Efficient load aware routing scheme

  • 313 views
Uploaded on

 

More in: Education , Technology
  • Full Name Full Name Comment goes here.
    Are you sure you want to
    Your message goes here
    Be the first to comment
No Downloads

Views

Total Views
313
On Slideshare
0
From Embeds
0
Number of Embeds
0

Actions

Shares
Downloads
0
Comments
0
Likes
1

Embeds 0

No embeds

Report content

Flagged as inappropriate Flag as inappropriate
Flag as inappropriate

Select your reason for flagging this presentation as inappropriate.

Cancel
    No notes for slide

Transcript

  • 1. IEEE TRANSACTIONS ON MOBILE COMPUTING, VOL. 9, NO. 9, SEPTEMBER 2010 1293 Efficient Load-Aware Routing Scheme for Wireless Mesh NetworksKae Won Choi, Wha Sook Jeon, Senior Member, IEEE, and Dong Geun Jeong, Senior Member, IEEE Abstract—This paper proposes a load-aware routing scheme for wireless mesh networks (WMNs). In a WMN, the traffic load tends to be unevenly distributed over the network. In this situation, the load-aware routing scheme can balance the load, and consequently, enhance the overall network capacity. We design a routing scheme which maximizes the utility, i.e., the degree of user satisfaction, by using the dual decomposition method. The structure of this method makes it possible to implement the proposed routing scheme in a fully distributed way. With the proposed scheme, a WMN is divided into multiple clusters for load control. A cluster head estimates traffic load in its cluster. As the estimated load gets higher, the cluster head increases the routing metrics of the routes passing through the cluster. Based on the routing metrics, user traffic takes a detour to avoid overloaded areas, and as a result, the WMN achieves global load balancing. We present the numerical results showing that the proposed scheme effectively balances the traffic load and outperforms the routing algorithm using the expected transmission time (ETT) as a routing metric. Index Terms—Wireless mesh network, load-aware routing, utility, dual decomposition. Ç1 INTRODUCTIONA wireless mesh network (WMN) consists of a number of wireless routers, which do not only operate as hostsbut also forward packets on behalf of other routers. WMNs over the traditional hop-count routing metric, they neglect the problem of traffic load imbalance in the WMN. In the WMN, a great portion of users intends tohave many advantages over conventional wired networks, communicate with outside networks via the wired gate-such as low installation cost, wide coverage, and robust- ways. In such environment, the wireless links around theness, etc. Because of these advantages, WMNs have been gateways are likely to be a bottleneck of the network. If therapidly penetrating into the market with various applica- routing algorithm does not take account of the traffic load,tions, for example, public Internet access, intelligent some gateways may be overloaded while the others maytransportation system (ITS), and public safety [1]. One of not. This load imbalance can be resolved by introducing athe main research issues related to WMNs is to develop the load-aware routing scheme that adopts the routing metricrouting algorithm optimized for the WMN. with load factor. When the load-aware routing algorithm is designed to maximize the system capacity, the major benefit In mobile ad-hoc networks, the primary concern of of the load-aware routing is the enhancement of the overallrouting has been robustness to high mobility. However, system capacity due to the use of underutilized paths.nodes in the WMN are generally quasi-static in their location. Although there have been some works on load-awareThus, the focus of the routing studies in the WMN has moved routing for mobile ad-hoc networks (e.g., [9], [10]) andto performance enhancement by using sophisticated routing WMNs (e.g., [11], [12], [13]), they simply include some loadmetrics [2], [3], [4], [5], [6], [7], [8]. For example, as the routing factors in the routing metric without consideration of themetrics, researchers have proposed the expected transmis- system-wide performance.sion number (ETX) [2], the expected transmission time (ETT) In this paper, we propose a load-aware routing scheme,and weighted cumulative ETT (WCETT) [5], the metric of which maximizes the total utility of the users in the WMN.interference and channel switching (MIC) [6], and the The utility is a value which quantifies how satisfied a user ismodified expected number of transmissions (mETX) and with the network. Since the degree of user satisfaction depends on the network performance, the utility can beeffective number of transmissions (ENTs) [8]. Although these given as a function of the user throughput. Generally, themetrics have shown significant performance improvement utility function is concave to reflect the law of diminishing marginal utility. To design the scheme, we use the dual. K.W. Choi is with the Department of Electrical and Computer Engineering, decomposition method for utility maximization [14], [16]. University of Manitoba, Room E2-390 EITC Building, 75A Chancellor’s Using this method, we can incorporate not only the load- Circle, Winnipeg, MB R3T 5V6, Canada. E-mail: kaewon.choi@gmail.com. aware routing scheme but also congestion control and fair. W.S. Jeon is with the School of Electrical Engineering and Computer rate allocation mechanisms into the WMN. Most notably, Science, Seoul National University, Gwanak-gu, Seoul 151-742, Korea. E-mail: wsjeon@snu.ac.kr. we can implement the load-aware routing scheme in a. D.G. Jeong is with the School of Electronics and Information Engineering, distributed way owing to the structure of the dual Hankuk University of Foreign Studies, Cheoin-gu, Yongin-si, Gyonggi-do decomposition method. 449-791, Korea. E-mail: dgjeong@hufs.ac.kr. In the proposed routing scheme, a WMN is divided intoManuscript received 7 Apr. 2008; revised 22 Oct. 2008; accepted 15 Dec. multiple overlapping clusters. A cluster head takes role of2009; published online 28 Apr. 2010. controlling the traffic load on the wireless links in its cluster.For information on obtaining reprints of this article, please send e-mail to:tmc@computer.org, and reference IEEECS Log Number TMC-2008-04-0129. The cluster head periodically estimates the total traffic loadDigital Object Identifier no. 10.1109/TMC.2010.85. on the cluster and increases the “link costs” of the links in the 1536-1233/10/$26.00 ß 2010 IEEE Published by the IEEE CS, CASS, ComSoc, IES, & SPS
  • 2. 1294 IEEE TRANSACTIONS ON MOBILE COMPUTING, VOL. 9, NO. 9, SEPTEMBER 2010cluster, if the estimated load is too high. In this scheme, each (ROMER) algorithm in [20] also uses opportunistic for-user chooses the route that has the minimum sum of the link warding to deal with short-term link quality variation. Thecosts on it. Thus, a user can circumvent overloaded areas in ROMER maintains the long-term routes and opportunisti-the network, and therefore, the network-wide load balance cally expands or shrinks them at runtime.can be achieved. The works in [21] and [22] focus on the applications The major advantages of the proposed load-aware accessing the wired gateways. The ad hoc on-demandrouting scheme can be summarized as follows: distance vector spanning tree (AODV-ST) in [21] is an adaptation of the AODV protocol to the WMN with the . Designed by the dual decomposition method, the wired gateways. The AODV-ST constructs a spanning tree of proposed load-aware routing scheme maximizes the which the root is the gateway. In [22], the authors propose a system-wide performance. routing and channel assignment algorithm for the multi- . The proposed scheme is scalable, has low control channel WMN. In this algorithm, a spanning tree is formed in and computation overheads, and can be easily such a way that a node attaches itself to the parent node. implemented by means of the existing ad hoc The load-aware routing protocols [9], [10], [11], [12], [13] routing protocols [17]. incorporate the load factor into their routing metrics. The The remainder of the paper is organized as follows: In dynamic load-aware routing (DLAR) in [9] takes as theSection 2, we briefly overview the related works. Section 3 routing metric the number of packets queued in the nodeoutlines the system model. In Section 4, we formulate the interface. The load-balanced ad hoc routing (LBAR) in [10]optimization problem and solve it by using the dual counts the number of active paths on a node and its neighbors,decomposition method. In Section 5, we explain how to and uses it as a routing metric. Both the DLAR and LBAR areimplement the proposed routing scheme in a distributed designed for the mobile ad hoc network, and aim to reduceway. Section 6 presents the numerical results which show the packet delay and the packet loss ratio. In [11], anthe performance of the proposed scheme. Finally, the paper admission control and load balancing algorithm is proposedis concluded with Section 7. for the 802.11 mesh networks. In this work, the available radio time (ART) is calculated for each node, and the route with the2 RELATED WORKS largest ART is selected when a new connection is requested. This algorithm tries to maximize the average number ofFor the WMN, a number of routing metrics and algorithms connections. In [12], the authors propose the WCETT loadhave been proposed to take advantage of the stationary balancing (WCETT-LB) metric. The WCETT-LB is thetopology. The first routing metric is the ETX [2], which is theexpected number of transmissions required to deliver a WCETT augmented by the load factor consisting of thepacket to the neighbor. In [3], the authors propose the average queue length and the degree of traffic concentration.minimum loss (ML) metric which is used to find the route The QoS-aware routing algorithm with congestion controlwith lowest end-to-end loss probability. In [4], the medium and load balancing (QRCCLB) in [13] calculates the numbertime metric (MTM) is proposed for the multirate network. of congested nodes on each route and chooses the route withThe MTM of a link is inverse proportional to the physical the smallest number of congested nodes.layer transmission rate of the link. The ETT in [5] is a Compared to these load-aware routing protocols, the proposed routing scheme has three major advantages. First,combination of the ETX and the MTM. The ETT is a required the proposed scheme is design to maximize the systemtime to transmit a single packet over a link in the multirate capacity by considering all necessary elements for loadnetwork, calculated in consideration of both the number of balancing, e.g., the interference between flows, the linktransmissions and the physical layer transmission rate. The capacity, and the user demand, etc. On the other hand, theauthors in [5] also suggest the routing metric and algorithm existing protocols fail to reflect these elements since they usefor the multiradio WMN, which are the WCETT and the heuristically designed routing metrics. For example, themultiradio link quality source routing (MR-LQSR), respec- DLAR, the ART, and the WCETT-LB do not take account oftively. The WCETT is a modification of the ETT to consider the interference between flows. Also, the link capacity is notthe intraflow interference. While the WCETT only considers considered by the DLAR, the LBAR, the ART, and thethe intraflow interference, the MIC [6] and the interference QRCCLB. Second, the proposed scheme can guaranteeaware (iAWARE) [7] take account of the interflow inter- fairness between users. When the network load is high, it isference as well as the intraflow interference. of importance for users to fairly share scarce radio resources. In [8], the mETX and the ENT are proposed to cope with However, the existing protocols cannot fairly allocatethe fast link quality variation. These routing metrics contain resources, since they are unable to distinguish which routethe standard deviation of the link quality in addition to the is monopolized by a small number of users. Third, theaverage link quality. The blacklist-aided forwarding (BAF) proposed scheme can provide routes stable over time. Sincealgorithm in [18] is proposed to tackle the problem of short- most of the existing protocols adopt highly variable routingterm link quality degradation by disseminating the black- metrics such as the queue length or the collision probability,list, i.e., a set of currently degraded links. The ExOR they are prone to suffer from the route flapping problem.algorithm in [19] decides the next hop after the transmis- We design the proposed routing scheme by using thesion for that hop without predetermined routes. The ExOR dual decomposition method for the network utility max-can choose the next hop that successfully received the imization. A brief introduction to this method can be foundpacket, and therefore, it is robust to packet error and link in [14], [15], and an elaborate explanation and a number ofquality variation. The resilient opportunistic mesh routing examples can be found in [16]. To use this method, one
  • 3. CHOI ET AL.: EFFICIENT LOAD-AWARE ROUTING SCHEME FOR WIRELESS MESH NETWORKS 1295 TABLE 1 Table of SymbolsFig. 1. Example mesh network.should formulate the global optimization problem that is to links in the network, respectively. In Table 1, we summarizemaximize the total system utility under the constraints on all mathematical notations introduced in this section.the traffic flows and the radio resources. After the The WMN under consideration provides a connection-constraints are relaxed by the Lagrange multipliers, the oriented service, where connections are managed in the unitwhole problem can be decomposed into the subproblems of a flow. A flow is also unidirectional. A user canwhich are solved by the different network layers in the communicate with the other user or the gateway node afterdifferent network nodes. In the decomposed problem, the setting up a flow connecting them. Since a user is connectedLagrange multipliers act as a interface between the layers to a unique node, the flow between a pair of users can alsoand the nodes, enabling the distributed entities to find the be specified by the corresponding node pair. The nodeglobal optimal solution only by solving their own subpro- where a flow starts (ends) will be called the sourceblems. Therefore, the dual decomposition method provides (destination) node of the flow. Fig. 1 shows an examplea systematical way to design a distributed algorithm which scenario where a user intends to send data to outsidefinds the global optimal solution. networks. As seen in this figure, if a flow conveys data to (from) outside networks, all gateway nodes can be the destination (source) node of the flow. We will identify a3 SYSTEM MODEL flow by an index, generally f, and define F as the set of the3.1 Mesh Network Structure indices of all flows in the network. Data traffic on a flow is conveyed to the destination nodeEach wireless router in a WMN is fixed at a location. Thus, through a multihop route. We only consider acyclic routes.the WMN topology does not change frequently and the Thus, a route can be determined by the set of all intermediatechannel quality is quasi-static. In addition, each wireless links that the route takes. We will index a route by r androuter serves so many subscribers (i.e., users) in general that define Dr as the set of the indices of all intermediate links onthe characteristic of the aggregated traffic is stable over time. the route r. For a flow, there can be a number of possibleTherefore, we design the routing scheme under the system routes that connect the source and destination nodes. Fig. 1model of which topology and user configuration are stable. shows some of the possible routes that a flow can take to In Fig. 1, we illustrate an example of the WMN. In this send data to the outside networks. Let Gf denote the set offigure, a node stands for a wireless router, which not only the indices of all possible routes for flow f.delivers data for its own users, but also relays data traffic for For mathematical development, we assume that a flow canother wireless routers. Among nodes, there are some utilize multiple routes simultaneously by dividing its datagateway nodes connected to the wired backhaul network. traffic into these routes. We limit the possible data rate of theEach user is associated with its serving node. In this paper, traffic conveyed by a flow on each route to control the amountwe do not deal with the interface between a user and its of traffic injected to the WMN. Let f;r denote the “flow dataserving node to focus on the mesh network itself. Through rate” which is defined as the maximum data rate at which thethe serving node, a user can send (receive) data traffic to flow f can send data traffic on the route r. We also define(from) the other user in the WMN or to (from) outside f :¼ ðf;r Þr2Gf as the “flow data rate vector” of flow f. Thenetworks via the gateway nodes. If node n can transmit data sum of all the components in a flow data rate vector is limitedto node m directly (i.e., without relaying), there exists a link to the “maximumP flow data rate,” denoted by max . That is, itfrom the node n to the node m. In this paper, we define a link should hold that r2Gf f;r max .as unidirectional. For the mathematical representation, we We will call f the “multipath flow data rate vector,” ifdefine N and L as the sets of the indices of all nodes and all f;r > 0 for more than one r in Gf . On the other hand, we
  • 4. 1296 IEEE TRANSACTIONS ON MOBILE COMPUTING, VOL. 9, NO. 9, SEPTEMBER 2010will refer to f as the “single-path flow data rate vector,” if transmission. If two links are adjacent enough to interferef; > 0 for only one “active route” and f;r ¼ 0 for the with each other, packets cannot be conveyed through theother rs in Gf . Since the multipath routing is hard to be two links at the same time. To incorporate this restrictionimplemented in a practical sense, we will focus on finding into the proposed scheme, we divide the WMN intothe single-path flow data rate vectors for all flows in the multiple overlapping clusters. A cluster includes the linksWMN. Let f denote the active route of the flow f that has adjacent enough to interfere with each other. Therefore, anythe single-path flow data rate vector. In case that all flows pair of links in the same cluster cannot deliver packets simultaneously. A cluster is generally indexed by c, and lethave the single-path flow data rate vector, we can let :¼ " C be the set of the indices of all clusters in the WMN. Weðf Þf2F denote the “active route vector.” also define Mc as the set of all links in the cluster c. Deciding a single-path flow data rate vector is equivalent The proposed scheme estimates the traffic load in eachto deciding an active route for the flow and the flow data cluster. The traffic load in a cluster is the sum of the trafficrate on the active route. An application using the flow can load on the links in the cluster. If the traffic load in a cluster issend user data through the active route at its flow data rate. estimated to be too high, the proposed scheme can redirectThe active route can be decided in such a way that the the routes passing through the overloaded cluster for loadglobal load balancing is accomplished. In addition, network balancing. The airtime ratio of a link represents the trafficcongestion can be controlled and fairness can be guaranteed load on the link. If the sum of the airtime ratios of the links inby deciding the flow data rate properly. Therefore, the load- a cluster exceeds a certain bound, the cluster can be regardedaware routing, congestion, and fairness problems can be as overloaded.solved at the same time, if we find a way to calculate Roughly, we assume that a fixed portion of the time canappropriate flow data rate vectors. be used for data transmission, while the remainder is used for the purpose of control, e.g., control message exchange3.2 Physical and Medium Access Control Layer and random back-off. Let
  • 5. denote the ratio of the time for Model data transmission to the whole time. Since only a link canThe proposed scheme can be implemented on top of convey data traffic at a time within a cluster, the sum of thevarious physical (PHY) and medium access control airtime ratios of the links in a cluster cannot exceed
  • 6. .(MAC) layer protocols that utilize a limited bandwidth Therefore, we have the following constraint:and divide the time for multiple access, for example, such Xas the carrier sense multiple access/collision avoidance al
  • 7. ; for all c 2 C: ð2Þ(CSMA/CA), the time division multiple access (TDMA), l2Mcand the reservation ALOHA (R-ALOHA). In Fig. 1, we give an example organization of clusters. The effective transmission rate of a link is defined as the Note that we do not draw all clusters to avoid over-number of actually transmitted bits divided by the time crowding. In this figure, four clusters are presented, each ofspent for data transmission, calculated in consideration of which is indicated by a dashed circle. Suppose that a clusterretransmissions due to errors. That is, the effective transmis- includes all incoming and outgoing links of the nodes in thesion rate can be calculated as the PHY layer transmission dashed circle. In this example, the clusters 1 and 2 cover therate times the probability of successful transmission. The areas around the gateway nodes 1 and 2, respectively.PHY layer transmission rate can be fixed, or can be When the estimated traffic load around the gateway node 1adaptively adjusted according to the channel quality by is too high, the user taking the route to the gateway node 1means of rate control schemes such as the receiver-based may not achieve high data rate due to the constraint (2) forautorate (RBAR) [23]. In the WMN under consideration, the cluster 1. In this case, if the gateway node 2 is lightly loaded,effective transmission rate of a link is assumed to be static for it is desirable for the user to choose the route to the gateway node 2 for higher data rate. Thus, it can be said that thea long time due to fixed locations of nodes. We define dl as traffic load is estimated and controlled in the unit of thethe effective transmission rate of the link l. cluster for global load balancing. If all flows convey data traffic through each route at their The notion of a cluster corresponds to a clique in theflow data rates, the sum of the data P P rates of traffic passing “conflict graph” introduced in [24]. In the conflict graph,through link l is calculated as r2Hl f2Qr f;r , where Hl is vertices correspond to the links in the WMN. An edge isdefined as the set of the indices of all routes passing through drawn between two vertices if the corresponding linksthe link l, i.e., Hl :¼ fr : l 2 Dr g, and Qr is the indices of all interfere with each other. Thus, an edge stands for conflictionflows that use the route r, i.e., Qr :¼ ff : r 2 Gf g. We define between two vertices. A clique in the conflict graph is a set ofthe “airtime ratio” of the link l, denoted by al , as the ratio of vertices that mutually conflict with each other. According tothe time taken up by the transmission to the total time of link [24], unless the conflict graph is a “perfect graph,” the cliquel. The airtime ratio of the link l can be calculated as the sum of constraints in (2) are not tight in the strict sense even when allthe data rates on the link l divided by the effective cliques (clusters) are taken into account. In [25], the authorstransmission rate of the link l. That is, propose the centralized algorithm that transforms the conflict X X f;r graph to a perfect graph by adding unnecessary edges to the al ¼ : ð1Þ conflict graph. This algorithm can also be applied to our r2Hl f2Q dl r routing scheme. However, from a practical point of view, this Now, we discuss the restriction on the radio resource algorithm is inefficient since it requires centralized controlallocation. For the protocols under consideration, time is the and can overly reduce spatial reuse. Therefore, in this paper,only radio resource, which is shared by links for data we recommend to use the clique constraints in (2) as it is.
  • 8. CHOI ET AL.: EFFICIENT LOAD-AWARE ROUTING SCHEME FOR WIRELESS MESH NETWORKS 1297  À1Actually, these clique constraints are enough to serve our duf ðxÞ ¼ xþ1 : ð5Þpurpose, i.e., identifying overloaded regions in the WMN to dx pfredirect the routes. Also, there can be too many cliques in the The marginal utility of a flow is the amount of the utility thatconflict graph, and therefore, considering all of them can the system can obtain by assigning a unit data rate to therender the proposed scheme highly complex. From a practical flow. Thus, to maximize the system utility, the proposedpoint of view, the clusters do not need to cover all possible scheme is likely to allocate more data rate to the flow withcliques, but it is enough for the clusters to be formed in such a large marginal utility. If is large, the marginal utility dropsway that the traffic load in each region of the WMN is quickly as the data rate increases. In this case, the data ratesseparately evaluated. tend to be fairly allocated to flows, since the incentive to3.3 Utility and Delay Penalty as Optimization Target assign more data rate to the flow which has already received a high data rate reduces. The flow with large value of pf hasThe flow with longer distance consumes generally more a large marginal utility, and therefore, it can receive moreairtime to convey the same amount of data. Therefore, if data rate.maximizing system throughput is the optimization target inthe WMN, the flows with short distance are likely to be When the flowP sends data at its flow data rate, the data f Pallowed to send much more data than those with long rate of flow f is r2Gf f;r . Note that r2Gf f;r is equal todistance are, which leads to unfair resource allocation among the data rate allocated to one active route, if f is a single-flows. Thus, we need an optimization target other than the path flow data rate vector. Hence, the utility of flow f is Psystem throughput to take into account the fairness among uf ð r2Gf f;r Þ. If we consider only the utility in the objectiveflows. We consider the utility of a flow that represents the function of the optimization problem, the routing algorithmdegree of satisfaction felt by the user using the flow. The is likely to ignore the end-to-end delay on a route. Since theutility is a highly desirable performance measure since the delay is also of great importance in the practical WMN, weuser satisfaction is the ultimate goal of the network design. incorporate the delay term into the objective function. To do The utility function defines the mapping between the this, we define the “delay penalty function” for each flow,data rate of a flow and the utility of that flow. Since the which penalizes the objective function for selecting theutility function quantifies the network performance per- route with long end-to-end delay. Since the time to transmitceived by users when a data rate is given, it can only be a packet of x bits through the link l is given as x=dl , we canestimated by a subjective survey, not by theoretical sayP the end-to-end delay on the route r is proportional thatdevelopment. In [26] and [27], the utility function for best to l2Dr 1=dl . The delay penalty function should have aeffort traffic is derived by analyzing the results from asubjective survey for various network data applications. larger value when a flow sends more data on the route withThe utility function of a single best effort user is given as long end-to-end delay. Therefore, the delay P P penalty func- tion for the flow f can be given as r2Gf f;r l2Dr 1=dl . uðxÞ ¼ 0:16 þ 0:8 lnðx À 3Þ; ð3Þ Since we have two different objectives (i.e., the utility and the delay penalty), we should find a Pareto optimal solutionwhere x is the data rate in the unit of Kilobits per second. such that no other solution can improve any objectiveThis utility function is a strictly concave function, and as a without worsening the other objective. To calculate a Paretoresult, it well shows the law of diminishing marginal utility. Although the utility function (3) might be best fitted to optimal solution, we use the scalarization technique [28,the survey results, it is not adequate for theoretical pp. 178-180] that merges multiple objectives into a singlederivation since it is not defined for x 3. Thus, we do objective by taking the weighted sum of the objectives. Wenot take the specific values in (3) but adopt only the log introduce the merged objective function as follows:form of (3) to reflect the law of diminishing marginal utility. 0 1In this paper, the utility function is designed so as to contain X X XX X 1the parameters related to system-wide fairness and priority Oð"Þ :¼ uf @ f;r A À  Á f;r ; ð6Þ f2F r2G f2F r2G l2D dlof flows. We define the utility function of flow f as follows: f f r   where :¼ ðf;r Þf2F ;r2Gf and  is the “delay penalty para- " pf uf ðxÞ :¼ ln xþ1 ; ð4Þ meter” that controls the relative importance of the delay pf penalty to the utility. We can reduce the end-to-end delay atwhere x is the data rate, is the system-wide fairness the expense of the utility by increasing the delay penaltyparameter, and pf represents the priority of flow f. The parameter. We will demonstrate the impact of the delayutility function enables us to control the trade-off betweenefficiency and fairness by adjusting . With a high value of penalty parameter in Section 6. , the system-wide fairness can be guaranteed at the cost ofthe system throughput. In other words, by increasing the 4 PROPOSED LOAD-AWARE ROUTING SCHEMEvalue of , the standard deviation of the flow data rates canbe decreased, but the average flow data rate also decreases. In this section, we design the proposed routing scheme byThe parameter pf controls the priority of flow f. The flow using the dual decomposition method. We first formulate thewith high priority pf is likely to enjoy a high data rate. optimization problem from the objective function and the To explain the reason behind the effect of the parameters, constraints introduced in the previous section, and derivewe introduce the marginal utility. The marginal utility can the dual problem. Next, we explain how to calculate the flowbe derived by differentiating the utility function by the data data rate vector for the given Lagrange multipliers andrate. That is, suggest the subgradient method to iteratively calculate the
  • 9. 1298 IEEE TRANSACTIONS ON MOBILE COMPUTING, VOL. 9, NO. 9, SEPTEMBER 2010 Xoptimal Lagrange multipliers. Finally, we propose the s:t: l À c ¼ 0; for l 2 L; ð13Þ c2V ldampening algorithm to alleviate the route flapping problem. c ! 0; for c 2 C: ð14Þ4.1 Problem Formulation à à ÃWe formulate the optimization problem from (1), (2), and Let w :¼ ð ;  Þ be any optimal solution of this dual  problem.(6) as follows: ! 4.2 Flow Data Rate Calculation for Given Lagrange X X XX X1 max uf f;r À  Á f;r ; ð7Þ Multipliers f2F r2Gf f2F r2Gf d l2Dr l Now, we calculate the flow data rates that maximize the X X f;r Lagrangian (10) for given Lagrange multipliers. To do this, s:t: al ¼ ; for all l 2 L; ð8Þ r2Hl f2Qr dl we find the maximizer of ð f ;  Þ over f 2 P f for each flow X f. Let P f ðÞ denote the set of such maximizers, i.e.,  al
  • 10. ; for all c 2 C; ð9Þ P f ð Þ :¼ arg maxf 2P f ð f ;  Þ. The following proposition  l2Mc holds for the set P f ðÞ:  Pwhere f;r ! 0 and r2Gf f;r max for all f and r. This Proposition 1. The set P f ð Þ contains at least one single-path optimization problem is feasible and convex. Let à be any f;r flow data rate vector.optimal solution of this optimization problem. We also Proof. The proof is provided in Appendix A. u tdefine the optimal flow data rate vector à :¼ ðà Þr2Gf . f f;r We solve the optimization problem by converting it to From now on, we will explain how to find a single-paththe dual problem according to the Lagrangian method in flow data rate vector in P f ðÞ. Let f ð;  Þ be the maximum [29]. The Lagrangian is given as follows: value of ðf ;  Þ when f is a single-path flow data rate Âð"; a; wÞ vector with the active route of . That is, ! & X X XX X1 X  þ l :¼ uf f;r ÀÁ f;r f ð; Þ :¼ max uf ðÞ À : ð15Þ d 0 max dl f2F r2Gf f2F r2Gf l2Dr l l2D X & X X f;r X & X þ  l al À þ c
  • 11. À al We define Rf ðÞ :¼ arg max2Gf f ð;  Þ as the set of the  l2L r2Hl f2Qr dl c2C l2Mc ð10Þ “optimal routes.” To maximize ðf ;  Þ, the active route ! X& X X X  þ l should be one of the optimal routes in Rf ðÞ. Since uf ðÞ is  ¼ uf f;r À f;r a concave function, f ð; Þ is larger for the active route P f2F r2Gf r2Gf l2Dr dl with the smaller l2D l =dl . Therefore, we can also write X & X X Rf ð Þ as follows:  þ l À c a l þ
  • 12. c ; l2L c2V l c2C X  þ l Rf ð Þ ¼ arg min  : ð16Þ 2Gf dlwhere a :¼ ðal Þl2L and V l denotes the set of the indices of all l2Dclusters that the link l belongs to (i.e., V l :¼ fc : l 2 Mc g). Itis noted that we have P We define ð þ l Þ=dl as the “link cost” of the link l. Since l2D ð þ l Þ=dl is the sum of link costs on the route , the X XX XX X optimal route is the route that minimizes the sum of link l f;r =dl ¼ f;r l =dl ; l2L r2Hl f2Qr f2F r2Gf l2Dr costs. Since most existing ad hoc routing algorithms can P P P P employ the sum of link costs as the routing metric, theand c2C c l2Mc al ¼ l2L al c2V l c . In (10), l and c optimal route can be found by applying those ad hocare the Lagrange multipliers corresponding to the con- routing algorithms.straints (8) and (9), respectively. We also define  :¼ ðl Þl2L , When f is a single-path flow data rate vector with the :¼ ðc Þc2C , and w :¼ ð ;  Þ.  active route of , we can maximize ðf ;  Þ if the flow data From the Lagrangian, we define the dual function as rate on the active route is equal togðwÞ :¼ max;a Âð"; a; wÞ. The dual function gðwÞ is defined " P f ð;  Þonly for the Lagrange multipliers such that l À c2V l c ¼ & X  þ l 0 for l 2 L, since we have gðwÞ ¼ 1 for the other Lagrange :¼ arg max uf ðÞ À multipliers. Then, the dual function is given as follows: 0 max l2D dl ð17Þ X X &  !þ pf 1 gðwÞ ¼ max ð f ;  Þ þ
  • 13. c ; ð11Þ ¼ min P À1 ; max ; l2D ð þ l Þ=dl f 2P f f2F c2C Pwhere P f :¼ ff : f;r ! 0 for r 2 Gf ; r2Gf f;r max g and w h e r e ½xŠþ ¼ maxf0; xg. L e t u s d e f i n e f ð; Þ :¼ P P Pðf ;  Þ :¼ uf ð r2Gf f;r Þ À r2Gf f;r l2Dr ð þ l Þ=dl . ðf;r ð; ÞÞr2Gf as follows: From the dual function, we define the following dual & f ð;  Þ; if r ¼ ;problem: f;r ð;  Þ ¼ ð18Þ 0; otherwise: min gðwÞ; ð12Þ Then, we have f ð; Þ 2 P f ðÞ if 2 Rf ð Þ.  
  • 14. CHOI ET AL.: EFFICIENT LOAD-AWARE ROUTING SCHEME FOR WIRELESS MESH NETWORKS 1299 ðjÞ4.3 Lagrange Multiplier Update jP f ðà Þj > 1, it is not guaranteed that f ðf ;  ðjÞ Þ converges From now on, we will find the solution of the dual problem to the optimal flow data rate vector. The cardinality of(12). The constraint (13) in the dual problem can be P f ð Ã Þ is closely related to the cardinality of Rf ð Ã Þ in (16).  incorporated into the dual function. We define the modified If jRf ð à Þj ¼ 1, we also have jP f ð à Þj ¼ 1. On the other  dual function hð Þ :¼ gðð ðÞ; ÞÞ, where  ð Þ :¼ ðl ðÞÞl2L      hand, if jRf ðà Þj > 1, the set P f ðÃ Þ contains multiple flow  satisfies the constraint (13) for given . That is, data rate vectors, which include the multipath flow data X rate vector distributing the data rates on the multiple routes l ð Þ ¼  c ; ð19Þ in Rf ð à Þ. This means that à may be the multipath flow  f c2V l data rate vector for the flow f such that jRf ðà Þj > 1. In this  ðjÞfor all l. Then, the dual optimal solution  à minimizes hðÞ  case, the single-path flow data rate vector f ðf ; ðjÞ Þover c ! 0 for all c 2 C. cannot converge to à . However, we can still use f ðjÞ We will use the subgradient method [29, p. 620] to f ðf ; ðjÞ Þ for all fs as a suboptimal solution very closecalculate the dual optimal solution. We can calculate the to the optimal one, since there are generally a small numbersubgradients of h from the theorem of dual derivatives [29, of the flows fs such that jRf ð à Þj > 1, and the flow data p. 604]. Recall that f denotes the active route of the flow f, rates in P f ð Ã Þ well approximate à even for fs such that  fand ¼ ðf Þf2F denotes the active route vector. We first " jRf ð à Þj > 1.  ðjÞcalculate al ð" ; Þ, which denotes the airtime ratio of link l Even when we use f ðf ;  ðjÞ Þ as a suboptimal solution,when and  are given. The airtime ratio al ð" ; Þ is " we still have the “route flapping problem.” The fact that ðjÞcalculated as follows: f ðf ; ðjÞ Þ converges to the set P f ðÃ Þ does not mean  ðjÞ that f ðf ; ðjÞ Þ has a limit in P f ð à Þ. Therefore, for fs such  X X f;r ðf ;  Þ ðjÞ al ð" ;  Þ :¼ : ð20Þ that jP f ð à Þj > 1, it is possible that f ðf ; ðjÞ Þ alternates  r2Hl f2Qr dl between some points in P f ðÃ Þ even after  ðjÞ is sufficiently  converged to  à . In this case, the active route can flap betweenFrom al ð" ;  Þ, we can calculate sð" ;  Þ :¼ ðsc ð" ; ÞÞc2C , some routes in Rf ð à Þ. We now suggest the “dampening which is defined as algorithm” to solve this route flapping problem. X sc ð" ; Þ :¼
  • 15. À al ð" ;  ðÞÞ:  ð21Þ 4.5 Dampening Algorithm: Solution to Route l2Mc Flapping ProblemIf f 2 Rf ððÞÞ for all f 2 F , the vector sð" ; Þ is the   The dampening algorithm should alleviate the route flappingsubgradient of hðÞ.  problem while keeping the solution in the close range of the Let  ðjÞ :¼ ððjÞ Þc2C be the estimation of à at the jth c optimal one. Moreover, the dampening algorithm should beiteration of the subgradient method. We also define ðjÞ :¼ " able to be implemented in a distributed way. To accomplish ðjÞ ðjÞðf Þf2F as the active route vector such that f 2 Rf ðððjÞ ÞÞ   these goals, the dampening algorithm prevents the routefor all f 2 F . The iteration begins when j ¼ 1. The initial flapping by changing the active route more conservativelyvalues should satisfy ð0Þ ! 0 for all c 2 C. The Lagrange c than the original algorithm does. When the original algo- ðjÞmultipliers are updated at the jth iteration as follows: rithm is used, we have f 2 Rf ð ðjÞ Þ. This means that, at the   À ÁÃþ jth iteration, the original algorithm finds any optimal route in  ðjþ1Þ ¼  ðjÞ À ðjÞ s ðjÞ ; ðjÞ " ; ð22Þ Rf ð ðjÞ Þ from (16), and immediately changes the active route  to the new optimal route. However, the dampening algo-where ½ðxc Þc2C Šþ ¼ ðmaxf0; xc gÞc2C and ðjÞ is the step size. rithm changes the active route only if the new route increasesThe step size can be the diminishing step size that satisfies P1 ðjÞ P1 ðjÞ 2 f ð; ðjÞ Þ by a certain margin.ðjÞ > 0, j¼0  ¼ 1, and j¼0 ð Þ < 1. In this case, Let us explain the operation of the dampening algorithm. ðjÞ converges to à . Alternatively, we can also use the We define  as the “dampening parameter” which controlsconstant step size, which makes ðjÞ converge to within the conservativeness in changing the route. The value of  is ðjÞsome range of  à . We define  ðjÞ :¼ ðl Þl2L ¼  ððjÞ Þ. Then,  between zero and one. If  is set to one, the dampening ðjÞ converges to  à when ðjÞ converges to à . algorithm is the same as the original algorithm. The active route is changed more conservatively with the smaller value4.4 Convergence of Flow Data Rate ðjÞ of . At the jth iteration, the dampening algorithm first findsWe take f ðf ;  ðjÞ Þ as the estimation of the optimal flow ðjÞ any optimal route in Rf ð ðjÞ Þ. Let yf denote the optimal route data rate vector à at jth iteration. We will discuss the f for the flow f, newly found at the jth iteration. Theconvergence of this flow data rate vector. Since the ðjÞ dampening algorithm decides the active route f accordingoptimization problem (7) is strictly feasible and the objective to the following rules:and constraint functions are concave, the strong duality ðjÀ1Þ ðjÞholds from the Slater’s constraint qualification [29, p. 520]. . If f ðf ;  ðjÞ Þ !  Á f ðyf ; ðjÞ Þ, the dampeningTherefore, à is included in the set P f ð à Þ. Moreover, f  algorithm does not change the active route, i.e., ðjÞ f ðf ;  ðjÞ Þ also converges to the set P f ð Ã Þ as j ! 1, since  ðjÞ ðjÀ1Þ ðjÞ f ¼ f .we have f ðf ; ðjÞ Þ 2 P f ð ðjÞ Þ for all j.  ðjÀ1Þ ðjÞ From the above statements, we can conclude that . If f ðf ;  ðjÞ Þ <  Á f ðyf ; ðjÞ Þ, the dampening ðjÞ algorithm changes the active route to the new f ðf ;  ðjÞ Þ converges to à as j ! 1 for the flow f such f ðjÞ ðjÞthat jP f ð à Þj ¼ 1. However, for the flow f such that  optimal route, i.e., f ¼ yf .
  • 16. 1300 IEEE TRANSACTIONS ON MOBILE COMPUTING, VOL. 9, NO. 9, SEPTEMBER 2010 Consequently, the dampening algorithm can alleviate the Theorem 1. If the dampening algorithm is used, we have ½jðjÞ À route flapping problem if the dampening parameter  is set  à j À ðފþ ! 0 as j ! 1, where ðÞ :¼ maxfj À à j : to a sufficiently small value. The stability comes at the cost  Á hðÞ hð à Þg.  of the suboptimal route selection. For all j, the suboptimal Proof. The proof is provided in Appendix B. u t ðjÞroute f satisfies the following condition: À ðjÞ Á According to Theorem 1,  ðjÞ converges to within the f f ;  ðjÞ !  Á max f ð; ðjÞ Þ: ð23Þ distance of ðÞ from  à . If we set  to a smaller value to 2Gf make the WMN more stable, the Lagrange multipliersExcept for the difference in deciding ðjÞ , the dampening " deviate more from à , since ðÞ becomes larger for thealgorithm uses the same equations (17)-(22) as the original smaller . Additionally, for the given Lagrange multipliers,algorithm to calculate the flow data rate vectors and update the dampening algorithm selects the suboptimal route thatthe Lagrange multipliers. satisfies the condition in (23). Therefore, we can conclude Now, we discuss the property of the dampening that the dampening algorithm finds the suboptimal solu-algorithm. For the given active route vector ¼ ðf Þf2F , " tion, to which the distance from the optimal solution iswe define the “fixed-route optimization problem” as the restricted by the condition in (23) and Theorem 1.modified version of the optimization problem (7) in such a Though it is verified in Theorem 1 that  ðjÞ converges to away that the flow data rate vector of flow f is restricted to certain range of à , we cannot guarantee that ðjÞ has a limitthe single-path flow data rate vector with the active route of within the range. However, ðjÞ can have a limit, if thef for all fs in F . Let hð" ; Þ denote the dual function of the following condition is met. Suppose that the dampeningfixed-route optimization problem with the active route algorithm selects a certain active route vector à :¼ ðf Þf2F " à ðjÞ Ãvector of . The dual function hð" ; Þ is given as " at the Jth iteration. It is possible that ¼ for j ! J, if " " X X we have f ðf ;  ðjÞ Þ !  Á max2Gf f ð;  ðjÞ Þ for all f 2 F for à hð" ;  Þ :¼ f ðf ;  ðÞÞ þ
  • 17.  c : ð24Þ j ! J. There can be à satisfying the above condition, if  is " f2F c2C sufficiently small. Let à ð" Þ denote the optimal Lagrange multiplier of the fixed-route optimization problem with theThe following propositions hold for the dual function active route vector of . From Proposition 4, the Lagrange "hð" ;  Þ: multiplier ðjÞ approaches à ð" ðjÞ Þ at the jth iteration. Proposition 2. hðÞ ¼ max hð" ;  Þ.  " Therefore, ðjÞ has a limit on à ð" à Þ, if we have ðjÞ ¼ à " " ðjÞProof. From Proposition 1, the single-path flow data rate for j ! J. Accordingly, f ðf ;  ðjÞ Þ also converges to the vector can be a maximizer of ðf ;  ð ÞÞ. Therefore, it  optimal solution of the fixed-route optimization problem holds that with the active route vector of à . In this case, we can " X X guarantee the lower bound on the performance of the hðÞ ¼  max f ð;  ðÞÞ þ
  • 18.  c dampening algorithm as stated in the following theorem: 2Gf f2F c2C Theorem 2. Suppose that there exist à and J such that ðjÞ ¼ " " ( ) X X à for j ! J. Then, we have the lower bound on the " ¼ max f ðf ; ðÞÞ þ
  • 19.  c ¼ max hð" ;  Þ: performance of the dampening algorithm as follows: " " f2F c2C t u lim Oð"ð" ðjÞ ; ðjÞ ÞÞ !  Á Oð"à Þ; ð25Þ j!1 where à :¼ ðà Þf2F ;r2Gf , ð" ;  Þ :¼ ðf;r ðf ;  ÞÞf2F ;r2Gf , " f;r " Proposition 3. hð" ðjÞ ; ðjÞ Þ !  Á hððjÞ Þ for all j.  and Oð"Þ is the objective function defined in (6). P ðjÞProof. For all j, we have hð" ; ðjÞ Þ ¼ f2F f ð f ; ðjÞ Þ þ ðjÞ Proof. The proof is provided in Appendix C. u t P P ðjÞ P
  • 20. c2C ðjÞ ! f2F  Á max2Gf f ð;  Þ þ
  • 21. c2C ðjÞ ! c c  Á hððjÞ Þ.  u t 5 DISTRIBUTED IMPLEMENTATIONProposition 4. sð" ; Þ is the subgradient of hð" ; Þ. The proposed routing scheme can be implemented in aProof. When  is given, f ðf ;  ðÞÞ is the maximizer of the  distributed way, which improves the scalability of the Lagrangian of the fixed-route optimization problem with WMNs. In this section, we discuss the distributed the active route vector of . Therefore, sð" ;  Þ is the " implementation of the proposed scheme. The flow data ðjÞ subgradient of the dual function of this optimization rate vectors f ðf ; ðjÞ Þs and the Lagrange multipliers ðjÞ s c problem from the theorem of dual derivatives. u t are distributively managed by the nodes in the WMN. The ðjÞ flow data rate vector f ðf ;  ðjÞ Þ is managed by the source ðjÞ From Proposition 4, the Lagrange multipliers are node of the flow f. Recall that f ðf ;  ðjÞ Þ is the single-path ðjÞupdated at the jth iteration, not by using the subgradient flow data rate vector with the active route of f , and theof hð ðjÞ Þ but by using the subgradient of hð" ðjÞ ; ðjÞ Þ, since  ðjÞ ðjÞ ðjÞ flow data rate on f is equal to f ðf ;  Þ. Therefore, the ðjÞf may not included in Rf ððjÞ Þ. In the dampening  source node of the flow f actually manages f and ðjÞ ðjÞ ðjÞalgorithm, the Lagrange multipliers converge to within a f ðf ;  Þ.certain range of the optimal Lagrange multiplier as stated in For implementation, one node within a cluster is desig-the following theorem: nated as the head of the cluster. The head of a cluster is
  • 22. CHOI ET AL.: EFFICIENT LOAD-AWARE ROUTING SCHEME FOR WIRELESS MESH NETWORKS 1301 By the above steps, not only the routing but also the link cost control and the flow control are performed. That is, . Each cluster head estimates the traffic load within P the cluster, that is, ðjÞ ðjÞ l2Mc al ð" ;  Þ for cluster c, and the nodes within the cluster adjust the link costs ðjÞ of its outgoing links, ð þ l Þ=dl for link l, on the basis of this estimated load as in steps 1-5 (i.e., link cost control). . Based on the link costs, the active route for each flow is updated in steps 6-8 (i.e., routing). . The flow data rate is calculated in step 9 (i.e., flow control). Theoretically, an operation can be performed only after the preceding operation has been completed. In this case, however, the proposed scheme may converge very slowly. When the dual decomposition method is used, different variables can be updated according to the different time schedules [15]. Therefore, in order to improve the conver- gence speed, in practice, different network entities (i.e., cluster heads, nodes in each cluster, and source nodes) carry out these operations asynchronously, by using currentlyFig. 2. Control information exchange for distributed implementation. available information. Though it is difficult to prove that the asynchronous operation leads to the exact solution, weassumed to be able to communicate with the transmitter have confirmed by the simulation that the solutionsnodes of the links in its cluster. Let us call the head of cluster c produced by the asynchronous and the synchronous opera-the “cluster head” c. The cluster head c takes the role of tions are the same in our routing problem. In the following,maintaining and updating ðjÞ . we describe three operations in more detail when they are c The proposed scheme takes the following steps at the jth implemented asynchronously.iteration of the subgradient method. Fig. 2 illustrates an . Link cost control: For link cost control, the cluster headexample of control information exchange for this operation. c gathers the information on the total load in the cluster c and adjusts c to control the load on the 1. The source node of flow f sends a message contain- ðjÞ ðjÞ cluster c. This process is as follows: Each node ing f ðf ;  ðjÞ Þ to the nodes on the active route f . estimates the total loads for all outgoing links and 2. Each node calculates al ð" ðjÞ ;  ðjÞ Þs for its all outgoing broadcasts the estimated loads periodically as in step 2. links from (20) and broadcasts them to the heads of Upon receiving the broadcasting message, the cluster clusters to which the links belong. head can compute the total airtime ratio consumed by 3. The cluster head c receives al ð" ðjÞ ;  ðjÞ Þs for all links the links in its cluster. If the total airtime ratio exceeds in its cluster and updates ðjÞ to ðjþ1Þ from (22). c c the available airtime ratio (i.e.,
  • 23. ), it means that the 4. The cluster head c broadcasts ðjþ1Þ to the transmit- c cluster is overloaded. Therefore, the cluster head c ter nodes of the links in its cluster. increases c as in step 3. If the cluster c is not ðjþ1Þ 5. Each node calculates l ¼ l ððjþ1Þ Þ from (19) and  overloaded, that is, its total airtime ratio is smaller ðjþ1Þ derives the link cost ð þ l Þ=dl for its all outgoing than
  • 24. , the cluster head c decreases c . The cluster links. head c periodically broadcasts c as in step 4. From 6. The source node of flow f finds the optimal route this broadcasting message, f þ l ðÞg=dl s are calcu-  ðjþ1Þ yf 2 Rf ððjþ1Þ Þ. An implementation example for  lated for all links within the cluster as in step 5. Since finding the optimal route, when each node main- l ðÞ is the sum of c s for the clusters around the link l  tains the updated link costs for its all outgoing links, (see (19)), the link cost of the link l (i.e., f þ l ð Þg=dl )  will be explained later. can be controlled by the cluster heads based on the 7. The source node of flow f is informed of the link inflicted load. ðjþ1Þ ðjÞ costs ð þ l Þ=dl s on the current active route f . Routing: The link cost of link l is calculated as ðjþ1Þ and the new optimal route yf . f þ l ðÞg=dl . Since dl is the effective transmission  ðjþ1Þ ðjþ1Þ 8. The source node of flow f sets f to yf , if it rate reflecting the PHY transmission rate as well as holds that the packet error probability, we can say that 1=dl is À ðjÞ Á À ðjþ1Þ Á equivalent to the ETT in [5]. Therefore, the link cost in f f ; ðjþ1Þ <  Á f yf ; ðjþ1Þ : the proposed scheme can be viewed as the ETT augmented with the load control variable (i.e., l ðÞ).  If not, the active route is not changed, i.e., To find the optimal route on which the sum of the ðjþ1Þ ðjÞ f ¼ f . proposed link cost is minimized, we can use either the 9. The source node of flow f calculates the flow data existing proactive routing protocols (e.g., the destina- ðjþ1Þ rate f ðf ; ðjþ1Þ Þ from (17). tion-sequenced distance vector (DSDV)) or reactive
  • 25. 1302 IEEE TRANSACTIONS ON MOBILE COMPUTING, VOL. 9, NO. 9, SEPTEMBER 2010 routing protocols (e.g., the AODV and the dynamic TABLE 2 source routing (DSR)). The source node of flow f Table of Simulation Parameters periodically finds the new optimal route yf by using these routing protocols as in step 6. The source node is aware of the link costs on the current active route from the periodic report, and is also informed of the link costs on the new optimal route by the routing protocol (step 7). Based on these link costs, the source node decides whether to change the active route or not as in step 8. . Flow/congestion control: As in step 9, the source node periodically recalculates the flow data rate f ðf ; ð ÞÞ by using the link costs on the active route  according to (17). As seen from (17), the source node lowers the flow data rate when the link costs on the active route is high. Since the source node limits the data rate of its traffic to the flow data rate, the source node can be quenched when the active route Let us explain the organization of clusters. We organize passes through the congested area of the network. clusters based on the following two rules, which define the interfering relationship between links. First, all incoming By the above three operations, network-wide load and outgoing links of a node interfere with each other. Thisbalance can be achieved. If an area of the network is is because a node cannot receive or transmit data throughoverloaded, the link costs around the area are increased by more than one link at the same time. Second, two linksthe link cost control operation. Then, the source node of the interfere with each other, if the transmitter or receiver nodeflow passing through the area reduces its flow data rate, or of one link is within the interfering range of the transmitterfinds another route that allows a higher flow data rate. or receiver node of the other link. This can be exemplified Fig. 2 illustrates an example of control information by the IEEE 802.11 protocol, in which the transmitter nodeexchange for a flow and a cluster. On the active route f , the sends a request-to-send (RTS) packet and the receiver nodeflow data rate f ðf ;  ðÞÞ and the link cost f þ l ð Þg=dl are   responds with a clear-to-send (CTS) packet to prevent theexchanged. This control information can be piggybacked on nodes close to the transmitter or receiver node fromthe data and acknowledgement (ACK) packets. In addition, interfering with packet transmission. On the basis of thesethe airtime ratio al ð" ;  ðÞÞ and the Lagrange multiplier c are  rules, we organize clusters as follows: for each pair of nodesexchanged between the cluster head and nearby nodes. If the m and n within the interfering range of each other, there is aproposed scheme is applied to the systems broadcasting the cluster including all the incoming and outgoing links ofbeacon message periodically, for example, such as IEEE the nodes m and n. We assume that two nodes are within802.11, 802.15.3, and 802.15.4 standards, this information can the interfering range of each other, if the received SNR at aalso be conveyed by these beacon messages without addi- node from the other node is over À7 dB.tional control packet transmission. It is assumed that the ratio of the time dedicated to data transmission (i.e.,
  • 26. ) is 0.8. The maximum flow data rate (i.e., max ) is set to 100 Mbps. The system-wide fairness6 NUMERICAL RESULTS parameter in the utility function (i.e., ) is set to 1, unlessThe numerical results presented below show that the noted otherwise. The priority of flow f (i.e., pf ) is set to 1 forproposed routing scheme effectively balances traffic load, all flows. For the subgradient method, the step size (i.e., ðjÞ )and consequently, outperforms the routing algorithm using is set to 0.01. Unless noted otherwise, the delay penaltythe ETT as a routing metric. parameter (i.e., ) is set to 1, and the dampening parameter (i.e., ) is set to 0.95.6.1 Environments and Parameters for Numerical We do not assume any specific PHY/MAC layer Results protocol. Instead, we assume that the WMN uses a simpleWe first describe the environment and parameters for getting reservation protocol (e.g., the mesh deterministic accessthe numerical results. In Table 2, we summarize the (MDA) of the 802.11s protocol [30]). The WMN firstsimulation parameters. We consider 8 km  8 km square calculates the required amount of time for each link toarea. There are 300 nodes, which are randomly located support the flow data rates of the flows on it. After that, itaccording to the uniform distribution. The network uses the assigns the calculated amount of time to each link by usingbandwidth of 20 MHz. The transmission power of a node is a reservation protocol. Each flow on a link reserves a24.5 dBm, and the noise density is assumed to be À163 dBm/ portion of the time assigned to the link. Then, each flow hasHz. We assume the line-of-sight (LOS) link between nodes. an end-to-end path with a reserved data rate. We assumeTherefore, for the channel fading model, we only consider the that a flow has an infinite backlog and sends data at its flowpath loss calculated as xÀ3:7 , where x is the distance between data rate through its end-to-end path. In this setting, we cannodes in meters. A link is established between two nodes if estimate the throughput of each flow from the reserved datathe signal-to-noise ratio (SNR) between them exceeds 0 dB. rate of its end-to-end path. In this paper, we do not use anThe effective transmission rate of link l (i.e., dl ) is assumed to event-based simulator since it is not necessary to simulatebe the Shannon capacity of the link. That is, dl ¼ 20  log2 ð1 þ the behavior of individual packets for estimating theSNRl Þ Mbps, where SNRl is the SNR of the link l. throughput of each flow.
  • 27. CHOI ET AL.: EFFICIENT LOAD-AWARE ROUTING SCHEME FOR WIRELESS MESH NETWORKS 1303Fig. 3. Gateway and no gateway scenarios. (a) Gateway scenario.(b) No gateway scenario. Fig. 5. System throughput according to the number of flows in the gateway scenario. Fig. 6. System throughput according to the load skewness  in the gateway scenario.Fig. 4. Routing examples in the gateway scenario. (a) Proposed,  ¼ 0.(b) ETT,  ¼ 0. (c) Proposed,  ¼ 1. (d) ETT,  ¼ 1. the degree that the source nodes of flows are concentrated in the shaded areas in Fig. 3a. The load skewness is defined We compare the performance of the proposed scheme as  :¼ ðS À N Þ=ðS þ N Þ, where S and N are thewith that of the routing algorithm using the ETT (i.e., 1=dl number of the source nodes in the shaded and nonshadedfor the link l) as the routing metric. For fair comparison, we areas, respectively.apply the same flow control mechanism as the one used by Fig. 4 shows the routing examples in the gatewaythe proposed scheme to the ETT-based routing algorithm. scenario when there are 100 flows. Figs. 4a and 4b show the routing of the proposed and the ETT-based schemes,6.2 Results and Discussions respectively, when the load skewness is zero, i.e., there areWe consider two scenarios, i.e., with and without gateway. the same number of the source nodes in the shaded andIn the gateway scenario, there are four gateway nodes and nonshaded areas. In this case, there is no notable differenceall users intend to communicate with the outside network in the routing of the proposed and the ETT-based schemes.via the gateway nodes. In the no gateway scenario, all users Figs. 4c and 4d present the routing examples with the loadcommunicate with each other without gateway nodes. skewness of one, i.e., all the source nodes are in the shaded area. In these figures, we can see that the proposed routing6.2.1 Gateway Scenario scheme balances the load on the gateway nodes by makingFig. 3a illustrates the gateway scenario. The whole area is some flows utilize the gateway nodes in the nonshadeddivided into four square areas. At the center of each divided area, while the ETT-based scheme focuses almost all loadsarea, a gateway node is placed. All users try to send data on the gateway nodes in the shaded area.traffic to the outside network. Therefore, they should find The load balancing capability of the proposed schemethe route to one of the four gateways. To model the load results in the throughput enhancement, as can be seen inimbalance situation, we introduce a parameter named the Figs. 5 and 6, which show the system throughput as a functionload skewness, denoted by . The load skewness represents of the number of flows and the load skewness, respectively.
  • 28. 1304 IEEE TRANSACTIONS ON MOBILE COMPUTING, VOL. 9, NO. 9, SEPTEMBER 2010Fig. 7. System throughput according to the system-wide fairnessparameter in the gateway scenario. Fig. 9. Routing examples in the no gateway scenario. (a) Proposed, nonconcentrated. (b) ETT, nonconcentrated. (c) Proposed, concen- trated. (d) ETT, concentrated. proposed scheme enhances the throughput from 131 to 223 Mb/s, while at the same time, it reduces the standard deviation from 1,040 to 809 kb/s. Therefore, the proposed scheme can ensure more fairness between flows than the ETT-based scheme does as well as it achieves higher throughput. Thus, we can say that the performance gain of the proposed scheme is not obtained by sacrificing fairness. 6.2.2 No Gateway Scenario In the no gateway scenario, all users communicate with the other user in the WMN. We use two models to arrange theFig. 8. Standard deviation of flow throughput according to the system- source and destination nodes of flows. In the nonconcen-wide fairness parameter in the gateway scenario. trated traffic model, the source and destination nodes of a flow are decided randomly among all nodes. In theIn Fig. 5, we can see that the proposed scheme increases the concentrated traffic model, all source and destination nodesthroughput by 18-64 percent over the ETT-based scheme are within the shaded circle in Fig. 3b. The concentratedwhen the load skewness is one. If the load skewness is zero, traffic model emulates the hot spot where the communica-the throughput gain of the proposed scheme is 10-18 percent. tion is very active.In both cases, the proposed scheme outperforms the ETT- Fig. 9 shows the routing examples in the no gatewaybased scheme, and the performance gap is greater when the scenario when there are 100 flows. Though it is difficult toload skewness is high. Fig. 6 clearly demonstrates the effect of recognize an individual route in this figure, we can observethe load skewness on the throughput. This figure shows that the geographical distribution of the routes. The routes of theas the load skewness increases, the throughput of theproposed scheme only slightly drops (i.e., from 246 to proposed and ETT-based schemes in the nonconcentrated239 Mb/s if the number of flows is 100), whereas the traffic model seem to be similar to each other, as shown inthroughput of the ETT-based scheme sharply drops (i.e., Figs. 9a and 9b. However, in Figs. 9c and 9d, showing thefrom 213 to 144 Mb/s if the number of flows is 100). This routing examples in the concentrated traffic model, we canmeans that the proposed scheme is robust to high load clearly see that the flows make a detour to avoid overloadedskewness owing to its load balancing capability. areas if the proposed scheme is used, while the flows with Recall that the system-wide fairness parameter controls the ETT-based scheme cannot.the fairness among flows. Fig. 7 shows the system through- Fig. 10 plots the system throughput of the proposed andput according to , when the number of flows is 100. To ETT-based schemes according to the number of flows. Thisquantify the fairness, we use the standard deviation in figure shows that the proposed scheme outperforms thethroughput of all flows, which is plotted according to in ETT-based schemes for both the nonconcentrated andFig. 8. In Figs. 7 and 8, we can see that the fairness can be concentrated traffic models. The proposed scheme increasesguaranteed at the expense of the throughput when is high. the throughput by 10-40 percent in the nonconcentratedIn these figures, it can also be seen that if ¼ 2 and  ¼ 1, the traffic model and by 25-40 percent in the concentrated
  • 29. CHOI ET AL.: EFFICIENT LOAD-AWARE ROUTING SCHEME FOR WIRELESS MESH NETWORKS 1305Fig. 10. System throughput according to the number of flows in the no Fig. 12. System throughput according to the delay penalty parameter gateway scenario. in the no gateway scenario.Fig. 11. Average hop count according to the delay penalty parameter  in Fig. 13. Convergence of the proposed routing scheme.the no gateway scenario. that the routes are stabilized as the number of iterationstraffic model. This performance gain is caused by the ability increases. Every 100 iterations, we count the number of flowsof the proposed scheme to go around overloaded areas. whose route is changed during the last 100 iterations. The Recall that the delay penalty parameter  can affect the simulation is performed in both the gateway and no gatewaylength of the route. In Figs. 11 and 12, we, respectively, show scenarios. The load skewness is one for the gateway scenario,the average hop count of routes and the system throughput and the concentrated traffic model is selected for the noaccording to , when the number of flows is 100. In Fig. 11, gateway scenario. The total number of flows is 100.we can see that the average hop count diminishes as the In this figure, we can see that almost all route changesvalue of  increases, especially in the concentrated traffic occur within 200 iterations for both scenarios. It is noted thatmodel. This shows that with a large value of , the proposed this number of iterations (i.e., 200) is required for conver-routing scheme tries to find a short route even when it runs gence from the initial state, where all flows start simulta-through a congested area. Therefore, by setting  to a large neously. Since there is only a small change in networkvalue, routes can be shortened in exchange for the systemthroughput, as can be seen in Fig. 12. configuration (e.g., addition/deletion of a flow) at a time in usual situation, much smaller number of iterations is needed6.2.3 Algorithm Convergence to converge in practice. In case of the gateway scenario, weNow, we study the convergence of the proposed routing observe that no route change takes place after 1,700scheme. To clearly show the convergence speed, we assume iterations. In the no gateway scenario, few routes changein this simulation that all operations are performed synchro- continuously due to the route flapping. However, this isnously at each iteration, instead of employing the asynchro- acceptable since the number of route flappings is very smallnous distributed implementation in previous simulations. compared to the total number of flows. As mentioned before,Note that one iteration includes one round of routing, link if needed, the routing scheme can be further stabilized bycost control, and flow/congestion control. In Fig. 13, we show decreasing the parameter .
  • 30. 1306 IEEE TRANSACTIONS ON MOBILE COMPUTING, VOL. 9, NO. 9, SEPTEMBER 2010  2 À ÁÀ Á7 CONCLUSION ¼ ðjÞ À  à  À 2ðjÞ s ðjÞ ; ðjÞ ðjÞ À à "  À Á2In this paper, we have developed a load-aware routing þ ðjÞ s ðjÞ ; ðjÞ  " ð29Þscheme for the WMN. We have formulated the routing  ðjÞ  È Á À ÁÉ  À  à 2 À 2ðjÞ hð" ðjÞ ; ðjÞ À h ðjÞ ; à "problem as an optimization problem, and have solved it by  À Á2using the dual decomposition method. The dual decom- þ ðjÞ s ðjÞ ; ðjÞ  " ð30Þ  ðjÞ  È À Á Éposition method makes it possible to design a distributed  À  à 2 À 2ðjÞ h ðjÞ ; ðjÞ À hðÃ Þ " routing scheme. However, there could be a route flapping  À Á2problem in the distributed scheme. To tackle this problem, þ ðjÞ s ðjÞ ; ðjÞ  : " ð31Þwe have suggested a dampening algorithm and have We have (28) since ½ÁŠþ is a projection onto the domain of . Weanalyzed the performance of the algorithm. The numerical also have (30) from Proposition 4 and (31) from Proposition 2.results show that the proposed scheme with a dampening If jðjÞ À à j > ðÞ, we have  Á hððjÞ Þ > hðà Þ. Since   algorithm well converges to a stable state and achieves hð" ðjÞ ; ðjÞ Þ !  Á hð ðjÞ Þ from Proposition 3, we have much higher throughput than the ETT-based scheme does hð" ðjÞ ; ðjÞ Þ À hðÃ Þ > 0 for ðjÞ such that jðjÞ À  à j > ðÞ.  owing to its load-balancing capability. In (31), we can see that if hð" ðjÞ ; ðjÞ Þ À hðÃ Þ > 0, the  The main advantage of the proposed routing scheme is distance between ðjÞ and à decreases, provided that thethat it is favorable to practical implementation although it is step size ðjÞ is sufficiently small. This means that ðjÞtheoretically designed. The proposed scheme is a practical approaches  à as long as jðjÞ À  à j > ðÞ. single-path routing scheme, unlike other multipath routing Suppose that we use the diminishing step size. Then, weschemes which are designed by using the optimization P can define ÁðiÞ :¼ 1 ððjÞ Þ2 . Since sð" ðjÞ ; ðjÞ Þ is bounded, theory. Also, the proposed scheme can easily be imple- j¼imented in a distributed way by means of the existing we can define S as the maximum value of jsð" ðjÞ ; ðjÞ Þj. The routing algorithms. The proposed scheme can be applied to theorem is proved if we prove that for an arbitrary small valuevarious single-band PHY/MAC layer protocols. In future , there exists J such that j ðjÞ À  à j À ðÞ  for all j ! J. work, we can extend the proposed scheme so that it can also For a certain i such that j ðiÞ À  à j À ðÞ 0, the maximum be applied to the multiband protocols, which can provide value of jðjÞ À  à j2 À 2 ðÞ over j ! i is 2ðiÞ ð1 À Þhð Ã Þ þ   ðiÞ 2 ðiÞ ðiÞ Ãlarger bandwidth to the WMN. Á S from (31), since hð" ;  Þ À hð Þ can be a negative  value as low as ð À 1Þhðà Þ. Therefore, if jðiÞ À  à j À ðÞ   0, we have maxj!i fjðjÞ À  à j À ðÞg ¼ f2ðiÞ ð1 À Þhð Ã Þ þ  APPENDIX A 1 ÁðiÞ S 2 þ 2 ðÞg2 À ðÞ. Since 2ðiÞ ð1 À ÞhðÃ Þ þ ÁðiÞ S 2 ! 0 PROOF OF PROPOSITION 1 as i ! 1, there exists I such that f2ðiÞ ð1 À Þhð Ã Þ þ  1Consider any flow data rate f 2 P f ðÞ. Let f :¼ ð^f;r Þr2Gf  ^ ÁðiÞ S 2 þ 2 ðÞg2 À ðÞ  for i ! I. We prove the theorembe the single-path flow data rate vector such that f; ¼ ^ in the following two cases. If there exists no i ! I such thatP P r2Gf f;r for the active route that minimizes l2Dr ð þ jðiÞ À à j À ðÞ 0, there exists J ! I such that jðjÞ À  à j À  l Þ=dl over r 2 Gf , and f;r ¼ 0 for the other rs in Gf . Then, ^ ðÞ  for all j ! J, since  ðjÞ approaches à as long aswe have jðjÞ À  à j > ðÞ. If there exists i ! I such that jðiÞ À  à j À   ðÞ 0, we have jðjÞ À  à j À ðÞ  for all j ! i.  X  þ l ð^f ;  Þ ¼ uf ð^f; Þ À f; ^ dl X  l2D APPENDIX C X X  þ l ¼ uf f;r À f;r PROOF OF THEOREM 2 r2Gf r2Gf l2D dl ð26Þ X  Let us define à ð" Þ :¼ ðà ð" ÞÞf2F ;r2Gf as the optimal flow " f;r X X  þ l ! uf f;r À f;r data rates of the fixed-route optimization problem with the r2Gf r2Gf l2Dr dl active route vector of . Since ð" ðjÞ ;  ðjÞ Þ ! à ð" Ã Þ as " " " j ! 1, we have limj!1 Oð"ð" ðjÞ ; ðjÞ ÞÞ ¼ Oð"à ð" à ÞÞ. We ¼ ðf ;  Þ: also have Oð"à ð" à ÞÞ ¼ hð" à ;  à ð" à ÞÞ, since the strong dua- Since f 2 P f ð Þ, we have f 2 P f ð Þ from (26). This means  ^  lity holds for the fixed-route optimization problem. Inthat there exists a single-path flow data rate vector in P f ð Þ.  addition, hð" à ; à ð" à ÞÞ ¼ limj!1 hð" ðjÞ ; ðjÞ Þ, since ðjÞ ! à " " and ðjÞ ! à ð" Ã Þ as j ! 1. From the above equalities, it holds that limj!1 Oð"ð" ðjÞ ; ðjÞ ÞÞ ¼ limj!1 hð" ðjÞ ; ðjÞ Þ. From APPENDIX B Proposition 3, it holds that limj!1 hð" ðjÞ ;  ðjÞ Þ ! limj!1  Á PROOF OF THEOREM 1 hð ðjÞ Þ !  Á hðà Þ. In addition, we have hð Ã Þ ¼ Oð"à Þ,    The following inequality holds for the distance to the since the strong duality holds for the optimization problemoptimal Lagrange multiplier: (7). Finally, we conclude that limj!1 Oð"ð" ðjÞ ; ðjÞ ÞÞ !  Á Oð"à Þ.  ðjþ1Þ 2  À à   À ÁÃþ 2 ¼  ðjÞ À ðjÞ s ðjÞ ; ðjÞ " À à  ð27Þ ACKNOWLEDGMENTS  ðjÞ À ðjÞ ðjÞ Á  à 2 This work was supported in part by the Ministry of  À  s ;  ðjÞ " À ð28Þ Knowledge Economy (MKE), Korea, under the Information
  • 31. CHOI ET AL.: EFFICIENT LOAD-AWARE ROUTING SCHEME FOR WIRELESS MESH NETWORKS 1307Technology Research Center (ITRC) support program [23] G. Holland, N.H. Vaidya, and P. Bahl, “A Rate-Adaptive MAC Protocol for Multi-Hop Wireless Networks,” Proc. ACM MobiCom,supervised by the National IT Industry Promotion Agency July 2001.(NIPA) (grant numbers NIPA-2010-C1090-1011-0004 and [24] K. Jain, J. Padhye, V.N. Padmanabhan, and L. Qiu, “Impact ofNIPA-2010-C1090-1011-0005), and in part by the Hankuk Interference on Multi-Hop Wireless Network Performance,” Proc.University of Foreign Studies Research Fund. ACM MobiCom, Sept. 2003. [25] B. Wang, M. Mutka, and E. Torng, “Optimization Based Rate Allocation and Scheduling in TDMA Based Wireless Mesh Networks,” Proc. IEEE Int’l Conf. Network Protocols, Oct. 2008.REFERENCES [26] Z. Jiang, H. Mason, B.J. Kim, N.K. Shankaranarayanan, and P.[1] R. Bruno, M. Conti, and E. Gregori, “Mesh Networks: Commodity Henry, “A Subjective Survey of User Experience for Data Multihop Ad Hoc Networks,” IEEE Comm. Magazine, vol. 43, no. 3, Applications for Future Cellular Wireless Networks,” Proc. Symp. pp. 123-131, Mar. 2005. Applications and Internet, Jan. 2001.[2] D. De Couto, D. Aguayo, J. Bicket, and R. Morris, “High- [27] Z. Jiang, Y. Ge, and Y. Li, “Max-Utility Wireless Resource Throughput Path Metric for Multi-Hop Wireless Routing,” Proc. Management for Best-Effort Traffic,” IEEE Trans. Wireless Comm., ACM MobiCom, Sept. 2003. vol. 4, no. 1, pp. 100-111, Jan. 2005.[3] D. Passos, D.V. Teixeira, D.C. Muchaluat-Saade, L.C.S. Magalhaes, [28] S. Boyd and L. Vandenberghe, Convex Optimization. Cambridge and C.V.N. Albuquerque, “Mesh Network Performance Measure- Univ. Press, 2004. ments,” Proc. Int’l Information and Telecomm. Technologies Symp., [29] D.P. Bertsekas, Nonlinear Programming. Athena Scientific, 1999. Dec. 2006. [30] G.R. Hiertz, Y. Zhang, S. Max, T. Junge, E. Weiss, and B. Wolz,[4] B. Awerbuch, D. Holmer, and H. Rubens, “The Medium Time “IEEE 802.11s: WLAN Mesh Standardization and High Perfor- Metric: High Throughput Route Selection in Multi-Rate Ad Hoc mance Extensions,” IEEE Network, vol. 22, no. 3, pp. 12-19, May/ Wireless Networks,” Mobile Networks and Applications, vol. 11, June 2008. no. 2, pp. 253-266, Apr. 2006.[5] R. Draves, J. Padhye, and B. Zill, “Routing in Multi-Radio, Multi- Kae Won Choi received the BS degree in civil, Hop Wireless Mesh Networks,” Proc. ACM MobiCom, Sept. 2004. urban, and geosystem engineering and the MS[6] Y. Yang, J. Wang, and R. Kravets, “Designing Routing Metrics for and PhD degrees in electrical engineering and Mesh Networks,” Proc. IEEE Workshop Wireless Mesh Networks, computer science from Seoul National Univer- Sept. 2005. sity, Korea, in 2001, 2003, and 2007, respec-[7] A.P. Subramanian, M.M. Buddhikot, and S.C. Miller, “Interference tively. From 2008 to 2009, he was with Aware Routing in Multi-Radio Wireless Mesh Networks,” Proc. Telecommunication Business of Samsung Elec- IEEE Workshop Wireless Mesh Networks, Sept. 2006. tronics Co., Ltd., Korea. He is currently a[8] C.E. Koksal and H. Balakrishnan, “Quality-Aware Routing postdoctoral researcher in the Department of Metrics for Time-Varying Wireless Mesh Networks,” IEEE J. Electrical and Computer Engineering, University Selected Areas in Comm., vol. 24, no. 11, pp. 1984-1994, Nov. 2006. of Manitoba, Winnipeg, Manitoba, Canada. His research interests[9] S.J. Lee and M. Gerla, “Dynamic Load-Aware Routing in Ad Hoc include wireless network optimization, radio resource management, Networks,” Proc. IEEE Int’l Conf. Comm., June 2001. wireless mesh networks, and cognitive radio.[10] H. Hassanein and A. Zhou, “Routing with Load Balancing in Wireless Ad Hoc Networks,” Proc. ACM Int’l Workshop Modeling Wha Sook Jeon received the BS, MS, and PhD Analysis and Simulation of Wireless and Mobile Systems, July 2001. degrees in computer engineering in 1983, 1985,[11] D. Zhao, J. Zou, and T.D. Todd, “Admission Control with Load and 1989, respectively, from Seoul National Balancing in IEEE 802.11-Based ESS Mesh Networks,” Wireless University, Korea, where she joined the faculty Networks, vol. 13, no. 3, pp. 351-359, June 2007. in 1999, and is currently a professor in the[12] L. Ma and M.K. Denko, “A Routing Metric for Load-Balancing in School of Electrical Engineering and Computer Wireless Mesh Networks,” Proc. Advanced Information Networking Science. From 1989 to 1999, she was with the and Applications Workshops, May 2007. Department of Computer Engineering, Hansung[13] W. Song and X. Fang, “Routing with Congestion Control and Load University, Korea. Her research interests include Balancing in Wireless Mesh Networks,” Proc. Int’l Conf. ITS resource management for wireless and mobile Telecomm., June 2006. networks, mobile communications systems, high-speed networks,[14] X. Lin, N.B. Shroff, and R. Srikant, “A Tutorial on Cross-Layer communication protocols, and network performance evaluation. She Optimization in Wireless Networks,” IEEE J. Selected Areas in currently serves on the editorial board of the Journal of Communications Comm., vol. 24, no. 8, pp. 1452-1463, Aug. 2006. and Networks (JCN). She is a senior member of the IEEE.[15] D.P. Palomar and M. Chiang, “A Tutorial on Decomposition Methods for Network Utility Maximization,” IEEE J. Selected Areas Dong Geun Jeong received the BS, MS, and in Comm., vol. 24, no. 8, pp. 1439-1451, Aug. 2006. PhD degrees from Seoul National University,[16] M. Chiang, S.H. Low, R.A. Calderbank, and J.C. Doyle, “Layering Korea, in 1983, 1985, and 1993, respectively. as Optimization Decomposition: A Mathematical Theory of From 1986 to 1990, he was a researcher with Network Architectures,” Proc. IEEE, vol. 95, no. 1, pp. 255-312, the R&D Center of DACOM, Korea. During Jan. 2007. 1994-1997, he was with the R&D Center of[17] E.M. Royer and C.K. Toh, “A Review of Current Routing Protocols Shinsegi Telecomm, Inc., Korea, where he for Ad Hoc Mobile Wireless Networks,” IEEE Personal Comm., conducted and led research on advanced vol. 6, no. 2, pp. 46-55, Apr. 1999. cellular mobile networks. In 1997, he joined the[18] S. Nelakuditi, S.H. Lee, Y. Yu, J. Wang, Z. Zhong, G. Lu, and Z. faculty at the Hankuk University of Foreign Zhang, “Blacklist-Aided Forwarding in Static Multihop Wireless Studies, Korea, where he is currently a professor in the School of Networks,” Proc. IEEE Int’l Conf. Sensors and Ad Hoc Comm. and Electronics and Information Engineering. His research interests include Networks, Sept. 2005. resource management for wireless and mobile networks, mobile[19] S. Biswas and R. Morris, “ExOR: Opportunistic Multi-Hop communications systems, communication protocols, and network Routing for Wireless Networks,” Proc. SIGCOMM, Aug. 2005. performance evaluation. He served as the program committee vice[20] Y. Yuan, H. Yang, S.H.Y. Wong, S. Lu, and W. Arbaugh, “ROMER: chair for IEEE VTC 2003-Spring. From 2002 to 2007, he served on the Resilient Opportunistic Mesh Routing for Wireless Mesh Net- editorial board of the Journal of Communications and Networks (JCN). works,” Proc. IEEE Workshop Wireless Mesh Networks, Sept. 2005. He is a senior member of the IEEE.[21] K.N. Ramachandran, M.M. Buddhikot, and G. Chandranme- non, “On the Design and Implementation of Infrastructure Mesh Networks,” Proc. IEEE Workshop Wireless Mesh Networks, Sept. 2005. . For more information on this or any other computing topic,[22] A. Raniwala and T. Chiueh, “Architecture and Algorithms for an please visit our Digital Library at www.computer.org/publications/dlib. IEEE 802.11-Based Multi-Channel Wireless Mesh Network,” Proc. IEEE INFOCOM, Mar. 2005.