Traffic balancing-based path recommendation mechanisms in vehicular networks Maram Bani Younes *, Azzedine Boukerche and Graciela Román-Alonso

WIRELESS COMMUNICATIONS AND MOBILE COMPUTING Wirel. Commun. Mob. Comput. 2016; 16:794 809 Published online 29 January 2015 in Wiley Online Library (wileyonlinelibrary.com)..2570 RESEARCH ARTICLE Traffic balancing-based path recommendation mechanisms in vehicular networks Maram Bani Younes *, Azzedine Boukerche and Graciela Román-Alonso PARADISE Research Laboratory, DIVA Strategic Research Center, University of Ottawa, Ottawa, Canada ABSTRACT In this paper, we propose both reactive and proactive balancing traffic path recommendation mechanisms, which we refer to as Bal-Traf and Abs-Bal, respectively. Bal-Traf is initiated when a certain output road segment located at any road intersection is detected in an overloaded situation. In the event that the existing traffic density of any output road exceeds its optimal capacity, Bal-Traf recommends that those vehicles that plan to pass over this road segment as next hop choose another, less congested output road segment. On the other hand, Abs-Bal is a proactive balancing traffic mechanism. Its main purpose is to distribute input traffic completely even among all output road segments at intersections. Moreover, Abs-Bal considers the best travel time of vehicles in addition to the goal of balancing traffic. From the experimental results, we can see that Bal-Traf eliminates the number of overwhelmed road segments over the road network in scenarios with only partial network congestion. It also decreases the number of congested road segments in scenarios with complete network congestion. However, it increases the density drastically over the remaining congested road segments in these scenarios. Abs-Bal performs well in decreasing the percentage of congested road segments and balancing traffic among road segments located throughout the road network, in the event of complete network congestion. Copyright 2015 John Wiley & Sons, Ltd. KEYWORDS road intersection; input road segment; output road segment; traffic congestion; path recommendation protocol; traffic balancing *Correspondence Maram Bani Younes, PARADISE Research Laboratory, DIVA Strategic Research Center, University of Ottawa, Ottawa, Canada. E-mail: mbani047@uottawa.ca 1. INTRODUCTION Several applications in the literature have been introduced with the purpose of decreasing traffic congestion and recommending the fastest path toward each targeted destination. The previously proposed path recommendation protocols can be classified into two main categories: centralized path recommendation and distributed path recommendation protocols. Centralized path recommendation protocols gather the real-time traffic data and the location data of destinations at a central database. A central processor (i.e., server) recommends the best path to travel toward each targeted destination according to the data gathered [1 7]. However, centralized mechanisms suffer from issues regarding scalability, on-time traffic evaluation, and on-time path recommendation which affect the superiority of the selected path. On the other hand, distributed path recommendation protocols construct the path toward each destination in a hop-by-hop fashion. No central data gathering function or central processor is required for distributed path recommendation protocols. At each road intersection, the next hop for each traffic flow is recommended according to the traffic characteristics of the surrounding road segments [8,9]. A set of Road Side Units (RSUs) must be installed over the road network, and an RSU is installed at each road intersection to evaluate and to make recommendations to the surrounding traffic flow locally. At each road intersection, the incoming traffic (i.e., input flow) is able to leave the intersection through three directions: straight, right, and left. The direction with the least amount of required travel time toward each destination is recommended for each traveling vehicle [8]. Distributed path recommendation protocols handle scalability and accuracy issues, but they generate additional overhead in terms of bandwidth consumption over the communications network. This is due to the large communication requirements among located RSUs in order to keep the gathered data synchronized among them. However, in some scenarios, the input flow is highly congested, and most of the traveling vehicles are heading toward the same destination or different destinations 794 Copyright 2015 John Wiley & Sons, Ltd.

M. Bani Younes, A. Boukerche and G. Román-Alonso Traffic balancing located close to each other, within downtown areas. In these scenarios, the existing bottleneck problem can be transferred from one road segment to the next road segment at each road intersection. This exaggerates the traffic congestion problem throughout the road network instead of solving or reducing it. In other scenarios, the bottleneck problem is generated by the path recommendation protocol over some output road segments. The scenarios of the latter case occur when most of the vehicles arriving from several input road segments, moving toward the road intersection, are directed to take the same output road segment toward their destinations. In this paper, we first propose a balancing traffic (Bal- Traf) mechanism that is intended to eliminate the bottleneck problem caused by the distributed path recommendation protocol, within downtown areas. In Bal-Traf, the estimated traffic density of each output road segment is compared with the saturated traffic density (Sd i )oftheroad segment in question. Sd i represents the acceptable density of each road that enables all vehicles to proceed smoothly (i.e., the optimal capacity). If the density of the investigated road segment exceeds Sd i, the road should be detected as an overloaded road segment. In this case, the input traffic of the road intersection is initially supposed to use this road segment as a next hop and is equitably redirected to other output road segments. This occurs even if all vehicles are heading toward the same destination. In general, Bal-Traf predicts the overwhelmed output segments and suggests that some vehicles change the next hop road segment in order to alleviate the traffic congestion conditions. Bal-Traf is expected to perform well in terms of eliminating the bottleneck problem, when the downtown area is partially congested. However, in scenarios with congestion through most of the road network, there are no road segments with less congested output among which traffic can be distributed. In these scenarios, Bal-Traf decreases the number of overwhelmed road segments occurring over the road network. However, it drastically increases the traffic congestion occurring over the output road segments that remain in overloaded conditions. Second, we propose an absolute balancing traffic mechanism (Abs-Bal) that distributes the input traffic at each road intersection evenly among the output road segments. In Abs-Bal, the RSU located at each road intersection computes the estimated traffic density for each output road segment. The RSU recommends that vehicles leaving the road intersection take a specific output road segment; these recommendations are made while it ensures that the same density among all output road segments is maintained. Vehicles supposed to travel through road segments with high traffic density are directed to choose another less congested output road segment. In scenarios where the entire road network is congested, Abs-Bal avoids the drastically congested output road segment situation. Output road segments at each road intersection should all experience the same level of traffic congestion during any period of time. However, in scenarios where a small number of road segments is congested, Abs-Bal is expected to produce an additional overhead in terms of travel time and travel distance; this occurs without Abs-Bal obtaining the benefits in terms of smoothing the traffic over the road network. In general, we propose two applications of the Abs-Bal and Bal-Traf mechanisms, in which each is more suited to a certain condition of the road network. Bal-Traf should work well when the road network is partially congested; meanwhile, Abs-Bal is the best option to run in the event that the entire road network is highly congested. Experimental scenarios have been presented in this paper to evaluate the performance of these mechanisms for different road network scenarios. The bottleneck problem creates hazardous areas that negatively affect drivers safety and causes extra harmful gas emissions in the atmosphere. Thus, the goal of the traveling authorities is to eliminate this problem in order to guarantee a smooth and safe environment to drive. The most suitable path recommendation algorithm for the road network situation should be selected and forced by authorities; then, impatient drivers that create bottlenecks in order to arrive at their destinations a few minutes faster will be forced to follow the rules. The remainder of this paper is organized as follows. In Section 2, we briefly present the previously proposed path recommendation protocols that use VANETs. The phases of typical distributed path recommendation protocols are presented in Section 3. In Section 4, we explain how to detect overloaded road segments and how to rank the output road segments at each road intersection according to the travel time parameter. The details of Bal-Traf and Abs- Bal mechanisms are proposed in Section 5. Section 6 illustrates the performance evaluations of Bal-Traf and Abs-Bal mechanisms; these are compared with a typical distributed path recommendation protocol known as intelligent path recommendation protocol (ICOD) [8]; the purpose of this section is to illustrate the advantages and drawbacks of each mechanism in different scenarios. Finally, Section 7 concludes the paper. 2. RELATED WORK Several research studies have investigated the best path toward each targeted destination in downtown and urban areas [6,8,10 14,20,21,24]. These studies mainly recommended that traveling vehicles avoid highly congested road segments; these studies also intended to find the best alternative path toward each targeted destination. Efficient path recommendation protocols enhance traffic fluency and then decrease the travel time of vehicles trips. As explained before, the previously proposed path recommendation protocols in VANETs can be classified into two main categories: centralized and distributed path recommendation protocols. Most of the centralized path recommendation protocols provide each vehicle with a static path toward their targeted destinations: each vehicle acquires the information pertaining to the entire path before it starts moving, and the configured path cannot be changed during the trip of the vehicle [1,3,4,12,22,23]. Decisions relating to change of direction at each road intersection Wirel. Commun. Mob. Comput. 2016; 16:794 809 2015 John Wiley & Sons, Ltd. 795

Traffic balancing M. Bani Younes, A. Boukerche and G. Román-Alonso are made based on recommendations acquired from the centralized processor. The aforementioned processor gathers the basic traffic data of all road segments over the area of interest and the location information of the targeted destinations [1,6,13]. Other dynamic centralized path recommendation protocols have used moving vehicles as mobile sensors to gather data on traffic conditions of paths traversed. Traveling vehicles report their travel history during their trips or at the end [11]. Each time the central processor receives the traffic characteristics of road segments traversed, it looks for the best path for traveling vehicles, which is confirmed by the latest collected traffic data [2,5,10,11,14,25,26]. The path of each vehicle can be changed during the vehicle trip, according to the most recently gathered traffic data [2]. However, the dynamic path recommendation protocols of the centralized category require high communication overhead in terms of the bandwidth consumptions and end-to-end delay metrics. Moreover, centralized dynamic path recommendations are not practical in extended road network environments. The second category in this field (i.e., distributed path recommendation protocols) produces more real-time and more accurate traffic congestion avoidance recommendations. The path leading toward each destination is recommended through cooperative communication among a set of distributed processors (i.e., RSUs); these are scattered throughout the area of interest [8,9,15,27]. Our previously proposed path recommendation protocol (ICOD) [9] provides vehicles with accurate information concerning recommendations for the least congested path leading toward the respective destinations, in a dynamic and distributed manner. Moreover, the ICOD protocol flexibly considers the existence of common targeted public services and the conditions of road segments while selecting the desired path. The path leading toward each destination can be changed at any road intersection because of the real-time traffic situation of the investigated road network. However, these protocols recommend that all vehicles target the same destination and arrive at the same road intersection in order to follow the same path toward the destination. A large number of vehicles targeting the same destination, or even destinations located closely together, is common within downtown areas. In these scenarios, the path recommendation protocol will generate and/or exaggerate the bottleneck problem in some road segments over the road network. Indeed, previously proposed path recommendation protocols recommend the best path toward each targeted destination, without considering the volume of generated traffic at the output road segments of road intersections. In this case, the path recommendation protocol may exaggerate the bottleneck problem over the road network instead of solving the traffic congestion problem. As a consequence, in this paper, we propose two Bal-Traf mechanisms, Bal-Traf and Abs-Bal, primarily to address the aforementioned problem in existing path recommendation protocols. These proposed mechanisms are designed to efficiently distribute the traffic over the road network toward located destinations. Bal-Traf is designed to eliminate the estimated overwhelmed road segment scenarios in road networks. Abs-Bal is designed to absolutely balance the traffic among the located road segments over the road network; it is intended to do so in scenarios where the traffic densities of all located output road segments are required to be the same for each road intersection. 3. DISTRIBUTED PATH RECOMMENDATION PROTOCOLS In this section, for comprehensiveness, we present the phases of the typical distributed path recommendation protocol. These phases are used to find the best path for each vehicle leading toward its targeted destination in a hopby-hop fashion. Real-time traffic distribution is the main factor used to select the optimal path toward each destination. In general, the more accurate the gathered traffic characteristics are, the more efficient the constructed paths become. The phases of gathering traffic characteristics and constructing the best path in a hop-by-hop fashion are presented in this section. 3.1. Traffic evaluation of located road segments In this phase, the traffic characteristics of each road segment are evaluated separately, over the road network. Each vehicle periodically broadcasts the basic traffic data including its speed, location, direction, and the targeted destination. Traveling vehicles on each road segment gather the basic traffic data of surrounding vehicles (i.e., located within the transmission range of the vehicle). The basic traffic data of traveling vehicles at each road segment is used to generate a real-time traffic report for such a road segment. For the scenarios in which the transmission range of traveling vehicles does not cover the entire road segment, the road segment is divided into a set of virtual adjacent clusters, and the traffic characteristics of each cluster are evaluated separately. Communication among traveling vehicles in these clusters helps to cooperatively generate a traffic report that summarizes the traffic characteristics of the entire road segment [16]. Moreover, the traffic characteristics should be evaluated to consider each traffic direction separately. Figure 1 illustrates the input and output flows of two adjacent road intersections, with an RSU installed at each (A, B). As we can see from Figure 1, each road intersection has four input flows and four output traffic flows. In this phase, the RSU at each road intersection gathers the destination reports from all input flows and the traffic characteristic reports of each output flow. An RSU is expected at each road intersection over the downtown grid-layout area. In general, each road segment starts from a certain road intersection and ends at the next road intersection. This means that an RSU exists at the starting point of such a road segment, while another RSU 796 Wirel. Commun. Mob. Comput. 2016; 16:794 809 2015 John Wiley & Sons, Ltd.

M. Bani Younes, A. Boukerche and G. Román-Alonso Traffic balancing Figure 1. Input and output traffic flows at road intersections. is installed at the end point as well. The starting and ending points are assigned based on the direction of traffic over the road segment. For example, in Figure 1, RSU (A) is located at the starting point of direction (D1), and RSU (B) is located at the end point. However, B is located at the starting point of D2, and A is located at the end point. The RSU at each road intersection is responsible for ranking the output road segments as an option toward each targeted destination (D k ). Ranking output road segments at each road intersection depends on the following: (i) the length of each road segment; (ii) the average speed driven on such a road segment; and (iii) the location of that road segment in respect to D k. At the same time, these RSUs gather destination reports from traveling vehicles at each input road segment. The destination reports categorize the traffic of the road segment based on the locations of targeted destinations for traveling vehicles. Thus, an appropriately located RSU can predict the best next hop to take toward each destination based on the local gathered information. It then sends a recommendation message to traveling vehicles; the details involved in selecting the best direction at each road intersection are explained in Section 3.2. 3.2. Constructing the best path toward each destination The distributed path recommendation protocols construct the path toward each destination in a hop-by-hop fashion. At each road intersection, recommendations are made to each traveling vehicle for the most efficient turn option toward their targeted destinations. The path construction process starts at the location of each destination and runs toward each road intersection within the road network. RSUs are installed at road intersections throughout the road network; each RSU gathers information concerning the traffic characteristics of the surrounding road segments, as illustrated in Section 3.1. Real-time communication among these RSUs then construct the best path toward each targeted destination. This is done while considering the real-time traffic characteristics of each road segment over the entire road network [8]. Each targeted destination (D i ) periodically broadcasts an advertisement (ADV i ) message, mainly declaring the following: its location, a time stamp for the broadcast message, the required travel time, and the required travel distance to reach D i. The last two fields in ADV i are set to zero when the message is sent by the original destination. The closest RSU to each destination updates and forwards the ADV i message to the neighboring RSUs over the network. The receiver RSUs should then update and forward these messages to the neighboring RSUs in only one of the following two cases: (a) when the ADV i message, originating from the destination D i, is received by the RSU in question for the first time (i.e., this RSU has never received any previous messages related to D i ); (b) when the received ADV i message recommends a better path to take toward D i, in terms of travel time and/or distance. These ADV messages are updated by adding the required travel time and travel distance of the road segment that connects between the sender RSU and the receiver RSU to the last two fields in the ADV i message. Otherwise, the ADV i message should be dropped instead of being forwarded to the next RSUs and consuming the bandwidth of the communication network. This protocol allows all located RSUs within the investigated area to obtain the best path toward each destination. Communication among RSUs updates the database of each RSU to construct the best path in terms of travel time, distance, or fuel consumption metrics, to mention a few elements. The distributed path recommendation protocol considers more real-time traffic characteristics of located road segments over the road network. This happens by forwarding the updated ADV i messages, which reflect the most accurate data gathered concerning road segments connected at these intersections. Wirel. Commun. Mob. Comput. 2016; 16:794 809 2015 John Wiley & Sons, Ltd. 797

Traffic balancing M. Bani Younes, A. Boukerche and G. Román-Alonso 4. OVERLOADED ROAD SEGMENTS AND ALTERNATIVE OPTIONS Traffic volume and traffic density parameters, compared with the road segment capacity, are the main factors used to detect highly congested (i.e., overloaded) road segments. In this section, we first explain the technique used to detect overloaded road segments. We then explain how to rank the output road segments at each road intersection that leads toward the targeted destinations, keeping in mind the parameter of travel time. 4.1. Detecting overloaded road segments An overloaded road segment can be caused or exaggerated by the path recommendation protocol in use. This occurs when vehicles within the road network are directed by the recommendation protocol to take the same path to a destination. In this section, we first describe how to detect the overloaded road segments that already exist within the road network. Then, we investigate the scenarios in which the path recommendation protocol being used causes overloaded road segments, and how to predict these overloaded output road segment scenarios. 4.1.1. Input overloaded road segments. Some road segments are overwhelmed by a large number of traveling vehicles. These roads experience high traffic density that exceeds their capacity; this causes low traffic speed. Traveling vehicles cannot proceed at the highest allowed speed on such road segments, because of the high traffic congestion status. In order to detect overloaded road segments, we investigate the saturation density (Sd i ) parameter of each road segment (i). The saturation density (Sd i ) is defined in this work as the maximum traffic density that can travel smoothly over each road segment. According to [17], the optimum density of US freeways is described as ranging [0.016 0.022] vehicle per meter per lane. When the optimum density occurs, the maximum traffic flow can proceed on the road at the maximum allowed speed [17]. On the other hand, the jam density is described as ranging [0.1 0.136] vehicle per meter per lane, according to [17]. In this case, extreme traffic density is associated with a completely stopped traffic flow. The interval of saturation density (Sd i )isdefinedin between the optimum density and the traffic jam density intervals; saturation density ranges [0.068 0.076] vehicle per meter per lane. In order to detect overloaded road segments, the current density (d i ) of each input road segment is compared with the saturation traffic density (Sd i ) of such a road segment. In the event that d i is higher than Sd i, the road segment i is defined as an overloaded road segment. 4.1.2. Output overloaded road segments. In some scenarios, congested road status is caused or generated by the path recommendation protocols. For example, in distributed path recommendation protocols, the RSUs located at each road intersection recommend that most of the traveling vehicles moving toward the intersection (i.e., not coming from the same road segment input) take the same output road segment. In this case, the overloaded road segment scenario is generated by the path recommendation protocol in use. In order to predict the overloaded output road segments, each RSU counts the number of vehicles that are supposed to travel toward each output of the road intersection. RSUs can count the number of vehicles at each output road segment according to the number of vehicles targeted at each existing destination, and the best output road segment is assigned to each destination. In the event that the estimated density (d i ) of any output road segment is heavier than the saturation density (Sd i ), of that particular road segment, the output road segment is predicted to be an overloaded output road segment. The proposed mechanisms in Section 5 are introduced to handle this problem. Directing traveling vehicles to other less congested output road segments helps decrease the effect of the congested output road segments. 4.2. Ranking the output road segments leading toward each destination RSUs located on road intersections compute the travel time associated with all next hop options leading toward each destination. Travel time is computed from the destination location to the road intersection location, and the same is done in the distributed path recommendation protocols in Section 3.2. From its internal database, the RSU in question records the cost of all candidate next hop options leading toward each destination. Three next hop options leading toward each destination are investigated for each input traffic flow, at each RSU: right turn (RT), Figure 2. Turn options toward each destination. LT, left turn; RT, right turn; DT; direct move. 798 Wirel. Commun. Mob. Comput. 2016; 16:794 809 2015 John Wiley & Sons, Ltd.

M. Bani Younes, A. Boukerche and G. Román-Alonso Traffic balancing direct move (DT), and left turn (LT). Figure 2 illustrates these options for any targeted vehicle destination at any road intersection. The RSU located at each road intersection ranks the output options leading toward each destination beginning with the option requiring the least amount of travel time and ending with the option that results in the longest travel time:(op 1, Op 2,andOp 3 ). The typical path recommendation protocols usually recommend that all vehicles travel through Op 1 toward their destinations. Op 2 and Op 3 may be recommended for some traveling vehicles for the purpose of eliminating predicted overwhelmed road segment scenarios generated by Op 1. Ranking turn options toward each destination facilitate the recommendation of the next best option; this is the case when the ideal turn is not applicable or it has generated overwhelmed road segment scenarios. 5. BALANCING TRAFFIC PATH RECOMMENDATION MECHANISMS In this section, we present the details for the proposed Bal-Traf based mechanisms (Bal-Traf and Abs-Bal). These mechanisms are designed to eliminate bottleneck problems and highly congested road segment scenarios; these factors are generated as a consequence of the path recommendation protocol in use. Bal-Traf and Abs-Bal both consider the generated traffic volume at output road segments in each road intersection. Bal-Traf first predicts the estimated overloaded output road segment, as explained in Section 4.1.2. Then, it distributes the estimated traffic of that overloaded road segment among other existing output road segments at each road intersection. Bal-Traf reactively helps to eliminate the bottleneck problem as it distributes the traffic only if it detects an overloaded scenario; otherwise, vehicles travel along the best option toward their targeted destinations. On the other hand, Abs-Bal recommends that vehicles leave the road intersection in an absolutely balanced manner (i.e., the same density is recommended for each output road segment). Abs-Bal proactively eliminates overloaded road segment scenarios, unless all output road segments are overloaded. The incoming traffic at each road intersection is distributed evenly among the outgoing traffic flows leading toward each targeted destination. This mechanism is intended to avoid generating drastically overloaded road segment scenarios. The details for these mechanisms are introduced in the rest of this section. 5.1. Balanced traffic path recommendation mechanism (Bal-Traf) In this mechanism, each RSU responds by running the Bal-Traf algorithm only if it is predicted that any of the outputting road segments is an overloaded road segment. The outgoing traffic flow is considered to be overloaded if the estimated density of traveling vehicles (Td i )atthatflowis more than Sd i ; this is explained in Section 4.1. First, if the Wirel. Commun. Mob. Comput. 2016; 16:794 809 2015 John Wiley & Sons, Ltd. 799

Traffic balancing M. Bani Younes, A. Boukerche and G. Román-Alonso outgoing flow i is detected to be an overloaded flow, the percentage (O per )ofi that is overloaded is computed using Equation 1. O per D Td i Sd i Td i (1) The RSU located at the road intersection where the overwhelmed output road segment (Or i ) is detected checks the list of destinations for which Or i is the best next hop. For each destination on this list, the difference between the cost of the best next hop option (Op 1 ) and the second option (Op 2 ), understood as the cost(op 2 ) cost(op 1 ), is computed. The cost of Op 1 is the required travel time from the road intersection to the destination (D k ) if the output Op 1 is taken as a next hop. However, cost(op 2 )isthe required travel time toward D k ; this is the case where Op 2 is the next-hop option to be followed at the road intersection. The recommendation is made for vehicles traveling toward this destination (D i ) should travel through Op 2 toward their targeted destinations only if Op 2 is not overloaded. In the case that Op 2 is an overloaded road segment, Op 2 is omitted and replaced by Op 3, and the traveling expenses for D k are incurred. The process of finding the destination to obtain the minimum value of cost(op 2 ) cost(op 1 ) is executed again in this scenario. The destination D k is found which has the lowest difference between the cost of Op 1 and Op 2, understood as Min(cost(Op 2 ) cost(op 1 )). The latter metric has been used to obtain the minimum average travel time for vehicles, while considering the Bal-Traf metric and traffic volume characteristics over the road network. The number of vehicles that are deported to travel through Op 2 or Op 3 should be configured to compute their density (Td Dk ) when taking the Op 2 or Op 3 road segment. For simplicity, we assume that all road segment flows have the same area A i. In the event that Td Di is more than (O per Td i ), then only.o per Td i / A i of these vehicles will be directed to choose the Op 2 ;.Td Dk.O per Td i // A i of the vehicles traveling toward D k will, on the other hand, be directed to pass by the Op 1 : the estimated overloaded road segment..td Dk.O per Td i //A i vehicles can travel through Op 1 without generating overloaded road segment scenarios. On the other hand, if Td Di is less than.o per Td i /,thersubeginstosearchfor the next destination D j ; the RSU will do this to recommend vehicles traveling toward this destination or some of them to travel through Op 2 to eliminate the overloaded status of Or i. However, sometimes eliminating the traffic congestion in one output road segment may cause a traffic congestion scenario in another output road segment. To address this problem, the evaluating and balancing algorithm should be run repeatedly until there are no longer any overloaded output road segments. Algorithm 1 illustrates the systematic procedure (Bal-Traf) used for eliminating the bottleneck and congestion problems in distributed path recommendation protocols. In the scenario in which the investigated road network is partially congested, Bal-Traf can eliminate the bottleneck problem and the case of highly congested road segments. However, in scenarios where the entire road network is congested, or where most of the road segments over the road network are congested, Bal-Traf cannot be expected to completely eliminate the bottleneck problem. This is due to the fact that the entire road network is overloaded, which definitely leads to the generation of overloaded output road segments at some road intersections. However, Bal-Traf decreases the total number of expected overloaded output road segments. 800 Wirel. Commun. Mob. Comput. 2016; 16:794 809 2015 John Wiley & Sons, Ltd.

M. Bani Younes, A. Boukerche and G. Román-Alonso Traffic balancing 5.2. Absolutely traffic balanced based path recommendation mechanism (Abs-Bal) This mechanism aims to keep the traffic throughout the road network absolutely balanced among all existing output road segments. At each road intersection, the respective RSU recommends that the arriving traveling vehicles leave the road intersection in an evenly balanced distribution. The RSU computes the traffic density of each output road segment of the road intersection: Td 1, Td 2, Td 3,:::Td n. The RSU then computes the average among all densities of the output road segments (Ad), which is done using Equation 2. Ad D Td 1 C Td 2 C Td 3 C :::C Td n O n (2) where O n is the number of output road segments at each road intersection. The default consideration of a typical road intersection scenario is four output road segments. The RSU recommends that some vehicles travel on the second (Op 2 )orthird(op 3 ) options toward the targeted destination. Abs-Bal is designed to keep the traffic density of all output road segments the same as the Ad value, regardless of the overloaded situation of the road segments. First, the RSU computes the difference between travel time required through Op 1 and Op 2, dif 1 Dk, toward each destination D k. It also computes the difference between the travel time that is required through Op 1 and Op 3, known as dif 2 Dk. Equations 3 and 4, respectively, illustrate how to compute dif 1 Dk and dif 2 Dk at each road intersection leading toward the destination D k.thecost(op 1 )isthe required travel time toward D k ;thisisthecaseifop 1 is the next hop at the respective road intersection. dif 1 Dk D Cost.Op 2 / Cost.Op 1 / (3) dif 2 Dk D Cost.Op 3 / Cost.Op 1 / (4) Second, the RSU sequentially checks the Td i of each output road segment. In the event that Td i of the output road segment i is more Ad, the RSU checks the list of destinations for which the respective road segment is recommended as the best next hop output road segment. The RSU selects the destination D j which has the lowest value of dif 1 Dj among all these destinations. It checks the traffic density of the second option (Op 2 ) road segment toward the D j,(td Op2 ). If Td Op2 is more than Ad, the RSU should remove dif 1 Dj from the comparison list, and it should add dif 2 Dj. It then iteratively finds the minimum value among the listed dif 1anddif 2 and compares the Td of the alternative option with the computed Ad value. The first detected alternative option (i.e., the option with minimum extra travel time overhead) that has a Td less than Ad is recommended as a next hop on which to travel toward the selected destination (D k ). Finally, in the event that the density of arriving vehicles traveling toward D k (Td Dk ) is greater than (Td i Ad), only.td i Ad/ A i of traveling vehicles toward D k are recommended to take the second option Op 2 ; the other vehicles, proceed on Op 1 (i.e., road segment i) towardd k, because the density of Op 1 should be Ad as well. On the other hand, if Td Dk is less than (Td i Ad), the RSU should select more destinations to which traveling vehicles can be deported, so they take the second or third options toward these destinations until the density of Op 1 is equal to Ad. The systematic explanation of the Abs-Bal mechanism is illustrated in Algorithm 2. 6. PERFORMANCE EVALUATION The performance of the proposed Bal-Traf mechanisms is evaluated in this section. All experiments were executed when traveling vehicles were moving toward one of the three defined destinations (A, B, and C); these are located in a grid layout scenario of a downtown area, as illustrated in Figure 3. We consider this scenario to be a simplified study that represents an extended situation in which more destinations are defined over the grid layout. Three different sets of experiments were conducted to investigate the benefits and overheads of Bal-Traf and Abs-Bal mechanisms. All experiments were implemented using NS 2 [18], and the results are introduced in this section. In the first set of experiments, different levels of traffic congestion were generated over half the located road networks. In the second set, a different number of road segments were considered to be congested in each experiment. Finally, the last set of experiments aimed to investigate the impact of the size of the downtown area using a medium level of traffic congestion. The parameters used in these experiments are illustrated in Table I. We used simulation of urban mobility (SUMO) [19] to generate the mobility scenarios. It configures random destinations on the map and selects random paths toward each destination, where the path stays inside the provided map, and the travel speed of vehicles does not exceed the maximum limit speed of each road segment. It presents a map-based mobility model that considers the provided map, where vertices of the map are possible destinations and the connected roads on the mapareallowedpaths. 6.1. Comparative performance of Bal-Traf and Abs-Bal to other path recommendation protocols In this section, we simulate different traffic scenarios in grid layout road networks, as shown in Figure 3. The generated scenarios can be ordered into five main categories: No congestion, Low, Medium, High, and Mixed congestion scenarios. In the No congestion scenario, the densities of all road segments over the road networks are within the optimal density range. The traffic speed of all road Wirel. Commun. Mob. Comput. 2016; 16:794 809 2015 John Wiley & Sons, Ltd. 801

Traffic balancing M. Bani Younes, A. Boukerche and G. Román-Alonso Figure 3. Grid-layout scenario with three destinations. Table I. Simulation parameters. Parameters Value Simulation area (m m) 1000 1000 10, 000 10, 000 Wireless medium IEEE802.11 No. of RSUs 16, 25, 36, 49, 64 No. of vehicles 200 2000 No. of destinations 3 Mobility model Map-based mobility Transmission range (m) 250 Map layout Grid-layout: 4 4, 5 5, 6 6, 7 7, 8 8 No. of road segments 40 160 bidirectional Simulation time (s) 5000 segments is close to or equal to the maximum allowed traffic speed. For the other generated scenarios (i.e., Low, Medium, High, and Mixed), half of the road segments on the road networks are in the optimal density range. The densities of the other half of the roads in the Low scenario are equal to saturation density. In the Medium scenario, the densities of the other half of the road network are greater than the saturated density range and less than the jam density range. The densities of the other half in the High scenario are in the jam density range. Finally, in the Mixed scenario, the densities of the second half of the road network are randomly set among the optimal density and the jam density. Moreover, greater congestion in each road segment leads to a reduction in traffic speed. 802 Wirel. Commun. Mob. Comput. 2016; 16:794 809 2015 John Wiley & Sons, Ltd.

M. Bani Younes, A. Boukerche and G. Román-Alonso Traffic balancing The performance of Bal-Traf and Abs-Bal is investigated for the defined scenarios. We compared these mechanisms with the typical distributed path recommendation protocol (ICOD) [8]; this was done in respect to bandwidth consumption, end-to-end delay, vehicle travel time and distance toward each destination, the number of overloaded output flows generated, and the average traffic density of each output road segment. Figure 4 illustrates the investigated parameters of Bal-Traf and Abs-Bal mechanisms compared with the typical distributed path recommendation protocol (ICOD) [8]. Each experiment was executed for 30 different scenarios. Figure 4 sketches the average of these experiments and the obtained interval of each experiment; the confidence interval of each experiment is 95%. The same number of packets are transmitted among located RSUs for each road network scenario. However, the size of each packet in Bal-Traf and Abs-Bal is larger than the transmitted packet in ICOD. This is because in the Bal-Traf and Abs-Bal mechanisms, each RSU sends the cost of all options toward each destination to its neighboring RSUs, unlike those in ICOD; with the latter, only the cost of the best option is transferred in each packet. Because of the packet size of each mechanism, as Figure 4. The comparative performance of balance trafficking (Bal-Traf) and absolute balancing traffic (Abs-Bal) mechanisms with ICOD [8]: (a) bandwidth consumption; (b) end-to-end delay; (c) average travel time toward each destination; (d) the average travel distance toward each destination; (e) the number of overwhelmed road segments; and (f) the average density of output road segments at each road intersection. Wirel. Commun. Mob. Comput. 2016; 16:794 809 2015 John Wiley & Sons, Ltd. 803

Traffic balancing M. Bani Younes, A. Boukerche and G. Román-Alonso illustrated in Figure 4(a), Bal-Traf and Abs-Bal both consume 30% more bandwidth than ICOD. Regarding the end-to-end delay, a short additional delay is required at each RSU to run the balancing mechanisms in each RSU. Thus, Bal-Traf and Abs-Bal require extra time compared with ICOD in order to obtain the path from each road intersection toward the targeted destination. From Figure 4(b), Bal-Traf consumes 2% more delay than ICOD, while Abs-Bal consumes 3% extra delay than ICOD. As we can see from Figure 4(c) and (d), Bal-Traf and Abs-Bal incur additional travel time and distance compared with ICOD. In general, Bal-Traf requires 5% more travel time than ICOD, while Abs-Bal requires on average 40% more time to travel in the same scenarios. Compared with ICOD, Bal-Traf requires an additional 10% travel distance, while Abs-Bal requires an additional 50%. As the network becomes more congested, the travel time and distance of these Bal-Traf schemes increase. This is due to the fact that a larger number of vehicles must choose a longer trip with more traveling time in order to ensure the road segments of the best path are less overloaded and to resolve the bottleneck problem. In low congestion road network scenarios, the performance of the Bal-Traf mechanism is very similar to the performance of ICOD. This fact is justified by the low number of output road segments as having detected an overloaded status; thus, no vehicles are required to change their path. On the other hand, Abs-Bal produces on average an extra 40% travel time and distance compared with ICOD, even in low congestion scenarios. This overhead is justified by the behavior of Abs-Bal, which distributes the traffic regardless of the existence of overloaded road segment conditions. From Figure 4(e), almost 40% of the road segments within the road network were detected as being overloaded when using ICOD. On the other hand, the Bal-Traf mechanism decreases the overloaded output road segments over the road network by an average of 70% compared with ICOD. The Abs-Bal mechanism also performs well in terms of its ability to decrease the number of overloaded road segments compared with ICOD. Bal-Traf achieves, on average, 5% better performance in terms of decreasing the number of overloaded road segments compared with Abs-Bal. However, in highly congested road network scenarios, Bal-Traf drastically increases the traffic density of some output road segments in order to eliminate other overloaded road segments. Finally, the traffic density of each output road segment decreases by 10% on average when the Bal-Traf mechanism is used compared with ICOD, as illustrated in Figure 4(f). Abs-Bal achieves the best performance in terms of decreasing and balancing the traffic density among the existing output road segments at each road intersection. On average, the Abs-Bal mechanism decreases the traffic density of the output road segments by 30% compared with ICOD. 6.2. Impact of the percentage of congestion over the road network In this section, we evaluate the performance of Bal-Traf and Abs-Bal mechanisms compared with ICOD [8] for different percentages of congested road network scenarios. We have generated different sets of scenarios where 0%, 25%, 50%, 75%, and 100% of the input road segments are congested. We set the traffic density and the estimated travel time of these congested road segments by the medium traffic congestion level, as defined earlier. The bandwidth consumption of Bal-Traf and the bandwidth consumption of Abs-Bal are the same, as appear in Figure 5(a). This is because the same number of packets are sent in each scenario, and the packet size of these mechanisms is the same. However, the bandwidth consumption of ICOD for the different congestion percentages is, on average, 30% less than the bandwidth consumptions of Bal-Traf and Abs-Bal, as expected. As illustrated in Figure 5(b), the delay of constructing the path toward each destination is slightly increased when the percentage of the congested network increases. This is true as long as the percentage of congested road segments is less than or equal to 75%. In these scenarios, RSUs consume more delay time looking for less congested road segments to recommend for vehicles. However, in the event that all road segments are congested on the road network (i.e., 100% of the road segments are congested), RSUs do not require extra time to find the path toward each destination, because all road segments have close congestion status. Thus, the delay in this scenario is the same as the delay when there is no congestion on the road network (i.e., 0% of the road network is congested). As we can see from Figure 5(c), the average travel time of vehicles toward each destination increases for all investigated mechanisms when the traffic congestion percentage is increased over the road network. In the case where the entire road network (i.e., 100% of the road network) is congested, the travel time increases drastically compared with other scenarios. This is justified by the fact that all vehicles need to travel through the highly congested road segments. In other scenarios where 25%, 50%, or 75% of the road network is not congested, most vehicles are directed to travel through these non-congested road segments. Abs- Bal requires the longest travel time, while ICOD requires the shortest time in all investigated scenarios. Regarding the travel distance of these mechanisms, Abs- Bal recommends the longest path while ICOD requires the shortest path in all investigated scenarios, as illustrated in Figure 5(d). The same travel time is required by the Abs-Bal scheme for the scenario in which there is no overloaded road segments within the road network and one in which all road segments are overloaded. The same result was observed for the ICOD protocol. This is justified by the fact that both Abs-Bal and ICOD recommend the exact same path for traveling vehicles in these scenarios, because all road segments have almost the same relative traffic status. However, Bal-Traf performs 804 Wirel. Commun. Mob. Comput. 2016; 16:794 809 2015 John Wiley & Sons, Ltd.

M. Bani Younes, A. Boukerche and G. Román-Alonso Traffic balancing Figure 5. The performance of balance trafficking (Bal-Traf) and absolute balancing traffic (Abs-Bal) mechanisms compared with ICOD [8] for different percentages of congested road network scenarios: (a) bandwidth consumption; (b) end-to-end delay; (c) the average travel time toward each destination; (d) the average travel distance toward each destination; (e) the number of overwhelmed road segments; and (f) the average density of output road segments at each road intersection. differently in a scenario in which all road segments are overloaded, compared with a scenario in which all road segments have optimal traffic density. This is due to the large number of overloaded output road segments detected in the scenario in which the entire road network is congested. Bal-Traf tries to eliminate the number of overloaded road segments; this produces different paths than paths produced in the case where none of the road segments are congested. Bal-Traf achieves the best performance in terms of its ability to decrease the number of overloaded output road segments, as shown in Figure 5(e). Bal-Traf decreases the number of overloaded output road segments by 10% compared with Abs-Bal, and by 50% compared with ICOD. Finally, Abs-Bal achieves the best performance in terms of decreasing the traffic density of each output road segment for all congestion scenario percentages, as illustrated in Figure 5(f). 6.3. Impact of the size of the road network In this section, we investigate the effects of road network sizes on the performance of the Bal-Traf and the Abs-Bal Bal-Traf mechanisms. In our experiments we have used a Wirel. Commun. Mob. Comput. 2016; 16:794 809 2015 John Wiley & Sons, Ltd. 805

Traffic balancing M. Bani Younes, A. Boukerche and G. Román-Alonso grid layout road network environment, in which the entire area ranges 1000 1000 m 2 10000 10000 m 2.However, traveling vehicles move only within existing road segments. Different number of road intersections are simulated in our road network map: 44, 55, 66, 77, and 8 8 road intersections. Half of the road segments existing over these road networks have the optimal traffic density status; the other half of the located road segments have a medium traffic congestion status. Increased bandwidth consumption coincides with an increase in the size of the investigated area. As we can infer from Figure 6(a), Bal-Traf and Abs-Bal both consume on average 45% more bandwidth consumption compared with ICOD because of the previously mentioned larger packet size in these mechanisms. This is for different road network sizes and different numbers of intersections on the road network. Abs-Bal requires the maximum delay time to give path recommendations to vehicles for their target destinations, Figure 6. The impact of the road network sizes on the performance of balance trafficking (Bal-Traf), absolute balancing traffic (Abs-Bal) schemes and ICOD protocol [8]: (a) bandwidth consumption; (b) end-to-end delay; (c) the average travel time toward each destination; (d) the average travel distance toward each destination; (e) the number of overwhelmed road segments; and (f) the average density of output road segments at each road intersection. 806 Wirel. Commun. Mob. Comput. 2016; 16:794 809 2015 John Wiley & Sons, Ltd.

M. Bani Younes, A. Boukerche and G. Román-Alonso Traffic balancing as shown in Figure 6(b). Bal-Traf requires on average 1% less delay time than Abs-Bal and 2% more delay than ICOD. As expected, the average travel time and average travel distance increased when the road network size for all investigated schemes was increased, as shown in Figure 6(c) and (d). Some interesting results have been inferred from these figures: the performance of Abs-Bal in terms of travel time and travel distance is closer to the performance of ICOD in large road networks (i.e., 88) than in small road networks (i.e., 4 4). However, the road network size does not have an impact on the performance of Bal-Traf or Abs-Bal in terms of the number of overloaded road segments or the traffic density of output road segments. These results are illustrated graphically in Figure 6(e) and (f), respectively. 7. CONCLUSION In this paper, we have proposed two Bal-Traf based mechanisms (Bal-Traf and Abs-Bal), for distributed path recommendation protocols in VANETs. These mechanisms are mainly intended to resolve the bottleneck problem and to enhance the smoothness mobility over the road network. The Bal-Traf mechanism detects and eliminates the expected overwhelmed output road segments at each road intersection. Typical path recommendation protocols recommend the best next hop toward each targeted destination with the least required travel time. However, Bal-Traf considers the traffic load of each output road segment; it does so while recommending the next hop for vehicles traveling toward their destinations. Some vehicles need to travel extra time and/or distance to avoid causing traffic congestion over the road network. As we can infer from the experimental results, the overwhelmed output road segments at each road segment decrease when using Bal-Traf by 70% compared with ICOD. However, the travel time increases by 10% and the travel distance increases by 15% on average compared with ICOD. On the other hand, the Abs-Bal mechanism is designed to evenly distribute the input traffic of each road intersection among the output road segments. The Abs-Bal mechanism performs the best in decreasing the traffic density of each output road segment over the road network. The traffic density of each output road segment while using Abs-Bal is, on average, 40% less than the traffic density of output road segments achieved using ICOD. However, the required travel time and travel distance of Abs-Bal is 40% and 60% greater than when using the ICOD protocol. ACKNOWLEDGEMENTS This work is partially supported by NSERC DIVA Strategic Network, Canada Research Chairs Program, and MRI/OIT Research funds. REFERENCES 1. Inoue S, Shozaki K, Kakuda Y. An automobile control method for alleviation of traffic congestions using inter-vehicle ad hoc communication in latticelike roads. In Globecom Workshops, 2007 IEEE, New Orleans, LA, Nov. 2007; 1 6, 26 30. 2. Lakas A, Cheqfah M. Detection and dissipation of road traffic congestion using vehicular communication. In Proc. of IEEE, Mediterrannean Microwave Symposium, Tangiers, November 2009. 3. Doolan R, Gabriel-Miro M. VANET-enabled ecofriendly road characteristics-aware routing for vehicular traffic. In Vehicular Technology Conference (VTC Spring), 2013 IEEE 77th, Dresden, 2013; 1 5. 4. Xiang Y, Liu Z, Liu R, Sun W, Wang W. GeoSVR: a map-based stateless VANET routing. Ad Hoc Networks 11, no 2013; 7(2013): 2125 2135. 5. De Felice M, Cerqueira E, Melo A, Gerla M, Cuomo F, Baiocchi A. A distributed beaconless routing protocol for real-time video dissemination in multimedia VANETs. Computer Communications 2014. 6. Muázu A, Jung T, Lawal I, Shah A. A QoS approach for cluster-based routing in VANETS using TDMA scheme. In ICT Convergence (ICTC), 2013 International Conference on, Jeju, Korea, 2013; 212 217. 7. Boukerche A. Algorithms and Protocols for Wireless and Mobile Ad Hoc Networks. Wiley & Sons, 2008. 8. Younes MB, Boukerche A, Román-Alonso G. An intelligent path recommendation protocol (ICOD) for VANETs. Computer Networks 2014; 64(2014): 225 242. 9. Younes MB, Boukerche A. Performance evaluation of a context-aware path recommendation protocol for VANETs. In IEEE Global Communications Conference, Exhibition and Industry Forum (GLOBECOM 2013), Atlanta (USA), December 2013; 9 13. 10. Morla R. Sentient future competition: vision of congestion-free road traffic and cooperating objects. Wireless Sensor Network 2006; 74: 1 8. 11. Collins K, Muntean GM. A vehicle route management solution enabled by wireless vehicular networks. In INFOCOM Workshops IEEE, Phoenix, USA, 2008; 1 6. 12. Zhang X, Hong J, Fan S, Wei Z, Cao J, Ren Y. A novel real-time traffic information system based on wireless mesh networks, Sept. 30 2007; 618 623. 13. Schunemann B, Wedel J, Radusch I. V2X-based traffic congestion recognition and avoidance. Tamkang Journal of Science and Engineering 2010; 13(1): 63 709. 14. Ou C-H, Chao C-F, Huang T-Y. Centralized trajectorybased routing protocol for rural vehicular ad hoc networks. IJCEBM 2012: 107. Wirel. Commun. Mob. Comput. 2016; 16:794 809 2015 John Wiley & Sons, Ltd. 807

Traffic balancing M. Bani Younes, A. Boukerche and G. Román-Alonso 15. Wedde HF, Lehnhoff S, Bonn B, Bay Z, Becker S, Bttcher S, Brunner C. Highly Dynamic and Adaptive Traffic Congestion Avoidance in Real-Time Inspired by Honey Bee Behavior, In Mobilitt und Echtzeit. Springer: Berlin Heidelberg, 2008, 21 31. 16. Younes MB, Boukerche A. A performance evaluation of an efficient traffic congestion detection protocol (ECODE) for intelligent transportation systems. Ad Hoc Networks 2015; 24: 317 336. 17. Lieu H. Traffic-flow Theory. Public Roads (US Dept of Transportation). (Vol. 62 No 4). (January/ February 1999). 18. Network Simulator nss-2. http://www.isi.edu/nsnam/ ns/. 19. Behrisch M, Bieker L, Erdmann J, Krajzewicz D. SUMO - simulation of urban mobility: an overview, SIMUL 2011. In The Third International Conference on Advances in System Simulation, Barcelona, Spain, October 2011; 63-68. 20. Chou LD, Li D, Chao H. Mitigation traffic congestion with virtual data sink based information dissemination in intelligent transportation system in ubiquitous and future networks (ICUFN), Dalian, 2011; 37 42. 21. Korkmaz G, Ekici E. Urban multi-hop broadcast protocol for inter-vehicle communication systems in VANET. Ekici. In Urban Multi-Hop Broadcast Protocol for Inter-Vehicle Communication Systems in VANET 04 Proceedings of the 1st ACM international workshop on Vehicular ad hoc networks, Philadelphia, Pennsylvania, 2004. 22. Nawaz A, Chan E, Li W. On scheduling data access with cooperative load balancing in vehicular ad hoc networks (VANETs). The Journal of Supercomputing 2014; 67(2): 438 468. 23. Aljeri N, Abrougui K, AlMulla M, Boukerche A. A performance evaluation of load balancing and QoSaware gateway discovery protocol for VANETs. In Advanced Information Networking and Applications Workshops (WAINA), 2013 27th International Conference, Barcelona, 2013; 90 94. 24.JuK,ChenL,WeiH,ChenK.Anefficientgatewaydiscovery algorithm with delay guarantee for VANET-3G heterogeneous networks. Wireless Personal Communications 2014: 1 18. 25. Sheu T-L, Shang Y-Y. Pipelined forwarding with energy balance in hexagonal wireless sensor networks. Wireless Communications and Mobile Computing 2014; 14(18): 1720 1731. 26. Chen J, Zhang H, Du X, Fang B, Yan L. Designing robust routing protocols to protect base stations in wireless sensor networks. Wireless Communications and Mobile Computing 2014; 14(17): 1613 1626. 27. Boukerche A, Zhang L. A performance evaluation of a pre-emptive on-demand distance vector routing protocol for mobile ad hoc networks. Wireless Communications and Mobile Computing 2004; 4(1): 99 108. AUTHORS BIOGRAPHIES Maram Bani Younes received the BSc and MSc degrees in Computer Science from University of Jordan, Amman, Jordan. She is currently working toward the PhD degree with the PARADISE Research Laboratory, University of Ottawa, Ottawa, ON, Canada. Her current research interest includes Traffic Efficiency Applications through vehicular ad hoc networks. Azzedine Boukerche is a Full Professor and holds a Canada Research Chair position at the University of Ottawa. He is the Founding Director of PARADISE Research Laboratory at Ottawa. Prior to this, he held a Faculty position at the University of North Texas, U.S.A., and he was working as a Senior Scientist at the Simulation Sciences Division, Metron Corporation located in San Diego. He was also employed as a Faculty at the School of Computer Science McGill University and taught at Polytechnic of Montreal. He spent a year at the JPL/NASA California Institute of Technology where he contributed to a project centered about the specification and verification of the software used to control interplanetary spacecraft operated by JPL/NASA Laboratory. His current research interests include wireless ad hoc and sensor networks, wireless networks, mobile and pervasive computing, wireless multimedia, QoS service provisioning, performance evaluation and modeling of large-scale distributed systems, distributed computing, large-scale distributed interactive simulation, and parallel discrete event simulation. He has published several research papers in these areas. He was the recipient of the Best Research Paper Award at IEEE/ACM PADS 97 and the recipient of the 3rd National Award for Telecommunication Software 1999 for his work on a distributed security systems on mobile phone operations, and has been nominated for the best paper award at the IEEE/ACM PADS 99, ACM MSWiM 2001, and ACM MobiWac 2006. He is a holder of an Ontario Early Research Excellence Award (previously known as Premier of Ontario Research Excellence Award), Ontario Distinguished Researcher Award, and Glinski Research Excellence Award. He is a cofounder of QShine Int l Conference on Quality of Service for Wireless/Wired Heterogeneous Networks (QShine 2004), served as a General Chair for the 8th ACM/IEEE Symposium on modeling, analysis and simulation of wireless and mobile systems, 808 Wirel. Commun. Mob. Comput. 2016; 16:794 809 2015 John Wiley & Sons, Ltd.

M. Bani Younes, A. Boukerche and G. Román-Alonso Traffic balancing and the 9th ACM/IEEE Symposium on distributed simulation and real time-application, a Program Chair for ACM Workshop on QoS and Security for Wireless and Mobile networks, ACM/IFIPS Europar 2002 Conference, IEEE/SCS Annual Simulation Symposium ANNS 2002, ACM WWW 02, IEEE MWCN 2002, IEEE/ACM MASCOTS 2002, IEEE Wireless Local Networks WLN 03 04; IEEE WMAN 04 05, ACM MSWiM 98 99, and a TPC member of numerous IEEE and ACM sponsored conferences. He served as a Guest Editor for the Journal of Parallel and Distributed Computing (JPDC) (special issue for routing for mobile ad hoc, special issue for wireless communication and mobile computing, and special issue for mobile ad hoc networking and computing), ACM/kluwer Wireless Networks and ACM/Kluwer Mobile Networks Applications, and the Journal of Wireless Communication and Mobile Computing. He serves as Vice General Chair for the 3rd IEEE Distributed Computing for Sensor Networks (DCOSS) Conference 2007, and as program cochair for Globecom 2007 Symposium on Wireless Ad Hoc and Sensor Networks. He serves as an Associate Editor for ACM/Kluwer Wireless Networks, Wiley International Journal of Wireless Communication and Mobile Computing, the Journal of Parallel and Distributed Computing, and the SCS Transactions on simulation. He also serves as a Steering Committee Chair for the ACM Modeling, Analysis and Simulation for Wireless and Mobile Systems Symposium, the ACM Workshop on Performance Evaluation of Wireless Ad hoc, Sensor, and Ubiquitous Networks and the IEEE Distributed Simulation and Real-Time Applications Symposium(DS-RT). Graciela Romaná-Alonso is a full professor at the Universidad Autónoma, Metropolitana-Iztapalapa, Mexico D.F. In 1997, she received her PhD from the Universitè de Technologie de Compiègne, France. She was an invited professor at the PARADISE Research Laboratory at the University of Ottawa, 2010 2012, Canada. Her research interests include dynamic load distribution, parallel and distributed computing systems, 3D streaming, P2P networks, and vehicular ad hoc Networks. Wirel. Commun. Mob. Comput. 2016; 16:794 809 2015 John Wiley & Sons, Ltd. 809