Congestion Aware Routing in Hybrid Wireless Mesh ... - CiteSeerX

Download PDF
3 downloads 132 Views 309KB Size Report
Comment
Abstract—Multi-radio Wireless Mesh Networks (WMN) are gaining lot of ..... It then computes the new CDCAi value for each wireless interface operating channel ...
Congestion Aware Routing in Hybrid Wireless Mesh Networks Asad Amir Pirzada and Ryan Wishart

Marius Portmann†

Research Laboratory, National ICT Australia Limited, Brisbane, QLD 4000, Australia. Email: {Asad.Pirzada, Ryan.Wishart}@nicta.com.au

School of Information Technology and Electrical Engineering, The University of Queensland, Australia, Brisbane, QLD 4072, Australia. Email: [email protected]

† Queensland

Abstract— Multi-radio Wireless Mesh Networks (WMN) are gaining lot of popularity owing to their increased application to community and public safety networks. WMNs form a static wireless backhaul to provide connectivity to mobile clients. The wireless medium, being shared and contended for, creates a number of hurdles including congestion, interference and noise. Multiradio nodes can take advantage of the wider frequency spectrum. However, current mesh technologies employ a simplistic approach by assigning one channel for client servicing and another for the backhaul network. The improper reuse of the same channel across multiple hops causes extensive co-channel interference leading to lower bandwidth. The problem is aggravated in a hybrid WMN where the mobile clients act as routers for other clients. In this paper, we propose a congestion aware routing protocol, which can successfully establish channel diverse routes through least congested areas of a hybrid WMN. The prime advantage of the protocol is its ability to discover optimal routes in a distributed manner without the requirement of an omniscient network entity. Simulation results show that the congestion aware routing protocol can successfully achieve a high packet delivery ratio with lower routing overhead and latency in a hybrid WMN.

Keywords: Multi-radio, routing, mesh, wireless, network I. I NTRODUCTION A WMN is formed with the help of two distinct types of nodes1 i.e. Mesh Routers and Mesh Clients. The Mesh Routers, with multi-radio transceivers and access to external power sources, form the multi-hop wireless backhaul network. This network is used by the Mesh Clients to communicate among each other and also to gain access to an external network through a gateway. The Mesh Routers are generally static and act as general packet forwarders, while the Mesh Clients portray a disparate mobility pattern and only communicate through the Mesh Routers. A Hybrid WMN is formed when, in addition to the Mesh Routers, the Mesh Clients also act as packet forwarders and assist in establishing the backhaul network [1]. Thus a Mesh Client in a Hybrid WMN episodically performs the role of a Mesh Router by executing a routing protocol. A Hybrid WMN is the most versatile form of autonomic network and depicts self-configuring, self-healing and self-optimising characteristics. Owing to the peculiar characteristics of Hybrid 1 In rest of the paper, the word ‘node’ will be used interchangeably for both Mesh Clients and Mesh Routers.

WMNs, these networks are considered a promising technology for Public Safety and Disaster Recovery communications. A typical Hybrid WMN is shown in Fig. 1. Routing protocols assure connectivity between the ClientRouter and Router-Router pairs. A number of approaches have been proposed for providing communication support in a Hybrid WMN. These approaches can be broadly categorised into three types i.e. Pro-active, Reactive and Hybrid [2]. In the pro-active approach, routes between node pairs are computed and maintained on a periodic basis by sharing link tables or distance vectors. The advantage of this approach is that all nodes have instantaneous access to other nodes in the network. The obvious disadvantage is the persistent overhead due to route management and the resulting limited scalability. The reactive approach is to find a route between nodes only when required. In this case, two or more communicating entities form a route on-the-fly. The route is maintained for the duration of the communication and obliterated upon culmination of the communication. The advantage of the reactive approach is the low routing overhead at the cost of increased route discovery latency [3].

       

 

 

 

 

 

Fig. 1.

A typical Hybrid Wireless Mesh Network

The hybrid approach generally employs proactive routing in the static portion of the network and reactive routing in the mobile portion of the network. It is essential to clarify here that the pro-active and reactive routing approaches are equally applicable to a hybrid WMN as is the hybrid routing approach. A Hybrid WMN is faced with a number of challenges including role-switching; mobility and meagre battery resources of Mesh Clients; shared wireless medium leading to cochannel and inter-channel interference; resilient noise in the ISM bands etc. Mesh Clients are typically resource constrained mobile devices and can perform limited functionality of a Mesh Router in a hybrid WMN. A Mesh Client may undergo frequent power-saving periods and thus cannot be used exclusively as a Mesh Router for sustaining long term data connections. As the 802.11 wireless medium is shared, dense node networks can cause extensive contention for the medium leading to lower data rates and high latency. Simultaneous transmissions on the same channel can cause collisions due to the hidden-node problem. In case virtual carrier sensing is not used (RTS/CTS), this problem becomes fairly obvious in chain topologies (Fig. 2), where a packet being sent from Node A to Node B interferes with a packet being sent from Node C to Node D due to the proximity of Node B to both Node A and Node C. This is also called intra-flow co-channel interference, which causes the multi-hop performance degradation dilemma.

A

C1

B Fig. 2.

C

C1

D

Intra-flow co-channel interference

Similarly, co-channel interference can occur between two or more distinct parallel or intersecting flows operating a common channel. For example two crisscrossing flows (A-BC and D-B-E), operating a common channel, interfere with each other while transmitting a single data packet on any hop (Fig. 3). This type of interference is called inter-flow co-channel interference, which also causes severe contention for the wireless medium resulting in packet losses and delays. Interference can also occur between different channels depending upon the physical wireless medium. For example the IEEE 802.11b standard supports simultaneous operation of only three orthogonal channels. In addition the shared use of the ISM bands and the ambient noise present in the wireless medium causes additional interference and poor Signal to Noise Ratios (SNR). Thus, the above-mentioned challenges make provisioning of reliable broadband access across multiple hops in a hybrid WMN both intricate and exigent. Contemporary routing protocols for WMNs make use of special routing metrics in order to counter one or more of these challenges. Most of these routing metrics target community mesh networks, which are to a large extent static with predictable traffic patterns. In addition, these metrics are developed for networks with nodes

D C1

A

C1

B

C1

C

C1

E Fig. 3.

Inter-flow co-channel interference

having a homogeneous configuration. Consequently, if these metrics are applied directly to hybrid WMNs, they are likely to cause sub-optimal utilisation of the available communication resources and imbalanced load distribution. In this paper we present a routing protocol that permits establishment of high performance routes in a hybrid WMN. The protocol is a derivative of the well known AODV routing protocol and, hence, inherits its core self-configuring and self healing properties. The protocol is very scalable and supports high mobility. The routing protocol makes use of an interference and congestion aware routing metric, which permits discovery of optimal routes in the network. The simulation results indicate a significant improvement over the multi-radio AODV in terms of packet delivery ratio, routing overhead and latency. In rest of the paper, we first explain some related work in Section II. The proposed protocol is then explained in Section III. Simulation environment and results are discussed in Section IV followed by conclusions in Section V. II. R ELATED W ORK A number of routing protocols have been developed for WMNs in recent years. All protocols use different routing metrics to establish optimal routes. The four main characteristics of any good routing metric are path length, link capacity, packet loss ratio and interference [4]. All four characteristics have direct impact on the latency and packet delivery ratio of the network. Standard AODV uses the hop-count routing metric and endeavours to minimise the path lengths. However, it does not take into consideration the other three characteristics. As AODV processes Route Request packets on a ‘first come fist serve basis’, the hop-count metric is often ignored in congestion scenarios. Therefore hop-count, without taking into account the other routing metric characteristics, is not a good path cost representative.

Another commonly used metric is the Expected Transmission Count (ETX) [5], which is the expected number of MAC layer transmissions required to successfully deliver a packet. The metric is generally computed using periodic probe packets that are broadcast every second. The ratio of successfully received probes from a neighbour gives the reverse link delivery ratio (dr ). Similarly, the ratio of successfully received probes by that neighbour gives the forward link delivery ratio (df ). The product dr ×df gives the probability of a data packet being successfully transmitted in a single attempt. The inverse of this probability gives the ETX value for that link. ET X =

1 df × dr

The main problem with ETX is the assumption that probe packets depict similar losses as the data packets. However, probe packets are actually very small in size and are broadcast at the lowest data rate. These packets thus have a better ETX count as compared to data packets. Another problem with the metric is that is does not take the link data rate into account. Thus a slow link with lower ETX is considered superior to a faster link with a higher ETX. The Expected Transmission Time (ETT) [6] is a derivative of the ETX metric and measures the expected time needed to successfully transmit a fixed-sized packet on a link, and is defined as follows: SZ ET T = ET X × BW where SZ is the packet size and BW represents the bandwidth of the link. By introducing the link bandwidth, two links with the same ETX value can be graded accurately. The disadvantage of ETT is that is does not take into consideration the effect of intra-flow interference. Thus simply minimising the cumulative ETT of a path does not guarantee channel diversity of that path. The Weighted Cumulative ETT (WCETT) [6] reduces intraflow interference by introducing a parameter Xj , which is the sum of ETT of all links operating the most used channel j. X W CET T = (1 − α) × ET T + α × M ax Xj where α is a tunable parameter varying from 0 to 1 and determines the weight given to intra-flow channel interference in the computation of the aggregate ETT of a path. The prime disadvantage of WCETT is that it penalises for channel reuse, without considering the spatial separation between the reused channels. Any path which reuses a channel every three hops is considered in par with a path in which a channel is reused in consecutive hops. The former path may not introduce any intra-flow interference because of the spatial separation of the channel reuse, while the latter path may induce excessive channel interference due to the proximity of channel reuse. Similarly, there are some other variants [4][7] of the WCETT routing metric, which try to minimise intra-flow and intra-flow interference. However, compared with WCETT, these metrics offer a marginal performance improvement.

In the following section we present a routing protocol, which uses the channel diversity and congestion levels as input to its routing metric. The metric has been formulated in such a way that it ensures diversified channel selection to minimise intra-flow interference. To minimise inter-flow interference, the routing metric makes use of the local congestion information during interface selection. III. C ONGESTION AWARE A D - HOC O N - DEMAND D ISTANCE V ECTOR ROUTING P ROTOCOL The Congestion Aware Ad-hoc On-demand Distance Vector (AODV-CA) routing protocol is a variant of the standard AODV protocol with an improved routing metric and support for multi-radio nodes. In order to explain the working of AODV-CA, we first briefly explain the standard AODV protocol in this section. AODV is a reactive distance vector routing protocol that has been optimised for mobile ad-hoc wireless networks [8]. AODV borrows basic route establishment and maintenance mechanisms from the Dynamic Source Routing (DSR) protocol [9], and hop-to-hop routing vectors from the DestinationSequenced Distance-Vector (DSDV) routing protocol [10]. Multi-path support can also be added to AODV using a number of extensions that provide loop-free and disjoint alternate paths [11]. To avoid the problem of routing loops, AODV makes extensive use of sequence numbers in control packets. When a source node intends to communicate with a destination node whose route is not known, it broadcasts a Route Request packet. Each Route Request contains an ID; source and destination node IP addresses and sequence numbers together with a hop-count and control flags. The ID field uniquely identifies the Route Request. The sequence numbers indicate the freshness of control packets, and the hop-count maintains the number of nodes between the source and the destination. Each recipient of the Route Request that has not seen the Source IP and Route Request ID pair, or does not maintain a fresher (with larger sequence number) route to the destination, rebroadcasts the same packet after incrementing the hop-count. Such intermediate nodes also create a reverse route to the source node for a certain interval of time. When the Route Request reaches the destination node, or any node that has a fresher route to the destination, a Route Reply packet is generated and unicast back to the source of the Route Request. Each Route Reply contains the destination sequence number, the source and the destination IP addresses, route lifetime together with a hop-count and control flags. Each intermediate node that receives the Route Reply, increments the hop-count, establishes a forward route to the source of the packet and transmits the packet on the reverse route. In case a link break is detected for a next hop of an active route, a Route Error packet is sent to its active neighbours that were using that particular route. The standard AODV protocol was originally developed for single-radio nodes in wireless ad-hoc networks. However, later on the protocol was modified to support nodes with multiple

radios. We refer the standard AODV protocol with support for multiple radios as AODV-MR in this paper. When using AODV-MR, each Route Request is broadcast on all the node’s interfaces. Intermediate nodes with one or more interfaces operating a common channel, receive the Route Request and create a reverse route that points towards the source node. If the Route Request is a duplicate, it is simply discarded. The first Route Request received by the destination, or any intermediary node, is selected and all subsequent Route Requests are discarded. The Route Reply is generated in response to the selected Route Request, and is sent back to the source node via the existing reverse route. The routing metric used by AODV-CA is called Channel Diverse Congestion Aware (CDCA) metric. CDCA is made up of two parts: CD and CG. CD represents the level of channel diversity present in a route while CG represents the congestion level of a route. We assume that all interfaces are tuned to fixed channels and are not changed during operation. Prior research shows that in a chain topology a channel can be reused after a gap of two or more hops [12]. If Ci represents a channel being used on hop i, the following condition must hold to avoid intra-flow co-channel interference: Ci ∈ / {Ci−1 , Ci−2 , Ci+1 , Ci+2 }

(1)

This is to ensure that a data packet sent by Node C on channel Ci is not interfered by an acknowledgement sent from Node B to Node A on channel Ci−2 as shown in Fig 4.

interface queues (IFQ) of contending nodes. The congestion in turn causes building up of the IFQ lengths. Similarly, if there is extensive noise in the wireless medium, the transmitted data packets may need to be retransmitted a number of times before being successfully acknowledged by the recipient node. A low SNR also causes elongation of the IFQs. Intra-flow and inter-flow interference essentially have the same effect on an IFQ. This impact is directly related to the incoming and outgoing rate of wireless frames and causes the IFQ to grow and shrink accordingly. Thus if due to intra-flow interference, the link capacity of channel Ci gets reduced by n, then the IFQ also builds up by an impact factor of k. The impact factor is primarily dependent upon the current flow rate and may be adjusted locally at each node sustaining an ongoing flow. If we assume that the current IFQ length of a channel Ci is represented by Qi , then for a channel bandwidth of BWi , the Queue Discharge Interval (QDIi ) of Ci is given by: Qi BWi The QDIi represents the time a packet has to remain in the IFQ before being transmitted on to the physical medium. By normalising Qi , we ensure that the QDI of different nodes with varying channel bandwidths are comparable. The cumulative congestion CGi at any link on channel Ci can be computed as follows: QDIi =

CGi =

i−2 X

QDIn

n=i

A

Ci-2 QDIi-2

B Fig. 4.

Ci-1 QDIi-1

C

Ci QDIi

D

Intra-Flow Interference

Thus to create a channel diverse chain of i hops, we need i number of channels for i ≤ 2 and a minimum of 3 orthogonal channels for i > 2. The channel diversity achieved using condition (1) is thus true only for a single unidirectional flow. For full-duplex operation and multiple co-located or intersecting flows, additional orthogonal channels are required to achieve optimal channel diversity. A big constraint on using the above approach is the number of simultaneous flows that a node can effectively support without introducing co-channel interference. For example a two-radio intermediary node with both radios tuned to orthogonal channels can only sustain a single flow in one direction at any one time. The IEEE 802.11 radios are half-duplex, thus to sustain a full duplex flow through a single node a minimum of four radios tuned to orthogonal channels are required. As the Mesh Clients in Hybrid WMN are constantly on the move, it is very likely that a number of simultaneous flows will be traversing a Mesh Router. As the wireless medium is shared, contention takes place when multiple nodes try to access it. This contention causes congestion in the transmit

If we assume that the flow rate is equal to the BWi and the impact factor k is equal to n number of simultaneous interfering flows, then the channel diversity CDi at any link operating channel Ci can be computed as follows: ¯ ¯ ¯ 1 if Ci 6= Ci−1 AND Ci 6= Ci−2 ¯¯ ¯ ¯ 2 if Ci = Ci−1 AND Ci 6= Ci−2 ¯¯ CDi = ¯¯ 2 if Ci 6= Ci−1 AND Ci = Ci−2 ¯¯ ¯ ¯ 3 if Ci = Ci−1 AND Ci = Ci−2 ¯ CDi = 1 means that there is no channel reuse across the last three hops of a single flow. CDi = 2 implies that the same channel is being used on two of the last three hops of a flow. While, CDi = 3 represents that the same channel is being used on all three hops of a flow. As discussed earlier, the QDI is proportional to the congestion in the network. We compute the CDCAi at any link operating channel Ci as follows: CDCAi = CDi · CGi The routing metric CDCAi for the path A-B-C-D, as shown in Fig. 4, is computed as follows: ¯ ¯ ¯ 1(QDIi + QDIi−1 + QDIi−2 ) if CDi = 1 ¯ ¯ ¯ ¯ 2(QDIi + QDIi−1 + QDIi−2 ) if CDi = 2 ¯ ¯ ¯ CDCAi = ¯ ¯ ¯ 2(QDIi + QDIi−2 + QDIi−1 ) if CDi = 2 ¯ ¯ 3(QDIi + QDIi−1 + QDIi−2 ) if CDi = 3 ¯

The CDCAi value shows that for the case where there is maximum channel diversity (CDi = 1), the cost of the route is equal to CGi . However, when there are two or more interfering links (CDi > 1) on the last three hops of a flow, the CGi is proportionally increased to the number of interfering links. To compute CDCAi a node needs the QDI and channel information of the preceding two nodes. We convey this information in the Route Request packets as four bytes. In addition each node also conveys the cumulative CDCA value as two bytes in the Route Request packet. Each node before forwarding a Route Request, first extracts this information. It then computes the new CDCAi value for each wireless interface operating channel Ci . Finally, it updates the cumulative CDCA, QDI and channel information fields in the Route Request packet. All nodes maintain a minimum cumulative CDCA (minCDCA) value along with each routing entry in the routing table. A Mesh Router sets the min-CDCA to the value received in the first Route Request (from the same source and having the same ID). All subsequent copies of the Route Request are forwarded only if their cumulative CDCA value is lower than the min-CDCA. If the value is lower, the current min-CDCA is replaced by the lower one. This ensures that the Route Request with the maximum channel diversity and least congestion is always forwarded and used for route creation. In worst case scenarios, it is possible that multiple copies of the same Route Request (with decreasing cumulative CDCA values) are received by a Mesh Router. Thus we will have additional Route Requests propagating in the network. However, we have found through simulations that the optimal Route Request (with least cumulative CDCA) is generally received earlier than those with higher cumulative CDCA values. This is due to the fact that the optimal Route Request traverses paths with maximum channel diversity and least loaded interface queues. IV. S IMULATION R ESULTS AND A NALYSIS A. Simulation Environment We have evaluated the efficiency of the AODV-CA protocol through extensive simulations in NS-2 [13], using the Extended Network Simulator (ENS) extensions [14]. A WMN covering an area of 1 square km is established using 25 static Mesh Routers, which are distributed in a uniform 5x5 grid and each equipped with six 802.11b radios. The network further consists of 50 mobile Mesh Clients, each equipped with a single radio and placed randomly in the simulation area. Concurrent UDP flows are established between 30 randomly selected source and destination Mesh Client pairs. The performance metrics are obtained by averaging the results from over 50 simulation runs. The simulation parameters are listed in Table I. B. Performance Metrics The simulations provide the following three performance metrics:

TABLE I TABLE 1: S IMULATION PARAMETERS Examined Protocol Simulation time Simulation area Propagation Model Mobility model for Mesh Clients Speed of Mesh Clients Transmission range Traffic Type No. of Flows Flow Rate

AODV-MR and AODV-CA 900 seconds 1000 x 1000 m Two-ray Ground Reflection Random waypoint 0, 5, 10, 15 & 20 m/s 250 m CBR (UDP) 30 128 kbps

Packet Delivery Ratio (PDR): The ratio between the number of data packets successfully received by destination nodes and the total number of data packets sent by source nodes. Routing Packet Overhead: The ratio of control packets generated to successfully received data packets. Average Latency: The mean time (in seconds) taken by the data packets to reach their respective destinations. C. Results and Analysis The simulation results under varying Mesh Client speeds are shown in Fig. 5. During route establishment, AODV-CA focuses on using minimally loaded channels for routing data traffic. This helps to sustain the 30 simultaneous 128 kbps connections. However, AODV-MR forms routes over multiple hops by randomly selecting the available channels [15]. Thus, a route may comprise of a large number of overlapping and saturated channels, resulting in severe packet losses. The lower packet losses incurred by AODV-CA enable it to achieve a significantly higher PDR. The PDR of AODV-CA drops from 67% to almost 58% when the maximum Mesh Client speed increases from 0 to 20 m/s. The PDR of AODV-MR drops from 55% to 43% for a similar increase in Mesh Client speeds. However, owing to its effective congestion aware routing, AODV-CA maintains a consistent PDR at 58% with increasing Mesh Client speeds. The multi-homed nodes, which execute the AODV-MR protocol, form routes using a random interface rather than selecting an optimal interface. As a result, links frequently get saturated and suffer from interference. This essentially causes the routes to sever, thereby, causing new route discoveries. These route discoveries increase the routing overhead of AODV-MR to more than ten control packets for each received data packet. On the other hand, AODV-CA selects optimal channels during the route establishment phase. This enables the routes to be more stable, which minimises the need for extraneous route discoveries. As the routes created using AODV-CA have been optimised for minimal interference using orthogonal channels, packets are sent promptly without incurring excessive contention for the physical medium. This has the effect of reduced average latency for the data packets traversing an average of four hops. The packets flowing on routes established using AODVMR face severe contention for the physical medium and are, therefore, significantly delayed at each hop. AODV-MR

65 60 55 50 45

11

0.4

10

0.35

Average Latency (seconds)

AODV−MR AODV−CA

Routing Packet Overhead

Packet Delivery Ratio (%)

70

9 8 7 6 5 4

40

0

5

10

15

20

2

0

5

Maximum Speed

10

15

Maximum Speed

Fig. 5.

0.25 AODV−MR AODV−CA

0.2 0.15 0.1 0.05

AODV−MR AODV−CA

3

0.3

20

0

0

5

10

15

20

Maximum Speed

Simulation Results

shows an average latency of more than 300 ms under varying Mesh Client speeds. The large delay is primarily due to the presence of extensive intra-flow and inter-flow interference in the network. As the routes established using AODV-CA have been optimised for maximum channel diversity and minimal interference, we observe significantly reduced latency in the order of 30 ms under all mobility conditions. V. C ONCLUSIONS Hybrid Wireless Mesh Networks, which present the most versatile form of mesh technology, enable mobile Mesh Clients to connect to a high speed wireless backhaul network formed using static Mesh Routers. A major advantage of the Hybrid Mesh Network is its ability to support the backhaul using Mesh Clients in addition to the Mesh Routers. A common problem observed in these networks is the performance degradation over multiple wireless hops. This occurrence is generally caused due to intra-flow and inter-flow co-channel interference. Although the Mesh Routers are equipped with multiple radios tuned to orthogonal channels, minimal effort is made to achieve channel diversification on a per flow basis. This in turn induces extensive contention for the physical medium causing significant packet losses. In this paper, we have proposed a routing protocol with a channel diverse and congestion aware metric. This metric assures channel diversity on a per flow basis. In addition, it utilises the local congestion information in scenarios where optimal channel diversity cannot be assured. By integrating channel diversification and congestion information, the metric is able to successfully trace out optimal routes in a mobile network. The simulation results indicate that the routing protocol is able to achieve a significantly high packet delivery rate with extremely low latency in a Hybrid Wireless Mesh Network. ACKNOWLEDGEMENTS National ICT Australia is funded by the Australian Government’s Department of Communications, Information Technology, and the Arts and the Australian Research Council through Backing Australia’s Ability and the ICT Research Centre of Excellence programs and the Queensland Government.

R EFERENCES [1] I. F. Akyildiz and X. Wang, “A Survey on Wireless Mesh Networks,” IEEE Communications Magazine, vol. 43, no. 9, pp. S23–S30, 2005. [2] E. M. Royer and C. K. Toh, “A Review of Current Routing Protocols for Ad hoc Mobile Wireless Networks,” IEEE Personal Communications Magazine, vol. 6, no. 2, pp. 46–55, 1999. [3] A. A. Pirzada, C. McDonald, and A. Datta, “Performance Comparison of Trust-Based Reactive Routing Protocols,” IEEE Transactions on Mobile Computing, vol. 5, no. 6, pp. 695–710, 2006. [4] Y. Yang, J. Wang, and R. Kravets, “Designing Routing Metrics for Mesh Networks,” in Proceedings of the IEEE Workshop on Wireless Mesh Networks (WiMesh). IEEE Press, 2005. [5] D. S. J. Couto, D. Aguayo, J. Bicket, and R. Morris, “A high-throughput path metric for multi-hop wireless routing,” Wireless Networks, vol. 11, no. 4, pp. 419–434, 2005. [6] R. Draves, J. Padhye, and B. Zill, “Routing in Multi-Radio, MultiHop Wireless Mesh Networks,” in Proceedings of the 10th Annual International Conference on Mobile Computing and Networking. ACM Press, 2004, pp. 114–128. [7] A. P. Subramanian, M. M. Buddhikot, and S. Miller, “Interference aware routing in multi-radio wireless mesh networks,” in 2nd IEEE Workshop on Wireless Mesh Networks, 2006, pp. 55–63. [8] J. Broch, D. A. Maltz, D. B. Johnson, Y. C. Hu, and J. Jetcheva, “A Performance Comparison of Multi-hop Wireless Ad hoc Network Routing Protocols,” in Proceedings of the 4th Annual International Conference on Mobile Computing and Networking (MobiCom). ACM Press, 1998, pp. 85–97. [9] D. B. Johnson, D. A. Maltz, and Y. Hu, “The Dynamic Source Routing Protocol for Mobile Ad hoc Networks (DSR),” IETF MANET, Internet Draft, 2003. [10] C. E. Perkins and P. Bhagwat, “Highly Dynamic Destination-Sequenced Distance-Vector Routing (DSDV) for Mobile Computers,” in Proceedings of the SIGCOMM Conference on Communications, Architectures, Protocols and Applications. ACM Press, 1994, pp. 234–244. [11] A. A. Pirzada, M. Portmann, and J. Indulska, “Performance Comparison of Multi-Path AODV and DSR Protocols in Hybrid Mesh Networks,” in Proceedings of the 14th IEEE International Conference on Networks (ICON), vol. 2, 2006, pp. 587–592. [12] P. Gupta and P. R. Kumar, “The Capacity of Wireless Networks,” IEEE Transactions on Information Theory,, vol. 46, no. 2, pp. 388–404, 2000. [13] NS, “The Network Simulator,” http://www.isi.edu/nsnam/ns/, 1989. [14] B. Raman and C. Chebrolu, “Design and Evaluation of a new MAC Protocol for Long-Distance 802.11 Mesh Networks,” in Proceedings of the 11th Annual International Conference on Mobile Computing and Networking (MobiCom). ACM Press, 2005, pp. 156–169. [15] A. A. Pirzada, M. Portmann, and J. Indulska, “Evaluation of MultiRadio Extensions to AODV for Wireless Mesh Networks,” in Proceedings of the 4th ACM International Workshop on Mobility Management and Wireless Access (MobiWac), 2006, pp. 45–51.