IEICE TRANS. COMMUN., VOL.E89–B, NO.4 APRIL 2006 1252 PAPER Support of Eﬃcient Route Discovery in Mobile Ad Hoc Access Networks Chun-Yen HSU†a) , Student Member, Jean-Lien C. WU†† , and Shun-Te WANG†† , Nonmembers SUMMARY The Public Wireless Local Area Network (PWLAN) is an emerging service for wireless access to the Internet. However, the service coverage of the PWLAN is limited by the deployment of access points (APs) because only those who stay near the AP can access the PWLAN. A feasible way of extending the service coverage of a PWLAN is to deploy mobile ad hoc access networks (MAHANs) so that users who are not in an AP’s radio coverage area can send their packets to the AP in a multihop manner. However, in a MAHAN, mobile nodes that intend to access the Internet have to discover routes to the AP first, which may result in considerable bandwidth cost. In this paper, we propose the Appointed BrOadcast (ABO) method to reduce the cost of route discovery in MAHANs. Using the ABO method can achieve this goal on the basis of packet overhearing. Functions that are necessary for network and data link layers to employ the ABO method are also discussed. Simulation results show that using the ABO method can significantly reduce the cost on route discoveries. Due to the widespread use of legacy IEEE 802.11 nodes, the problem of how ABO-enhanced and legacy IEEE 802.11 nodes can coexist in a MAHAN is also discussed. key words: mobile ad hoc access network (MAHAN), public wireless local area network (PWLAN), IEEE 802.11, routing, packet overhearing, Internet 1. Introduction With the evolution of wireless devices as well as wireless technologies, accessing the Internet via laptops or PDAs instead of desktop computers as we use today is expected to be an everyday occurrence in the future. The public wireless local area network (PWLAN), which enables user to access the Internet via WLANs ubiquitously or at least at some hotspots such as airports, stations, libraries or museums, emerges with this trend –. However, the service coverage of a PWLAN is limited by the deployment of access points (APs), i.e., only users that stay near an AP can connect to the AP directly. In order to extend the service coverage of PWLANs, the mobile ad hoc access network (MAHAN), as shown in Fig. 1, can be a feasible solution because of the following two reasons. The first one is that there is no need of infrastructures. In a MAHAN, each node acts as an end point as well as a router for other nodes. Those who are not in the radio coverage of the AP can rely on their Manuscript received May 18, 2005. Manuscript revised October 5, 2005. † The author is with the Department of Computer Science and Information Engineering, National Taiwan University of Science and Technology, Taipei, Taiwan, R.O.C. †† The authors are with the Department of Electronic Engineering, National Taiwan University of Science and Technology, Taipei, Taiwan, R.O.C. a) E-mail: [email protected] DOI: 10.1093/ietcom/e89–b.4.1252 Fig. 1 The mobile ad hoc access network. (MAHAN) neighbors to forward packets to the AP in a multihop manner. The second one is that using the MAHAN can mitigate the eﬀect of dead spot. This is valuable especially for the urban environment where the shadowing eﬀect could be a problem. The MAHAN is a special case of mobile ad hoc networks (MANETs)  that has two characteristics: • There is a hot node (i.e., the AP) that all traﬃcs flow through it. • This hot node is fixed and has no energy consumption concerns. While many routing protocols for MANETs also work in MAHANs, they can be further optimized –. Because of the node mobility, route failures may occur abruptly and frequently, therefore routing protocols must be adaptive. However, the limited radio bandwidth as well as the imperfect channel quality dictate the amount of routing information exchanged and make routing in MAHANs a laborious work. Routing protocols for such a dynamic network environment can be either proactive or reactive according to the maintenance strategy of routing tables. A detailed description and comparison between proactive and reactive routing protocols can be found in . In this paper, reactive routing is considered. A typical reactive routing protocol is described as follows. When a packet arrives at the network layer of a source node that has no valid route information to the destination, a route request (RREQ) message will be broadcasted. Having received the RREQ message, either the destination node or an intermediate node that has valid route information to the destination can reply a route reply (RREP) message to the source node using unicast. When the source node receives the RREP message, it can begin to forward packets along the discovered route. If the route fails because of link failures, the upstream node of the failed link will initiate a route error (RERR) message to notify the source node. The forwarding of RERR message can be either broadcast if it is used to notify multiple nodes, or uni- c 2006 The Institute of Electronics, Information and Communication Engineers Copyright HSU et al.: SUPPORT OF EFFICIENT ROUTE DISCOVERY IN MOBILE AD HOC ACCESS NETWORKS 1253 cast otherwise. In this paper, we propose the Appointed BrOadcast (ABO) method to minimize the route discovery cost. The ABO method achieves this goal on the basis of packet overhearing. Details of the ABO method are provided in Sect. 3. Because of the widespread use of legacy IEEE 802.11 nodes , the coexistence situation of both ABO-enhanced nodes and legacy IEEE 802.11 nodes in a MAHAN is also discussed. The rest of this paper is organized as follows. Section 2 provides the concepts of packet overhearing in IEEE 802.11 MAC protocol. The proposed ABO method and its performance evaluation are provided in Sects. 3 and 4, respectively. Finally we conclude this paper in Sect. 5. 2. Packet Overhearing in IEEE 802.11 MAC Protocol A route discovery process is necessary in ad hoc networks but it may involve large amount of control messages (i.e., the RREQ and consequent RREP messages) and thus introduce considerable amount of overhead. To reduce the overhead on route discovery, one approach is to allow nodes to overhear packets sent by their neighboring nodes. Routing information may be obtained in advance by analyzing the overheard packets and hence the frequency of route discoveries is reduced. It has been shown in the literature that packet overhearing can significantly reduce the number of control messages produced for route discovery , . In the Adhoc On-demand Distance Vector with Backup Routes routing protocol (AODV-BR) , backup routes are used to recover from route failures. By overhearing packets, backup routes can be constructed in the process of route discovery without producing extra control messages. Packet overhearing is helpful to reduce route discovery cost, however, in current IEEE 802.11 MAC design, frames that are not targeted to the node will be discarded. If a node needs to receive packets that are not targeted to it, it is essential to set the node’s network interface into promiscuous mode. In this mode, the MAC address filtering function that discards frames not targeted to the node is disabled. The drawback of using promiscuous mode is that, once it is set, all frames that can be correctly received will be passed to the upper layer and the node may eventually be drowned by overheard packets. We propose the ABO method as an alternative to achieve packet overhearing. The concept of the ABO method is that, packet overhearing should be performed in a controlled manner, in other words, the network layer should be able to actively select critical packets that need to be disseminated to neighbors. In this way, nodes only overhear critical packets, and then the load on processing overheard packets is reduced while the benefit of packet overhearing is retained. 3. The Proposed ABO Method In the ABO method, a special type of frame called the ABO frame, which is a modified IEEE 802.11 data frame and can http://folk.uio.no/paalee/ Fig. 2 The IEEE 802.11 frame format in ad hoc mode. Fig. 3 The ABO frame structure. coexist with legacy IEEE 802.11 data frames, is introduced. The ABO method takes advantage of the ABO frames to achieve packet overhearing. At the MAC layer, the packets that need to be disseminated to neighbors are carried by the ABO frame. Necessary functions in the logic link control as well as network layers in order to obtain maximum benefit available from the ABO method are also described in the following subsections. 3.1 MAC Layer Design The IEEE 802.11 data frame structure in ad hoc mode is shown in Fig. 2, where the duration field is used to update the network allocation vector (NAV), address 1 holds the address of the receiver, address 2 holds the address of the transmitter and address 3 holds the basic service set identification (BSSID) of the network . An ABO frame, as shown in Fig. 3, is transmitted in the way of IEEE 802.11 MAC layer broadcast. However, broadcast frames in IEEE 802.11 MAC design are acknowledge-free and thus unreliable. To keep the transmission to the target node reliable, the target node’s MAC address is recorded in the intended receiver address (IRA) appended in the frame body. Upon receiving an ABO frame, the node whose MAC address matches the IRA will reply an ACK frame to the sender. The ABO frame, as depicted in Fig. 3, therefore has the following diﬀerences compared to legacy IEEE 802.11 data frame: 1. The IRA field is appended at the end of the frame body. 2. Address 1 holds the broadcast address. 3. The duration field contains the short interframe space (SIFS) time plus the ACK frame transmission time. Note that a legacy IEEE 802.11 broadcast frame has a zero value in the duration field. Hence, when a node receives a frame, it can check whether the frame is an ABO frame or not. The node examines if the address 1 field holds a broadcast address and whether the duration field contains a nonzero value. If all positive, the frame is recognized as an ABO frame, otherwise it is recognized as a legacy IEEE 802.11 frame. Figure 4 depicts the corresponding processes when an IEEE 802.11 node or an ABO-enhanced node receives a data frame. When an ABO-enhanced node receives a data frame, the node first checks the address 1 and the duration fields to IEICE TRANS. COMMUN., VOL.E89–B, NO.4 APRIL 2006 1254 Fig. 4 Processing of received data frames. see if it is an ABO frame. If true, the following steps will be triggered, otherwise, the received frame is processed in the legacy way. Step 1: Check if the IRA field matches its own MAC address, if true, prepare the ACK frame and transmit after a SIFS time period. Step 2: Trim the IRA field. Step 3: Send the MSDU to the upper layer. In generating an ABO frame, the IRA field must be included in the frame check sequence (FCS) calculation to prevent the receiving node from mistaking the ABO frame as a frame error. Similarly, if encryption is needed, the IRA field is treated as a part of the plaintext. In the IEEE 802.11 MAC design, two functions are prohibited, namely, fragmentation and the RTS/CTS transaction, for broadcast frames. The transmissions of ABO frames also obey these rules. The no-fragment rule avoids the potential problem that an overhearing node fails to receive all fragments and then requests for the lost ones. This rule is convenient because in most systems, the maximum transmission unit (MTU) is set to be 1500 bytes, which is smaller than the frame body length limitation of 2312 bytes in the IEEE 802.11 standard. The noRTS/CTS rule potentially increases the probability of collisions, so a short ABO frame is preferred. 3.2 Logical Link Control and Network Layers The ABO method gives routing protocols the flexibility to selectively disseminate packets to neighbors. Packets that are to be disseminated to neighbors are carried by the ABO frame. This can be done by sending these packets via a specific service access point (SAP) to the logical link control (LLC) layer and then to the MAC layer as shown in Fig. 5. The way to select packets that are to be disseminated to neighbors could be a strategy. Routing protocol can determine the critical messages to be disseminated to neighboring nodes based on its own routing metrics and strategies. For example, in AODV-BR , nodes just need to overhear the Fig. 5 Protocol stack of the ABO method. RREP messages, so only RREP messages are transmitted using the ABO method while other packets are transmitted with original processes. On receiving an overheard packet, a node should first extract the routing information embedded in the packet and then discard the packet. According to the send/receive operations particularly in the ABO method, the network layer will be enhanced in two aspects in advance to take advantage of the ABO method. The first one is the awareness of the specific ABO SAP described above. The second one is that, because the next hop address in the overheard packet will not match the current node’s address, the routing protocol must be able to extract the routing information embedded in the packet rather than just trigger an error and then discard it. Some routing protocols, such as the Dynamic Source Routing (DSR) , can handle the overheard packets and benefit from it. Other protocols, e.g., AODV , however, are not designed to handle overheard packets and thus enhancement is required. In most protocol designs, the network layer, LLC functions and network interface card drivers are implemented using software, so they can be enhanced easily. HSU et al.: SUPPORT OF EFFICIENT ROUTE DISCOVERY IN MOBILE AD HOC ACCESS NETWORKS 1255 3.3 The Reply Storm Problem Many routing protocols allow intermediate nodes that have valid route information to the destination node to reply RREP messages upon receiving the RREQ message in order to reduce the number of broadcast RREQ messages. However, if the number of intermediate nodes that have a valid route to the destination node is large, a considerable number of RREP messages will be sent. Because all the RREP messages are sent to the source node, congestion may occur at or near the source node and hence the reply storm problem occurs , . The reply storm problem may appear in MAHANs because all nodes have the same destination node (i.e., the AP) rather than peer-to-peer communications as in MANETs. The reply storm problem can be even worse if all RREP messages are transmitted using the ABO method because more nodes have route information to the AP. Besides, since RREP messages are carried by the ABO frame, which can be received by all the neighboring nodes, a large number of RREP messages can introduce considerable processing load to the network layer. Therefore, schemes to eliminate the reply storm problem are desired. Two schemes can be employed at the network layer to eliminate the reply storm problem. The first one is the ABO frame prohibition (AFP) scheme, which reduces the frequency on using the ABO frame. If an intermediate node has learned a new route, it is allowed to transmit an RREP message using the ABO frame for once to disseminate the new route information. Otherwise, the use of ABO frame is prohibited. The second one is that, similar to , when an intermediate node decides to send a RREP message to the source node, the RREP message will be deferred for a time period of DRREP before sending. During the deferring period, if the node receives a RREP message heading to the same source node, the node will discard the RREP message it prepares. DRREP can be determined according to the routing protocol’s major concern. For example, in the Load-Balanced Ad hoc Routing (LBAR) , traﬃc load is the major concern, so the deferring delay can be determined according to the load along the route. In routing protocols such as ADOV and DSR, route length is the major concern. It is intuitive to assign a smaller DRREP for more preferable routes. In this way, transmission of RREP messages for minor route will have larger deferring delay and is more likely to be discarded. 3.4 Coexistence with Legacy IEEE 802.11 Nodes Because of the widespread use of the IEEE 802.11 devices, the coexistence of both the ABO-enhanced and legacy IEEE 802.11 nodes in one MAHAN must be considered. In the ABO method, the ABO frame has the same frame control values as that in IEEE 802.11 frame, so upon receiving an ABO frame, a legacy IEEE 802.11 node will take it as a normal broadcast frame. In this case, the IRA field will not be trimmed and will be passed to the upper layer as part of the Fig. 6 Handling of the legacy IEEE 802.11 and ABO frames. frame body. This is safe because the IRA field will be simply trimmed by the packet length verification at the network layer , thus the higher layer transactions is prevented from error. How legacy IEEE 802.11 and ABO-enhanced nodes handle their respective frames are depicted in Fig. 6, where the dotted lines between the NSDU and NPDU implies that they are present only at the source node and the destination node. Note that legacy IEEE 802.11 nodes transmit standard IEEE 802.11 frames only, thus the higher the proportion of ABO-enhanced nodes in a network, the better the dissemination eﬃciency. In case an ABO-enhanced node transmits an ABO frame to a legacy IEEE 802.11 node, no IEEE 802.11 ACK will be returned even if the ABO frame is received correctly. In this case, the transmitter will time out and take it as a transmission error, then enter the retransmission phase. However, the transmitter will eventually receive no ACK frame from the legacy IEEE 802.11 node even in retransmissions. This issue is solved as follows. If an ABO frame experiences a transmission error, the subsequent retransmissions of this frame should use the legacy IEEE 802.11 frame. In this way, the desired information is disseminated to neighbor nodes in the transmission of the ABO frame while the subsequent retransmissions ensure reliable transmissions to the target node. The ABO method solves the coexistence problem at the cost of more MAC layer transmissions. Because larger proportion of ABO-enhanced nodes implies more eﬃcient route information dissemination, the additional MAC layer load can be compensated by the reduction of control messages. The simulation results on the coexistence scenario presented in the following section will verify this point. 4. Simulation Study In this section, we evaluate the performance of the ABO method using simulation. 4.1 Simulation Model In the simulation, we tested two typical routing protocols based on two diﬀerent concepts: DSR  and AODV , and three diﬀerent MAC protocols: legacy IEEE 802.11, ABO and promiscuous mode. DSR is a source routing protocol that the source node knows the complete hop-by-hop IEICE TRANS. COMMUN., VOL.E89–B, NO.4 APRIL 2006 1256 route to the destination node while AODV is a destinationoriented routing protocol that each node only knows the next hop to the destination node. In DSR, the route is specified by the source node in the packet header while in AODV, the destination node address is the only guidance to which the packet should be forwarded. Both the routing protocols are enhanced with functions described in Sects. 3.2 and 3.3 to employ the ABO method and to handle overheard packet properly. Note that the AFP scheme described in Sect. 3.3 is applicable to the ABO method but is ineﬀective to the other two MAC protocols. A variation of the ABO method by releasing the AFP scheme will have similar performance as the promiscuous mode because all control messages will be overheard by neighboring nodes. The only diﬀerence between using the AFP-released ABO method and using promiscuous mode is that nodes using promiscuous mode overhear data packet while nodes using the AFP-released ABO method do not. We employ the expending ring search mechanism, which is specified in both AODV and DSR, to reduce the number of RREQ messages in the network. Unicast control messages, including RREP and RERR messages, are selected to transmit using the ABO method. The deferring time DRREP is based on the following function proposed in : DRREP = H × (h − 1 + r) where H is a small constant delay, h denotes the number of hops along the route being replied, and r is a random number uniformly distributed between 0 and 1. It is suggested that H is at least twice the maximum wireless link propagation delay . We evaluated the eﬀects of diﬀerent H values as shown in Fig. 8 and find that when H is 10 ms, the control cost is the lowest. Similar simulation configurations are used as described in , where the channel bit rate is 2 Mb/s and the radio range is 250 meters. An AP is placed at the center of the simulation area where 50 mobile nodes are placed randomly. Three scenarios are studied: 300 × 1500 m2 , 600 × 1500 m2 and 1000 × 1000 m2 . We find that similar results can be drawn and here we only show the results using the first scenario. The data packet size is 512 bytes and the inter-arrival time of packets is 0.4 second. The simulation time is 900 seconds. The random waypoint model  is used for the mobility model. Each mobile node is independent and no group movement is presented. Mobile nodes are stationary (i.e., no mobility is presented) for a period of time t pause at the beginning of simulation. t pause is exponentially distributed with a mean pause time. Larger mean pause time means more stationary nodes and lower topology-changing rate. When t pause times out, the node will move to a randomly selected new location with a moving speed uniformly distributed with a range of (0, 20) m/s. When the node reaches the new location, it picks a new t pause value and remains stationary before t pause times out. The pause time ranges from 0 to 900 seconds. Note that a zero pause time implies that mobile nodes are always moving and the net- work topology is highly dynamic, while pause time of 900 seconds implies that all nodes are stationary during simulation. For fair comparisons, the mobility and traﬃc patterns are the same among the compared protocols. The presented simulation results come from a mean of 50 runs. 4.2 Performance Comparisons Table 1 summarizes performance metrics that are used for comparisons. Since the AP is assumed to have no energy concerns, the cost spent by the AP to send or receive packets and frames is not included in the simulation results. In this way, we emphasize on the costs incurred in mobile nodes. Power consumption in memory access is not considered since it is insignificant compared with that of CPU operations and wireless transmissions . We use the model developed in  to measure the CPU time required at the network layer. Pradhan and Chiueh  investigated CPU time required for four packet lengths: 100, 192, 500 and 1000 bytes. We supplement the CPU time required for other packet lengths by linear regression as depicted in Fig. 7. The model developed by Feeney , where the corresponding parameters are given in Table 2, is used to measure the enTable 1 The IEEE 802.11 frame format in ad hoc mode. Performance metrics Description Normalized network The number of control packets (i.e., the layer control cost RREQ, RREP and RERR packets, but not data packets) sent or received at the network layer per data packet delivered to the AP. Network layer cost The total number of packets (including data and network layer control packets) sent and received at the network layer. MAC layer cost The number of frames (including data and MAC layer control frames) sent and received at the MAC layer. Normalized CPU time Required CPU time at the network layer per data packet delivered to the AP. Normalized energy conEnergy consumed at the MAC layer per sumption data packet delivered to the AP. Packet delivery ratio The number of data packets received by the AP to the number of packets generated by all traﬃc sources. Average end-to-end data The average end-to-end delay of data packet delay packets. Fig. 7 CPU time required with respect to packet length. HSU et al.: SUPPORT OF EFFICIENT ROUTE DISCOVERY IN MOBILE AD HOC ACCESS NETWORKS 1257 Table 2 Parameter btranctl brecvctl btran mtran brecv mrecv bdiscard mdiscard brecv prom mrecv prom Parameters for energy consumption computation . Description Fixed energy consumption on transmitting a control frames. Fixed energy consumption on receiving a control frames. Fixed energy consumption on transmitting a data frame. Incremental energy consumption on transmitting a data frame. Fixed energy consumption on receiving a data frame. Incremental energy consumption on receiving a data frame. Fixed energy consumption on discarding a frame. Incremental energy consumption on discarding a frame. Fixed energy consumption on receiving a data frame in promiscuous mode. Incremental energy consumption on receiving a data frame in promiscuous mode. Value 120 mW·s 29 mW·s 246 mW·s 1.89 mW·s/byte 56.1 mW·s 0.494 mW·s/byte 97.2 mW·s -0.49 mW·s/byte Fig. 8 Diﬀerent deferring delay. (pause time=300 s, CBR sources=25, MAC=ABO) 136 mW·s 0.388 mW·s/byte ergy consumption at the MAC layer† : At the transmitter, btranctl + brecvctl + mtran × size + btran + brecvctl . At the destination node of the data frame, Fig. 9 brecvctl + btranctl + mrecv × size + brecv + btranctl . Packet delivery ratio. (pause time=300 s) At non-destination node that discards the frame, bdiscardctl + bdiscardctl n∈T + n∈D (mdiscard × size + bdiscard ) + n∈T bdiscardctl . n∈D At non-destination node that operating in promiscuous mode, bdiscardctl + bdiscardctl n∈T + n∈T n∈D (mrecv prom × size + brecv prom ) + bdiscardctl , n∈D where T and D denote respectively the set of neighboring node of the transmitting and destination nodes. Figure 8 shows the normalized network layer control cost using diﬀerent H values. H=0 implies that nodes do not defer before replying RREP message back to the source node. An underestimated H value cannot prevent the network from the reply storm problem. On the contrary, an overestimated H value may result in that the source node route discovery times out, and then route discovery is initiated again, which introduces more number of control messages. It is interesting that AODV and DSR have similar performance while they are diﬀerent in routing concepts. The power of DSR is that nodes can learn routes to other Fig. 10 Average end-to-end data packet delay. (pause time=300 s) nodes from the packet header for future use. This is useful in peer-to-peer communications scenes . However, in MAHANs, all source nodes aim at the same destination node, i.e., the AP, learning routes to nodes other than the AP does not bring extra benefit. In the following, only the result of AODV is presented. The tags 802.11, ABO and Prom in Figs. 9 to 16 indicate the corresponding results of using legacy IEEE 802.11 † The measurement of power consumption is based on an assumption that an ABO-enhanced network interface card consumes equivalent power compared with a legacy one. IEICE TRANS. COMMUN., VOL.E89–B, NO.4 APRIL 2006 1258 MAC, the ABO method and promiscuous mode, respectively. The delivery ratio and end-to-end delay subject to diﬀerent number of CBR sources are depicted in Figs. 9 and 10, respectively. Promiscuous mode can achieve the highest performance in terms of delivery ratio and end-to-end delay because that it consumes the least bandwidth on routing control, thus more bandwidth is left for data transmissions and consequently the performance is improved. Using the ABO method can also achieve remarkable performance im- Fig. 11 Packet delivery ratio. (CBR sources=25) Fig. 13 provement compared with that of legacy IEEE 802.11 MAC. However, using the ABO method does not give as much improvement as that of using promiscuous mode because the AFP scheme causes the eﬃciency degradation on disseminating route information. Similar results can be observed when using diﬀerent pause time as shown in Figs. 11 and 12, respectively. Figure 13 compares the number of control and data packets sent and received using the three MAC proto- Fig. 12 Average end-to-end data packet delay. (CBR sources=25) Network layer costs. (CBR sources=25) HSU et al.: SUPPORT OF EFFICIENT ROUTE DISCOVERY IN MOBILE AD HOC ACCESS NETWORKS 1259 cols. Note that even when all nodes are stationary (pause time=900 s), a small amount of RERR messages are issued. This is because, sometimes, the MAC layer transmits a frame but receives no ACK because of collisions. The MAC layer will retry until an ACK is received or the retry limit is reached. If the latter occurs, the MAC layer will signal the network layer that the wireless link is not available. The network layer then recognizes this as a route failure and initiates an RERR message to the corresponding source nodes. In Fig. 13(a), compared to legacy IEEE 802.11 MAC, using the ABO method can significantly reduce the number of control packets. Using promiscuous mode can obtain even more improvements. As shown in Fig. 13(b), using promiscuous mode results in a large number of control packets arrived at the network layer since the MAC layer does not filter out frames. On the contrary, the network layer receives the least number of control packets when the ABO method is employed, because when the ABO method is used, only those packets that are carried by ABO frames (unicast RREP and RERR messages in this case) can be overheard by the network layer. Note that the AFP scheme limits the frequency on using the ABO frame, so when the ABO method is used, the number of control packets arrived at the network layer is less than that when promiscuous mode is used. On the other hand, compared with legacy IEEE 802.11 MAC, the ABO method issues 50% less control packets as shown in Fig. 13(a). The reduction of control packets can compensate the increased number of received control packets caused by the use of the ABO frames. The number of control packets arrived at the network layer of the ABO method is less than that of the legacy IEEE 802.11 MAC. On the number of data packets sent/received at the network layer, although using promiscuous mode results in the least cost on sending data packets as shown in Fig. 13(c), a large number of data packets are received at the network layer—about 20 times as many as that of using the legacy IEEE 802.11 MAC and the ABO method as depicted in Fig. 13(d). This is inevitable because using promiscuous mode, the MAC layer receives data frames and passes all Fig. 15 of them to the upper layer without filtering. The result indicates that nodes using promiscuous mode have to spend much more computation power to deal with the overheard packets, as shown in Fig. 14. Figure 15 depicts the simulation results on the number of frames transmitted/received at the MAC layer. The number of frames transmitted is directly related to the number of packets sent at the network layer, and the number of frames received is directly related to the number of frames transmitted. Because promiscuous mode has the least number of control packets issued, it introduces the least number of frames transmitted/received at the MAC layer. However, less number of transmitted/received frames does not always imply lower energy consumption. It is reported that, for a typical wireless LAN card such as the Lucent WaveLAN IEEE 802.11 PC card, receiving a frame is more expensive than discarding it because the network interface card can enter a reduced energy consumption mode during the frame transmission that will be discarded by the node . The normalized energy consumption of the three MAC protocols is depicted in Fig. 16, where energy consumption is further classified according to 1) overheard frames in promiscuous mode (Overhear), 2) accepted frames that received by the node and then the MSDU is sent to the upper layer (Accept) and 3) discarded frames that are received and then discarded MAC layer costs. (CBR sources=25) Fig. 14 Normalized CPU time. IEICE TRANS. COMMUN., VOL.E89–B, NO.4 APRIL 2006 1260 by the node (Discard). In promiscuous mode, data frames are not discarded and thus more energy is consumed compared with the other two MAC protocols. Figures 17 to 21 show the simulation results of a MAHAN that consists of both ABO-enhanced and legacy IEEE 802.11 nodes. The pause time is 300 seconds and the number of CBR sources is 25. CBR sources are evenly distributed among all mobile nodes. At the network layer, Fig. 16 Normalized energy consumption. (CBR sources=25) Fig. 18 performance metrics including packet delivery ratio, endto-end delay, number of control/data packets sent/received and CPU time have better improvement when more ABOenhanced nodes are present, as shown in Figs. 17 to 19. Note that the MAC layer cost as well as the energy consumption show a minor improvement when 20% of mobile nodes are ABO-enhanced, as shown in Figs. 20 and 21, because that the ABO method solves the coexistence problem at the Fig. 17 Delivery ratio and delay. (coexist, pause time=300 s, CBR sources=25) Network layer costs. (coexist, pause time=300 s, CBR sources=25) HSU et al.: SUPPORT OF EFFICIENT ROUTE DISCOVERY IN MOBILE AD HOC ACCESS NETWORKS 1261 cost of introducing more MAC layer transmissions. The reduction of network layer control packets can compensate the increased MAC layer load, however, when the proportion of ABO-enhanced nodes is small, the compensation effect is not obvious. As a result, the improvement on MAC layer cost is minor. For larger proportion of ABO-enhanced nodes, the eﬀect of MAC layer cost compensation is more apparent. 5. Conclusion Reactive routing protocols typically rely on broadcast-based route discovery messages (i.e., RREQ) to discover routes. When the network is highly dynamic, route failures would occur frequently and introduce considerable costs on routing control. Packet overhearing can help reduce the number of control packets and leave more bandwidth for data transmissions. Promiscuous mode is generally used to achieve packet overhearing. In promiscuous mode, the MAC layer transfers all received packets to the upper layer without filtering. Although it can achieve the highest delivery ratio and the lowest end-to-end delay while producing minimal number of control messages, using promiscuous mode will make the network interface card always in active mode and results in large amount of packets arrived at the network layer. Thus the node is forced to consume more energy and take Fig. 19 Normalized CPU time. (coexist, pause time=300 s, CBR sources=25) Fig. 20 more CPU time on processing overheard packets. Consider that wireless devices are usually equipped with lower level CPU and are powered by batteries, using promiscuous mode may cause disinclined performance degradation at the system level. In this paper, we propose the ABO method as an alternative to achieve packet overhearing. In the ABO method, packet overhearing is performed in a controlled manner that routing protocols can determine which packet to be disseminated to neighbors. The determination of disseminated packets is dependent on the routing protocol’s strategy. We have shown that system and network performance, including CPU time, energy consumption, packet delivery ratio and end-to-end delay, can be significantly improved when only the route reply (RREP) and route error (RERR) messages are disseminated using the ABO method. Besides, considering the widespread used legacy IEEE 802.11 standard, the coexistence problem of the ABO-enhanced and legacy IEEE 802.11 nodes is also discussed. A solution to the coexistence problem is provided and evaluated in this paper. Acknowledgments The authors acknowledge the very useful feedback provided during discussions with the members of Computer Network Laboratory, NTUST. This work was supported by National Science Council under Contract Nos. NSC 93-2213-E-011007 and NSC 94-2213-E-011-009. Fig. 21 Normalized energy consumption. (coexist, pause time=300 s, 25 CBR sources) MAC layer costs. (coexist, pause time=300 s, 25 CBR sources) IEICE TRANS. COMMUN., VOL.E89–B, NO.4 APRIL 2006 1262 References  A. Acharya, C. Bisdikian, A. Misra, and Y.-B. Ko, “ts-PWLAN: A value-add system for providing tiered wireless services in public hot-spots,” Proc. IEEE ICC 2003, vol.1, pp.193–197, Anchorage, Alaska, USA, May 2003.  http://www.ericsson.com/products/hp/PWLAN pa.shtml  http://www.hpintelco.net/portfolio/pwlan hotspots.htm  http://www.gii.co.jp/press/ba11884 en.shtml  Mobile Ad-hoc Networks (manet) Charter, http://www.ietf.org/html. charters/manet-charter.html  H. Li and D. Yu, “Comparison of ad hoc and centralized multihop routing,” Proc. WPMC 2002, vol.2, pp.791–795, Honolulu, Hawaii, Oct. 2002.  Y. Sun and E.M. Belding-Royer, “Application-oriented routing in hybrid wireless networks,” Proc. IEEE ICC 2003, vol.1, pp.502– 506, Anchorage, Alaska, USA, May 2003.  M. Benzaid, P. Minet, K.A. Agha, C. Adjih, and G. Allard, “Integration of mobile-IP and OLSR for a universal mobility,” Wireless Networks, vol.10, no.4, pp.377–388, July 2004.  S.H. Bae, S. Lee, and M. Gerla, “Unicast performance analysis of extended ODMRP in a wired-to-wireless ad-hoc network testbed,” Proc. IEEE MILCOM 2002, vol.2, pp.1228–1232, Anaheim, California, USA, Oct. 2002.  P.E. Engelstad, A. Tonnesen, A. Hafslund, and G. Egeland, “Internet connectivity for multi-homed proactive ad hoc networks,” Proc. IEEE ICC 2004, pp.4050–4056, Paris, France, June 2004.  Y.-C. Tseng, C.-C. Shen, and W.-T. Chen, “Integrating mobile IP with ad hoc networks,” Computer, vol.36, no.5, pp.48–55, May 2003.  J.-H. Song, V.W.S. Wong, and V.C.M. Leung, “Eﬃcient on-demand routing for mobile ad hoc wireless access networks,” IEEE J. Sel. Areas Commun., vol.22, no.7, pp.1374–1383, Sept. 2004.  M. Abolhasan, T. Wysocki, and E. Dutkiewicz, “A review of routing protocols for mobile ad hoc networks,” Ad Hoc Networks, vol.2, no.1, pp.1–22, Jan. 2004.  “IEEE 802.11 local and metropolitan area networks: Wireless LAN medium access control (MAC) and physical (PHY) specifications,” ISO/IEC 8802-11: 1999(E).  T. Camp, J. Boleng, B. Williams, L. Wilcox, and W. Navidi, “Performance comparison of two location based routing protocols,” Proc. IEEE INFOCOM 2002, vol.3, pp.1678–1687, New York, USA, June 2002.  L.M. Feeney, “An energy consumption model for performance analysis of routing protocols for mobile ad hoc networks,” Mobile Networks and Applications, vol.6, no.3, pp.239–249, June 2001.  S.-J. Lee and M. Gerla, “AODV-BR: Backup routing in ad hoc networks,” Proc. IEEE WCNC 2000, vol.3, pp.1311–1316, Chicago, USA, Sept. 2000.  IETF draft, http://www.ietf.org/internet-drafts/draft-ietf-manet-dsr10.txt  IETF RFC 3561, http://www.ietf.org/rfc/rfc3561.txt  M.K. Marina and S.R. Das, “Impact of caching and MAC overheads on routing performance in ad hoc networks,” Comput. Commun., vol.27, no.3, pp.239–252, Feb. 2004.  H. Hassanein and A. Zhou, “Routing with load balancing in wireless ad hoc networks,” Proc. ACM International Workshop on Modeling Analysis and Simulation of Wireless and Mobile Systems 2001, pp.89–96, Rome, Italy, July 2001.  G.R. Wright and W.R. Stevens, TCP/IP Illustrated, Volume II: The Implementation, p.216, Addison Wesley, 2003.  D.B. Johnson and D.A. Maltz, “Dynamic source routing in ad hoc wireless networks,” Mobile Comput., pp.153–181, Kluwer Academic Publishers, Boston, USA, 1996.  J.R. Lorch and A.J. Smith, “Software strategies for portable computer energy management,” IEEE Pers. Commun., vol.5, no.3, pp.60–73, June 1998.  P. Pradhan and T.-C. Chiueh, “Real-time performance guarantees over wired/wireless LANs,” Proc. IEEE RTAS 1998, pp.29–38, Denver, Colorado, USA, June 1998. Chun-Yen Hsu received the B.S. and M.S. degrees from National Taiwan Ocean University in 1997 and 1999, respectively. He is working toward the Ph.D. degree in the Department of Computer Science and Information Engineering at the National Taiwan University of Science and Technology. His research interests include ad hoc networks, wireless LAN and wireless MAN. Jean-Lien C. Wu received the BSEE from National Taiwan University, Taiwan, in 1970, and the Ph.D. degree in EE from Cornell University, Ithaca, NY, in 1976. From 1978 to 1984, she was Member of the Technical Staﬀ, Bell Laboratories, Holmdel, NJ. Since 1984, she has been a professor in the Electronic Engineering Department, National Taiwan Univ. of Science and Technology (NTUST), Taipei, Taiwan. She was Head of the department from 1987 to 1990, Dean of the College of Electrical and Computer Engineering from 1998 to 2001 and Dean of the Academic Aﬀairs from 2003 to 2005. She is currently Vice President of the university. Dr. Wu is the coordinator of the Communication Engineering Program of the Engineering and Applied Sciences Dept. of the National Science Council (NSC), Taiwan. She is also a member of the advisory board of the Science and Technology Advisory Oﬃce of the Ministry of Education, Taiwan. Her current research interests are in wireless networks, broadband Internet, and mobile multimedia networks. Dr. Wu is a member of the IEEE communication society. She is also a member of the Chinese Institute of Engineerings society and the Chinese Institute of Electrical Engineerings society. Shun-Te Wang received his B.E. degree in mechanical engineering from National Taiwan University of Science and Technology, Taipei, Taiwan, in 1996. He received the M.S. degree in engineering science from National Cheng Kung University, Tainan, Taiwan, in 1999. He has been a software engineer at Institute of Information Science, Academia Sinica, Taiwan, since September 1999. Currently he is working toward a Ph.D. in the Department of Electronic Engineering at National Taiwan University of Science and Technology. His main areas of interest include pervasive computing, sensor networks, ad hoc networks, wireless networks, mobile computing, communication protocols, component software, software warehouse, authoring system, and Java collaborative computing.