Efficient Broadcasting in IEEE 802.11 Networks ... - Semantic Scholar

Efficient Broadcasting in IEEE 802.11 Networks through Irresponsible Forwarding Stefano Busanelli† , Gianluigi Ferrari,† and Sooksan Panichpapiboon‡ Ad-hoc and Sensor Networks (WASN) Lab, Dept. of Information Eng., University of Parma, Italy & CNIT Research Unit of Parma, Italy ‡ Faculty of Information Technology, King Mongkut’s Institute of Technology Ladkrabang, Bangkok, Thailand † Wireless

Abstract—In a self-organizing vehicular network, vehicles share and distribute information by rebroadcasting a received information packet to their neighbors. An efficient broadcast technique can offer a high reactivity without sacrificing the communication reliability. Therefore, broadcast techniques are particularly suitable for safety-related vehicular transmissions, whose goal is reaching reliably the widest area in the shortest time. Among the numerous solutions appeared in the literature, the probabilistic broadcast approaches seem to be promising and not yet accurately analyzed. Since the interaction between a high level broadcasting protocol with the lower layers cannot be ignored, in this work we analyze the behavior of a recently proposed broadcast technique, denoted as the Irresponsible Forwarding (IF), in IEEE 802.11 networks. Our attention concentrates on the Medium Access Control (MAC) layer, which is affected by some critical impairments for broadcasting, such as the hidden terminal problem and self-interference. In this work, we evaluate the benefits brought by the use of IF to perform efficient broadcasting in IEEE 802.11 networks.

I. I NTRODUCTION In a self-organizing vehicular network, vehicles share and distribute information by rebroadcasting. Typically, rebroadcasting results in redundant retransmissions of the same information packet, thus leading to a useless occupation of the radio channel, and depleting the available network resources. Minimizing the redundancy, while still guaranteeing complete reachability throughout the network, is one of the main objectives in multi-hop broadcasting. In recent years, many solutions have been proposed by the research community. In [1], one can find a possible classification of these protocols. Position-based broadcast protocols (see [2] and the references therein) have been considered sufficiently promising to be used in some practical vehicular networks, as in the Network on Wheels (NoW) German project [3]. Despite its good performance, the position-based broadcasting approach needs information concerning the network topology and the geographic characteristics of the scenario where the nodes are located, often collected by a dedicated logical channel [3]. Since collecting this information may be very expensive, several probabilistic broadcasting protocols have been recently proposed with the goal of achieving the same performance level of position-based protocols without the need of major information exchange [4]–[9]. In general terms, the problem of designing efficient probabilistic rebroadcasting protocols consists in assigning to the nodes the “best” rebroadcast probability, according to proper

application-related criteria. In [10], the authors propose an innovative probabilistic broadcasting scheme, denoted as Irresponsible Forwarding (IF), such that the rebroadcast probability depends on a series of topological parameters, such as the statistical spatial distribution of the vehicles, the transmission range, and the inter-vehicle distance. The key idea is that a singularly irresponsible behavior may be networkwide responsible. In [10], the performance of IF schemes is analyzed in “ideal” communication conditions, i.e., without any need for the use of channel contention mechanisms and in the absence of collisions. In the current paper, we apply the IF approach to design efficient broadcasting schemes in IEEE 802.11 networks [11]. In particular, we analyze the behavior of the IF protocol as a function of the traffic load and of the node density. Since we consider a single traffic generating node, the collisions are due to consecutive packets (e.g. with different sequence numbers) generated by the source or to broadcasted duplicates of the same packet. The second type of collisions will be indicated to as self-generated (by the source). The performance of IF broadcasting schemes is analyzed considering various performance metrics, including (i) the average number of rebroadcasts, (ii) the reachability (RE), (iii) the number of saved rebroadcast (SRB), (iv) the average number of hops, and (v) the end-to-end delay. The remainder of this paper is organized as follows. In Section II, we recall the principles of the IF protocol introduced in [10]. In Section III, we describe the considered IEEE 802.11 network simulation scenario and considered performance metrics. Results are presented and discussed in Section IV. Finally, conclusions are drawn in Section V. II. I RRESPONSIBLE F ORWARDING A. Reference Scenario As in [10], we consider a static one-dimensional wireless network, with a single source placed at the left vertex of the network line (identified by the index 0) and a series of N potential rebroadcasting nodes positioned along the line itself (each node is uniquely identified by an index j ∈ {1, 2, . . . , N }).1 The number of nodes N is not a-priori fixed. Instead, we set the network size (the line length) to 1 This model applies also to one-dimensional vehicular networks where the vehicles move in the same direction with approximately the same speed.

978-1-4244-4148-8/09/$25.00 ©2009 This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE "GLOBECOM" 2009 proceedings.

L (dimension: [m]) and we iteratively generate the nodes’ positions according to a one-dimensional Poisson distribution with parameter ρs > 0 (dimension: [veh/m]). We denote as dj,k the (positive) distance between the two nodes j and k. The generation process is stopped as soon as d0,k > L, where k is the index of the last generated node. Each node has the same transmission range z (dimension: [m]) and is equipped with a Global Positioning System (GPS) receiver. In this work, in order to carry out a fair comparison between the results obtained with different z values, we denote as normalized network size the positive real number l  L/z. Finally, we introduce the concept of last reachable node (lrn). For every pair of two consecutive nodes, say (j, j + 1), since dj,j+1 ∼ exp (ρs ), there is the probability e−ρs z > 0 of having dj,j+1 > z. Among all the consecutive nodes pairs (j, j + 1), j = 1, 2, . . . , N − 1, we denote as j ∗ the minimum node index such as dj ∗ ,j ∗ +1 > z. Clearly, if j ∗ exists, than the (j ∗ + 1)-th node is unreachable from any node j such as j < j ∗ + 1. Therefore, the network is said topologically disconnected and lrn  j ∗ . Conversely, if j ∗ does not exist, then the network is topologically connected and lrn  N . B. Protocol description The IF protocol is a probabilistic forwarding technique in which every node computes its own transmission probability in a per-packet manner. The initial transmission of a new packet from the source is denoted as the 0-th hop transmission, while the source itself identifies the so-called 0-th transmission domain. The packet is then received by a subset of the source neighbors, that are the potential rebroadcasting nodes for the 1-st hop. Hence, their union constitutes the 1-st transmission domain. All nodes of the 1-st transmission domain rebroadcast with a probability computed with a proper probability assignment function, introduced in [10] and recalled in Subsection II-C. If an intermediate node receives more than one copy of a packet, it will make the rebroadcast decision only upon the reception of the first copy of the packet. All the successive copies are automatically discarded to reduce the network traffic and avoid self-loops. All the nodes that receive a “fresh” packet by a node belonging to the 1-st transmission domain, contribute to form the 2-nd transmission domain. This happens recursively, until the packet is not rebroadcast or reaches the physical network limit. We remark that in the ad-hoc mode, the IEEE 802.11 standard [11] mandates the use of the Distributed Coordination Function (DCF)—a Carrier Sense Multiple Access with Collision Avoidance (CSMA/CA) protocol—as the medium access mechanism. Hence, due to the probabilistic nature of the IF protocol itself and to the DCF, each information packet sees its own sequence of transmission domains. Those sets are mutually exclusive, since a node can belong to just a single transmission domain for a given packet. Since a transmission domain can be reinterpreted as a reception codomain, in an ideal scenario the union of all transmission domains contains all network nodes. This may not be the case for a realistic

network protocol, where not all nodes necessarily receive every packet, due to the presence of unavoidable impairments. C. Probability Assignment Function The probability assignment function of the IF scheme introduced in [10] is p

= e−

ρs (z−d) c

(1)

where d, ρs , and z were introduced in Subsection II-A, while c is a “shaping” coefficient. The rationale behind (1) is the fact that in order to reduce the packet redundancy, a vehicle should relegate the responsibility of rebroadcasting the packet to another vehicle downstream. As a result, our probability assignment function is derived from the probability that there is another vehicle, able to rebroadcast the packet. In other words, each receiving vehicle computes the rebroadcast probability based on the likelihood of finding another vehicle downstream able to rebroadcast the packet. Note that with the probability assignment in (1), the vehicle spatial density is also taken into account. When the network is sparse, the overall rebroadcast probability should be high (e.g., even if the receiving vehicle is close to the transmitter) in order to ensure complete reachability. In addition, the coefficient c is also effective at shaping the rebroadcast probability, as the overall rebroadcast probability can be increased by increasing the value of c. D. Analytical Performance Evaluation In [10], the following closed-form expression for the average number of rebroadcast events at the first hop (or in the first transmission domain) is derived: ⎧  ∞ (ρs z)n e−ρs z n (ρs z)i e−ρs z ⎪ ⎪ n=1 i=1 n! i! ⎪ ⎪ ⎪ ⎪ 1 ⎪ · ⎪  (ρs z)j ⎪ 1−e−ρs z −e−ρs z i−1 ⎪ j=1 j! ⎪ ⎪ ⎪ if c = 1 ⎪ ⎪ ⎨ (2) E[M1 ] = ∞ (ρs z)n e−ρs z n  c i − ρs z ⎪ ⎪ c ⎪ e ⎪ n=1 i=1 c−1 n! ⎪ ⎪ ⎪ ⎪ 1−e−ρs z −e−ρs z  i−1 (ρs z)j ⎪ ⎪ j=1 j! ⎪ ⎪ ·  i−1 (ρs z)j ⎪ −ρs z −e−ρs z 1−e ⎪ j=1 j! ⎪ ⎩ if c > 1.

where ρs = ρs 1 − 1c . In [10], Matlab simulations are also carried out in an ideal (no collision) scenario, and it is observed that the average number of rebroadcasts in the i-th transmission domain (i ≥ 1) converges to a saturation value for high ρs z. Moreover, this saturation value coincides with the shaping parameter c for the 1-st transmission domain, while it becomes smaller and smaller with increasing domain indexes. III. IEEE 802.11 N ETWORK S IMULATION S ETUP A. IEEE 802.11 Networks Unlike [10], where the simulations are performed in an ideal scenario (without interference, collisions and channel access delay), in this work we investigate the impact of a realistic

978-1-4244-4148-8/09/$25.00 ©2009 This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE "GLOBECOM" 2009 proceedings.

wireless communication protocol stack, namely that of an adhoc IEEE 802.11 network [11]. Since the communications are intrinsically broadcast, correct packet reception cannot be acknowledged. Therefore, the packets are never retransmitted and the Contention Window (CW) of the backoff procedure of the CSMA/CA MAC protocol is constant and equal to the value specified by the parameter CWMIN of the IEEE 802.11 standard [11]. Moreover, it is well known that the broadcast communications cannot exploit the Ready To Send / Clear To Send (RTS/CTS) mechanism foreseen by the IEEE 802.11 standard. Hence, they suffer of the hidden terminal problem, that inevitably reduces the number of packets able to propagate till the farthest node of the network. We also remark that the DCF mechanism of the IEEE 802.11 standard obviously introduces an additional channel access delay somehow proportional to the traffic load and to the CWMIN parameter, thus increasing the end-to-end delay. We remark that while the traffic load is ineffective in the ideal configuration considered in [10], in the scenario of interest in this paper it will have a relevant impact. B. Simulations Setup We employ the same topology setup previously described in Subsection II-A, setting l = 15, and averaging the obtained results over 1000 independent scenarios. The source sends a burst of 1000 packets according to a Poisson transmission distribution with parameter λ (dimension: [pck/s]).2 We adopt two different values for λ: 0.1 pck/s for low intensity traffic and 100 pck/s for medium-high intensity traffic. Two values of the parameter c, namely 1 and 5, are chosen as representative of feeble and aggressive rebroadcasting policies, respectively. The results are obtained for a fixed node spatial density ρs = 0.01 veh/m, while the values of the transmission range z are selected in order to have the desired value of ρs z and are listed in Table I.3 We insert the IF protocol on top of the IEEE 802.11 model present in Network Simulator 2 (ns-2.31 [12]) after fixing the bugs reported in [13]. Since this work does not focus on physical layer issues, we adopt a simple Friis freespace propagation model [14]. The relevant parameters of the IEEE 802.11 network and of the IF protocol are listed in Table I. In particular, we remark the choice of a small packet size (105 bytes), in order to avoid fragmentation. Finally, in order to have a performance benchmark, we have also performed simulations of the classical flooding protocol (denoted as “flood” in the following figures), whose forwarding policy is trivial: every node rebroadcasts fresh packets with probability equal to 1 (i.e., always). 2 Our simulations show that the numbers of the generated scenarios (1000) and of the transmitted packets (1000) are sufficient to guarantee a confidence interval larger than 95%. We will thus omit any further discussion in our results analysis. 3 For small values of z, the network is rarely connected since P r{d > ρs −1 } is relatively high. On the other hand, the network gets connected with a high probability (almost 1), if z is longer than 750 m.

TABLE I M AIN SIMULATIONS PARAMETERS .

c λ ρs z l Packet Size Carrier Freq. Data rate CWMIN

{1, 5} {0.1, 100} pck/s 0.01 veh/m {100, 300, 500, 750, 1000, 1500, 2000} m 15 105 bytes 2.4 GHz 1 Mbps 31

C. Collected Statistics We define two categories of statistics: per-hop, that are relative to the behavior shown in a single transmission domain allowing; and global, that offer a more general characterization of the performance and do not depend on a specific transmission domain. In particular, referring to [1], we introduce three global statistics: the REeachibility (RE), the number of Saved ReBrodascast (SRB), and the end-to-end delay. The latter is the duration of the packet trip between the source and the lrn.4 The RE is the fraction of node that receives the source packet among the set of the reachable nodes, i.e., those topologically connected to the source, whose cardinality coincides with lrn. Intuitively, for a probabilistic protocol the RE is inversely proportional to the normalized distance between the source and the lrn, given by d0,lrn /z < l, since at every hop there is a non-zero probability of having zero transmissions cutting off the packet flow. For the same reason, paradoxically, less connected networks (with d0,lrn /z