A Distributed Mechanism for Handling of Adaptive ... - Springer Link

Gunasekaran R, Uthariaraj V R, Yamini U et al. A distributed mechanism for handling of adaptive/intelligent selfish misbehaviour at MAC layer in mobile ad hoc networks. JOURNAL OF COMPUTER SCIENCE AND TECHNOLOGY 24(3): 472–481 May 2009

A Distributed Mechanism for Handling of Adaptive/Intelligent Selfish Misbehaviour at MAC Layer in Mobile Ad Hoc Networks Raja Gunasekaran1 , Member, IEEE, Vaidheyanathan Rhymend Uthariaraj2 , Member, IEEE Uamapathy Yamini1 , Member, IEEE, Rajagopalan Sudharsan1 , Member, IEEE and Selvaraj Sujitha Priyadarshini1 1

Department of Information Technology, Anna University, Chennai-600044, India

2

Ramanujan Computing Centre, Anna University, Chennai-600025, India

E-mail: [email protected]; [email protected]; {yamini.mit, sudhar86, sujibh}@gmail.com Received May 25, 2008; revised January 7, 2009. Abstract Medium access control (MAC) protocols such as IEEE 802.11 are used in wireless networks for sharing of the wireless medium. The random nature of the protocol operation together with the inherent difficulty of monitoring in the open poses significant challenges. All nodes are expected to comply with the protocol rules. But, some nodes in order to gain greater benefits misbehave by not complying with the rules. One such selfish misbehavior is waiting for smaller back-off intervals when compared to the other nodes in the same subnet. Such selfish misbehavior is being tackled in this paper. A diagnosis scheme and a penalty scheme are being proposed for overcoming such selfish-misbehavior at MAC layer of mobile ad hoc networks which could be extended to other types of networks also. Keywords

1

WLANs, MAC layer, DCF, contention, backoff

Introduction

A mobile ad hoc network (MANET) is a type of ad hoc network that can change locations and configure itself on the fly. The network’s wireless topology may change rapidly and unpredictably[1] . IEEE 802.11 is being used as the Medium Access technique for distributed contention resolution of the wireless channel. The distributed and random nature of the protocol makes it more prone to attacks. Though all nodes are expected to comply with the protocol rules, some do not follow it and make use of the loopholes in the protocol to gain greater benefits and higher bandwidth than the other nodes in the network. Several such selfish misbehaviors have been discovered[2] such as hosts intentionally dropping packets from other nodes without forwarding them in order to save battery power, jamming up packets from other nodes, and manipulating back-off timer values (the misbehavior addressed in this paper). Such selfish hosts at MAC (Medium Access Control) layer which use smaller back-off values deny the channel access to other well behaved hosts. Also there is a problem of colluding selfish nodes where both the sender and Regular Paper

receiver nodes mutually agree to misbehave[3] . The rest of the paper is organized as follows. Section 2 summarizes related work. Section 3 presents a basic introduction to the MAC protocol 802.11 DCF. Section 4 presents a proposal that detects and prevents misbehaving hosts, Section 5 details the simulation results, and Section 6 discusses about the future research directions and concludes the paper. 2

Related Work

This section gives a brief overview of the current research work on detection and prevention of selfish misbehavior at MAC layer. Most researches addressing selfish misbehavior assume that selfish hosts misbehave primarily to improve their own performance[4] (throughput, latency, energy, etc.). In [5], non-cooperative user behavior in randomaccess wireless networks is discussed, in which users have freedom to choose their back-off contention window size according to network’s congestion status. A non-cooperative game is formulated existence, and the uniqueness of its equilibrium point is shown. An iterative method leading to the equilibrium point of the game is proposed with alternative game formulation in

Raja Gunasekaran et al.: Alleviating Adaptive Selfish Misbehaviour in MANET

the same problem context. A nonparametric batch and sequential detectors based on the Kolmogorov-Smirnov (K-S) statistics[2,6] that do not require any modification on the existing CSMA/CA protocols are developed, and applied it to detect misbehaviors in an IEEE 802.11 DCF network using the ns-2 simulator. The performance of the proposed detectors and the optimum detectors with perfect information about the misbehavior strategy, for both the batch case (based on the Neyman-Pearson test), and the sequential case (based on Wald’s sequential probability ratio test) are compared. It is shown that the proposed nonparametric detectors have a performance comparable to the optimum detectors for the majority of misbehaviors (the more severe) without any knowledge of the misbehavior strategies. Determining how well or how bad the MAC layer protocol IEEE 802.11 DCF performs is studied in [7] and this question is studied by modeling the selfish MAC protocol as a non-cooperative repeated game where players follow the TIT-FOR-TAT (TFT) strategy which is regarded as the best strategy in such environments. For single-hop ad hoc networks the game admits a number of Nash Equilibria (NE), then NE refinement to eliminate the inefficient NE is performed and shows that there exists one efficient NE maximizing both local and global payoff. An algorithm to approach the efficient NE is also proposed. It is extended to multi-hop cases by showing that the game converges to an NE which may not be globally optimal but quasi-optimal in the sense that the global payoff is only slightly less than that of the optimal case. As conclusion, the posed question is answered by showing that selfishness does not always lead to network collapse. On the contrary, it can help the network operate at an NE globally which is optimal or quasi-optimal under the condition that players are long-sighted and follow the TFT strategy. [8] reverse-engineers backoff-based random-access MAC protocols in ad-hoc networks. It is shown that contention resolution algorithm in such protocols is implicitly participating in a non-cooperative game. Each link attempts to maximize a selfish local utility function, whose exact shape is reverse-engineered from the protocol description, through a stochastic subgradient method by which the link updates its persistence probability based on its transmission success or failure. It is proved that existence of a Nash equilibrium is guaranteed in general. The minimum amount of backoff aggressiveness needed, as a function of density of active users, for the uniqueness of Nash equilibrium and convergence of the best response strategy is established. Convergence properties and connection with

473

the best response strategy are also proved for variants of the stochastic-subgradient-based dynamics of the game. Together with known results in reverse-engineering TCP and BGP, this paper further advances the recent efforts in reverse-engineering layers 2∼4 protocols. In contrast to the TCP reverse-engineering results in earlier literature, MAC reverse-engineering highlights the non-cooperative nature of random access. Game-theoretic techniques[1] have been used to develop protocols which are resilient to misbehavior. Game theoretic approach assumes that all users are selfish, protocols are designed that reach an equilibrium state called the “Nash equilibrium” where a selfish host cannot gain more benefits than a well behaved host. A combination of deterministic and statistical methods that facilitate detection of malicious nodes causing bandwidth starvation and hence, a denial of service to legitimate nodes is proposed[9] . With this approach, each of the nodes is made aware of the pseudo-random sequences that dictate the back-off times of all its onehop neighbors. A blatant violation of the timer is thus immediately detected. In certain cases, a node may be unable to monitor the activities of its neighbor and therefore deterministically ascertain if the neighbor is misbehaving. To cope with such cases, a statistical inference method is proposed, wherein based on an autoregressive moving average (ARMA) of observations of the system state, a node is able to estimate if its neighbor is indulging in misbehavior. Simulation results show that with this method, it is possible to detect a malicious node with a probability close to one. Furthermore, the probability of false alarms is lower than 1%. In [10] the authors have proposed a modification of the existing DCF[11] functionality. Here the receiver is a trusted host namely a base station and it assigns the backoff value for which the senders should wait, before transmitting the next packet. Several possible schemes of node misbehavior in 802.11 for achieving a higher throughput are presented already. The detection of such misbehavior is achieved through a system called DOMINO[12] . However, their detection scheme for backoff manipulation, based on comparing average values of the backoff to given thresholds, is a suboptimal detection technique for every strategy of the greedy users. In [13] any one of the stations is assumed to be honest (either the sender or the receiver), here both the stations agree to a back-off value through a public discussion. This algorithm does not handle the problem of colluding selfish nodes. To address this problem entropy estimation is being used along with some statistical methods.

474

3

J. Comput. Sci. & Technol., May 2009, Vol.24, No.3

IEEE 802.11 Distributed Coordination Function

Table 1. Neighbor List Example

Distributed Coordination Function (DCF) is the fundamental MAC technique of the IEEE 802.11 wireless LAN standard[11] . DCF employs a CSMA/CA distributed algorithm and an optional virtual carrier sense using RTS and CTS control frames. DCF requires a station wishing to transmit to listen for the channel status for a DIFS interval. If the channel is found busy during the DIFS interval, the station defers its transmission. In a network where a number of stations contend for the multi-access channel, if multiple stations sense the channel busy and defer their access, they will also virtually simultaneously find that the channel is released and then try to seize the channel. As a result, collisions may occur. In order to avoid such collisions, DCF also specifies random backoff, which forces a station to defer its access to the channel for an extra period of time. The length of the backoff period is determined by the following equation: Backoff = integer(CW × rand × slot-time).

(1)

DCF also has an optional virtual carrier sense mechanism that exchanges short Request-to-Send (RTS) and Clear-to-Send (CTS) frames between source and destination stations during the intervals between the data frame transmissions. DCF includes a positive acknowledge scheme, which means that if a frame is successfully received by the destination it is addressed to, the destination needs to send an ACK frame to notify the source of the successful reception. DCF is defined in sub clause 9.2 of the IEEE 802.11 standard and is the de-facto default setting for Wi-Fi hardware. 4 4.1

Neighbor Nodes

Expected Backoff (ExpBOV)

Egocentricity (ε)

Access Values as Sender (S) (Sendacc)

Access Values as Receiver (R) (Recvacc)

1 2 3 4 5

10 15 20 13 18

3 4 5 1 2

10 13 14 20 5

12 5 0 14 2

The neighboring nodes, list all possible neighbors that the node can transmit within the BSS. The egocentricity (ε) defines the degree of selfishness of every node and takes a value between 0 (minimum) and 10 (maximum). At first for every node its egocentricity (ε) is set to 0. Whenever a node decides another node to be selfish it raises its egocentricity (ε) by 1 and decreases it by 1 when it senses a normal transmission. The access value of the nodes determines the number of times a node is accessing the channel either as a sender (Sendacc) or as a receiver (Recvacc). The access threshold (accthresh) is the average number of transmissions in the BSS (base station subsystem), i.e., the ratio between the total number of transmissions to the total number of nodes in the BSS. Whenever a node sends a DATA its Sendacc is increased by 1 and whenever a node receives DATA (i.e., sends ACK) its Recvacc is increased by 1. Whenever the egocentricity (ε) of a node is above maximum (10) it is judged to be selfish. SendAccRatio(SAR) = Sendacc RecvAccRatio(RAR) = Recvacc

Distributed Coordination Function/Selfish Misbehavior Avoidance (DCF/SMA) Neighbor List

In this scheme every node in the BSS maintains a data structure (neighbor list) shown in Table 1 containing • neighboring nodes; • egocentricity (ε) of each node; • the expected backoff value (ExpBOV) for the transmission; • access values for sender(S) and receiver(R) (Sendacc for S, Recvacc for R) informing the number of medium access of the node (i.e., the number of times a node transmits/receives within the BSS (basic service set)).

.X

.X

Sendacc, (2) Recvacc. (3)

RTS and CTS are broadcast messages and the entire neighborhood listens to it. The sender calculates the backoff value (BOV) from RecvBOV and SendBOV and so the neighboring nodes calculate the Expected Backoff (ExpBOV) for the sender. Using this information neighborhood manipulates its Neighbor list and thereby monitors the sender S during its next transmission. 4.2

Backoff Calculation

In CSMA/CA, every node having a packet to transmit will wait for a period of time (backoff time + DIFS time) before initiating transmission. A node is judged to be selfish if it chooses a small backoff value every


475

time and thereby hindering the channel access to every other node in the BSS (basic service set) whereas a node is judged to be misbehaving if it waits for a shorter interval than its accepted value (backoff) and accessing the channel more than its fair share. The proposed modification (discussed below) to IEEE 802.11 DCF ensures that the backoff value selected by each node for its transmission averages to an optimum value that prevents selfish-misbehavior. The modified backoff value (BOV) is being calculated as follows. A node to transmit sends the RTS incorporated with a backoff value (SendBOV). The SendBOV is calculated using the deterministic function (1) (discussed in DCF). An RTS frame structure is shown in Fig.1.

(5)

SendBOV = Random Number × slot-time × CW . (calculated by sender)

(4)

Fig.1. RTS frame.

Fig.2. CTS frame.

incorporated with a backoff value (RecvBOV) chosen by them shown in Fig.2. The RecvBOV is calculated using the deterministic function (1) (discussed in DCF), discussed in Fig.3. RecvBOV = Random Number × slot-time × CW . (calculated by receiver)

The RA (receiver address) of the CTS frame is copied from the TA (transmitter address) Field of the immediately previous RTS frame to which the CTS is a response. Now the intended receiver (R) is decided by the sender (S) upon analyzing the RA (Receiver address) and the RecvBOV (backoff value suggested by the receiver) sent by them in their CTS frames (i.e., the receiver (R) is the destination to which the packet is forwarded or the node that demands least backoff in the CTS frame among all other nodes) and transmits the DATA which ends with an ACK (Acknowledgment) in Figs. 4, 5. Now the sender (S) calculates its backoff value (BOV) for its next transmission (i.e., for its next RTS)

Fig.4. Data frame.

Fig.5. ACK frame.

Fig.3. Backoff calculation.

The receiving nodes respond with a CTS frame

Fig.6. Allocation of network allocation vector.

476


using (6) and (7) and saves with it (Fig.6). If (ε > 0) BOV = SendBOV ⊕ RecvBOV ε = ε + roundl(SAR + RAR),

(6)

else, BOV = Previous BOV + (SendBOV ⊕ RecvBOV), ε = ε + round(SAR + RAR) + 1.

(7)

The other nodes by overhearing the RTS and CTS frames can uniquely identify the sender (S), receiver (R) and its backoff values (SendBOV from S in RTS, RecvBOV from R in CTS). These nodes will calculate the expected backoff value (ExpBOV) using the same equations (6) and (7) given above and record it in the table (NEIGHBOR LIST) and increases the Sendacc of S and Recvacc of R by 1 after its DATA and its ACK transmission (discussed below). 4.3

Detection and Penalizing of Selfish Nodes

Whenever a node selects a backoff value less than every other node in the BSS for its continuous transmission attempts, it is marked to be a selfish node. Every node before starting the transmission waits for a period of time, BOV + DIFS. The DIFS time is more or less the same for all nodes. It is the BOV value that determines which node has to transmit. Now the selfishmisbehaving nodes will manipulate this BOV value to increase their performance. Here the BOV calculation of (6) and (7) itself ensures that it prevents selfishness to an extent (i.e., the BOV is calculated by manipulating both the sender and receiver).

every DATA transmission (i.e., before a node replies with CTS and before a sender determines its receiver) it checks the egocentricity (ε) and the access values of the sender/receiver (respectively) maintained in its neighbor list. Whenever a node is marked as selfish, it is penalized by assigning a greater BOV as in (6) and (7). Adding round (SAR + RAR) increases the probability that the medium is shared equally across the nodes which is shown in Fig.7. 4.4

Detection and Penalizing of Misbehaving Nodes

The sender may misbehave by backing off for a less duration then the accepted value. Hence all the neighboring nodes count the number of idle slots (BOV by S, ExpBOV by all the other nodes) on the channel, during the interval between the sending of ACK by R, and the reception of the next RTS from S. The sender is designated as deviating from the protocol if the observed number of idle slots BOV is lesser than its backoff value ExpBOV (calculated using the value in the table), i.e., BOV < ExpBOV. Now the sender is detected as misbehaving node. Once the sender is detected as a misbehaving node no other nodes send CTS to the sender. Now the transmission from the sender is blocked, and thus penalized shown in Fig.8.

Fig.8. Penalizing of misbehaving nodes.

Fig.7. Penalizing of selfish nodes.

Since the BOV is an XORed parameter of two values from two different sources, it is quarantined for an optimum value that prevents misbehavior. Further, before

A deviation does not necessarily indicate that the sender/receiver are misbehaving as the channel conditions seen by the sender and other nodes may be different. For example, if the sender senses the channel to be idle and counts down its backoff timer, while the other nodes senses the channel to be busy and does not count down its timer, then the transmission from the sender may be falsely designated as a deviation.


4.5

Handling Packet Retransmissions

477

a = 5,

probability of colliding again. When the sender uses a deterministic function f to compute the backoff value X after a collision, the receiver on reception of a packet from the sender can calculate this backoff value X used by the sender, by applying the same deterministic function f . If such a deterministic function is not used by the sender, the receiver cannot easily estimate the backoff value used by the sender after a collision. When an RTS is successfully received at the receiver (after possibly multiple transmission attempts by the sender), the receiver can estimate the number of retransmission attempts by using the attempt number field included in the RTS. An attempt number value greater than 1 indicates that there was at least 1 unsuccessful transmission attempt by the sender. The receiver can then estimate the total time, BOV (10), which the sender was expected to backoff (using the above deterministic function f (9)) This estimated backoff is then used in checking for possible deviation. It may be possible for the sender to provide incorrect attempt number values in the RTS. To ensure that senders provide correct attempt numbers, the receiver can sense the channel to identify high collision intervals (when the channel is mostly busy but few transmissions are successful). During these intervals, the receiver can analyze the traffic to identify any sender S achieving larger number of successful transmissions than other nodes, or having smaller average attempt values than other nodes. If such a sender S exists, the receiver can intentionally drop RTS packets from S occasionally, and verify that S increments the attempt number in the retransmission of RTS. Even a single failure by S to increment the attempt number in the retransmission is an immediate proof of misbehavior by S. As S does not know which RTS packets are lost due to collisions and which are intentionally dropped by the receiver, it will be harder for such misbehaving senders to persistently send incorrect attempt numbers without being detected. Dropping RTS packets occasionally will not significantly affect the throughput of S.

c = 2 × attempt + 1, and

4.6

Fig.9 demonstrates the working of the protocol after a collision. The number in parenthesis next to the RTS is the value of the attempt number. When an RTS is unsuccessful, it selects a new backoff using a deterministic function f (using the backoff previously assigned and concluded by the sender and the receiver (BOV)), the unique node identifier (nodeId), the attempt number (attempt) and contention window (CW) as follows. BOV = BOV+Σi=2

to attempt (f (BOV , nodeID, i)×CW i

(8) where, attempt is the attempt number in the received RTS, BOV is the backoff assigned by both the sender and the receiver, nodeId is the sender’s identifier, contention window, CW i = min((CWmin + 1) × 2i+1 − 1, CW max ).

Fig.9. Handling packet retransmissions.

The function f used by the sender for computing backoff values for retransmission attempt[10] is given by f (BOV, nodeId, attempt) = (aX + c) mod (CW min + 1), (9) where,

X = (BOV + nodeId) mod (CW min + 1). The function f generates a uniform random number between [0, CW min ] and dividing the number by CW min gives the required number between 0 and 1. The deterministic function f that we use has been carefully chosen to ensure that after collisions the colliding senders will select different backoff values with high probability. After a collision, the sender has to compute a new backoff value, from a larger range, to reduce the

Reducing Misdiagnosis

In IEEE 802.11, the channel is said to be busy in a slot in following cases: • when the slot has been reserved by an RTS or CTS, or a host is receiving a packet; • when the strength of the received signal on the channel (including noise and interference) is above a threshold called the “carrier sense threshold” (even if the Packet is not decoded correctly). This enables a host to sense transmissions originating from its two-hop

478


neighbors, and avoid colliding with them. Due to the hidden terminal problem, the sender can sense the medium to be idle and the receiver on the other hand could sense the medium to be busy. So the receiver freezes its counter and sender counts down to zero and transmits. The receiver thinks that the sender did not wait for the specified backoff time and so is misbehaving. But actually the sender is not misbehaving. So possible solutions to overcome this problem could be: i. The receiver classifies a slot be busy, only when overheard RTS/CTS has reserved the slot or a packet is received. The receiver does not classify a slot to be busy when only sensing a transmission but not receiving anything correctly. If this rule is followed then the receiver will in most cases identify the slot as busy or idle as the sender senses it. ii. The receiver decrements its counter value by monitoring the sender’s counter. So the two counters are synchronized. So when the sender’s counter reaches zero the receiver’s counter also should reach zero. If not the receiver could report the sender is misbehaving by following the above mentioned diagnosis scheme.

to it. This is the case of the intelligent attacker in which a node tries to misbehave more and obtain higher throughput share when not being penalized and reduce its misbehavior on the event of detection and penalization. Experimental analysis is done on the basis of both the misbehavior models. Fig.10 shows the comparison of backoff values computed by trusted nodes in IEEE 802.11 DCF and DCF/SMA. The graph is plotted with transmission number in X axis and backoff values in Y axis for a sample of 100 transmissions. From the graph it is clearly inferred that DCF/SMA mechanism reduces the overall backoff of trusted nodes by diagnosing the selfishmisbehavior nodes.

The first solution is being analyzed in this paper. 5

Simulation

DCF/SMA function is implemented with the help of ns2[13] simulator by extending the IEEE 802.11 standard. Here UDP is used as the transport layer protocol and CBR as the traffic source between the sender and the receiver. The simulation topology is randomly generated, and also the misbehaving nodes are selected at random. Each simulation with 70 nodes is being run for 30s and the results are being averaged across 10 simulations. The protocol is evaluated on the basis of the following metrics, namely percentage of correct diagnosis, percentage of misdiagnosis throughput of well-behaved hosts and throughput of misbehaving hosts. A misbehavior model in ns2 is implemented. In this model each node is assigned with a misbehavior percentage. If this misbehavior percentage is 100% then the node does not backoff at all. Say if it is 80% then the node backs off for only 20% of its actual assigned back-off value. So this problem of misbehavior has serious effects on the well-behaved hosts. Two misbehavior models have been used. One is called the persistent model where the misbehaving node does not change its percentage of misbehavior even on the event of detection and penalization. Another misbehavior model is being proposed called the Adaptive misbehavior model in which a host changes its magnitude of misbehavior based on the penalty assigned

Fig.10. Comparison of backoff values of trusted nodes computed using IEEE 802.11 DCF and DCF/SMA.

Fig.11 compares the fairness properties of DCF/SMA function with the IEEE 802.11 fairness properties using Jain et al.’s Fairness Index as the fairness metric which is defined as follows: ³ X ´.³ ³ X ´´ Tf Tf2 , (10) Fairness Index = N× f

f

where Tf represents the throughput of a flow (between a sender S and a receiver R) and N is the total number of flows. Fairness Index values close to 1 indicate better fairness. Fig.11. compares the fairness index of IEEE 802.11 and DCF/SMA scheme. The fairness index value of the DCF/SMA scheme is even slightly less than that of the IEEE 802.11 standard but the difference is very small and hence is tolerable.

Fig.11. Comparison of the fairness index between IEEE 802.11 and DCF/SMA.


Fig.12 plots the correct diagnosis percentage and the misdiagnosis percentage for various misbehavior percentages. When the misbehavior percentage is close to 0 the correct diagnosis percentage is not very high. The percentage of correct diagnosis is quite high once the extent of misbehavior increases. For all our simulations we use degree of selfishness ε = 10, and then we allow an amount of tolerance when comparing the backoff time for which a sender waits actually to reduce misdiagnosis, and that is the tolerance parameter called alpha. For our simulations we use alpha = 0.8. The percentage of Misdiagnosis reduces sharply as the percentage of Misbehavior increases.

479

higher when the misbehaving host uses an aggressive misbehavior policy of adding larger additive increase values. Although the diagnosis accuracy is lesser than that of persistent model, the throughput gain obtained by the misbehaving host is as same as that of the misbehaving nodes under the persistent model. The throughput gain is the same because the misbehaving hosts are being penalized as soon as they are diagnosed to be misbehaving and so retransmission occurs. Fig.14 shows the plot of throughput vs. additive increase value.

Fig.14. Average throughput of well-behaved hosts and misbehaving hosts. Fig.12. Diagnosis accuracy for varying magnitudes of misbehavior.

The Adaptive misbehavior model is as described. Initially to begin with the value of the variable, Percentage of Misbehavior PM is set to 0. Then its value is being increased by an additive value in steps. This is a slow increase phase. When a misbehaving node is being diagnosed and penalized by denial of CTS, then that node reduces its percentage of misbehavior value by half and then again repeats the same procedure of additive increase in steps. This reducing by half each time it is being penalized is called Exponential decrease.

The modifications given for reducing misdiagnosis are tested under both the adaptive and persistent misbehavior model. The modified scheme is quite successful in reducing misdiagnosis. Fig.15 plots the correct diagnosis percentage and misdiagnosis percentage using the persistent misbehavior model. The correct diagnosis accuracy does not change under the persistent model using this extension. The misdiagnosis percentage drops to zero.

Fig.15. Percentage of correct diagnosis and misdiagnosis using persistent model. Fig.13. Percentage of correct diagnosis for adaptive misbehavior model.

For experimental purposes using an additive increase value of 5, 10, . . . , 25 the protocol performance is evaluated. The following figure (Fig.13) plots the diagnosis accuracy of the protocol vs. the additive increase value using the adaptive misbehavior model. It could be inferred that the percentage of correct diagnosis is a little less when compared to the persistent misbehavior model for less additive increase values but it becomes

Fig.16 plots the correct diagnosis percentage and misdiagnosis percentage for the modified protocol under the adaptive misbehavior model. As it is seen the correct diagnosis percentage is lower than that obtained under the base scheme under the adaptive misbehavior model. The percentage of misdiagnosis has dropped to zero. The low diagnosis accuracy is due to classifying some busy slots to be idle, thereby not detecting some instances of misbehavior. The throughput curves for the extended scheme is similar to that of the base scheme, so the extension for reducing misdiagnosis does

480

not allow the misbehaving hosts to gain any additional advantages.

Fig.16. Percentage of correct diagnosis and misdiagnosis using adaptive model.

Fig.17 and Fig.18 compare the sensitivity of correct diagnosis and misdiagnosis respectively with different values of degree of selfishness (ε) and tolerance parameter (alpha) in a random scenario. As it is seen from the results, the correct diagnosis percentage is not very sensitive to variations in parameter values. The misdiagnosis percentage is higher when a small degree of selfishness (ε = 1) value is chosen, but is not very sensitive to exact ε value or alpha value when larger values are chosen. When ε value is small, channel variations may increase the possibility of misdiagnosis, but, with larger values such an effect is minimized.

Fig.17. Sensitivity of DCF/SMA with varying parameters (correct diagnosis).

Fig.18. Sensitivity of DCF/SMA with varying parameters (misdiagnosis).


6

Conclusion

Handling MAC Layer misbehavior is an important requirement in guaranteeing service availability. An MAC protocol which simplifies misbehavior detection for non-colluding and colluding selfish nodes in ad hoc networks has been proposed. Plans are there to augment the existing scheme with mechanisms to detect a misbehaving node that gains more bandwidth by using multiple MAC Addresses and to handle selfish misbehavior at MAC layer in which nodes choose a smaller SIFS value, there by deceiving the already existing techniques for detecting misbehavior at MAC layer and other higher layers. References [1] Moscibroda T, Wattenhofer R. The complexity of connectivity in wireless networks. In Proc. IEEE INFOCOM, Barcelona, Spain, April 23–29, 2006, pp.1–3. [2] Alberto Lopez Toledo, Xiaodong Wang. Robust detection of selfish misbehavior in wireless networks. IEEE Journal on Selected Areas in Communications, August 2007, 25(6): 1124–1134. [3] Lopez Toledo A, Vercauteren T, Wang X. Adaptive optimization of IEEE 802.11 DCF based on Bayesian estimation of the number of competing terminals. IEEE Transaction on Mobile Computing, Nov. 2006, 5(9): 1283–1296. [4] Li M, Prabhakaran B. MAC Layer Admission Control and Priority Re-Allocation for Handling QoS Guarantees in NonCooperative Wireless LANs Mobile Networks and Applications. Springer Science & Business Media, Inc., 2005, pp.947– 959. [5] Jin Y, Kesidis G. Distributed contention window control for selfish users in IEEE 802.11 wireless LANs. IEEE Journal on Selected Areas in Communications, August 2007, 25(6): 1113–1123. [6] Lopez Toledo A, Wang X. A robust Kolmogorov-Smirnov detector for misbehavior in IEEE 802.11 DCF. In Proc. IEEE International Conference on Communications (ICC’07), Glasgow, UK, June 24–28, 2007, pp.1564–1569. [7] Chen L, Leneutre J. Selfishness, not always a nightmare: Modeling selfish MAC behaviors in wireless mobile ad hoc networks. In Proc. the 27th International Conference on Distributed Computing Systems (ICDCS’07), Toronto, Canada, June 25–29, 2007, p.16. [8] Jang-Won Lee, Ao Tang, Jianwei Huang, Mung Chiang, A Robert Calderbank. Reverse-engineering MAC: A noncooperative game model. IEEE Journal on Selected Areas in Communications, August 2007, 25(6): 1135–1147. [9] Venkata Nishanth Lolla, Lap Kong Law, Srikanth V Krishnamurthy. Detecting MAC layer back-off timer violations in mobile ad hoc networks. In Proc. the 26th IEEE International Conference on Distributed Computing Systems (ICDCS 2006), Lisbon, Portugal, July 4–7, 2006, pp.63–73. [10] Kyasanur P, Vaidya N. Selfish MAC layer misbehavior in wireless networks. IEEE Transactions on Mobile Computing, September 2005. [11] Mackenzie B, Wicker S B. Game theory and the design of self-configuring, adaptive wireless networks. IEEE Communications Magazine, 2000, 39(11): 126–131. [12] Raya M, Hubaux J P, Aad I. DOMINO: A system to detect greedy behavior in IEEE 802.11 hotspots. In Proc.


[13]

[14] [15]

[16]

[17]

MOBISYS’04, Boston, USA, June 6–9, 2004, pp.84–97. Cardenas A A, Radosavac S, Baras J S. Detection and prevention of MAC layer misbehavior for ad hoc networks. Technical Report, Univ. Maryland, 2004. Fall K, Varadhan K. Ns notes and documentation. UC Berkeley, LBL, USC/ISI, Xerox PARC, 2003. Jin Y, Kesidis G. Charge sensitive and incentive compatible end to end window-based control for selfish users. IEEE JSAC Special Issue on Network Economics and Pricing, May 2006, 24(5): 952–961. Konorski J. Multiple access in ad-hoc wireless LANs with noncooperative stations. In Proc. NETWORKING, Coimbra, Portugal, LNCS 2345, Springer, May 15–19, 2002, pp.1141– 1146. MacKenzie B, Wicker S B. Stability of multipacket slotted aloha with selfish users and perfect information. In Proc. Infocom 2003, San Francisco, CA, IEEE, March 30–April 3, 2003, pp.1583–1597.

Raja Gunasekaran received the B.E. degree in computer science and engineering from the University of Madras, Chennai, India, and the M.E. degree in computer science and engineering from the Bharathiyar University, Coimbatore, India. Since 2003 he has been with Anna University, Chennai, India, where he is currently a lecturer in the Department of Information Technology. He is currently working toward the Ph.D. degree in computer science and engineering in Anna University. His research interests include mobile ad hoc networks, mobile communications and WiMAX. He is associated with Journal of Network and Computer Applications, Elsevier Publications as a reviewer. He is a member of IEEE and ISTE. Vaidheyanathan Rhymend Uthariaraj received the M.E. degree in computer science and engineering from Anna University, Chennai, India and the Ph.D. degree in computer science and engineering from the Anna University Chennai, India. Since 1986 he has been with Anna University, Chennai, India where he is currently a professor in the Department of Computer Science and Engineering. He is also the Director Ramanujan Computer Centre and Secretary, Tamilnadu Engineering Admissions. He was the coordinator of the campus wide networking for MIT campus in Anna University Chennai, India. His research interests include operation research, computer networks, pervasive computing and mobile communications. He is a member of IEEE, CSI and ISTE.

481 Uamapathy Yamini received the B.E. degree from Anna University, Chennai, India in computer science and engineering. She is working as a software engineer with Cisco Systems, India. She has worked on projects related to quality of service, priority scheduling, energy conservation in mobile ad hoc networks. She is a member of ISTE and CSI. Rajagopalan Sudharsan received the B.E. degree from Anna University, Chennai, India in computer science and engineering. He is working as a software engineer with Cisco Systems, India. He has worked on projects related to service differentiation and MAC admission control in mobile ad hoc networks. He is member of IEEE and ISTE. Selvaraj Sujitha Priyadarshini received the B.E. degree in computer science and engineering from Anna University, Chennai, India. She is working as a software engineer with Tata Consultancy Services, India. She has worked on projects related to quality of service and service differentiation in WiMAX. She is a member of ISTE and CSI.