughput becomes low and a node has a (CTP) [32]. The CTP protocol is efficient, robust, and reliable list of low-trust neighbors, the node will exclusively use the in a network with highly dynamic link topology. It quantifies trust as the criterion to evaluate those neighbors for routing link quality estimation in order to choose a next-hop node. decisions. As shown in Fig. 5, it uses trust/cost as a criteria The software platform is TinyOS 2.x. To perform the only when the candidate has a trust level above certain integration, after proper interface wiring, invoke the Trust- threshold. The reason is, the sole trust/cost criteria could be Control.start command to enable the trust evaluation; exploited by an adversary replaying the routing information call the Record.addForwarded command for a nonroot from a base station and thus pretending to be an extremely node to add forwarded record once a data packet has been attractive node. As for Step 2, compared to the CTP forwarded; call the Record.addDelivered command for a implementation, we add two more circumstances when a root to add delivered record once a data packet has been node decides to switch to the optimal candidate found at received by the root. Finally, inside the CTPs task to update Step 1: that candidate has a higher trust level, or the current the routing path, call the Record.getTrust command to next-hop neighbor has a too low trust level. This new implementation integrating TARF requires retrieve the trust level of each next-hop candidate; an algorithm taking trust into routing consideration is executed moderate program storage and memory usage. We implemented a typical TinyOS data collection application, to decide the new next-hop neighbor (see Fig. 5). Similar to the original CTPs implementation, the im- MultihopOscilloscope, based on this new protocol. The plementation of this new protocol decides the next-hop MultihopOscilloscope application, with certain modified neighbor for a node with two steps (see Fig. 5): Step 1 sensing parameters for our later evaluation purpose, traverses the neighborhood table for an optimal candidate periodically makes sensing samples and sends out the for the next hop; Step 2 decides whether to switch from the sensed data to a root via multiple routing hops. Originally, current next-hop node to the optimal candidate found. For MultihopOscilloscope uses CTP as its routing protocol. Step 1, as in the CTP implementation, a node would not Now, we list the ROM and RAM sizes requirement of both consider those links congested, likely to cause a loop, or implementation of MultihopOscilloscope on nonroot Telosb having a poor quality lower than a certain threshold. This motes in Table 1. The enabling of TARF in MultihopOscillonew implementation prefers those candidates with higher scope increases the size of ROM by around 1.3 KB and the trust levels; in certain circumstances, regardless of the link size of memory by around 1.2 KB. quality, the rules deems a neighbor with a much higher trust level to be a better candidate (see Fig. 5). The preference of 5.3 Empirical Evaluation on Motelab highly trustable candidates is based on the following We evaluated the performance of TARF against a combined consideration: on the one hand, it creates the least chance sinkhole and wormhole attack on Motelab [36] at Harvard

ZHAN ET AL.: DESIGN AND IMPLEMENTATION OF TARF: A TRUST-AWARE ROUTING FRAMEWORK FOR WSNS

193

TABLE 1 Size Comparison of MultihopOscilloscope Implementation

University. One-hundred eighty-four TMote Sky sensor motes were deployed across many rooms at three floors in the department building (see Fig. 6), with two to four motes in most rooms. Around 97 nodes functioned properly while the rest were either removed or disabled. Each mote has a 2.4 GHz Chipcon CC2420 radio with an indoor range of approximately 100 meters. In Fig. 6, the thin green lines of Motelab (not indicate the direct (one-hop) wireless connection between Fig. 6. Connectivity map Motelab [Motelab].including the inter-floor connectivity), adapted from motes. Certain wireless connection also exists between nodes from different floors. 7c separately. On each floor, without any adversary, at least We developed a simple data collection application in 24 CTP nodes were able to find a successful route in each six TinyOS 2.x that sends a data packet every five seconds to a minute. However, with the five fake base stations in the base station node (root) via multihop. This application was wormhole, the number of CTP nodes that could find a executed on 91 functioning nonroot nodes on Motelab. For successful route goes down to nine for the first floor; it comparison, we used CTP and the TARF-enabled CTP decreases to no more than four for the second floor; as the implementation as the routing protocols for the data worst impact, none of the nodes on the third floor ever collection program separately. The TARF-enabled CTP found a successful route. A further look at the data showed has a TARF period of 30 seconds. We conducted an attack that all the nine nodes from the first floor with successful with five fake base stations that formed a wormhole. As in delivery record were all close to the real base station. The Fig. 6, whenever the base station sent out any packet, three CTP nodes relatively far away from the base station, such as fake base stations which overheard that packet replayed the those on the second and the third floor, had little luck in complete packet without changing any content including making good routing decisions. When TARF was enabled http://ieeexploreprojects.blogspot.com the node id. Other fake base stations overhearing that on each node, most nodes made correct routing decisions replayed packet would also replay the same packet. Each circumventing the attackers. That improvement can be fake base station essentially launched a sinkhole attack. Note verified by the fact that the number of the TARF-enabled that there is a distinction between such malicious replay nodes with successful delivery record under the threat of and the forwarding when a well-behaved node receives a the wormhole is close to that of CTP nodes with no attackers, broadcast from the base station. When a well-behaved node as shown in Figs. 7a, 7b, and 7c. forwards a broadcast packet from the base station, it will 5.4 Application: Mobile Target Detection in the include its own id in the packet so that its receivers will not Presence of an Antidetection Mechanism recognize the forwarder as a base station. We conducted To demonstrate how TARF can be applied in networked the first experiment by uploading the program with the sensing systems, we developed a proof-of-concept resilient CTP protocol onto 91 motes (not including those five application of target detection. This application relies on a selected motes as fake bases in later experiments), and no deployed wireless sensor network to detect a target that could attack was involved here. Then, in another experiment, in move, and to deliver the detection events to a base station via addition to programming those 91 motes with CTP, we also multiple hops with the TARF-enabled CTP protocol. For programmed the five fake base stations so that they stole simplification, the target is a LEGO MINDSTORM NXT 2.0 the id the base station through replaying. In the last vehicle robot equipped with a TelosB mote that sends out an experiment, we programmed those 91 motes with the Active Message AM packet every three seconds. A sensor TARF-enabled CTP, and programmed the five fake base node receiving such a packet from the target issues a detection stations as in the second experiment. Each of our programs report, which will be sent to the base station with the aforementioned TARF-enabled CTP protocol. run for 30 minutes. The experiment is set up within a clear floor space of 90 As illustrated in Fig. 7a, the existence of the five wormhole attackers greatly degraded the performance of CTP: the by 40 inches with 15 TelosB motes (see Fig. 8a). To make the number of the delivered data packets in the case of CTP multihop delivery necessary, the transmission power of all with the five-node wormhole is no more than 14 percent that the Telosb motes except two fake base stations in the in the case of CTP without adversaries. The TARF-enabled network is reduced through both software reduction and CTP succeeded in bringing an immense improvement over attenuator devices to within 30 inches. The target uses an CTP in the presence of the five-node wormhole, almost antidetection mechanism utilizing a fake base station close doubling the throughput. That improvement did not show to the real base station, and another remote base station any sign of slowing down as time elapsed. The number of close to the target and mounted on another LEGO vehicle nodes from each floor that delivered at least one data packet robot. The two fake base stations, with a transmission range in each six-minute subperiod is plotted in Figs. 7a, 7b, and of at least 100 feet, collude to form a wormhole: the fake base

194

IEEE TRANSACTIONS ON DEPENDABLE AND SECURE COMPUTING,

VOL. 9,

NO. 2,

MARCH/APRIL 2012

http://ieeexploreprojects.blogspot.com

Fig. 8. Deployment of a TARF-enabled wireless sensor network to detect a moving target under the umbrella of two fake base stations in a wormhole.

Fig. 7. Empirical comparison of CTP and TARF-enabled CTP on Motelab: (a) number of all delivered data packets since the beginning; number of nodes on (b) the first floor, (c) the second floor and (d) the third floor that delivered at least one data packet in subperiods.

station close to the base station replays all the packets from the base station immediately; the remote fake base station, after receiving those packets, immediately replays it again. This antidetection mechanism tricks some network nodes into sending their event reports into these fake base stations instead of the real base station. Though the fake base station

close to the real base station is capable of cheating the whole network alone by itself with its powerful radio for a certain amount of time, it can be easily recognized by remote nodes as a poor next-hop candidate soon by most routing protocols based on link quality: that fake base station does not acknowledge the packets sent to it from remote nodes with a weak radio via a single hop since it cannot really receive them. Thus, the antidetection mechanism needs to create such a wormhole to replay the packets from the base station remotely. The target node 14 and the fake base station 13 close to it move across the network along two parallel tracks of 22 inches back and forth (see Fig. 8b); they travel on each forward or backward path of 22 inches in around 10 minutes. The experiment lasts 30 minutes. For comparison, three nodes 9, 10, and 11 programmed with the CTP protocol are paired with another three nodes 6, 7, and 8 programmed with the TARF-enabled CTP (see Fig. 8b); each pair of nodes are physically placed close enough. All the other nodes, except for the fake base stations and the target node, are programmed with the TARF-enabled CTP. To fairly compare the performance between CTP and the TARF-enabled CTP, we now focus on the delivered detection reports originating from these three pairs of nodes: pair (9, 6), (10, 7), and (11, 8). For the time stamp of each detection report from these six nodes, we plot a corresponding symbol: a purple circle for the nodes with the TARF-enabled CTP; a black cross for the CTP nodes. The resulting detection report is visualized in Fig. 9a. Roughly, the TARF nodes report the existence of the target seven times as often as the CTP nodes do. More specifically,

ZHAN ET AL.: DESIGN AND IMPLEMENTATION OF TARF: A TRUST-AWARE ROUTING FRAMEWORK FOR WSNS

195

on trust-aware secure routing that is evaluated only through computer simulation, such as [38]. There are certain existing secure routing solutions for WSNs based on trust and reputation management; however, they rarely address the identity theft exploiting the replay of routing information. Two such representative solutions are ATSR [22] and TARP [23]. Neither ATSR nor TARP offers protection against the identity deception through replaying routing information. ATSR [22] is a location-based trust-aware routing solution for large WSNs. ATSR incorporates a distributed trust model utilizing both direct and indirect trust, geographical information as well as authentication to protect the WSNs from packet misforwarding, packet manipulation, and acknowledgments spoofing. Another trust-aware routing protocol for WSNs is TARP [23], which exploits nodes past routing behavior and link quality to determine efficient paths.

CONCLUSIONS

Fig. 9. Comparison of CTP and the TARF-enabled CTP in detecting the moving target.

as shown in Fig. 9b, in the pair (9, 6), no report from CTP node 9 is delivered while 46 reports from TARF node 6 is delivered; in the pair (10, 7), no report from CTP node 10 is delivered while 80 reports from TARF node 7 is delivered; in the pair (11, 8), 40 reports from CTP node 11 is delivered while 167 reports from TARF node 8 is delivered. Taking into account the spatial proximity between each pair of nodes, the TARF-enabled CTP achieves an enormous improvement in target detection over the original CTP. The demonstration of our TARF-based target detection application implies the significance of adopting a secure routing protocol in certain critical applications. The experimental results indicate that TARF greatly enhances the security of applications involving multihop data delivery.

http://ieeexploreprojects.blogspot.com
1.

We have designed and implemented TARF, a robust trustaware routing framework for WSNs, to secure multihop routing in dynamic WSNs against harmful attackers exploiting the replay of routing information. TARF focuses on trustworthiness and energy efficiency, which are vital to the survival of a WSN in a hostile environment. With the idea of trust management, TARF enables a node to keep track of the trustworthiness of its neighbors and thus to select a reliable route. Our main contributions are listed as follows: Unlike previous efforts at secure routing for WSNs, TARF effectively protects WSNs from severe attacks through replaying routing information; it requires neither tight time synchronization nor known geographic information. The resilience and scalability of TARF are proved through both extensive simulation and empirical evaluation with large-scale WSNs; the evaluation involves both static and mobile settings, hostile network conditions, as well as strong attacks such as wormhole attacks and Sybil attacks. We have implemented a ready-to-use TinyOS module of TARF with low overhead; as demonstrated in the paper, this TARF module can be integrated into existing routing protocols with the least effort, thus producing secure and efficient fully functional protocols. Finally, we demonstrate a proof-of-concept mobile target detection application that is built on top of TARF and is resilient in the presence of an antidetection mechanism that indicates the potential of TARF in WSN applications.

2.

3.

RELATED WORK

We discuss more related work here in addition to the introduction in Section 1. It is generally hard to protect WSNs from wormhole attacks, sinkhole attacks, and Sybil attacks based on identity deception. The countermeasures often requires either tight time synchronization or known geographic information [4]. FBSR [37], as a feedback-based secure routing protocol for WSNs, uses a statistics-based detection on a base station to discover potentially compromised nodes. But the claim that FBSR is resilient against wormhole and Sybil attacks is never evaluated or examined; the Keyed-OWHC-based authentication used by FBSR also causes considerable overhead. There also exists other work

4.

ACKNOWLEDGMENTS
This work is in part supported by AFRL contract FA865010-C-1740 and US National Science Foundation (NSF) Career Award CCF-0643521. The authors also would like to thank the program manager Mr. John Woods for his great support as well as the anonymous reviewers for their constructive comments and suggestions.

196

IEEE TRANSACTIONS ON DEPENDABLE AND SECURE COMPUTING,

VOL. 9,

NO. 2,

MARCH/APRIL 2012

REFERENCES
[1] [2] [3] [4]

[5]

[6]

[7]

[8]

[9]

[10]

[11]

[12]

[13]

[14]

[15]

[16]

[17]

[18]

[19]

[20]

[21] [22]

[23] A. Rezgui and M. Eltoweissy, Tarp: A Trust-Aware Routing Protocol for Sensor-Actuator Networks, Proc. IEEE Intl Conf. G. Zhan, W. Shi, and J. Deng, Tarf: A Trust-Aware Routing Mobile Adhoc and Sensor Systems (MASS 07), 2007. Framework for Wireless Sensor Networks, Proc. Seventh European [24] A. Abbasi and M. Younis, A Survey on Clustering Algorithms for Conf. Wireless Sensor Networks (EWSN 10), 2010. Wireless Sensor Networks, Computer Comm., vol. 30, pp. 2826F. Zhao and L. Guibas, Wireless Sensor Networks: An Information 2841, Oct. 2007. Processing Approach. Morgan Kaufmann, 2004. [25] S. Chang, S. Shieh, W. Lin, and C. Hsieh, An Efficient Broadcast Authentication Scheme in Wireless Sensor Networks, Proc. ACM A. Wood and J. Stankovic, Denial of Service in Sensor NetSymp. Information, Computer and Comm. Security (ASIACCS 06), works, Computer, vol. 35, no. 10, pp. 54-62, Oct. 2002. pp. 311-320, 2006. C. Karlof and D. Wagner, Secure Routing in Wireless Sensor Networks: Attacks and Countermeasures, Proc. First IEEE Intl [26] K. Ren, W. Lou, K. Zeng, and P. Moran, On Broadcast Authentication in Wireless Sensor Networks, IEEE Trans. Wireless Workshop Sensor Network Protocols and Applications, 2003. Comm., vol. 6, no. 11, pp. 4136-4144, Nov. 2007. M. Jain and H. Kandwal, A Survey on Complex Wormhole [27] P. De, Y. Liu, and S.K. Das, Modeling Node Compromise Attack in Wireless Ad Hoc Networks, Proc. Intl Conf. Advances in Spread in Wireless Sensor Networks Using Epidemic Theory, Computing, Control, and Telecomm. Technologies (ACT 09), pp. 555Proc. Intl Symp. World of Wireless, Mobile and Multimedia Networks 558, 2009. (WoWMoM 06), pp. 237-243, 2006. I. Krontiris, T. Giannetsos, and T. Dimitriou, Launching a [28] A. Woo, T. Tong, and D. Culler, Taming the Underlying Sinkhole Attack in Wireless Sensor Networks; The Intruder Side, Challenges of Reliable Multihop Routing in Sensor Networks, Proc. IEEE Intl Conf. Wireless and Mobile Computing, Networking and Proc. First ACM Intl Conf. Embedded Networked Sensor Systems Comm. (WIMOB 08), pp. 526-531, 2008. (SenSys 03), Nov. 2003. J. Newsome, E. Shi, D. Song, and A. Perrig, The Sybil Attack in Sensor Networks: Analysis and Defenses, Proc. Third Intl [29] S. Ganeriwal, L. Balzano, and M. Srivastava, Reputation-Based Framework for High Integrity Sensor Networks, ACM Trans. Conf. Information Processing in Sensor Networks (IPSN 04), Apr. Sensor Networks, vol. 4, pp. 1-37 2008. 2004. L. Bai, F. Ferrese, K. Ploskina, and S. Biswas, Performance [30] G. Zhan, W. Shi, and J. Deng, Poster Abstract: SensortrustA Resilient Trust Model for WSNs, Proc. Seventh Intl Conf. Analysis of Mobile Agent-Based Wireless Sensor Network, Proc. Embedded Networked Sensor Systems (SenSys 09), 2009. Eighth Intl Conf. Reliability, Maintainability and Safety (ICRMS 09), [31] C. Perkins and P. Bhagwat, Highly Dynamic Destinationpp. 16-19, 2009. Sequenced Distance-Vector Routing (dsdv) for Mobile CompuL. Zhang, Q. Wang, and X. Shu, A Mobile-Agent-Based ters, ACM SIGCOMM Computer Comm. Rev., vol. 24, no. 4, Middleware for Wireless Sensor Networks Data Fusion, Proc. pp. 234-244, 1994. Instrumentation and Measurement Technology Conf. (I2MTC 09), [32] O. Gnawali, R. Fonseca, K. Jamieson, D. Moss, and P. Levis, pp. 378-383, 2009. Collection Tree Protocol, Proc. Seventh ACM Conf. Embedded W. Xue, J. Aiguo, and W. Sheng, Mobile Agent Based Moving Networked Sensor Systems (SenSys 09), pp. 1-14, 2009. Target Methods in Wireless Sensor Networks, Proc. IEEE Intl Symp. Comm. and Information Technology (ISCIT 05), vol. 1, pp. 22- [33] G. Zhan, W. Shi, and J. Deng, Design, Implementation and Evaluation of Tarf: A Trust-Aware Routing Framework for 26, 2005. Dynamic WSNs, Technical Report MIST-TR-2010-003, Wayne J. Hee-Jin, N. Choon-Sung, J. Yi-Seok, and S. Dong-Ryeol, A State Univ., http://mine.cs.wayne.edu/~guoxing/tarf.pdf, Oct. Mobile Agent Based Leach in Wireless Sensor Networks, Proc. http://ieeexploreprojects.blogspot.com 2010. 10th Intl Conf. Advanced Comm. Technology (ICACT 08), vol. 1, [34] Q. Zheng, X. Hong, and S. Ray, Recent Advances in Mobility pp. 75-78, 2008. Modeling for Mobile Ad Hoc Network Research, Proc. 42nd Ann. J. Al-Karaki and A. Kamal, Routing Techniques in Wireless Southeast Regional Conf. (ACM-SE 42), pp. 70-75, 2004. Sensor Networks: A Survey, Wireless Comm., vol. 11, no. 6, pp. 6[35] X. Hong, M. Gerla, G. Pei, and C. Chiang, A Group Mobility 28, Dec. 2004. Model for Ad Hoc Wireless Networks, Proc. Second ACM Intl C. Karlof, N. Sastry, and D. Wagner, Tinysec: A Link Layer Workshop Modeling, Analysis and Simulation of Wireless and Mobile Security Architecture for Wireless Sensor Networks, Proc. ACM Systems (MSWiM 99), pp. 53-60, 1999. Intl Conf. Embedded Networked Sensor Systems (SenSys 04), Nov. [36] Motelab, http://motelab.eecs.harvard.edu, 2005. 2004. [37] Z. Cao, J. Hu, Z. Chen, M. Xu, and X. Zhou, FBSR: FeedbackA. Perrig, R. Szewczyk, W. Wen, D. Culler, and J. Tygar, SPINS: Based Secure Routing Protocol for Wireless Sensor Networks, Security Protocols for Sensor Networks, Wireless Networks J., Intl J. Pervasive Computing and Comm., vol. 4, pp. 61-76, 2008. vol. 8, no. 5, pp. 521-534, Sept. 2002. [38] T. Ghosh, N. Pissinou, and K. Makki, Collaborative Trust-Based R. Watro, D. Kong, S. Cuti, C. Gardiner, C. Lynn, and P. Kruus, Secure Routing against Colluding Malicious Nodes in Multi-Hop Tinypk: Securing Sensor Networks with Public Key Technology, Ad Hoc Networks, Proc. 29th Ann. IEEE Intl Conf. Local Computer Proc. Second ACM Workshop Security of Ad Hoc and Sensor Networks Networks, pp. 224-231, Nov. 2004. (SASN 04), pp. 59-64, 2004. A. Liu and P. Ning, Tinyecc: A Configurable Library for Guoxing Zhan received the MS degree in Elliptic Curve Cryptography in Wireless Sensor Networks, mathematics from Chinese Academy of Proc. Seventh Intl Conf. Information Processing in Sensor Networks Sciences in 2007 and another MS degree in (IPSN 08), pp. 245-256, 2008. computer science from Wayne State University I.F. Akyildiz, W. Su, Y. Sankarasubramaniam, and E. Cayirci, A in 2009. He is currently working toward the PhD Survey on Sensor Networks, IEEE Comm. Magazine, vol. 40, no. 8, degree in the Department of Computer Science pp. 102-114, Aug. 2002. at Wayne State University. He is interested in H. Safa, H. Artail, and D. Tabet, A Cluster-Based Trust-Aware research on participatory sensing, wireless Routing Protocol for Mobile Ad Hoc Networks, Wireless Netsensor network, mobile computing, networking, works, vol. 16, no. 4, pp. 969-984, 2010. and systems Security, trust Management, and W. Gong, Z. You, D. Chen, X. Zhao, M. Gu, and K. Lam, Trust information processing. Several of his research papers have been Based Routing for Misbehavior Detection in Ad Hoc Networks, J. presented at international conferences or published in journals. Networks, vol. 5, no. 5, pp. 551-558, May 2010. Additionally, he has instructed a few computer science laboratories Z. Yan, P. Zhang, and T. Virtanen, Trust Evaluation Based consisting of interactive short lectures and hands-on experience. Security Solution in Ad Hoc Networks, Proc. Seventh Nordic Workshop Secure IT Systems, 2003. J.L.X. Li and M.R. Lyu, Taodv: A Trusted Aodv Routing Protocol for Mobile Ad Hoc Networks, Proc. Aerospace Conf., 2004. T. Zahariadis, H. Leligou, P. Karkazis, P. Trakadas, I. Papaefstathiou, C. Vangelatos, and L. Besson, Design and Implementation of a Trust-Aware Routing Protocol for Large WSNs, Intl J. Network Security and Its Applications, vol. 2, no. 3, pp. 52-68, July 2010.

ZHAN ET AL.: DESIGN AND IMPLEMENTATION OF TARF: A TRUST-AWARE ROUTING FRAMEWORK FOR WSNS

197

Weisong Shi received the BS degree from Xidian University in 1995 and the PhD degree from the Chinese Academy of Sciences in 2000, both in computer engineering. He is an associate professor of computer science at Wayne State University. His current research focuses on computer systems, mobile and cloud computing. He has published more than 100 peer reviewed journal and conference papers. He is the author of the book Performance Optimization of Software Distributed Shared Memory Systems (High Education Press, 2004). He has served the program chairs and technical program committee members of several international conferences. He is a recipient of the US National Science Foundation (NSF) CAREER award, one of 100 outstanding PhD dissertations (China) in 2002, Career Development Chair Award of Wayne State University in 2009, and the Best Paper Award of ICWE 04 and IPDPS 05. He is a senior member of the IEEE.

Julia Deng received the PhD degree in the Department of Electrical Engineering from the University of Cincinnati in 2004, majoring in communications and computer networks. She is currently a principal scientist at Intelligent Automation, Inc. Her primary research interests include protocol design, analysis, and implementation in wireless ad hoc/sensor networks, network security, information assurance, and network management. At IAI, she serves as the PI and leads many network and security-related projects, such as secure routing for airborne Networks, network service for airborne Networks, DoS mitigation for tactical networks, trust-aware querying for sensor networks, trusted routing for sensor networks, agent-based intrusion detection system, just to name a few.

. For more information on this or any other computing topic, please visit our Digital Library at www.computer.org/publications/dlib.

http://ieeexploreprojects.blogspot.com