Exploiting Context Severity to Achieve Opportunistic Service ...

IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 63, NO. 6, JULY 2014

2901

Exploiting Context Severity to Achieve Opportunistic Service Differentiation in Vehicular Ad hoc Networks Mohammad A. Salahuddin, Student Member, IEEE, Ala Al-Fuqaha, Senior Member, IEEE, and Mohsen Guizani, Fellow, IEEE

Abstract—Wireless Access in Vehicular Environment (WAVE) dictates primitives for intelligent transportation system (ITS) applications and services. The medium access control (MAC) layer in WAVE facilitates service differentiation. This offers quality of service (QoS) by prioritizing traffic through different access categories (ACs), which are based on application-requested priorities. However, it is oblivious to network load, to delay requirements for ITS safety applications, and to the “severity” of vehicles. In this paper, we propose a novel opportunistic service differentiation (OSD) scheme as an enhancement to WAVE. We define a fuzzy inference system (FIS) to deduce a context severity metric (CSM) that relates the driving behavior of a vehicle to its environment. Our OSD traffic distribution heuristic prioritizes vehicle traffic, with respect to CSM, and accounts for network load and link layer bounds. Furthermore, the OSD scheme guarantees that ACs do not exceed maximum allowable delay for ITS applications. Our methodology entails defining and designing a linear programming (LP) model for verification and validation of the OSD scheme. We perform a comparative study of OSD-enhanced WAVE and classical WAVE analytically and in a vehicular ad hoc network (VANET) simulator. We show that both simulation and analytical results substantiate our claim of improvement in performance of OSD-enhanced WAVE over classical WAVE. Index Terms—Context severity metric (CSM), intelligent transportation system (ITS), opportunistic service differentiation (OSD), vehicular ad hoc networks (VANETs), Wireless Access in Vehicular Environment (WAVE).

I. I NTRODUCTION

I

NTELLIGENT transportation system (ITS) applications and services are envisioned to increase safety and reduce congestion and pollution [1]. ITS applications can be divided into different classes, such as safety, efficiency, convenience, and infotainment. IEEE 802.11p and IEEE 1609.x form the standard for Wireless Access in Vehicular Environment (WAVE) in ITSs. IEEE 802.11p standard facilitates dedicated short-range communication (DSRC) and governs the medium access control (MAC) and physical (PHY) layers, whereas IEEE 1609.x spans over other layers. WAVE nodes consist of roadside units (RSUs) and onboard units (OBUs). RSUs are ITS infrastructure nodes installed Manuscript received June 22, 2013; revised October 5, 2013; accepted November 16, 2013. Date of publication December 18, 2013; date of current version July 10, 2014. The review of this paper was coordinated by Dr. S. Zhong. M. A. Salahuddin and A. Al-Fuqaha are with the Department of Computer Science, Western Michigan University, Kalamazoo, MI, 49009 USA (e-mail: [email protected]; [email protected]). M. Guizani is with the Qatar University, Doha 2713, Qatar (e-mail: [email protected]). Color versions of one or more of the figures in this paper are available online at http://ieeexplore.ieee.org. Digital Object Identifier 10.1109/TVT.2013.2295299

alongside the road, for example, on traffic lights and road signs. An OBU is comprised of localization systems (e.g., Global Positioning System, inertial measurement unit, etc.), processing units, and radio transceivers that are mounted on vehicles. Vehicle-to-vehicle (V2V) and vehicle-to-infrastructure communications are dictated by IEEE 802.11p in WAVE. WAVE addresses the unique challenges in vehicular ad hoc networks (VANETs) and ITS applications, such as fastmoving nodes, multipath environment, and applications with QoS requirements [1]. The communication channel in DSRC is divided into seven subchannels, of 10 MHz each, consisting of one control channel (CCH) and six service channels (SCHs). This multichannel approach increases the efficiency of the channel and addresses a high data transfer rate in fast and mobile environments. A channel is further divided into four access categories (ACs), each with a different priority for accessing the medium. This offers low latency communication, with service differentiation for applications. The multichannel coordination mechanism is addressed in IEEE 1609.4, as a sublayer between logical link control and MAC. It employs four critical services to manage channel coordination and MAC service data unit (MSDU) data transfer [1]. The channel routing service (channel router) routes data packets to the designated channel. The user priority service selects an AC within the channel, which is based on the application-defined priority. The channel coordination and MSDU data transfer service are responsible for channel selection and MSDU data delivery, respectively [1], as shown in Fig. 1. (For details of DSRC and multichannel operation in WAVE, see [1]–[4].) In WAVE, ITS applications prioritize their application packets for one of the four ACs, i.e., based on their QoS requirement. To provide QoS, each AC is prescribed a different set of parameter values to control its channel access and contention mechanism. This is adapted from the IEEE 802.11e enhanced distributed channel access (EDCA) [2]. However, this priority to AC mapping does not account for network load, link layer bounds, and the context of a vehicle, with respect to its severity and that of its neighbors or the road infrastructure information. These shortcomings deteriorate the performance of WAVE in high-traffic/dense vehicle scenarios [5] and cause time-critical ITS applications to exceed maximum allowable delays [6]. This undermines the effectiveness of time-critical lifesaving safety applications for ITS. We propose a novel opportunistic service differentiation (OSD) scheme, which complements WAVE and overcomes these limitations. The OSD scheme extends WAVE nodes and primitives by incorporating a fuzzy inference system

0018-9545 © 2013 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.

2902

IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 63, NO. 6, JULY 2014

Fig. 1. OSD service replaces user priority service in WAVE.

(FIS) in the OBU and an OSD traffic distribution heuristic on the RSU, and by replacing the user priority service with OSD service in the IEEE 1609.4 multichannel operation sublayer, as shown in Fig. 1. The FIS infers the context severity metric (CSM) for a vehicle. The CSM is based on the integration of the driving parameters of the host vehicle and neighboring vehicles with the ideal parameters from the road infrastructure. The driving parameters can be captured by a vehicle’s speed, acceleration, and directional stability. The ideal road infrastructure parameters can include speed limits, weather-related hazards, and other environmental advisories. These are typical parameters learned over time for a given geographic region. The OSD traffic distribution heuristic on the RSU uses all the vehicles CSMs, network load, link layer bounds, and delay bounds on ACs to deduce a context-severity-based priority for each vehicle. This careful priority assignment for vehicle load ensures that ACs do not exceed delay bounds. The OSD service for a vehicle uses this context-severity-based priority to select an AC for all its traffic. This offers context-severity-aware service differentiation. Our novel OSD scheme has multiple improvements over classical WAVE. First, we use CSM that would lend WAVE critical information about its vehicles. The metric integrates vehicle driving parameters, such as speed, acceleration, and directional stability with those of its neighboring vehicles. It is also augmented with road infrastructure information to build a context for the severity of a vehicle. This metric is used to prioritize traffic based on context severity of vehicles in the network, such that vehicles with higher CSM are given better QoS with respect to delay. Second, the OSD traffic distribution heuristic leverages network load and link layer bounds to opportunistically prioritize and distribute traffic among ACs to maximize the utilization of higher priority ACs and their inherent lower delays. The traffic is systematically redistributed among ACs such that the next best QoS is provided to the vehicles. Third, the OSD

traffic distribution heuristic overcomes a major limitation of unbounded delays of WAVE [4]. We will show that the OSD traffic distribution mechanism not only improves the performance and increases the integrity of time-critical safety applications but also guarantees delay bounds on ACs. The remainder of this paper is organized as follows. In Section II, we discuss related work and compare this paper to instigate its novelty. In Section III, we give an overview of the OSD traffic distribution mechanism. In Section IV, we use linear programming (LP) to formulate the OSD traffic distribution mechanism as an optimization problem. This is not only a deterministic and reproducible approach but also serves as a benchmark for developing an efficient heuristic. A thorough analysis of the OSD traffic distribution mechanism in Section V forms the basis of our OSD traffic distribution heuristic, as given in Section VI. Sections VII and VIII illustrate our major contributions through analytical and simulation studies, and shows convergence of the analytical model with simulation results. We use EstiNet Network Simulator and Emulator [7], [8] for simulating a small-scale real-life scenario to compare the performance of classical and OSD-enhanced WAVE. Finally, we conclude in Section IX with an overview of our contributions and discussion of our future research directions. II. R ELATED W ORK Here, we briefly discuss some related research work that studied the performance of WAVE and instigate the need to improve MAC protocols in WAVE, either by devising new MAC protocols or by improving the current MAC protocol in WAVE. We then compare these studies to our OSD scheme and highlight our major contributions. Many researchers [9]–[11] have proposed new and ingenious MAC layer protocols that are more suitable, efficient, or secure for WAVE, whereas Mittag et al. [2] discuss different MAC approaches for WAVE. Based on these studies, we utilize the widely accepted and standard 802.11a MAC, which is the basis of IEEE 802.11p MAC. Therefore, we propose an enhancement in the multichannel operation sublayer atop MAC in WAVE, which provides service differentiation with respect to context severity of vehicles. Others [12], [13] have modeled IEEE 802.11a and IEEE 802.11b MAC distributed coordination functions and scrutinized their performance with respect to the throughput and delay. IEEE 802.11p MAC uses the carrier-sense multiple access with collision avoidance (CSMA/CA) mechanism of IEEE 802.11a and service differentiation EDCA from IEEE 802.11e. Wu et al. [14] and Huang and Liao [15] scrutinize QoS parameters of throughput and delay in IEEE 802.11e EDCA. Huang and Liao [15] propose the use of an internal contention parameter that accounts for collision among different ACs, which is often overlooked. Malik et al. [16] analyze WAVE MAC by developing a VANET model in Matlab and by evaluating channel access delay and probability of channel access. In this paper, we use tractable equations for estimating delay, which reflect the intrinsic behavior of delays incurred by packets in the different ACs. Eichler [17] and Ferreira and Fonseca [18] study WAVE

SALAHUDDIN et al.: EXPLOITING CONTEXT SEVERITY TO ACHIEVE OSD IN VANETS

to gain insight into collision probability, throughput, and delay. There is an increase in delay in dense and high-load scenarios, as studied in [17]. Wang et al. [19] also noticed similar deterioration in performance of WAVE since a backoff window size does not adapt to increase in number of vehicles. They propose both centralized and distributed solutions to overcome these limitations by centrally calculating an optimal backoff window size or by adapting the window size locally. We propose a global approach, such that the OSD traffic distribution mechanism distributes vehicle traffic based on the CSM. This promotes better QoS for vehicles with higher context severity. Various authors have devised mechanisms for improving QoS for safety and nonsafety applications. Amadeo et al. [10] and Zhou et al. [20] aim to provide QoS for nonsafety applications. The former used an innovative MAC layer protocol that enhances the ability of WAVE to improve QoS, whereas the latter studied a content dissemination technique that employs timestamp-based reward/user satisfaction optimization. Di Felice et al. [21] develop enhancements to WAVE to reduce the delay for safety applications. Our OSD traffic distribution mechanism strives to improve the QoS for both safety and nonsafety applications. Our contribution is to extend the underlying communication protocols in WAVE by reprioritizing traffic to different ACs, to provide context-severity-aware service differentiation. Furthermore, Amadeo et al. [22] use the position of vehicles to ensure effective utilization of the channel resource, whereas we use the contextual information inferred from the position of a vehicle in our approach. However, both approaches bound the delay from above, unlike the unbounded delay of classical WAVE, which compromises time-sensitive applications. Chrysostomou et al. [23] design a dynamic tuning of contention window parameters to offer service differentiation in WAVE and improve the QoS with respect to the throughput. Qian et al. [11] also design adaptations of the WAVE MAC protocol by using dynamic contention windows and a vehicle mobility parameter of average speed to provide service differentiation. Others have analyzed and evaluated the MAC layer parameters for channel access delay and probability of channel access [16]. However, OSD uses the standard WAVE contention window and MAC layer parameters. It uses more complex vehicle mobility parameters, such as acceleration and directional stability, to conjure a CSM for vehicles in the network. Our major contribution is the OSD scheme that provides service differentiation based on the CSM, network load, and delay bounds on the ACs. We show an improvement over classical WAVE by using performance metrics, such as throughput, delay, number of collisions, packet loss ratio, and retransmission ratio. OSD achieves this by using a novel technique to distribute load opportunistically among different ACs, optimally utilizing the higher priority ACs, and guaranteeing minimum delay experience. III. O PPORTUNISTIC S ERVICE D IFFERENTIATION E NHANCED WAVE Here, we describe our OSD scheme, as shown in Fig. 2. The OSD scheme works in both unicast and broadcast scenarios. The OBU and the RSU nodes are enhanced with additional

2903

Fig. 2. OSD scheme with sequence of interactions between RSU and OBUs within communication range.

functionalities, and the user priority service is replaced with the OSD service in the multichannel operations sublayer in WAVE. The RSU and OBU are augmented with an OSD traffic distribution heuristic and a FIS, respectively. In step 1, the RSU broadcasts road infrastructure information, specifically the ideal speed, ideal acceleration, and ideal directional stability. The directional stability measures road lane departures. These are the RSU suggested ideal parameters since they are based on the integration of speed limit with realtime weather and road condition advisories (e.g., construction zones, congestion, etc.). In step 2, every OBU broadcasts its location and driving parameters. The rule-based FIS uses this to deduce the vehicle’s CSM, which is relayed to the RSU in step 3, along with vehicle’s traffic load. The OSD traffic distribution heuristic in the RSU computes the priority for each vehicle and communicates the new priorities to them in step 4. The OSD traffic distribution heuristic provides contextseverity-aware service differentiation, while accounting for network load, link layer bounds, and the delay thresholds on ACs. The output from the OSD traffic distribution heuristic is the priority used by the OSD service for AC selection, in the multichannel operation shown in Fig. 1. In this manner, OSDenhanced WAVE overrides application priorities, such that the context severity of a vehicle determines the priority for all its traffic through an AC. Thus, OSD traffic distribution heuristic assigns vehicle traffic to an AC, which is a tradeoff, since it may not fully utilize AC capacity. In the following, we will delineate the details of the FIS in the OBU and give an overview of the OSD traffic distribution mechanism on the RSU. A. Fuzzy Inference System A FIS is installed in the vehicle’s OBU to deduce the CSM for a vehicle; this is inspired by the work of Elbes et al. [24]. Our novel CSM gauges the severity of a vehicle relative to its neighborhood and environment. The FIS employs

2904

IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 63, NO. 6, JULY 2014

rule-based logic to map neighborhood and environment input data to a CSM output. In this manner, a vehicle is associated with urgency of the vehicles in its neighborhood. Environment parameters constitute factors such as traffic congestion, construction, weather elements, etc. The RSU integrates these factors with the road infrastructure information and broadcasts the ideal speed, ideal acceleration, and ideal directional stability ys , ya , and yd , respectively. Similarly, all vehicles vj in a neighborhood broadcast their location and driving parameters of speed, acceleration, and directional stability, v v v xsj , xaj , and xdj , respectively. Vehicle vi uses a lookup table to find its stopping distance based on its driving parameters, This stopping distance is used to demarcate an area of critical interest, as shown in Fig. 2, where vi gives more weight wij to the driving parameters of all vehicles vj that fall in this area. Each vehicle vi uses the following to compute weights wij , for all neighbors vj :   1, EuclidianDistance(vi , vj ) ≤ StoppingDistancevi vi wij = . StoppingDistance EuclidianDistance(vi ,vj ) , otherwise

Fig. 3. Membership functions for the input and output variables. (a) Context speed. (b) Context acceleration. (c) Context directional stability. (d) Context severity.

(1) The relative computation of speed, acceleration, and directional stability with respect to its neighboring vehicles and the RSU suggested ideal parameters are termed context speed, context acceleration, and context directional stability. Each vehicle uses the following to deduce its context speed, acceleration, and directional stability:  v V  xs j −ys w × ij j=1 10 (2) ContextSpeedvi = V   vj V wij × xa 10−ya j=1 ContextAccelerationvi = (3) V   v V xdj −yd j=1 wij × 10 ContextDirStabilityvi = . (4) V Evidently, vehicles within the stopping distance have a higher weight and, hence, a greater influence on vehicle’s contextual information. Consequentially, the context driving parameter of a vehicle is the weighted average of the deviation of all neighborhood vehicles from the RSU suggested ideal parameters. The context speed, context acceleration, and context directional stability are inputs to the FIS, which is used to deduce the output, i.e., the CSM of a vehicle. The FIS uses membership functions (MFs) to transform the crisp input values into a value between 0 and 1, which denotes the degree of membership to a fuzzy set. The IF - THEN rules defined in the FIS are used to infer a fuzzy set. This fuzzy set is retransformed using MFs into a crisp output value. Fig. 3 shows the MFs we used for our system inputs and output. The MFs in a FIS can be decomposed into simpler triangular and trapezoidal MF or more complex Gaussian and bell-shaped functions. The MF chosen represent the degrees of membership into the fuzzy set. We use the simpler triangular MF for context speed since it can be defined by three points,

Fig. 4.

FIS rules to deduce CSM.

Fig. 5.

FIS input variables deduce context severity of an OBU.

i.e., slow, average, and fast. However, the MF for context acceleration and context directional stability are bell-shaped, to utilize their smoothness. Similar to context speed, context severity is defined by MFs using two discrete points, i.e., normal and urgent. A set of rules are used in the FIS. These rules are delineated in Fig. 4 and conjure that a vehicle’s context severity is influenced by the deviation of the driving parameters of the vehicles in its neighborhood, including itself. Note that higher magnitude of the CSM implies higher severity of the vehicle within its context. Fig. 5 shows an example that depicts the relation of the input variables to the output variable in our FIS. It shows the crisp values for context speed, context acceleration, and context directional stability, and the MFs. Each row, numbered 1–4, in Fig. 5 shows the effect of each rule in Fig. 4 on the respective MF.

SALAHUDDIN et al.: EXPLOITING CONTEXT SEVERITY TO ACHIEVE OSD IN VANETS

B. OSD Traffic Distribution Mechanism All the vehicles will periodically transmit their context severity and load to the RSU performing the traffic distribution using the OSD traffic distribution heuristic. The RSU will use this information to form a global snapshot of the context severity and load of the vehicles. It will assign the vehicle to an AC based on its context severity, such that all traffic from the vehicle is transmitted through the assigned AC. The vehicleto-AC assignment is carefully achieved to ensure that delay bounds on the ACs are not exceeded and that vehicles with higher CSM are given better QoS with respect to delay. Hence, OSD provides service differentiation based on context severity of vehicles. It also overcomes unbounded delay, which is a major limitation in WAVE [4]. The OSD traffic distribution heuristic is our major contribution and the highlight of this paper. We validate our claims of performance improvement with OSD-enhanced WAVE, by formulating an unambiguous optimization problem. We define lemmas and a theorem and prove them to instigate the accuracy of our OSD traffic distribution heuristic. Furthermore, we use simulation tools to study the effect of context-severity-aware service differentiation in WAVE.

2905

is based on the CSMA/CA mechanism, where packets are transmitted when the channel is sensed to be idle. If the channel is busy, a retransmission is attempted after a specified backoff period. Similar to [26], we also assume that collision resolution time is included in the backoff period and, without loss in accuracy, use exponential distribution for backoff periods, with a mean duration of 1/β. Each vehicle generates packets according to a Poisson distribution with a probability density function, as described in (5). We assume that all vehicles are generating the same number of packets, each of the same size; then, the probability of finding the channel idle α is a constant for a given number of vehicles. This means that the number of backoffs k before a successful transmission is geometrically distributed [26] with a probability mass function α(1 − α)k−1 . Thus, the total backoff period is a convolution of possibly infinite number of backoffs [26]. Using Laplace transform backoff periods can be modeled as in (6), and for k successive backoffs, it becomes (β/β + s)k . Then, the Laplace transform for the total backoff period is as given in (7). Consequentially, a backoff period that was exponentially distributed has a mean rate of αβ [26]. f (t) = βe−βt ∀t ≥ 0 ∞ ∞ −st L [f (t)] = f (t)e dt = βe−βt e−st dt

IV. O PPORTUNISTIC S ERVICE D IFFERENTIATION T RAFFIC D ISTRIBUTION –L INEAR P ROGRAMMING F ORMULATION

0

0  ∞ β β e−(β+s)t = = −(s + β) β + s  0 k ∞ β F (s) = α(1 − α)k−1 β+s k=1  k−1 ∞ (1 − α)β αβ = β+s β+s k=1 l ∞  (1 − α)β αβ = β+s β+s

Here, we will present the problem statement and the delay model, and discuss the LP formulation. The LP formulation defines the mathematical model for OSD-enhanced WAVE, which is used for verification and validation. A mathematical model is deterministic and unambiguous with reproducible results, and serves as a baseline for the design of OSD traffic distribution heuristic. We develop a delay model that approximates the delay incurred by traffic in each AC. This enables us to use delay in a tractable and linear constraint in the LP formulation. A. Problem Statement Given a set of vehicles {v1 , v2 , . . . , vV } with context severity {S v1 , S v2 , . . . , S vV }, traffic loads {λv1 , λv2 , . . . , λvV }, and N ACs, each with a delay dk bounded by Tkmax . We need to find an optimal assignment of vehicle loads to the ACs, such that it provides QoS with respect to context severity while not exceeding the delay bounds Tkmax for ACk , ∀ 1 < k ≤ N . B. Delay Model We develop a simplified model to approximate the delay, similar to [25], on an AC. It appears as a linear constraint in our LP formulation. This model allows us to compare the performance of OSD-enhanced WAVE with classical WAVE for analytical and simulation results. It is not in the scope of this paper to design an intricate delay analysis model as presented in [26] and [27]. Instead, we approximate the process to illustrate the benefits of OSD-enhanced WAVE. We model MAC access delay, which is the time from when a packet becomes the head of the MAC queue to the time it is successfully transmitted. Recall that WAVE MAC access

(5)

(6)

(7)

l=0

where

∞

− α)β/β + s)l is a geometric series, then

   αβ 1 αβ F (s) = = . β + s 1 − (1−α)β αβ + s l=0 ((1

β+s

Our delay model approximates the delay for a packet to be the time spent in the total backoff period. Therefore, the delay is given as 1/αβ. Intuitively, 1/α is the mean number of retries and 1/β is the mean delay per retry. Recall that WAVE service differentiation is achieved by using different channel contention parameters for the ACs. Therefore, we formulate the delay incurred by traffic in each ACk as 1 . (8) dk = αk βk The probability of finding the channel idle αk is as follows: αk =

interarrival time interarrival time + transmission time V 1

=

V

i=1

1

i=1

v

λki

v

λki

+ Δt

=

1 + Δt

1 V i=1

λvki

(9)

2906

IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 63, NO. 6, JULY 2014

where the interarrival time is inversely proportional to the load on the AC and transmission time Δt = packet size/data rate. Each vehicle vi generates packets for each ACk according to a Poisson distribution with mean rate λvki ; then, the load on ACk  is Vi=1 λvki . The backoff mechanism in WAVE dictates the backoff period βk for each ACk . A vehicle wanting to transmit a packet from ACk selects a random backoff time if it senses the channel is idle for arbitration interframe space (AIFSk ) in (10) from [28]. The backoff time is a random number from the uniformly distributed interval [0, CWk + 1], where the initial contention window (CWk ) is equal to CWmin k . However, if the subsequent transmission attempt fails, the interval is doubled by increasing . The backoff value is the CWk value until it reaches CWmax k decreased when the channel is free. Therefore, we model the backoff period as in (11) in the following: AIFSk = SlotTime × AIFSNk + SIFSTime βk =

1 AIFSk + CWmin k (SlotTime)

(10) (11)

TABLE I N ETWORK AND WAVE Q O S PARAMETERS

C. Linear Programming Model We formulate the LP problem with continuous variables. In the following, we define the parameters and variables and present the formulation along with its discussion. 1) Parameters: N V Tkmax Δt

where CWmin k (SlotTime) is the lower bound on the contention window size for ACk . Table I [28]–[31] delineates the WAVE parameters used to compute βk . By substituting (9) into (8), we get the delay on ACk as

dk =

1 + Δt

V i=1

λvki

βk

.

(12)

βk λvki S vi 2) Variables: λvkidemo j

number of ACs; number of vehicles; upper bound on delay for ACk ; transmission time, packet size (B)/ data rate(M b/s); rate of backoff for ACk ; mean rate of packet generation, hereafter referred to as load from vi for ACk ; context severity of vi .

λvkipromo j Similar to [25], we do not account for hidden node terminals or saturated/unsaturated traffic to approximate the process. We validate our delay model by illustrating that this analytical delay follows the delay in simulation, as shown in Fig. 14(b). In this paper, we study the performance of OSD in VANETs, by taking a snapshot of the scenario and assuming a perfect channel [25]. Our model can be extended to account for factors such as vehicular mobility and channel fading.

i λvk,k−1,k−2,...,j ≡ λvkidemo j , load demoted from ACk to ACj , 1 ≤ k, j ≤ N ; i λvk,k+1,k+2,...,j ≡ λvkipromo j , load promoted from ACk to ACj , 1 ≤ k, j ≤ N ; delay for ACk

dk 3) Formulation: The formulation is as in (13)–(18), shown at the bottom of the page. 4) Discussion: We formulate context-severity-aware service differentiation by employing severity delay products and a systematic load

Objective min max{S vi dk ∀ vi , 1 ≤ k ≤ N } Subject to 1 + Δt dk = dk ≤ Tkmax

V i=1



(13)

 N  vi vi λvki + N j=k+1 λj demo k − j=k λj demo k−1  k vi vi + k−1 j=1 λj promo k − j=1 λj promo k+1 βk

∀1 < k ≤ N

λvkipromo k+1 + λvkidemo k−1 λvkidemo j−1 ≤ λvkidemo j λvkipromo j ≥ λvkipromo j+1

≤

∀1 ≤ k ≤ N

(14) (15)

λvki

∀ vi , ∀ vi ,

∀ vi ,

1≤k≤N

1 < k ≤ N;

1 λ. Therefore, the contrary must be  vk true. Similarly, assume that f ( M k=1 λ1 , 1) − d1 > f ((N − 1)λ, 1) is true. Then, the load on AC1 is  M vk hence, k=1 λ1 = (h(d1 , 1)/λ) λ + (N − 1)λ; (h(d1 , 1)/λ) λ + (N − 1)λ − h(d1 , 1) > (N − 1)λ. Thus, the contrary must be true. VI. O PPORTUNISTIC S ERVICE D IFFERENTIATION T RAFFIC D ISTRIBUTION –H EURISTIC We use the lemmas and a theorem in Section V to develop an efficient OSD traffic distribution heuristic. The pseudocode of this heuristic is shown in Fig. 7. First, the OSD traffic distribution heuristic aims to prioritize vehicles based on their CSM, as in Lemma 2. Then, it assigns vehicles to higher priority AC first, from Corollary 1, to leverage the inherent lower delays, thus offering higher QoS, with respect to delay, for higher context severity vehicles. This heuristic has a global snapshot of vehicles, their context severity, and load. It begins by sorting vehicles with respect to

Fig. 7.

Pseudocode for OSD traffic distribution heuristic.

severity so that they can be assigned an AC, as in Lemma 2. The heuristic deduces the delay on ACs based on Corollary 2 (23) and (24), and computes the maximum load (load cap) (19), based on this delay. It assigns the vehicles to AC, as in Corollary 1, until it reaches the load cap, before moving to the next best AC, with respect to priority for channel access. This ensures that the next best QoS is offered to vehicle traffic. Generally, the optimal solution occurs when there is equalization in the delay across all ACs, with higher context severity load assignment to a higher priority AC. This equalization for ACi can be computed from (22), where delay dEQ i h(dj , j) is the load on ACj with delay dj , ∀ i + 1 ≤ j ≤  N , g(λ − N j=i+1 h(dj , j), i) is the portion of load on ACi  when load λ − N j=i+1 h(dj , j) has to be distributed among ACi to AC1 to achieve equalization in delay, and f (g(λ − N h(dj , j), i), i) is the delay on ACi with load g(λ − j=i+1 N j=i+1 h(dj , j), i). Therefore, delay for ACi , ∀ 1 < i ≤ N , either achieves equalization across ACi−1 to AC1 or is bounded by Timax from (15) and is given in (23). However, the delay on the lowest priority AC is unbounded to accommodate the traffic overflow and is given as in (24). The OSD traffic distribution heuristic assigns the entire vehicle load to an AC in order of CSM. This may not fully utilize an AC. However, the Theorem quantifies this loss in utilization and shows that, given equal load per vehicle and entire vehicle load assignment to an AC, each AC will be underutilized by at most one vehicle load. Therefore, we have shown that our OSD traffic distribution heuristic efficiently achieves near optimal results, with respect to vehicle assignment to AC. In the following, we will show analytical and simulation results to demonstrate that, when service differentiation in WAVE is enhanced by our OSD traffic distribution mechanism, we can improve the QoS with respect to delay for ACs and for vehicles with respect to context severity. VII. A NALYTICAL R ESULTS Here, we will compare the OSD-enhanced WAVE with the classical WAVE and show that it not only outperforms classical WAVE but also accounts for network load, maximum allowable

SALAHUDDIN et al.: EXPLOITING CONTEXT SEVERITY TO ACHIEVE OSD IN VANETS

Fig. 8.

Delay comparison of OSD-enhanced WAVE with classical WAVE.

Fig. 9.

Load redistribution among ACs.

2909

Fig. 10. Load decomposition in ACs, with respect to severity, in OSD traffic distribution heuristic.

Fig. 11. QoS with respect to delay bounds on ACs for 1000 vehicles.

delay bounds on ACs, and the critical CSM for VANET vehicles. We use the parameters presented in Table I ([28]–[31]) for analytical comparison of classical WAVE with our optimal OSD traffic distribution formulation and heuristic. WAVE has been designed to operate at different data rates, i.e., 3–27 Mb/s. In our analytical study, we use 3 Mb/s, similar to studies in [32] and [33]. In these scenarios, each vehicle generated the same load comprising of randomly distributed payload across ACs. lp_solve [34] was used to solve the OSD formulation. Fig. 2 shows that the OSD scheme entails communication and computation complexities. However, as shown in Fig. 8, OSD-enhanced WAVE promises improvement in delay across AC, when compared with classical WAVE. This is achieved by increasing the utilization of higher priority ACs and leveraging their inherent lower delay. It is also evident that our heuristic achieves near optimal results. Furthermore, as the number of vehicles increases, the load increases, yielding higher delay. However, OSD-enhanced WAVE has a smaller rate of delay increase compared with WAVE. Figs. 9 and 10 show opportunistic load redistribution for 200 vehicles and four context severity levels S1 , S2 , S3 , and S4 , in decreasing order of severity. Note that the higher priority AC was assigned more traffic than the lower priority AC. Fig. 10 shows that vehicles with higher CSMs were assigned to higher

Fig. 12. Traffic demotion to achieve better QoS with respect to delay for 240 vehicles.

priority ACs. This ensures minimum severity delay products and optimal utilization of higher priority ACs. Fig. 11 shows how the OSD traffic distribution guarantees QoS with respect to delay for AC, bounded by Timax , ∀ 1 < i ≤ N , whereas WAVE exceeds the maximum acceptable delay for AC4 , AC3 , and AC2 . Fig. 12 shows the benefit of using demotion of load to achieve better overall QoS for all vehicles.

2910

IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 63, NO. 6, JULY 2014

Assuming that only AC4 has a high load, OSD traffic distribution will opportunistically redistribute the load, systematically demoting it to lower priority ACs and improving QoS with respect to delay.

TABLE II S IMULATION PARAMETERS

VIII. S IMULATION AND R ESULTS We use EstiNet 7.0 Network Simulator and Emulator [7], formerly NCTUns (National Chiao Tung University Network Simulator) [8]. It is a prevalent VANET simulator and emulator [35], which provides an integrated framework for traffic flow and network simulation. Typical simulators combine various functions, such as links with varying delays and bandwidths, IP packet routers, TCP/IP hosts, and mobility and network traffic generators into a single monolithic program. EstiNet has independent modules for these functions and executes them concurrently in a BSD Unix environment [36]. Here, we will delineate the scenario, simulation topology, and analysis setup, and present the performance comparison of classical WAVE with OSD-enhanced WAVE in the VANET environment. A. Scenario Let us consider the performance of a safety application, such as a cooperative collision avoidance system, where OBUequipped vehicles communicate with each other and RSUs to collectively exchange data with neighboring nodes and infrastructure to circumvent accidents. In WAVE, VANET nodes interact with each other via services that can be produced or consumed. We scrutinize the V2V communication aspect of the safety application by deploying services on OBU-equipped vehicles. In WAVE, all vehicles prioritize their safety application packets so that they are sent via AC4 . Whereas, OSD-enhanced WAVE reprioritizes the application requested priorities based on the context severity of the vehicle that hosts the application. Table III lists the context severity of the vehicles. For simplicity, we map the continuous context severity levels into four discrete values S1 , S2 , S3 , and S4 , arranged in order of decreasing severity. Simulation logs and statistics are used to evaluate and/or infer performance and QoS metrics of throughput, packet collisions, packet loss and retransmission ratio, and delay, to compare the performance of classical WAVE with OSD-enhanced WAVE. B. Topology This V2V scenario is simulated using EstiNet, by randomly deploying agent-controlled ITS OBUs with 802.11p interface as vehicles. Each vehicle is deployed with a default application, i.e., CarAgent, which manages the driving behavior of the vehicle, based on its profile, on the road infrastructure. A car profile is assigned to a vehicle and denotes its driving parameters, such as speed and acceleration. Each vehicle is equipped with an omnidirectional antenna, and its physical layer parameters are configured to ensure that vehicles are within the communication range of each other. Our vehicle parameters are listed in Table II.

Fig. 13. EstiNet simulation scenario consisting of one consumer (vehicle ID 2) and 30 provider OBUs.

The V2V communication in the network is established by creating a subnet and by assigning all vehicles to the same subnet. EstiNet uses this subnet assignment to generate IP and MAC addresses of the vehicles. Vehicles exchange data in the WAVE mode by joining a WAVE-mode basic service set (WBSS). The service provider broadcasts its service in the CCH via wave service advertisements (WSAs) and a consumer joins the WBSS of the provider. Since simulation time is directly proportional to the number of vehicles deployed, we simulated our scenario with 30 vehicles providing services to our consumer in vehicle ID 2, as shown in Fig. 13. The simulation parameters are listed in Table II. Similar to the work done in [32] and [33], we use a 3 Mb/s data rate in our simulation. The WAVE short message (WSM) application of EstiNet is added to the vehicles to generate prioritized data for a consumer and transmit it over a SCH. Each service provider uses WSM to send 15 data packets/s, of 400 B each, at a transmission data rate of 3 Mb/s on SCH 172, via unicast communication. These network and WAVE parameters are listed in Table II; note that we use EstiNet default WAVE parameters. Since we are simulating vehicles for a safety application, such as a

SALAHUDDIN et al.: EXPLOITING CONTEXT SEVERITY TO ACHIEVE OSD IN VANETS

TABLE III V EHICLE A SSIGNMENT TO AC S W ITH R ESPECT TO S EVERITY

cooperative collision avoidance system, all service providers are generating data for the highest priority AC, i.e., AC4 . In EstiNet, priority 7 is used for AC4 . However, the heuristic for OSD-enhanced WAVE assigns vehicles to different ACs with respect to their severity. This is to ensure that AC delays do not exceed thresholds (T4max , T3max , T2max ), delineated as OSD parameters in Table II. This reassignment of vehicles to ACs based on severity is given in Table III. C. Analysis Setup We monitor the network traffic using the packet trace file (ptr) logged by EstiNet and the packet statistics logging offered in the MAC module of a vehicle. The packet statistics logging records the out throughput (kB/s) and the number of collisions at the providers. The ptr file records the network traffic, and the packet statistics are sampled at 1 s intervals. The ptr file is used to deduce packet loss and packet retransmission ratio and log files for evaluating throughput and collisions. The QoS metric of delay perceived by a provider is inferred from the throughput. There are multiple replications of a simulation run to increase the accuracy of our results. The packet loss (25) and retransmission ratio (26) are calculated using the average number of packet transmissions TX, packet retransmissions RTX, and packets received RX. The average throughput and number of collisions for a provider are computed by excluding data from the warm-up and shutdown times, assumed to be 1 s each. The delay perceived by the providers is computed based on (27) in the following: TX + RTX − RX TX + RTX RTX packet retransmission ratio = TX 1 delay(ms) = × 1000. throughput (kB/s)×1024 packet size (B) packet loss ratio =

(25) (26) (27)

D. Results and Discussion Recall that, in our safety application simulation scenario, for classical WAVE, all 30 service providers are vehicles generating data for the highest priority AC, i.e., AC4 , and transmitting it in

2911

the same SCH, i.e., SCH 172. This simulates a high-contention scenario, where vehicles are forced to compete for the physical channel. It ensures that vehicles use the EDCA settings in WAVE to backoff and retry for the channel. EDCA parameters for AC4 have been selected so that vehicles in AC4 have the highest priority to access the channel and thus incur the lowest latency. However, this service differentiation in WAVE does not account for network traffic and load on ACs, and hence cannot guarantee delay bounds for time-critical safety applications. We will show that OSD-enhanced WAVE accounts for network traffic and load on ACs to fully utilize all ACs and can also guarantee meeting delay requirements for high-priority ACs. In our high-contention scenario for WAVE, all vehicles, providing services in AC4 , are constantly contending for the channel. This is evident in the high 383 packets/s average collisions in AC4 , as shown in Fig. 14(c). When the vehicles incur a collision, they backoff and retry for the channel. In the case of a collision, AC4 is given priority to reduce its traffic latency. However, in this scenario, all vehicles contending for the channel belong to AC4 ; hence, the priority of the AC does not benefit the vehicles. This is shown in the high retransmission ratio of 2.6 in Fig. 14(e), which captures the multiple retransmissions that must occur for one successful transmission. The packet loss ratio in Fig. 14(d) for vehicles providing services in AC4 , in classical WAVE, also concurs with the high collision rate and retransmission ratio. These collisions and retransmissions reduce the effective throughput of the vehicles providing services in AC4 , as shown in Fig. 14(a), where the average classical WAVE throughput is low (1.8 kB/s). A low throughput translates to a high delay for vehicles operating in the highest priority AC, i.e., AC4 . This undermines the service differentiation in classical WAVE and diminishes the effectiveness of time-critical safety and lifesaving applications, such as cooperative collision avoidance applications. Fig. 14(b) shows that vehicles, in classical WAVE, providing services in AC4 , incur an average delay of 288 ms, which exceeds T4max = 70 ms. Typically, the maximum allowable latency requirement for safety applications, such as cooperative collision avoidance, is approximately 100 ms [30], [31]. As shown, we overcome this limitation of WAVE, using OSD. We also show that our analytical delay closely follows the simulation delay, thereby validating our delay model. In our novel approach, we not only consider the load in the network and in the ACs but also account for the context severity of the vehicle and the maximum allowable delay requirements. Recall that our heuristic assigns a vehicle and its entire load to an AC, with respect to its context severity and the delay bounds on the ACs. This assignment of vehicles to ACs with respect to severity is delineated in Table III. The heuristic assigns vehicles to the ACs, based on the network load and delay bounds of 70, 120, and 180 ms, on AC4 , AC3 , and AC2 , respectively. OSD-enhanced WAVE distributed AC4 load among the other ACs to utilize the inherent service differentiation mechanism of WAVE, so that there is differentiation in the backoff and retry mechanism of all the service providers. This reduces the average collision, packet loss, and retransmission ratios, as shown in Fig. 14(c)–(e), respectively. Consequently, as vehicles incur less contention on the channel, their throughput increases, as shown

2912

IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 63, NO. 6, JULY 2014

Fig. 14. Simulation performance comparison of classical WAVE with OSD-enhanced WAVE with 95% confidence intervals. (a) OSD redistributes load, improving throughput in comparison to WAVE. (b) Analytical delay model follows simulation results, and OSD overcomes the unbounded delay in WAVE with Tmax for AC4 , AC3 , and AC2 . (c) Load distribution across ACs utilizes inherent service differentiation in WAVE to reduce contention and collision. (d) High-throughput AC4 shares higher packet loss ratio. (e) Packet retransmission ratio consequentially follows packet loss ratio.

in Fig. 14(a). It is important to note that in OSD-enhanced WAVE, some vehicles were assigned to lower priority AC to achieve the delay guarantees in AC4 , AC3 , and AC2 . Fig. 14(b) captures the innovation and essence of our OSD technique. It not only guarantees delay bounds on the three higher priority ACs but also improves the QoS metric of delay perceived by all service providers in the network. It decreases the average delay for service providers in AC4 from 288 to 70 ms from classical WAVE to OSD-enhanced WAVE, respectively. Now, the safety application packets from service providers in AC4 can ensure that latency does not exceed the maximum allowable delay of 70 ms. Fig. 14(b) explicitly shows that, even service providers in AC3 and AC2 in OSD-enhanced WAVE incur a

lower latency than service providers in AC4 in classical WAVE. The worst-case delay incurred by service providers in AC1 is approximately the same as the average delay incurred by service providers in AC4 in classical WAVE. Therefore, in our simulation scenario, classical WAVE performs as worst as the lowest priority AC in OSD-enhanced WAVE. However, it is important to note that this is not our claim. In fact, the performance of OSD-enhanced WAVE primarily depends on the delay bounds set on the ACs and the network load. If there is a low network load or no delay bounds on the ACs, OSD-enhanced WAVE distributes the load such that the delay incurred by all vehicles across all ACs is equalized. In Fig. 14, we show that OSD-enhanced WAVE can provide delay

SALAHUDDIN et al.: EXPLOITING CONTEXT SEVERITY TO ACHIEVE OSD IN VANETS

2913

Fig. 15. Performance of vehicles with respect to severity in OSD-enhanced WAVE with 95% confidence intervals. (a) ACs are utilized in order of CSM. (b) Higher CSM vehicles have smaller delay. (c) All ACs are optimally utilized to distribute load and share channel. (d) Higher CSM vehicles are assigned to higher throughput ACs, sharing a higher percentage of packet loss. (e) Packet retransmission ratio is consequential to the packet loss ratio.

guarantees for ACs and improve QoS metric of delay for all vehicles with a 95% confidence interval range. Table III lists the vehicle IDs of the service providers in the four different severity levels. The severity of a vehicle is different from the priority of the application or service that the vehicle is providing. Therefore, a low-severity vehicle can generate high-priority data. OSD-enhanced WAVE distributes traffic across the ACs with respect to vehicle severity, in an effort to provide the best possible QoS to the higher severity vehicles. The distribution of vehicles based on severity among the ACs allow us to provide service differentiation in WAVE based on the critical ITS parameter of vehicle severity. Our OSD technique for WAVE ensures the highest throughput and the

lowest delay for the highest severity vehicles in Fig. 15(a) and (b), respectively. The 95% confidence intervals are also shown for average collisions, packet loss, and packet retransmission ratios for the four severity levels in Fig. 15(c)–(e), respectively. IX. C ONCLUSION We proposed a novel OSD scheme to enhance WAVE. The proposed OSD is a traffic distribution mechanism that promotes service differentiation while accounting for vehicular and communication network parameters. In this paper, we have emphasized the need for a CSM that can accurately depict the urgency of vehicle traffic. We proposed the use of a FIS to deduce the

2914

IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 63, NO. 6, JULY 2014

severity of a vehicle, in the context of its surroundings, such as other vehicles and real-time road conditions. We use this CSM to provide service differentiation in WAVE, by using our OSD traffic distribution heuristic. Our traffic distribution mechanism has multiple benefits over the classical WAVE. First, it assigns vehicle traffic to ACs with respect to context severity. It provides higher QoS with respect to delay, for vehicles with higher context severity. Second, it apportions the ACs to vehicle traffic based on the network load, such that it can guarantee delay bounds. For example, users of the OSD scheme can specify a delay bound for traffic in AC4 ; the OSD traffic distribution heuristic will utilize this AC while ensuring that traffic in AC4 meets the delay bound. This is a significant improvement over classical WAVE and its unbounded delay. We also modeled the OSD traffic distribution mechanism as an LP problem to validate our claims. We derived lemmas and a theorem that form the basis for our traffic distribution heuristic. Our analytical comparisons show the improvement in QoS with the OSD-enhanced WAVE and the near optimal performance of the OSD traffic distribution heuristic. We also simulate a VANET with classical and OSD-heuristic-enhanced WAVE. The results show that, when network traffic is opportunistically distributed among ACs, with respect to CSM, it improves the QoS with respect to delay. This traffic distribution mechanism increases the utilization of ACs while meeting the upper bounds on the AC delay. Our future work entails implementing the OSD traffic distribution mechanism and the OSD service for extensive simulation and real-time performance analysis with respect to QoS, density, and time. Furthermore, scrutiny of the scalability and latency in OSD traffic distribution mechanism would enable a detailed analysis for deployment of the OSD scheme. R EFERENCES [1] R. Uzcategui and G. Acosta-Marum, “WAVE: A tutorial,” IEEE Commun. Mag., vol. 47, no. 5, pp. 126–133, May 2009. [2] J. Mittag, F. Schmidt-Eisenlohr, M. Killat, M. Torrent-Moreno, and H. Hartenstein, “MAC layer and scalability aspects of vehicular communication networks,” in VANET: Vehicular Applications and Inter-Networking Technologies. Hoboken, NJ, USA: Wiley, 2009, pp. 219–273. [3] IEEE Standard for Wireless Access in Vehicular Environments (WAVE)—Multichannel Operation, IEEE Std 1609.4-2010 (Revision of IEEE 1609.4-2006), 2011. [4] Y. Morgan, “Managing DSRC and WAVE standards operations in a V2V scenario,” Int. J. Veh. Technol., vol. 2010, pp. 797405-1–797405-18, 2010. [5] V. Harigovindan, A. Babu, and L. Jacob, “Ensuring fair access in IEEE 8002.11p-based vehicle-to-infrastructure,” EURASIP J. Wireless Commun. Netw., vol. 2012, no. 168, 2012. [6] R. Chen, D. Ma, and A. Regan, “TARI: Meeting delay requirements in VANETs with efficient authentication and revocation,” in Proc. Int. Conf. Wireless Access Veh. Environ., 2009. [Online]. Available: http://www.researchgate.net/publication/228895115_TARI_Meeting_ Delay_Requirements_in_VANETs_with_Efficient_Authentication_and_ Revocation/file/504635215028065cd2.pdf [7] EstiNet 7.0 Network Simulator and Emulator 2012. [Online]. Available: http://www.estinet.com/products.php?lv1=1&sn=2 [8] S. Y. Wang, C. L. Chou, C. H. Huang, C. C. Hwang, Z. M. Yang, C. C. Chiou, and C. C. Lin, “The design and implementation of the NCTUns 1.0 network simulator,” Comput. Netw., vol. 42, no. 2, pp. 175–197, Jun. 2003. [9] Y. Zang, L. Stibor, B. Walke, H. Reumerman, and A. Barroso, “A novel MAC protocol for throughput sensitive applications in vehicular environments,” in Proc. IEEE VTC, Dublin, Ireland, 2007, pp. 2580–2584.

[10] M. Amadeo, C. Campolo, A. Molinaro, and G. Ruggeri, “A WAVEcompliant MAC protocol to support vehicle-to-infrastructure non safety applications,” in Proc. IEEE ICC, Dresden, Germany, 2009, pp. 1–6. [11] Y. Qian, K. Lu, and N. Moayeri, “A secure VANET MAC protocol for DSRC applications,” in Proc. IEEE GLOBECOM, New Orleans, LA, USA, 2008, pp. 1–5. [12] M. Nekovee, “Quantify performance requirements of vehicle-to-vehicle communication protocols for rear-end collision avoidance,” in Proc. IEEE VTC, 2009, pp. 1–5. [13] J. Vardakas, I. Papapanagiotou, M. Logothetis, and S. Kotsopoulos, “On the end-to-end delay analysis of the IEEE 802.11 distributed coordination function,” in Proc. Internet Monitor. Protect., San Jose, CA, USA, 2007, p. 16. [14] H. Wu, X. Wang, Q. Zhang, and X. Shen, “IEEE 802.11e enhanced distributed channel access (EDCA) throughput analysis,” in Proc. IEEE ICC, Istanbul, 2006, pp. 223–228. [15] C. Huang and W. Liao, “Throughput and delay performance of IEEE 802.11e Enhanced distributed channel access (EDCA) under saturation condition,” IEEE Trans. Wireless Commun., vol. 6, no. 1, pp. 136–145, Jan. 2007. [16] S. Malik, M. Shah, S. Khan, M. Jahanzeb, U. Farooq, and M. Khan, “Performance evaluation of IEEE 802.11p MAC protocol for VANETs,” Aust. J. Basic Appl. Sci., vol. 4, no. 8, pp. 4089–4098, 2010. [17] S. Eichler, “Performance evaluation of the IEEE 802.11p WAVE communication standard,” in Proc. IEEE VTC, Baltimore, MD, USA, 2007, pp. 2199–2203. [18] N. Ferreira and J. Fonseca, “On the end-to-end delay analysis for an IEEE 802.11P/WAVE protocol,” in Proc. 18th IEEE SCVT, Nov. 22–23, 2011, pp. 1–6. [19] Y. Wang, A. Ahmed, B. Krishnamachari, and K. Psounis, “IEEE 802.11p performance evaluation and protocol enhancement,” in Proc. IEEE ICVES, Columbus, OH, USA, 2008, pp. 317–322. [20] L. Zhou, Y. Zhang, K. Song, W. Jing, and A. Vailakos, “Distributed mediaservice scheme for P2P-based vehicular networks,” IEEE Trans. Veh. Technol., vol. 60, no. 2, pp. 692–703, Feb. 2011. [21] M. Di Felice, A. Ghandour, H. Artail, and L. Bononi, “Enhancing the performance of safety applications in IEEE 802.11p/WAVE Vehicular Networks,” in Proc. IEEE Int. Symp. WoWMoM, San Francisco, CA, USA, 2012, pp. 1–9. [22] M. Amadeo, C. Campolo, and A. Molinaro, “Enhancing IEEE 802.11p/WAVE to provide infotainment applications in VANETs,” Ad Hoc Netw., vol. 10, no. 2, pp. 253–269, Mar. 2012. [23] C. Chrysostomou, C. Djouvas, and L. Lambrinos, “Dynamically adjusting the min-max contention window for providing quality of service in vehicular networks,” in Proc. 11th Annu. Med-Hoc-Net, Ayia Napa, Cyprus, 2012, pp. 16–23. [24] M. Elbes, A. Al-Fuqaha, M. Guizani, A. Rayes, and J. Oh, “A new hierarchical and adaptive protocol for minimum-delay V2V communication,” in Proc. IEEE GLOBECOM, Honolulu, HI, USA, 2009, pp. 1–6. [25] X. Chen, H. Refai, and X. Ma, “A quantitative approach to evaluate DSRC highway inter-vehicle safety communication,” in Proc. IEEE Global Telecommun. Conf., Washington, DC, USA, 2007, pp. 151–155. [26] J. Zhou and K. Mitchell, “A scalable delay based analytical framework for CSMA/CA wireless mesh networks,” Comput. Netw., vol. 54, no. 2, pp. 304–318, Feb. 2010. [27] Y. Yao, L. Rao, X. Liu, and X. Zhou, “Delay analysis and study of IEEE 802.11p based DSRC safety communication in a highway environment,” in Proc. IEEE INFOCOM, Turin, Italy, 2013, pp. 1591–1599. [28] C. Han, M. Dianati, R. Tafazolli, R. Kernchen, and X. Shen, “Analytical study of the IEEE 802.11p MAC sublayer in vehicular networks,” IEEE Trans. Intell. Transp. Syst., vol. 13, no. 2, pp. 873–886, Jun. 2012. [29] IEEE Std. for Information Technology- Telecommunications and Information Exchange Between Systems-Local And Metropolitan Area Networks-Specific Requirements. Part 11: Wireless LAN MAC and PHY Specifications. Amendment 6: Wireless Access in Vehicular Enviro, IEEE Std 802.11p 2010. [30] M. Boban, G. Misek, and O. K. Tonguz, “What is the best achievable QoS for unicast routing in VANETs?” in Proc. IEEE GLOBECOM Workshops, New Orleans, LA, USA, 2008, pp. 1–10. [31] “Vehicle Safety Communications Project—Task 3 Final Report—Identify Intelligent Vehicle Safety Applications Enabled by DSRC,” Nat. Highway Traffic Safety Admin., U.S. Dept. Transport., Washington D.C., USA, 2005. [32] Q. Wang, S. Leng, H. Fu, and Y. Zhang, “An IEEE 802.11p-based multichannel MAC scheme with channel coordination for vehicular ad hoc networks,” IEEE Trans. Intell. Transp. Syst., vol. 13, no. 2, pp. 449–458, Jun. 2012.

SALAHUDDIN et al.: EXPLOITING CONTEXT SEVERITY TO ACHIEVE OSD IN VANETS

[33] C. Campolo and A. Molinaro, “DREAM: IEEE 802.11p/WAVE extended access mode in drive-thru vehicular scenarios,” in Proc. IEEE ICC, Ottawa, ON, USA, 2012, pp. 5301–5305. [34] GNU LGPL (Lesser General Public License) M. Berkelaar, K. Eikland, and P. Notebaert, Open source (Mixed- Integer) Linear Programming system (lp_solve) 2004, GNU LGPL (Lesser General Public License). [35] F. J. Martinez, C. K. Toh, J. Cano, C. T. Calafate, and P. Manzoni, “A survey and comparative study of simulators for vehicular ad hoc networks (VANETs),” Wireless Commun. Mobile Comput., vol. 11, no. 7, pp. 813–828, Jul. 2011. [36] S. Y. Wang and H. T. Kung, “A simple methodology for constructing extensible and high-fidelity TCP/IP network simulators,” in Proc. INFOCOM, New York, NY, USA, 1999, pp. 1134–1149.

Mohammad A. Salahuddin (S’09) received the B.S. degree from the University of Karachi (FASTICS), Karachi, Pakistan, in 1999; the M.S. degree from Shaheed Zulfiqar Ali Bhutto Institute of Science and Technology, Karachi, in 2001; and the M.S. degree from Western Michigan University, Kalamazoo, MI, USA, in 2003, all in computer science. He is currently working toward the Ph.D. degree with the Department of Computer Science, Western Michigan University. As a Systems Analyst and Software Engineer for Flint Group and Quark, Inc., respectively, he has extensive experience in all phases of software product development. His research methodologies include operations research, evolutionary computing, and high-performance computing using Compute Unified Device Architecture/graphics processing units. His research interests include wireless sensor networks, indoor and outdoor localization techniques, quality of service and quality of experience in vehicular ad hoc networks, and service placement and routing for cloud resource management. He has developed business intelligence and data analytics software to facilitate decision-making on a research grant from Scenaria, Inc.: a subsidiary of AVL North America. Mr. Salahuddin serves as a Reviewer for IEEE conferences.

2915

Ala Al-Fuqaha (S’00–M’04–SM’09) received the M.S. and the Ph.D. degrees in electrical and computer engineering from the University of Missouri, Columbia, MO, USA, in 1999 and 2004, respectively. He is currently an Associate Professor with the Department of Computer Science, Western Michigan University, Kalamazoo, MI, USA. His research interests include intelligent network management and planning, quality-of-service routing, and performance analysis and evaluation of high-speed computer and telecommunication networks. Dr. Al-Fuqaha has served as a Technical Program Committee member and a Reviewer of many international conferences and journals. He is currently serving on the editorial board for John Wiley’s Security and Communication Networks Journal and the Industrial Networks and Intelligent Systems Journal.

Mohsen Guizani (S’85–M’89–SM’99–F’09) received the B.S. (with distinction) and M.S. degrees in electrical engineering and the M.S. and Ph.D. degrees in computer engineering in 1984, 1986, 1987, and 1990, respectively, from Syracuse University, Syracuse, NY, USA. He is currently a Professor and the Associate Vice President for Graduate Studies at Qatar University, Doha, Qatar. He was the Chair of the Computer Science Department at Western Michigan University from 2002 to 2006 and Chair of the Computer Science Department at the University of West Florida from 1999 to 2002. He also served in academic positions at the University of Missouri-Kansas City, the University of Colorado-Boulder, Syracuse University, and Kuwait University. His research interests include computer networks, wireless communications and mobile computing, and optical networking. He currently serves on the editorial boards of six technical Journals and is the Founder and Editorin-Chief of the John Wiley journal Wireless Communications and Mobile Computing. (http://www.interscience.wiley.com/jpages/1530-8669/). He is the author of eight books and more than 300 publications in refereed journals and conferences. He has guest edited a number of special issues in IEEE Journals and Magazines. He also served as member, Chair, and General Chair of a number of conferences. Dr. Guizani served as the Chair of IEEE Communications Society Wireless Technical Committee (WTC) and Chair of the TAOS Technical Committee. He was an IEEE Computer Society Distinguished Lecturer from 2003 to 2005. He is a Senior Member of the Association for Computing Machinery.