Flow rank based probabilistic fair scheduling for wireless ad hoc networks

Transcription

1 Wireless Netw (2010) 16: DOI /s Flow rank based probabilistic fair scheduling for wireless ad hoc networks Md. Mamun-Or-Rashid Æ Muhammad Mahbub Alam Æ Md. Abdul Hamid Æ Choong Seon Hong Published online: 13 February 2009 Ó Springer Science+Business Media, LLC 2009 Abstract Fair scheduling is an ideal candidate for fair bandwidth sharing and thereby achieving fairness among the contending flows in a network. It is particularly challenging for ad hoc networks due to infrastructure free operation and location dependent contentions. As there is no entity to serve coordination among nodes, we need a mechanism to overcome inherent unreliability of the network to provide reduced collision and thereby higher throughput and adequate fair allocation of the shared medium among different contending flows. This paper proposes a flow rank based probabilistic fair scheduling technique. The main focus is to reduce the collision probability among the contending flows while maintaining the prioritized medium access for those flows, which ensures a weighted medium access control mechanism based on probabilistic round robin scheduling. Each flow maintains a flow-table upon which the rank is calculated and backoff value is assigned according to the rank of the flow, i.e., lower backoff interval to lower ranked flow. However, flow-table instability due to joining of a new flow, partially backlogged flow, hidden terminal and partially overlapped region exhibits a challenging problem that needs to be mitigated for our mechanism to work Md. Mamun-Or-Rashid M. M. Alam Md. Abdul Hamid C. S. Hong (&) Networking Lab, Kyung Hee University, Seocheon, Giheung, Yongin, Geyonggi , Korea cshong@khu.ac.kr Md. Mamun-Or-Rashid mamun@netwroking.khu.ac.kr M. M. Alam mahbub@netwroking.khu.ac.kr Md. Abdul Hamid hamid@netwroking.khu.ac.kr properly. We take appropriate measures to make the flowtable stabilized under such scenarios. Results show that our mechanism achieves better throughput and fairness compared to IEEE MAC and existing ones. Keywords Fairness Ad hoc networks IEEE MAC Round-robin scheduling 1 Introduction Ad hoc network can be envisaged as a collection of wireless nodes without the support of any infrastructure. Therefore, a node to communicate with other nodes, no access point controlling medium access is available. Two nodes communicate directly if they are within the transmission range. If two nodes are not within the transmission range, they can communicate with the help of other intermediary node which can forward the message. Therefore, an ad-hoc wireless network can be either a single-hop or multi-hop network. In a multi-hop network, the network node may be spread in a wide range that is typically larger than the transmission range of a node and all nodes may not be directly reached in one hop. Ad hoc network is envisioned to support various applications, e.g., video conferencing in a workshop seminar or interactive lecture, emergency disaster management, communication in a battle field. These types of diverse mission critical applications require certain QoS and delay guarantee. Fair scheduling is an ideal candidate for fair bandwidth sharing and thereby assuring certain level of QoS guarantee and bounded delay access. It requires support from both medium access and packet scheduling algorithms. Therefore, we first describe the short-term unfairness problem ingrained in IEEE MAC. Particularly we describe

2 714 Wireless Netw (2010) 16: A B C Fig. 1 Hidden terminal in wireless network the short term unfairness that has been characterized by Li et al. [1] in the presence of hidden terminals using Fig. 1. Consider the situation that nodes A and C in Fig.1 are having initial contention window value (e.g., CW min ). A partially overlapping RTS transmission from both the nodes in such situation results collision. Consecutive collisions under the scenario contribute to large contention window before getting the access of the medium by any of the node. Let us consider node A chooses a small backoff value from its CW, while the C chooses a larger value. The difference between these two values may be large enough to satisfy that there will be no RTS/CTS collision. Once A completes a successful transmission, according to IEEE DCF it resets CW and chooses backoff value before the initiation of another transmission. While, the backoff counter of node C may be large compared to the backoff counter of node A which is taken from the initial contention window value. Such situation allows A and B to exchange several packets before node C s backoff timer reduces to zero. This process (i.e., several packet transmissions by node A, followed by collisions, and then packet transmissions by node A again) may repeat several times, leading to starvation at node C for a long period. This shows how much short-term unfairness is ingrained in the current IEEE , which is unacceptable for jittersensitive traffic. The above scenario is described in the presence of hidden terminals. However, the same situation may occur in a co-located scenario if two or more nodes choose the same backoff value. Contention window reset enhances the probability of consecutive transmission for a node while starving other nodes within a co-located scenario. As a consequence, traffic flows in the starving nodes will be unable to achieve the sustainable level of QoS. Therefore, a fair medium access is very necessary to avoid such situations in ad hoc networks. Ensuring fair medium access among the nodes is challenging because of the unique issues in wireless ad hoc network. These issues include shared channel among multiple contending nodes, the sender of a flow does not have explicit information regarding the contending flows generated from other nodes and finally wireless channel capacity is a scarce resource. A great magnitude of attention has been paid to resolve these issues by developing fair queuing algorithms. All of these existing algorithms follow the notion of Fluid Fair Queueing (FFQ) model [2] and achieve fairness through packet level implementation. However, these algorithms work fine under accurate flow information of the network which is little unrealistic assumption. As because a flow joins or leaves in the network dynamically resulting flow information instability frequently. Also, these algorithms maintain strict round robin by restricting the randomness of medium access by the flows and thereby reducing overall throughput under dynamic scenarios (e.g., joining of new flow, partially backlogged flow). In this paper, we aim to provide fairness in multi-hop wireless ad hoc network for multiple weighted flows. To accomplish this we propose a flow-rank based distributed scheduling which ensures fair access of the flows that contend for the shared medium. Each flow is associated with a service-tag and it will be updated after each successful transmission. Rank of a flow is calculated based on the service-tag (Sect. 4.2). Access priority for the shared medium depends on the rank of the flows, i.e., the flow with minimum rank gets higher priority to access the medium. Each flow probabilistically calculates backoff value based on rank after each transmission (Sect. 4.3). Note that the fairness of the flows entirely depends on the stability of the flow-table. Hence, we identify that flowtable instability is a major issue that needs to be addressed. We start with a very simple scenario termed as co-located scenario, identifying the reasons of flow-table instability and present corresponding solutions (Sect. 4.5). Then we extend the scenario into general (multi-hop) topology that experiences flow-table instability due to hidden terminal and partially overlapped regions and we provide solutions for both the issues (Sect. 4.6). Finally, we incorporate spatial reuse to improve the overall throughput of the network. The rest of the paper is organized as follows. Section 2 briefly articulates related works. Section 3 states the network model and assumptions. Section 4 describes our protocol in details. We show a probabilistic analysis in Sect. 5 followed by performance evaluation in Sect. 6. Finally Sect. 7 concludes our work. 2 Related works A great magnitude of research effort has been performed on fair queueing algorithms for achieving a fair allocation of bandwidth on a shared link. One of the common properties of these proposed protocols are they maintain several queue (or flows) which store packets to be transmitted on an output link. A fair queueing algorithm is used to determine which flows to serve next so as to satisfy a certain fairness criterion. By design, these fair queueing algorithms are centralized, i.e., the algorithm is executed on a single node (e.g., access point, base station) and

3 Wireless Netw (2010) 16: therefore will be unable to work under ad hoc network environment. Fair queueing has been a popular paradigm for providing fairness, minimum throughput assurance and guaranteed delay in wired network [2], and in packet cellular networks [3 5]. Existing works on fairness in ad hoc network through fair scheduling includes [6 8]. Among them [7] is a subsequent work of [6, 8] and provide a detailed design of packetized fair queueing algorithm. It also ensures coordinated fair channel access among spatially contending flows and spatial reuse of bandwidth among the non-contending flows. Flow contention (accomplished by flow contention graph) and flow weight is propagated once when the flow was established the first time. Whether a flow is backlogged and transmission of flow is updated dynamically using a conflict free multicast tree. Using the flow control information, packets are scheduled to transmit using a lookahead window. The fairness model performs well in ad hoc network environment with such assumption of two logical channels on a single shared physical channel. Some of the research works also incorporated both QoS and fair queueing in ad hoc networks. Both QoS guarantee and fair queueing in ad hoc networks have been proposed in [9] and [10, 11]. Also some of the works provide fairness with error compensation, e.g., [12, 13] and most of these are proposed based on the support of base stations. Another notion of achieving fairness, fair scheduling, can be classified into two groups, timestampbased [5, 7, 9] and credit-based [8, 10]. Timestamp-based protocols convert a node graph into flow graph and for each newly arrived packet two timestamps are assigned, namely start tag and finish tag. The start tag is set either to the system time at which the packet arrives, or to the finish tag of its previous packet, depending on which value is larger. The finish tag is set to the predicted finishing time, which is equal to the start tag plus the estimated packet transmission time. Either timestamp can serve as the service-tag. A back-off value is set based on the service-tag and it determines when the packet will be sent. The back-off timer is decremented by one at each time slot until it reaches zero. If the node with a zero timer finds that the channel is free, the packet is transmitted. Nodes with zero timers do not coordinate before transmission and thus collision may occur. Credit-based protocol [10, 12] assumes the network is divided into clusters and for each cluster there is a cluster head. Each flow simply maintains a counter to record the transmission credit, instead of using two tags as in timestampbased mechanisms. The basic scheduling concept is the less excess in usage value, the higher the transmission priority. The clustering approach is used to implement spatial channel reuse. However, both the protocols in [10] and [12] assume TDMA based shared channel accessed by all nodes and to avoid collision, code-division multiple access system is overlaid on the top of TDMA system. In CSMA/CA based shared medium access, DCF (distributed co-ordination function) performs the task of channel access. The transmission of packets depends on how frequently a station is accessing the shared channel. Therefore, the fair scheduling of transmission among the contending nodes can be controlled using contention window adjustment of IEEE DCF mechanism.chen and co-workers [14, 15] and [16] implement such notion to achieve higher throughput and short-term fairness. In [14] each AP (Attach Point) monitors number of nodes attached with it and defines contention window as a multiple of number of nodes. It also modifies the basic contention resolution by increasing the contention window linearly rather than exponentially. To ensure fairness the contention window is reset to the previous value in case of a successful transmission. The mechanism experiences less collision in low network load and achieves higher throughput. In [15] contention window adjustment technique is proposed based on transmission bit rate achieved by the nodes. Therefore, it achieves higher throughput and fairness if the nodes in the network have multiple physical layer bit rate due to Automatic Rate Fallback (ARF) mechanism. However, if large number of nodes operates in the same bit rate, the protocol suffers from throughput degradation. In [16] fairness is achieved by allocating bandwidth in proportion of the weights of the flows in the network. Distributed co-ordination function (DCF) of IEEE is modified and each flow chooses the backoff value proportionate to their flow weights to get medium access. Therefore, flows with higher weights have higher probability to access the medium. However, the protocol is designed for wireless LAN and not suitable for ad hoc network environment. Our proposed mechanism takes contention window value as a prime concern to provide fair access of the shared medium and thereby ensuring fairness among the contending flows. Also, it reduces the collision probability and thereby achieving higher throughput by introducing flow tank based probabilistic medium access (Table 1 in Sect. 5 shows the collision probability compared to IEEE ). Accordingly, we modify the basic access mechanism of IEEE by introducing flow rank based probabilistic medium access. Flows with the lower rank value choose backoff from relatively smaller contention window than flows with higher rank value. Also, it ensures higher probability of accessing the medium for un-served flows by adjusting contention window after

4 716 Wireless Netw (2010) 16: each successful transmission. Therefore, the collision probability is reduced and thereby ensuring higher throughput in the network. 3 Network model and assumptions We consider a multi-hop network consisting of N nodes. As it is ad hoc network, there is no base station and intermediate nodes acts as the forwarder between any two source destination pair. Also, the wireless medium is shared among multiple contending nodes. Transmissions are locally broadcast and nodes within the transmission range are able to receive it. When a receiver is in the reception range of two simultaneously transmitting nodes, collision occurs and maybe unable to receive signal cleanly from either of them. We assume the packet size is variable, which is a reasonable assumption to justify weighted fairness in wireless networks. However in simulation we have considered fixed size packet and all flows having flow weight 1. To understand the flow and contention we have articulated some basic concepts as follows: Flow (f): A set of packets transmitting from one node to another is called a flow. Flows conflict with one another if packets from these flows cannot be scheduled for transmission simultaneously. Flows are said to be conflict-free if they do not conflict with each other. Flow contention graph: A flow contention graph (or flow graph) is defined as an undirected graph, G = (V, E), where V is the set of all flows and an edge ðf i ; f j Þ belongs to E if and only if flow f i conflicts with flow f j. In a flow graph, each vertex denotes a backlogged flow, and an edge between two vertexes indicates that the two flows are contending with each other. A flow graph (Fig. 2 left) and the corresponding flow contention graph (Fig. 2 right) are shown in Fig. 2. The flow contending graph explicitly describes which flows are contending and which flows can be concurrently transmitting. For example, from flow contention graph, a node can be aware that flows f 2 and f 3 are not contending. However, flow f 0 is contending with all the other three nodes as it has an edge between itself and the other flows. Each node maintains a table termed as flow-table which is consisting of the tuple \ node-id, flow-id, service-tag, temp-tag [. Where node-id and flow-id stands for their conventional meaning, service-tag is associated with each flow which will be calculated based on the start time of a packet of the corresponding flow (calculation is shown in Sect. 4). Propagation of tuple information associated with each flow is discussed in Sect We have introduced temp-tag to reset service-tag without hampering the fairness and will be discussed in Sect Proposed protocol In this section we present our proposed mechanism in details. 4.1 Priliminaries In this section, we describe service-tag assignment and flow-table update operation. For each flow f we use the SFQ [17] algorithm to assign tags for the arriving packets: a start-tag and a finish-tag. But this tagging operation is dependent on the system virtual time [3]. However, in a distributed environment, this information is not available at each node. Allowing a system wide flood of the virtual time is too costly. Instead, we use a localized virtual time in the local neighborhood. During each transmission, each node can piggyback the current service-tag (start-tag will be considered as the service-tag of a flow) with the packet, while the neighboring nodes overhearing the packet keep a copy of the service-tag in order to determine the local virtual time. The local virtual time obviously may differ from the global virtual time. For the head-of-line (HOL) packet k of flow f, which arrives at time Aðt f kþ and packet size is L p, its start-tag and finish-tag are assigned as follows: (i) Start-tag: S f k ¼ Ff k 1 ð1þ If flow f is continuously backlogged, then S f k ¼ max g2sfv 2 gðaðt f k ÞÞg ð2þ Fig. 2 Flow graph and corresponding flow contention graph If flow f is newly backlogged, then (ii) Finish-tag: F f k ¼ Sf k þ L p w f ð3þ

5 Wireless Netw (2010) 16: where S consists of all flows stored in the flow-table of node n, and Vg(t) is the flow g 0 s virtual time at t. The starttag is used to find the transmission order of the HOL packets of a flow. So we use start-tag as the service-tag of the corresponding flow. Message exchange sequence: Each data transmission follows a basic sequence of RTS-CTS-DATA-ACK handshake, and this message exchange is preceded by a backoff of certain number of mini-slot times. At the beginning of each transmission slot, each node chooses a backoff value from and range based on the rank of the flow. The flow with the minimum service-tag will get the highest transmission priority based on the rules explained in Sect. 4.3.If the node does not hear any transmission then it decreases backoff value by one in each mini-slot. If the backoff timer of a node expires without overhearing any ongoing transmission, it starts RTS to initiate the handshake. If the node overhears some ongoing transmission, it cancels its backoff timer and defers until the ongoing transmission completes. In the meantime, it updates its local tables for the tags of the ongoing neighboring transmitting flow. When other nodes hear a RTS, they defer for one CTS transmission time to permit the sender to receive a CTS reply. Once a sender receives the CTS, it cancels all remaining backoff timers (for other flows). When hosts hear either a CTS, they will defer until the DATA-ACK transmission completes. Propagation of updated service-tag: In order to propagate a flow s service-tag to all its one-hop neighbors and reduce the chance of information loss due to collisions during the propagation, we attach the service-tag, node-id, and flow-id for flow f i in all four packets RTS, CTS, DATA and ACK. However, we do not use the updated tags for flow in RTS and CTS packets, since RTS and CTS do not ensure a successful transmission. When the handshake of RTS and CTS is completed, we attach the updated service-tag in ACK, to inform neighboring nodes about the new updated information of the current transmitting flow f i. 4.2 Rank calculation As described in Sect. 3, each node has a flow-table which consists of flow-id, node-id, service-tag and temp-tag. Note that a flow with lower service-tag has the higher priority to access the medium. Therefore, a flow having lower service-tag should have highest rank. Rank of a flow i, r(f i ), will be used to calculate backoff interval for the respective flow. Following algorithm calculates rank of a flow. Algorithm 1: Rank calculation BEGIN 1. rðf i Þ 0 2. FOR j = 1 to n where n is the number of flows in the flow-table 3. IF(j = = i) 4. j j þ 1 5. ENDIF 6. IF(service-tag(f i ) [ service-tag(f j ) 7. rðf i Þ rðf i Þþ1 8. ENDIF 9. ENDFOR END Algorithm 1 outputs the rank of a flow in an increasing order based on the service-tag value. Backoff value calculation based on rank is presented in the next section. 4.3 Rank based probabilistic medium access In IEEE , when a node wishes to transmit a packet, it chooses a back off interval equal to backoff (BO) slots specially BO is chosen uniformly distributed in the interval ½0 CW 1Š, where CW is the size of the so called contention window. Contention window at node i is reset to a value CW min at the beginning of time and also after its successful transmission of a data packet by node i. However this mechanism of assigning BO value creates short term and even long term unfairness of the medium. We explicitly overcome this very problem by assigning BO value based on the rank of a flow derived from the flow-table. Therefore each flow f i chooses its BO value using following equation: BO ¼ rand½0 frðf i ÞCW 1gŠ ð4þ where r(f i ) is the rank of the flow f i. According to Eq. 4 each flow gets a BO value based on its rank and thus a flow with lowest rank has higher probability to access the medium. Also after each transmission service-tag will be updated according to Eq. 1 which in turn ensures fair medium access among the flows in a round robin fashion. However, as the access mechanism is random, probability exists that two nodes choose same BO value and thus collide each other. After the collision both the nodes will again contend for the medium and the process continues. To overcome this problem we need to increase the range of the random value that a node may choose. Hence, we modify Eq. 4 to adapt colliding situations as BO ¼ rand½0 f2 c rðf i ÞCW 1gŠ ð5þ

6 718 Wireless Netw (2010) 16: where c is the number of collisions experienced. Equation 5 elongates the range of contention window to choose BO thereby, reducing the collision probability. Thus the transmission of the HOL packets of each flow is scheduled according to the rank of the corresponding flow. After each successful transmission each node updates their corresponding flow-table and chooses a new backoff value according to the mechanism described above. Our backoff assignment policy not only ensures fairness but also reduces the collision probability and thereby increase throughput of the network. We have stated probabilistic analysis and results of reduced collision and increased throughput in Sect Service-tag re-initialization service-tag associated with each flow increases infinitely as the network operation time increases. This requires reinitialization of virtual clock. However, virtual clock cannot be reinitialized until system becomes empty. Because if re-initialization is allowed the flow that first reinitializes has the minimum service-tag. This forces repeated transmission of the flow of itself until its service-tag is greater than the minimum of any other flows. Allowing a single flow (having maximum service-tag) does not solve the problem. This requires a simultaneous re-initialization of the flow-table by all the nodes. Following algorithm resets service-tag without hampering the fairness of the existing flows. Algorithm 2: Service-tag reset BEGIN 1. service-tag max 2. FOR all flows f i {i [1,2,3...n} 3. temp-tag(f i )=current-tag(f i )-service-tag max 4. ENDFOR 5. reset_flag_counter = 0 6. FOR all flows f i {i [1,2,3...n} 7. IF (temp-tag(f i ) [ 0) 8. reset_flag_counter = reset_flag_counter? 1 9. ENDIF 10. ENDFOR 11. IF(reset_flag_counter = = n) 12. FOR all flows f i {i [1,2,3...n} 13. current-tag(f i ) = temp-tag(f i ) 14. ENDFOR 15. ENDIF END Algorithm 2 re-initializes the service-tag simultaneously when the value of all the flows exceeds the predefined threshold ðservice tag max Þ and thereby solves the infinite increase of service-tag value. 4.5 Co-located scenario In a co-located scenario, all nodes are one hop neighbors of each other and hence a transmission is heard by all other nodes as shown in Fig. 3. As the flows are having rank based access to the medium, each flow will maintain a weighted round robin scheduling to access the medium. Therefore, the above mechanism of assigning backoff value and thereby medium access works under ideal situation, where ideal situation means flow-tables of all nodes are consistent. However, due to ad hoc network dynamics, new node/flow may join in the network and also, a flow might be partially backlogged and these make the flowtable instable. Next, we show how instability occurs due to above mentioned reasons and what are the mechanisms to resolve the instability problem. Case 1. Joining of a new flow: If a new flow starts in the network, then the new flow is unable to inform the other nodes about the service-tag. Moreover, if the flow starts from a newly joined node, it will consider itself as the only flow in the network and continuously contend for the medium with rank 1. On the other hand, if the flow starts from an existing node which already has a flow-table and the knowledge of ongoing transmissions, is unable to inform the newly joined flow information. The flow-table of all other nodes remains unstable until the first packet of the newly joined flow is transmitted. When the first packet of the newly generated flow is transmitted all other nodes add the flow information in their table. When a flow is new but the node has flow-table (i.e., the node is old) the newly joined flow is assigned its service-tag based on Eq. 2. However, flow-table of other nodes still remains unstable Fig. 3 Co-located scenario of nodes in the network

7 Wireless Netw (2010) 16: as the first packet of the newly joined flow has yet to be transmitted. This results in higher probability of collisions. To avoid such situation we dedicatedly assign the first free slot of the contention period to the newly joined flow. Thus all nodes in the network choose BO value using following equation: BO ¼ rand½1 f2 c rðf i ÞCW 1gŠ ð6þ Again when a flow is new and so the node then there is no flow-table. In this case if medium is idle, the node immediately transmits the first packet of the new flow and all other nodes update their respective flow-tables. However, if the medium is busy newly joined node will hear the ongoing transmission of a node and adds the first entry of its own flow-table and waits until it hears the transmission of that node again (newly joined node has to wait at least one round before transmitting its first packet of the first flow). This will eventually built the flow-table of the existing flows in the network for the newly joined node. When the flow-table is built up it will behave like the scenario of flow in an existing node explained above. Case 2. Partially backlogged flows: A flow is said to be partially backlogged, if it is not backlogged for all times. Such a backlogged flow can make the flow-table unstable. When the queue of a partially backlogged flow is empty, it does not transmit and therefore flow-table is not updated. Eventually when the flow has maximum service-tag and other flows contend for the medium with less priority repeatedly. One simple solution of this problem could be, when the flow transmits the last packet (that is queue is temporarily empty) it may set a negative service-tag for the next packet which may act as an indication of last packet to others and all the other nodes remove the entry from the flow-table. But this may create another problem, when this flow receives further packets it may behave like a new flow explained earlier. To overcome this we use a simple solution. If a node does not receive an update for a flow which has rank 1, for n transmission slots only then it removes from the flowtable. If the transmitter of the flow does not receive a packet for transmission slots (when the queue is empty) it also removes the flow information from the flow-table. If transmitter receives packet after the removal of flow information, then it is considered as new flow. Fig. 4 General topology However, in multi-hop scenario where all nodes are not the neighbor of the sender, the above mentioned mechanism suffers from two major problems. First, flow-table instability due to the presence of a hidden node, and second, flow-table instability due to partially overlapped region. We present both the problem scenario especially how those are taken place and accordingly we present solutions to those problems. Case 1. Flow-table instability due to the presence of a hidden node: A station encounters two types of collisions, (a) collision due to simultaneous transmission by at least one of its neighbors and (b) collision due to the transmission of hidden terminals. Collision due to former reason is unavoidable and obviously there is no flow-table instability. However, collision due to hidden terminal causes flowtable instability and this must be taken care of. So, at first we identify why collision takes place due to hidden terminal problem and a mechanism is presented to resolve this issue. The transmission of a RTS frame of a particular station can collide with the RTS frame of a hidden terminal and this can happen during the transmission period of the RTS frame. We denote this period for an RTS frame as the critical period which is depicted in Fig. 5, where node A is 4.6 General topology So far we have devised the fairness mechanism for co-located scenario. In case of general topology shown in Fig. 4, our proposed scheme will achieve fairness by splitting the network into multiple co-located scenarios. Fig. 5 Critical period for a transmitter, where it can encounter a collision due to the transmission of hidden terminals

8 720 Wireless Netw (2010) 16: the transmitter, node B is the receiver and H 1 and H 2 are hidden nodes. Though the RTS/CTS handshaking mechanism is introduced to combat with hidden terminals, collision may occur during the transmission of the RTS frame. Note that node B does not send the CTS frame immediately after receiving the very first bit of the RTS frame from A. Instead it receives the whole RTS frame, waits a period of SIFS, then if the medium is found free, sends the CTS frame, which stops the hidden nodes from transmission and informs them that the medium is busy. As shown in Fig. 4, node A sends the RTS frame at time t 1, when its backoff counter reaches zero. After successfully receiving the RTS frame and waiting an SIFS period, at time t 2 node B sends the CTS frame. If the propagation delay is assumed to be zero, H 1 and H 2 receives the CTS frame at time t 2 and be aware about the transmission of node A. However, if backoff counter of any of the nodes H 1 and H 2 reaches zero, during the period ðt 2 t 1 Þ; they start their RTS transmission and a collision occur at node B. The delay between reception of the first bit of the RTS frame and transmission of the first bit of the CTS frame is the reason of the collision. Thus a collision can only be avoided if and only if the backoff counter of hidden terminals are greater than that of the transmitter by an amount ðt 2 t 1 Þ; that is, the condition that a transmission of RTS does not experience a collision in the presence of hidden terminal is d [ critical period ð7þ where d ¼ min i2ha ððd i Þ d A Þ and d i is the value of the backoff counter of node i, H A is set of all hidden nodes. Generally any node chooses backoff value according to Eq. 6. From the RTS/CTS handshaking, a node can identify itself whether it is hidden and to whom it is hidden. Suppose a node identifies itself as hidden to some other nodes and the rank of nodes to whom it is hidden are greater, choosing backoff value using Eq. 6 may collide with those nodes. Therefore, to avoid collision in the presence of hidden node, the difference of backoff value between hidden node and other node should be greater than critical period. BO ¼ rand½1 f2 c rðf i ÞCW 1gŠ þ critical period ð8þ Case 2. Fow-table instability due to partially overlapped region: Partially overlap region is identified when a node resides within two or more transmitter s common ranges as shown in Fig. 6. As can be seen from the figure, two simultaneous transmissions from senders S 1 and S 3 are heard by the node S 2. As service-tags are sent using Fig. 6 Partially overlapped region RTS mechanism, in case of node s 2, two situations may occur. IfS 2 is very close to any of the senders (S 1 or S 3 ), it will be able to receive RTS frame successfully due to high SNIR value. If both RTS collide, S 2 is unable to receive any of the RTS sent by S 1 or S 3. The former scenario does not create any problem as node S 2 can successfully retrieve the service-tag from the received RTS. However, the later case causes flow-table instability problem. Eventually the scheduling will be unfair. Node S 2 in the partial overlapped region will listen to the medium and if the medium is busy for more than an RTS frame transmission time, S 2 will know that there is an ongoing flow. Therefore, node S 2 will add a dummy flow in its flow-table. As node S1 and S 3 are aware of node S 2,so S 2 having rank 1 will get chance to transmit in its next round and hence fairness won t be hampered. 4.7 Spatial reuse Location dependent contention and the nature of multi-hop ad-hoc networks make it possible for spatial reuse. Any two flows can potentially transmit data packets simultaneously if they are not interfering with each other. Consider the example in Fig. 2 flow f 0 is contending with all the other three nodes as it has an edge between itself and the other flows. However, flows f 2 and f 3 are not contending and can transmit simultaneously to improve the overall throughput of the network. Therefore, to allow spatial reuse rank should be calculated among the contending flows. Hence, we modify our previous rank calculation algorithm in the following way:

9 Wireless Netw (2010) 16: Algorithm 3: Modified rank calculation BEGIN 1. r(f i ) = 1; 2. FOR j = 1 to n where n is the number of flows in the flow-table 3. IF(j = = i) 4. j j þ 1 5. ENDIF 6. IF(f i is contending with f j ) 7. IF(service-tag(f i ) [ service-tag(f j ) 8. rðf i Þ rðf i Þþ1 9. ENDIF 10. ENDIF 11. ENDFOR END Algorithm 3 outputs the rank of a flow in an increasing order based on the service-tag value among the contending flows. Accordingly, rank calculation (and so the calculation of backoff interval) of non-contending flows will not be affected by the contending flows. Therefore, non-contending flows will be able to transmit simultaneously resulting a higher throughput of the network. 5 Probabilistic analysis Since we have modified the basic backoff policy according to the rank of a flow, different flows have different contention window value to access the transmission medium according to the rank of the flow. Therefore we claim that our proposed backoff policy reduces the collision probability and thereby achieves higher throughput. For the simplicity of the analysis we have considered single flow in a node and a transmission is unsuccessful if there is a collision. In every transmission slot, every node chooses its backoff value based on its rank calculated from the flow-table according to Eq. 4. After choosing the backoff value each node decrements the backoff counter in every slot if the transmission medium is found to be free. Suppose p be the probability that the medium busy in a slot time. Also, after each successful transmission every node updates their flowtable and chooses a new backoff value. Then the system can be modeled as discrete time Markov chain as given in Fig. 7. The non-zero one step transition probabilities of the Markov chain is given in Eq. 9. Pfkj0g ¼ 1=CW; k 2ð0; CW 1Þ Pfkjkg ¼ p=cw; k 2ð1; CW 1Þ ð9þ Pfkjk þ 1g ¼ 1 p; k 2ð0; CW 2Þ Fig. 7 Markov model for backoff counter Following a similar analysis given in [18], each flow having the rank r, will have an expected backoff value given by: E½BOŠ ¼ r CW þ 1 r ¼ 1; 2; 3::::n 1 ð10þ 2 As our backoff policy introduces lower contention window to the lower ranked nodes, the probability that a node transmits in a randomly selected slot is given by 2 s ¼ r ¼ 1; 2; 3::::n 1 ð11þ r CW þ 1 So, a node with lower rank transmits with a higher probability than others. However in general s depends on conditional collision probability. That is if more than one node chooses the same s value there will be a collision (Table 1). As the contention window of each node varies with rank the probability that more than a node transmits in a randomly selected slot becomes: s ¼ 1 r 2 r ¼ 1; 2; 3::::n 1 ð12þ r CW þ 1 Now, the probability that there is no transmission in a randomly selected slot is given by P n r 1 ð1 s iþ ð13þ Therefore, the probability that there is at least one transmission in a randomly selected slot is P trans ¼ P n r 1 ð1 s iþ ð14þ Now the probability that there is exactly one transmission in a randomly selected slot is given by Table 1 Collision probability Contending nodes Proposed scheme IEEE

10 722 Wireless Netw (2010) 16: P one ¼ Xn 1 r s i P n 2 j¼0;j6¼i ð1 s jþ ð15þ Then the probability that there is a successful transmission in a given slot, conditioned on the fact that there is at least one transmission in that slot is P suc ¼ P one P trans ¼ P n 1 r s ip n 2 j¼0;j6¼i ð1 s jþ P n r 1 ð1 s iþ ð16þ And there will be a collision if more than one node transmits in a single slot that is simultaneous transmission by more than one node Therefore the collision probability can be given as: P col ¼ P trans ð1 P suc Þ ð17þ The following table shows the comparison of collision probability of IEEE and our proposed scheme. We have taken the transmission probability in a randomly selected slot (s) of IEEE from the analysis given in [19]. In case of our proposed scheme s is calculated from Eq. 12. On an average our proposed mechanism reduces collision probability by a factor of Finally the throughput can be calculated as: P trans P suc E½PŠ tput ¼ ð1 P trans Þr þ P trans P suc T s þ P col T c ð18þ where r,t s, T c and E[p] are the slot time, successful transmission time, collision period and expected packet size respectively. As our proposed mechanism reduces the amount collision greatly, it achieves a better throughput than IEEE Performance evaluations 6.1 Simulation environment and parameters We have evaluated the performance of our proposed protocol using ns-2.30 [20]. To implement our proposed mechanism we have modified the basic implementation of IEEE in ns-2 by adding a flow-table at the MAC layer. Also, the necessary modifications are employed in transmission of RTS, CTS, DATA and ACK packets to update the flow-table entries. The radio model is based on the existing ns-2 architecture with a wireless transmission range of 250 m and channel capacity of 2 Mbps. It is to be mentioned that a transmission range of 250 m yields an interference range of 550 m. We run our simulation for both co-located scenario and general topology 15 times and plotted the results in the figures. For the simulation of colocated scenario we have set 32 nodes in an area of and nodes are assigned CBR traffic with a packet size of 512 bytes. On the other hand, for general topology 64 nodes are placed in an grid within the area of and as in co-located scenario nodes are equipped with CBR traffic with a packet size of 512 bytes. To saturate the channel we have generated 100 packets per second from each CBR sources. For both the scenarios Ad Hoc on-demand Routing (AODV) protocol [21] is used as routing protocol. The following metrics are used to utilize the performance of proposed schemes Throughput: It is measured as the sum of received packets for all flows per unit time. Number of collisions experienced: It indicates the amount of collisions experienced per second. Fairness Index: It indicates how fairly the channel bandwidth is shared among the flows. We calculate the fairness index for each flow as in [16], i.e., P i ðx iþ 2 ðn P ðx i Þ 2 Þ: where x i is the throughput achieved by flow i and n is the total number of flows in the network. Finally, we plot fairness index by averaging it over the total number of flows. The performance comparisons of the following mechanisms are carried out Basic IEEE : Experiments are carried out using the basic implementation of IEEE in ns2. It follows basic RTS-CTS-DATA-ACK handshake and backoff procedure to access the channel. Luo et al. [7]: To achieve fairness it proposed packetized fair queueing algorithm. Scheduling of packet transmission is accomplished by flow information and lookahead window. Finally, it enables spatial reuse using flow contention graph and achieved improved throughput. Chen et al. [14]: In this mechanism a central node, i.e., AP supervises the number of nodes and defines the contention window as a multiple of the number of nodes. BEB procedure is modified by increasing the contention window linearly rather than exponentially. Yassine et al. [15]: It proposed contention window adjustment technique based on transmission bit rate achieved by the nodes. TBCP [12]: A central node, i.e., clusterhead keeps track of transmission credit of the flows and implements the less excess in usage value, the higher the transmission priority for scheduling.

11 Wireless Netw (2010) 16: Fig. 8 Throughput of four flows in a co-located scenario Fig. 10 Throughput of eight flows in a co-located scenario 6.2 Simulation results Our first set of experiment shows the fairness achieved by our proposed mechanism under increasing number of flows in a co-located scenario. In Fig. 8 the throughput of each flow differs significantly. This explicitly shows the amount of unfairness in IEEE where different flow gets varying amount of channel access and achieves different amount of throughput. In case of Luo et al. [7], the collision resolution is not explained. Therefore if a collision occurs due the random access to the channel by the different flows, the flow experiencing a collision engenders repeated collisions in the networks. As a result, there will be throughput degradation and at the same time flow Fig. 11 Throughput of 10 flows in a co-located scenario Fig. 9 Throughput of six flows in a co-located scenario experiencing repeated collision gets less share of the channel. Also, in Figs. 9, 10 and 11, fairness which is expressed in terms of throughput achieved by the flows, degrades when the number of flow increases. On the other hand, our proposed mechanism achieves fairness under increasing number of flows. It is natural that per flow throughput decreases when we increase the number of flows in the network. However, throughput achieved by each flow gets almost equal share in our mechanism. The reason behind this is after each transmission, when a flow updates its own service-tag it will get a higher rank and thereby, lower priority to access the medium. A flow gets chance again only when all of the other flows transmit at least once, unless otherwise all flows have weight 1. When a flow is partially backlogged, it means the flow does not have packets for all the times. Therefore, network

12 724 Wireless Netw (2010) 16: Fig. 12 Throughput in presence of partially backlogged flows should achieve the throughput according to the number flows existing in the network. Figure 12 shows the trend of achieved throughput for partially backlogged flows. For first 50 s flow f 1 and f 2 is active, while flow f 1, f 2 and f 3 is active for s, at 30 s we stopped flow f 3 and continued up to 40 s and finally 4 flows are active during 40 to 50 s. Note that, in Luo et al. [7] flows are scheduled centrally and for partially backlogged flows the channel remains idle if there is no packet for particular time. Also, if a flow arrives newly in the network, it causes a certain amount of collision due to centralized scheduling. As a consequence, there is drastic degradation of throughput for Luo et al. [7]. On the other hand, we have handled new flows by allocating the very first transmission slot and partially backlogged flows through overhearing the other flows in the channel. This reduces the amount of collisions and thereby achieves higher throughput. To observe the short-term fairness among the flows in a co-located scenario we have counted the number of received packets of the flows for 3 s duration while 25 s duration for long-term fairness. Basic mechanism of IEEE follows exponential increase of contention window for collision resolution and contention window reset after a successful transmission. As a consequence, the number of received packets of different flows in short interval (3 s) varies significantly. Our proposed mechanism controls the transmission of frames by providing lower contention window to the un-served flows and higher contention window to the served flows. Therefore, the variation of number of received packets for different flows is very less. Luo et al. [7] adopts flow contention graph and schedules the transmission of the flows with the help of a sliding window in a round robin fashion. As a consequence, the variation in number of received packets of different flows for short interval is less. In case of number of received packets for a long interval of 25 s, all flows achieve almost equal number of received packets and thereby ensuring long-term fairness. However, our proposed scheme achieves much better throughput due to improved collision resolution scheme and spatial reuse (Tables 2 and 3). Figure 13 shows fairness index of the proposed mechanism, Luo et al. [7] and TBCP [12] in a co-located scenario. Luo et al. [7] experience higher amount of collision and thereby throughput degradation for higher amount of flows in the network. As a consequence, its Table 2 Short-term fairness Time Number of received packets Proposed mechanism IEEE Luo et al. [7] Flow1 Flow2 Flow3 Flow4 Flow1 Flow2 Flow3 Flow4 Flow1 Flow2 Flow3 Flow Table 3 Long-term Fairness Time Number of received packets Proposed Mechanism IEEE Luo et al. [7] Flow1 Flow2 Flow3 Flow4 Flow1 Flow2 Flow3 Flow4 Flow1 Flow2 Flow3 Flow

13 Wireless Netw (2010) 16: Fig. 13 Comparison of fairness index in co-located scenario Fig. 14 Number of collisions experienced varying contention window and flows in a co-located scenario fairness index reduces with the increase of number of flows. TBCP [12] implements the notion of scheduling described in [17] and determines the packet transmission order based on slots, queue size and the service-tag of the packet. Therefore, in a co-located scenario, i.e., in one cluster TBCP [12] achieves fairness index of almost 1. In our proposed scheme, flow rank based probabilistic access mechanism ensures the priority of transmission for served and un-served flows and thereby achieving a fairness index of almost 1. Figure 14 compares proposed scheme with existing mechanisms in terms of number of collisions experienced. Since the individual flows have different rank value and hence the contention window, our proposed scheme will experience less collision than IEEE Luo et al. [7] used flow information, i.e., service-tag and centralized Fig. 15 Aggregated throughput comparison under increasing number of flows in co-located scenario scheduler for the transmission of packets. In case of newly joined flow and partially backlogged flow, it performs scheduling with inadequate flow information and experiences huge collisions. Chen et al. [14] modifies the collision resolution scheme of IEEE by increasing the contention window linearly rather than exponentially. Such contention window adjustment works better with the low network load as the collision probability is less. However, the amount of collision increases with the increased number of flows in the network. Yassine et al. [15] works under different channel rates and contention window varies with the transmission rates. In our simulation, we have considered a fixed channel rate of 2 Mbps, which forces all nodes of [15] to have same contention window value. As a consequence, for higher number of flows, the protocol experience more collision than IEEE Excepting the proposed mechanism and Chen et al. scheme [14] all previous mechanisms showed higher collisions in our simulation results. We have also executed simulations for varying CWmin values for our proposed mechanism and as the graph depicts CWmin =16 yields minimum amount of collisions. TBCP [12] uses TDMA based shared channel access mechanism and hence we exclude its comparison in terms of number of collisions. Aggregated throughput under increasing number of flows is depicted in Fig. 15. Our proposed mechanism experiences less collision and backoff assignment procedure reduces the number of empty slots in the system. As a result proposed mechanism achieves relatively higher throughput than IEEE802.11, Luo et al. [7] method and Yassine et al. [15]. Chen et al. [14] experiences less number of collisions for reduced number of flows in the network. Therefore, it performs better than our proposed scheme when the number of flows is smaller in the network. In [15] the same contention

14 726 Wireless Netw (2010) 16: Fig. 16 Throughput of 6 flows in a general topology Fig. 17 Throughput of 8 flows in a general topology Fig. 18 Comparison of fairness index in general topology thereby results degradation of fairness property in multihop scenario. However, our proposed mechanism maintains stable information of the flow-table and tries to assign equal channel access among the flows in a round robin fashion. Therefore, proposed mechanism achieves almost equal throughput for all the flows in the network. Figure 18 shows the comparison of fairness index in general topology. Luo et al. [7] and TBCP [12] extends the centralized scheduling mechanism for multi-hop flows. There are possibilities of instable flow information due to hidden nodes. Also, the spatial reuse cannot be maximized due to partially overlapped region among the flows. Our proposed mechanism ensures flow-table instability due to presence of hidden nodes and partially overlapped region. As a consequence, in case of multi-hop flows in a general topology our proposed mechanism achieves better fairness index than others. window value for all the nodes contributes much collision and the number of collision increases with the number of flows in the network. Hence, throughput achieved by Yassine et al. is less than that achieved by IEEE for higher number of flows in the network. As, TBCP [12] uses TDMA based shared channel access mechanism it achieves better throughput than other protocols in a co-located scenario for backlogged environment. Figures 16 and 17 show the fairness comparison of our proposed mechanism with other schemes in a general topology. Note that, throughput decreases drastically in the multi-hop scenario. Medium access of different flows in IEEE varies significantly and throughput achieved by different flows varies a lot. This variation increases when the number of flow is increased in the network. Luo et al. [7] experiences flow information instability and Fig. 19 Number of collisions experienced per second varying contention window and flows in general topology

15 Wireless Netw (2010) 16: Fig. 20 Number of collisions experienced per second for increasing number of hidden flows The amount of collision per second in general topology is shown in Fig. 19. It is much higher than co-located scenario due to the existence of hidden terminals and partially overlapped flows. Our mechanism tries to keep the flow-table stability in presence of hidden terminals and partially overlapped flows, hence, experienced much less collision than IEEE and Luo et al. [7]. Figure 20 shows the amount of collisions experienced in presence of varying hidden terminals. flow-table instability in Luo et al. [7] is a major drawback and flow-table instability occurs due to RTS collisions in the presence of hidden terminals. Under higher number of hidden flows, Luo et al. [7] experienced much higher collisions than IEEE The reason is collision resolution method is absent and after a collision a flow is scheduled with the same contention window value. On the other hand, in IEEE contention window is doubled when it experience a collision. This in turn reduces the chance of further collisions. Our proposed scheme addressed the issue of hidden terminals which schedules hidden flows by adding extra 19 slots and thereby ensuring reduced collision due to hidden terminals. Finally we show the throughput comparison of IEEE , Luo et al. [7] and TBCP [12] with our proposed fair scheduling mechanism for multi-hop environment in Fig. 21. Our proposed mechanism achieves better throughput even under higher number of flows in the network. The reason behind this is, we are keeping flow-table stability thereby reducing collisions and allowing spatial reuse. The non-contending flows are transmitting simultaneously resulting higher throughput. While in IEEE and Luo et al. [7] the amount of collision is very high in multi-hop scenario resulting huge throughput degradation. In case of general topology where multi-hop flows exists TBCP s [12] throughput is constrained by the bottleneck cluster. Therefore, it achieves higher throughput in the low congested area and performs better where nodes have mobility. However, in our static simulation scenario our proposed mechanism performs better than TBCP [12]. 7 Conclusions The paper presents a fair scheduling mechanism in ad hoc network environment. We have modified the basic backoff assignment technique to reduce the collision probability among the contending flows while maintaining the prioritized medium access for those flows. This also ensures a weighted fair medium access control mechanism based on probabilistic round robin scheduling. Rank is assigned based on the service-tag found in the flow-table and updated after each transmission. Therefore, existing flows in the network get the access of the medium in a round robin fashion. A major difficulty with the proposed protocol is to maintain the flow-table stability under different network scenarios. We identified the flow-table instability problem for a simple co-located scenario and then for general (multi-hop) topology. We have presented each of the network scenarios where flow-table could be unstable and corresponding solution for that. Acknowledgements This Research was supported by the MKE under the ITRC support program supervised by the (IITA (C )) and Korea Science and Engineering Foundation (KOSEF) grant funded by the Korea government (MEST) (No. R ). References Fig. 21 Aggregated throughput comparison under increasing number of flows in general topology 1. Li, Z., Nandi, S., & Gupta, A. K. (2006). Modelling short-term unfairness of IEEE in presence of hidden terminal.

16 728 Wireless Netw (2010) 16: International Journal on Performance Evaluation, 63(4), Demers, A., Keshav, S., & Shenker, S. (1989). Analysis and simulation of a fair queueing algorithm. Proceedings on communications architectures and protocols, ACM SIGCOMM, pp Lu, S., Bharghavan, V., & Srikant, R. (1999). Fair scheduling in wireless packet networks. IEEE/ACM Transaction On Networking, 7(4), Ng, T. S., Stoica, I., & Zhang, H. (1998). Packet fair queueing algorithms for wireless networks with location-dependent errors. Proceedings of IEEE INFOCOM, San Francisco, CA, pp Ramanathan, P., & Agrawal, P. (1998). Adapting packet fair queueing algorithms to wireless networks. Proceedings of ACM MobiCom, pp Luo, H., & Lu, S. (2001). A self-coordinating approach to distributed fair queueing in ad hoc wireless networks. IEEE INFOCOM, Anchorage, AK, pp Luo, H., & Lu, S. (2005). A topology-independent fair queueing model in ad hoc wireless networks. IEEE Journal on Selected Areas in Communications, 23,(3), Chao, H. L., & Liao, W. (2002). Credit-based fair scheduling in wireless ad hoc networks. Proceedings of IEEE vehicular technology conference, Vol. 3, pp Cheng, J., & Lu, S. (2003). Achieving delay and throughput decoupling in distributed fair queueing over ad hoc networks. Proceedings of IEEE ICCCN, pp Chao, H. L., & Liao, W. (2004). Fair scheduling with QoS support in wireless ad hoc networks. IEEE Transaction on Wireless Communication, 3(6), Alam, M. M., Rashid, M. M., & Hong, C. S. (2006). Distributed coordination and fair queueing in wireless ad hoc networks. Lecture Notes in Computer Science, 3981, Chao, H. L., & Liao, W. (2005). Fair scheduling in mobile ad hoc networks with channel errors. IEEE Transaction of Wireless Communications, 4(3), Alam, M. M., Rashid, M. M., & Hong, C. S. (2006). QoS-aware fair scheduling in wireless ad hoc networks with link errors. Lecture Notes on Computer Science, 4238, Chen, W. T. (2008). An effictive medium contention method to improve the performance of IEEE Wireless Network, 14 (6), Chetoui, Y., & Boubdallah, N. (2007). Adjustment for the IEEE contention window: An efficient bandwidth sharing scheme. Computer Communications, 30(13), Vaidya, N., Dugar, A., Gupta, S., & Bhal, P. (2005). Distributed fair scheduling in a wireless LAN. IEEE Transaction on Mobile Computing, 4(6), Goyal, P., Vin, H. M., & Chen, H. (1996). Start-time fair queueing: A scheduling algorithm for integrated service access. ACM SIGCOMM, pp Rashid, M. M., Alam, M. M., Razzaque, M. A., & Hong, C. S. (2007). Congestion avoidance and fair event detection in wireless sensor networks. IEICE Transactions on Communications, E90-B (12), Bianchi, G. (2000). Performance analysis of the IEEE distributed coordination function. IEEE Journal on Selected Areas on Communications, 18(3), The Network Simulator ns-2. html. 21. Perkins, C., Belding-Royer, E., & Das, S. (2003). Ad hoc ondemand distance vector (AODV) routing, RFC Author Biographies Md. Mamun-Or-Rashid received the B.Sc. and M.Sc. degree in Computer Science from the University of Dhaka, Bangladesh, in the year 2000 and 2002, respectively. He has received Ph.D. in February 2008 from Kyung Hee University, Korea. During , he worked as a faculty member in Department of Computer Science and Engineering, University of Dhaka, Bangladesh. His research interest is in the area of modeling, analysis and optimization of protocols and architectures for wireless sensor networks, distributed systems, etc. He is now working as postdoctoral researcher in Institute of Multimedia Technology, Kyung Hee University, South Korea. Muhammad Mahbub Alam received the B.Sc. degree in Applied Physics and Electronics and M.Sc. degree in Computer Science from the University of Dhaka, Bangladesh in 1998 and 2000, respectively. His research interests include wireless and mobile networking and performance modeling and analysis of networking systems. Now he is a PhD student in the Department of Computer Engineering, Kyung Hee University, Korea. Md. Abdul. Hamid received his B. E. in Computer & Information Engineering in 2001 from International Islamic University Malaysia (IIUM). In 2002, he joined as a lecturer in the Computer Science & Engineering Department, Asian University of Bangladesh, Dhaka. Currently he is pursuing his Ph.D. in Department of Computer Engineering at Kyung Hee University, South Korea. His research interest includes Security in Wireless Mobile Ad Hoc and Sensor Networks, especially on Secure routing and Key management in Wireless Sensor Networks.

17 Wireless Netw (2010) 16: Choong Seon Hong received his B.Sc. and M.Sc. degrees in electronic engineering from Kyung Hee University, Seoul, Korea, in 1983, 1985, respectively. In 1988 he joined KT, where he worked on Broadband Networks as a member of the technical staff. From September 1993, he joined Keio University, Japan. He received the Ph.D. degree at Keio University in March He had worked for the Telecommunications Network Lab, KT as a senior member of technical staff and as a director of the networking research team until August Since September 1999, he has been working as a professor of the School of Electronics and Information, Kyung Hee University. He has served as a Program Committee Member and an Organizing Committee Member for International conferences such as NOMS, IM, APNOMS, E2EMON, CCNC, ADSN, ICPP, DIM, WISA, BcN and TINA. His research interests include ad hoc networks, network security and network management. He is a member of IEEE, IPSJ, KIPS, KICS and KISS.