QoS Extensions to Mobile Ad Hoc Routing ... - Semantic Scholar

QoS Extensions to Mobile Ad Hoc Routing Supporting Real-Time Applications K´aroly Farkas, Dirk Budke, Bernhard Plattner Computer Engineering and Networks Laboratory (TIK) Swiss Federal Institute of Technology Zurich (ETH Zurich) Gloriastrasse 35, CH-8092 Zurich, Switzerland Email: {farkas, budke, plattner}@tik.ee.ethz.ch Oliver Wellnitz, Lars Wolf Institute of Operating Systems and Computer Networks (IBR) Technische Universit¨at Braunschweig M¨uhlenpfordtstr. 23, D-38106 Braunschweig, Germany Email: {wellnitz, wolf}@ibr.cs.tu-bs.de

Abstract The increasing number and diversity of wireless mobile devices are driving a revolutionary change in the way users are accessing and using services and resources. Mobile ad hoc networks (MANETs), temporarily created by mobile devices, offer opportunities for sharing services and resources made available by member nodes. Especially, real-time applications have great potentials in mobile ad hoc environments as MANETs form an ideal technological basis for ad hoc real-time interaction among people. However, real-time applications require low latency connections with a minimum of jitter and packet loss rate from the underlying network. To meet these demands is a challenging task in mobile ad hoc networks due to the high level of node mobility, properties of the wireless communication channel and the lack of central administration. In this paper, via a simulation study we analyze and evaluate some Quality of Service (QoS) extensions, like priority queuing or rate control policies, to ad hoc routing in IEEE 802.11 wireless mobile ad hoc networks from the aspects of real-time applications. We show how these QoS extensions can improve the performance of AODV routing protocol and why we selected AODV as the starting point of our work.

1 Introduction Mobile ad hoc networks (MANETs) have been receiving much attention nowadays due to the continuously growing mobile device penetration, like notebooks, smart phones or palmtops, and their immense field of application. These

networks are often built from a collection of diverse mobile nodes connected to each other via wireless links forming a self-organizing wireless mobile multihop ad hoc environment. Real-time applications, like multimedia entertainment, group-work, multiplayer games or ‘edutainment’, are the most attractive candidates to be used over such networks because of the potentials this technology offers for ad hoc real-time interaction among people [7]. However, providing real-time services in such environment is a difficult task due to the mobility of the nodes, limited device resources, properties of the wireless channel and the lack of central administration. The routing mechanism used to transfer packets via multiple nodes from the source to the destination must cope with these difficulties. Moreover, it also has to be able to satisfy the Quality of Service (QoS) requirements of real-time applications such as connections with tight delay sensitivity and low packet loss rate [12]. In this paper, we analyze and evaluate via simulations, using the NS-2 network simulator [6], some Quality of Service extensions, such as priority queuing, hop-constrained queuing timeouts, real-time neighbor aware rate control policies, broken link detection and RTS/CTS (Ready To Send/Clear To Send) adaptation, to ad hoc routing in IEEE 802.11 [5] mobile ad hoc networks focusing on the demands of real-time applications. Moreover, we show why we selected the AODV [8] routing protocol as the starting point of our work. Though, in this paper we restrict our investigations to the individual impacts of the QoS extensions on the routing performance, a more comprehensive study considering even the combined impact of these and some other QoS mechanisms from the aspects of real-time multiplayer games can be found in our companion paper [2].

Application E-Mail File Transfer Web Remote Access Chat Audio streaming Video streaming Voice over IP Video Conference Multiplayer Games

Delay ◦ ◦ ◦ + + +

Jitter ◦ ◦ + + + + +

Bandwidth ◦ ◦ ◦ + + -

Reliability + + + + + ◦

Table 1. Typical Requirements of Popular Networked Applications The rest of the paper is organized as follows: In Section 2, we discuss the issues of QoS provisioning for real-time applications. Then, we introduce and discuss the different QoS extensions in Section 3. In Section 4, we show why we selected the AODV routing protocol as the starting point of our work and evaluate the QoS extensions via simulations. We survey related approaches in Section 5 and conclude the paper in Section 6.

2 Quality of Service Provisioning for RealTime Applications Providing Quality of Service for real-time applications in mobile ad hoc networks using wireless connections is especially challenging because wireless links suffer from high bit error rates and frequent link failures, mainly caused by the mobility of the nodes, interferences and multipath fading. At the same time, the characteristics of the applications often demand reliable data transfer, flow synchronization and delay sensitivity. In this section, we give a short introduction into the relevant QoS principles, discuss the substantial QoS requirements of real-time applications then describe our objective in regard to QoS provisioning in mobile ad hoc environments.

2.1 Quality of Service Various types of popular networked applications have emerged in the last couple of years with different demands on the network, such as delay, jitter, bandwidth and reliability. Delay: determines the end-to-end time it takes to transmit a packet from the source to the destination. This is often called latency, too. It is important to distinguish between end-to-end delay and round-trip time (RTT) delay which measures the time it takes to send a packet to the destination and back to the source. In particular

in wireless networks, the delay might not be symmetric and a connection can experience a higher delay in one direction than in the other direction. Jitter: describes how much the packets vary in latency and is determined by calculating the standard deviation of latency. Bandwidth: defines the maximum amount of data the network is able to transmit within a certain time frame. Reliability: specifies to which degree the network prevents transmission errors and thus garbled packets. The aim of Quality of Service provisioning is to assure that these QoS requirements are met by the network. Table 1 illustrates typical networked applications and their Quality of Service demands (low: -, medium: ◦, high: +) grouped into three categories: (1) high reliability, (2) minimal jitter and (3) high demands on latency and jitter. The first five applications require a strictly reliable transport mechanism, for example, transmission errors during a file transfer session yielding in a compromised copy of the transferred file are not acceptable. Audio and video streaming applications demand very low jitter while reliability is of less importance. Interactive real-time applications such as IP telephony and video conferencing have high demands on both delay and jitter but can cope with garbled packets to some extent. Real-time multiplayer games have similar demands but are more sensitive to garbled packets.

2.2 Requirements of Real-Time Applications As we can see in Table 1, the end-to-end communication delay, jitter and reliability (packet loss) to a certain degree are the most important QoS attributes of the network in case of interactive real-time applications due to the end users’ perception of interaction, while the available network bandwidth is of less importance (except Video Conferencing which has also high bandwidth demand).

Category Traffic Management Mac Layer Support

QoS Mechanism Priority Queuing Timeouts Rate Control Broken Link Detection RTS/CTS Adaptation

Mobility

√

Congestion √ √ √ √

Shared Medium √ √

Table 2. QoS Extensions For example, the upper bound of the affordable skew for interactive audio and video applications is estimated around 120 ms. For car racing games, a network delay in the range of 50 to 100 ms is already noticeable but is still tolerated by most players, whereas a delay of more than 150 ms results in remarkable degradation of end users’ performance. Similar results are applied for first-person shooter games, where a noticeable degradation can be perceived when the network delay exceeds 200 ms. However, the effects of jitter are not clear, yet. Rather, jitter and latency are strongly coupled for real-time applications and the users’ perception of jitter is application dependant. High level of jitter usually leads to packets not arriving in time thus requiring the use of prediction mechanisms. As these prediction mechanisms cannot always anticipate users’ actions accurately, high level of jitter usually degrades the end users’ experience. Hence, the jitter level must be kept as low as possible. Moreover, packet loss rate has similar effects and it also should be kept low (e.g., in the range of 3 – 5 % for real-time multiplayer games) [12].

2.3 Objective and Challenges Our objective in this work is to extend mobile ad hoc routing (we selected the reactive routing protocol AODV as the starting point of our work based on the simulation results shown in Section 4.2) with QoS mechanisms by which the performance of the ad hoc routing protocol and the network can meet or at least can get closer to the demands of real-time applications. However, to give QoS guarantees in MANETs is a difficult task and the QoS extensions have to face the following challenges: Mobility - The topology of the ad hoc network might change unpredictably resulting in broken links and stale routes. Congestion - Real-time traffic must arrive in-time even if the network is highly loaded. Shared Medium - IEEE 802.11 wireless networks operating in ad hoc mode do not provide any QoS guarantees at the MAC layer due to the applied contention based medium access mechanism [5].

3 Mobile Ad Hoc Routing Extended by QoS Mechanisms In this section, we overview and discuss the introduced QoS mechanisms in two categories such as traffic management and MAC (Medium Access Control) layer support, respectively. The different challenges of the mobile ad hoc environment which the various QoS mechanisms are supposed to address are depicted in Table 2.

3.1 Traffic Management The goal of traffic management is to differentiate among various types of traffic and give a higher level of support to high priority traffic (such as real-time traffic) at the cost of lower priority traffic (like background traffic). To achieve this, priority queuing can be used what we extended with hop-constrained queue timeouts and real-time neighbor aware rate control policies. Employing priority queuing, real-time packets are preferably transmitted to make sure that even in situations with higher load real-time packets do not become obsolete. Still, it might happen that realtime packets have to wait too long in the high priority queue. To prevent for transmission of outdated packets, we introduced hop-constrained queue timeouts dropping obsolete real-time packets and thus saving bandwidth. In addition, to limit the amount of low priority traffic, we used real-time neighbor aware rate control policies, which prevent the occupation of the communication channel by nodes sending low priority traffic if other nodes have high priority traffic to be sent. 3.1.1 Priority Queuing We have used and modified the interface queue sublayer of NS-2 to implement priority queuing mechanisms. In our implementation, the interface queue consists of three subqueues with different priorities and every packet is classified into one of three categories, as shown in Figure 1, employing the TOS1 field in the IP header. All data packets that do not have any particular real-time QoS demands are 1 Type Of Service - it is an 8 bit long field in the IPv4 packet header which can be, although it is rarely used for traffic management

marked as low priority, for instance, file transfer and e-mail traffic. AODV routing management packets such as R REQ, G-R REP2 and R ERR packets are marked as medium priority. And finally, real-time data packets are marked as high priority as well as AODV R REP messages if the high priority flag had been set in the corresponding R REQ. In contrast to medium and low priority, high priority packets are quickly outdated and dropped if they cannot be delivered in-time. Low priority packets might on the one hand experience higher delays, but on the other hand they are forwarded and delivered to the destination more reliably. Every subqueue has a limited size and allows a packet to be queued for a certain time interval. In order to limit the amount of real-time and routing packets a node is allowed to transmit, the size of the high and medium priority sub-queues has been restricted to 10 slots. A much longer queue for realtime packets would result in more outdated packets. As the packets are marked with different priorities they can be handled specifically by the routing protocol and the interface queue.

Figure 1. Design of the Priority Queue

3.1.2 Hop-Constrained Queue Timeouts As we discussed in Section 2.2, real-time applications demand low latency connections with minimum of jitter. Therefore, we drop packets which take more than 100 ms one-way and are clearly obsolete for real-time applications (assuming symmetric latencies and 200 ms as the maximum tolerable round-trip delay). Outdated real-time packets needlessly consume bandwidth and delay other real-time packets. Therefore, every packet that is stored in the queue is marked with a time-stamp. When the packet is dequeued the time the packet spent in the queue is compared with the queue policy. The maximum timeout for high priority traffic has been set to 100 ms. This timeout interval is further decreased depending on the number of hops the packet already traversed and still has to go. This information is re2 In

order to establish a bidirectional route between the source and destination nodes, an intermediate node that replies to an R REQ carrying the gratuitous flag sends an additional route reply message G-R REP to the destination

trieved from the local routing table. For every hop the maximal timeout is reduced by 10 ms which approximates the time it takes to process and forward the packet in the optimal case. The medium priority sub-queue has a timeout of 500 ms which has shown reasonable performance in our simulations. Since both priority sub-queues are rather small and the packets are outdated quickly compared to the low priority sub-queue, additional mechanisms that prevent for starvation of low priority traffic have not been added.

3.1.3 Real-Time Neighbor Aware Rate Control The current IEEE 802.11 protocol standards do not support QoS at the MAC layer and the medium access is carried out by the Distributed Coordination Function (DCF) handling every node equally [5]. All nodes have the same probability to gain access to the wireless channel and DCF does not distinguish between high priority and low priority data traffic. Furthermore, there is no guarantee that a node will send even high priority data packets within a certain time frame. Therefore, nodes that send high amounts of low priority data might consume most of the shared bandwidth. Due to the contention mechanism in DCF, other nodes which send high priority data might wait too long to access the channel and cannot transmit their high-priority data in time, although priority queuing has been used. The reason behind this is, that priority queuing works only locally and does not take neighboring nodes into account. To solve this problem at the MAC layer is one of the aims of the new QoS IEEE 802.11e [5] protocol. Even with the advent of IEEE 802.11e the current standard will not be displaced immediately and thus, a mechanism is required that enables nodes that send real-time data to refrain other nodes from accessing the channel. Our proposal is to limit the amount of low priority packets a node is allowed to transmit within a time interval depending on the amount of nodes handling real-time traffic in its neighborhood. Therefore, we have modified the interface queue to have access to the actual number of neighbors sending high priority traffic and thus adapt the rate control system of low priority traffic.

3.2 MAC Layer Support Unfortunately, in today’s networks many features, already implemented in the IEEE 802.11 MAC layer, remain unused in higher layers and are implemented again, however, less efficiently. With broken link detection based on Link Layer Feedback (LLF) the routing protocol can be notified instantly if packets cannot be sent any longer over a certain link. In addition, IEEE 802.11 relies on RTS/CTS (Ready To Send/Clear To Send) mechanism to avoid the

hidden terminal problem3 . However, RTS/CTS adaptation to the application’s requirements is essential to not degrade the overall performance. In our work, we also investigated the impacts of using broken link detection and adopting the RTS/CTS mechanism. 3.2.1 Broken Link Detection For broken link detection we employed the LLF mechanism of the AODV implementation. If a packet cannot be transmitted successfully, that is the sender does not receive an ACK from the destination within a certain time interval, the packet is retransmitted. Excessive retransmissions, however, can be reported by the MAC layer using LLF and this information can be propagated to the routing protocol to deal with the broken link. This mechanism has two benefits. First, it reacts to broken links by usually one order of magnitude quicker than relying on H ELLO messages. Second, H ELLO messages used for broken link detection are not required any more which reduces the routing overhead. 3.2.2 RTS/CTS Adaptation NS-2 supports a certain packet size threshold to indicate if RTS/CTS should be used. By default the threshold is 0 and RTS/CTS is always enabled. We disabled RTS/CTS for real-time packets. Without RTS/CTS the transmission channel is shared equally among the contenting nodes and it requires less bandwidth and presumably reduces end-toend delay and delay jitter.

4 Evaluation In this section first, via simulations, we discuss why we selected the AODV ad hoc routing protocol as the starting point of our work, then we evaluate the introduced QoS mechanisms.

4.2 Ad Hoc Routing Protocol Comparison We have run an initial simulation study to find out, which ad hoc routing protocol has the best performance and can be used as the starting point of our work. We investigated four different protocols, namely AODV (Ad Hoc On-Demand Distance-Vector), DSR (Dynamic Source Routing), DSDV (Destination-Sequenced Distance Vector) and OLSR (Optimized Link State Routing) [8]. We used NS-2’s built-in implementations of DSR and DSDV, AODV-UU [3] from University of Uppsala and UM-OLSR [11] from University of Murcia in the simulations. In order to evaluate the performance of ad hoc routing protocols we used the following metrics: Latency: the average time in ms it takes to transmit a packet from the source to the destination. Jitter: it describes how much the packets vary in latency and is determined by calculating the standard deviation of the latency. Loss rate: the loss rate determines the amount of sent packets in relation to the amount of packets that have not been received successfully at the destination. Parameter Simulation Time Nodes Mobility Model Area Speed Pause Time Traffic Type Connections Packet Size

Value 600 s 20 Random Way Point (RWP) 650x650 m2 0-3 m/s 180 s CBR with 20 packet/s 15 parallel unidirectional, 150 s 64 bytes

Table 3. Parameters of the Ad Hoc Routing Protocol Comparison

4.1 Simulation Environment We have used, in our simulations, the discrete event network simulator NS-2 [6] including the wireless extensions from the Monarch group [1] to model the IEEE 802.11 MAC layer, node mobility, radio network interfaces and physical layer. Throughout the simulations, each mobile node shares a 2 Mbit/s radio channel with its neighbouring nodes, using the IEEE 802.11 MAC protocol and tworay ground reflection model [10]. The transmission range of each node is 250 m, which is a typical value for WLAN in a free area without any obstacles. 3 The hidden terminal problem arises if two nodes transmit concurrently to the same destination without knowing of each other

4.2.1 Simulation Settings The simulation scenario consists of 20 nodes in an area of 650x650 m2 . All nodes follow the Random Way Point (RWP) mobility model [1] with a uniformly distributed speed between 0-3 m/s and a pause time of 180 seconds. On the average 15 randomly selected unidirectional UDP data flows transmit 20 packet/second Constant Bit Rate (CBR) traffic with 64 byte packet size for 150 seconds. For every routing protocol 60 sessions have been simulated with a duration of 600 seconds using different seed values. The detailed simulation parameters are depicted in Table 3. We

averaged the gained simulation results which are shown in the following subsections. 4.2.2 Latency and Jitter Figure 2 illustrates the performance of latency and jitter for the four routing protocols. AODV shows the lowest average latency of 125 ms. All other protocols have much higher latency values, in particular DSR, that has the highest latency of 218 ms. When comparing jitter, the reactive protocols DSR and AODV provide the lowest delay jitter around 300 ms while the proactive protocols OLSR and DSDV suffer from higher jitter values of 400 ms and 450 ms, respectively.

4.2.4 Selected Routing Protocol With regard to latency and jitter, AODV showed the best average performance while all other protocols produced much higher values in the range of some hundred milliseconds. DSR’s performance was by far the worst in these comparisons. Concerning loss rate, the reactive protocols (AODV and DSR) showed the best performance with less than 10 % packet loss while the proactive protocols (DSDV and OLSR) produced significantly higher losses. Based on these initial simulation results, we selected AODV as the routing protocol of choice for the following simulations.

4.3 Impacts of the QoS Mechanisms In the following, we show and discuss our simulation results investigating the impacts of the introduced QoS mechanisms. We have run the simulations independently to see the single impact of each QoS extension. We have examined latency, jitter and packet loss in these simulations, too. 4.3.1 Simulation Settings

Figure 2. Comparison of Latency and Jitter

4.2.3 Loss Rate The reactive protocols DSR and AODV show the lowest loss rate, namely 7 %, as depicted in Figure 3. The proactive protocols OLSR and DSDV have a loss rate of 13 % and 19 %, respectively, which is significantly higher than in case of the reactive protocols. Better loss rate results for the proactive protocols might be achieved by increasing the rate of periodic route advertisements to propagate link changes faster. Reactive protocols automatically adapt to the given scenario and send more routing messages in case of broken links to discover a new route.

We used the same simulation scenario as in case of the routing protocol comparison, but with different traffic properties (see Table 4). We differentiated between real-time or high priority, and background or low priority traffic. We simulated three real-time connections at a time as bidirectional UDP data flows with a CBR traffic of 20 packet/sec and a packet size of 64 bytes. In addition, we used always five parallel data flows between any two random nodes as background traffic employing the same traffic pattern as in case of the real-time traffic. Every real-time session took 600 seconds whereas the background traffic sessions took 150 seconds. We repeated all the simulations 10 times with different seed values and averaged the results. Parameter Traffic Type Real-Time Connections Background Traffic Packet Size

Value CBR with 20 packet/s 3 bidirectional, 600 s 5 bidirectional, 150 s 64 bytes

Table 4. Traffic Parameters for the QoS Simulations

4.3.2 Priority Queuing

Figure 3. Comparison of Loss Rate

By applying priority queue latency and jitter of high priority (real-time) traffic, as depicted in Figure 4 and 5, remain almost unchanged for two-hop connections. For connections with more than two hops latency and jitter have been increased by 5 ms. In case of four-hop connections this effect

is alleviated and even with priority queueing many packets do not reach the destination in-time. At the same time, the average loss rate has been reduced by 15 % from 55 % to below 40 %, as depicted in Figure 6. In particular, three-hop connections benefit from priority queueing and the loss rate has been reduced by almost 30 %. However, regarding low priority (background) traffic, latency and jitter have been more than doubled and show values around 2400 ms. The loss rate of low priority traffic has been increased by 1 % (see Figure 7). These results for the low priority traffic have been expected since the priority queue prefers high priority in favor of low priority traffic. 4.3.3 Hop-Constrained Queue Timeouts The queue timeout mechanism is based on the priority queue and therefore the simulation results can only be compared to the configuration that uses the priority queue. The average loss rate of high priority traffic is reduced from 40 % to 32 % by applying queue timeouts while latency and jitter have been slightly increased in all cases (see Figure 4, 5 and 6). At the first glance this might be surprising since the queue timeout mechanism dropped high priority packets on purpose and at the same time the loss rate has been reduced. However, only already outdated packets are dropped. Queue timeouts have a very positive effect on low priority traffic since latency and jitter can be reduced by 50 %. Furthermore, the loss rate is reduced from 30 % to 25 % as shown in Figure 7.

4.3.6 RTS/CTS Mechanism It is an interesting observation, that the RTS/CTS mechanism degrades the performance of the routing protocol. The latency and jitter, and even the loss rate of the real-time traffic are significantly higher if RTS/CTS is enabled (see Figure 4, 5 and 6). The reason is that the simulated data packets are very small in our scenario and the RTS/CTS mechanism induces a huge overhead in this case decreasing the performance in all metrics substantially. As a result, RTS/CTS should not be activated by default and instead a threshold should be maintained by the system that depends on the number of neighbors and the size of the data packets that have to be transmitted.

Figure 4. Latency of Real-Time Traffic

4.3.4 Real-Time Neighbor Aware Rate Control The rate control system relies on the priority queue as well and is compared to the performance results of using the queuing mechanism. The loss rate of real-time traffic is further reduced by 3 % in the average case whereas the loss rate of low priority traffic is increased by 1 % as shown in Figure 6 and 7, respectively. Latency and jitter of realtime traffic don’t experience any changes (Figure 4 and 5), since latency of low priority traffic remains unchanged at a very high level of 2500 ms and its jitter has been slightly increased (Figure 7). This outcome has been expected since the rate control mechanism limits low priority traffic.

Figure 5. Jitter of Real-Time Traffic

4.3.5 Broken Link Detection Using the LLF mechanism of the MAC layer broken links can be detected very quickly. In case of real-time traffic, the average latency can be reduced to below 25 ms and the jitter to below 17 ms if LLF is activated (see Figure 4 and 5). However, the packet loss rate can slightly increase as depicted in Figure 6. Moreover, the amount of routing traffic is much smaller if LLF is used since AODV does not require H ELLO messages any more for broken link detection.

Figure 6. Loss Rate of Real-Time Traffic

Acknowledgment This work has been partly supported by the European Union under the E-Next NoE FP6-506869 project.

References

Figure 7. Latency, Jitter and Loss Rate of Background Traffic

5 Related Work Many QoS architectures have been proposed for MANETs. Proposals like INSIGNIA [14], FQMM [13], or CEDAR [9] use a reservation-oriented approach and keep per-flow state information at the mobile nodes. Other approaches like SWAN [4] use a stateless feedback-based mechanism to achieve soft real-time services. All of these QoS architectures work at the networking layer or above, so they make little use of information which is available at the lower layers. They also provide a self-contained solution to the QoS problem in MANETs. However, the focus of our paper is on real-time applications to determine which individual QoS mechanism is helpful in reducing latency, jitter and packet loss in MANETs. Most of these mechanisms can be used in addition to the QoS architectures mentioned above.

6 Summary In this paper, via simulations we analyzed and evaluated some QoS extensions to mobile ad hoc routing. We grouped these extensions into two categories like traffic management mechanisms and MAC layer support mechanisms. In the simulations, we used AODV routing protocol as the starting point of our work because AODV had shown the best initial performance compared to some other ad hoc routing protocols such as DSR, DSDV and OLSR. We showed, that in case of real-time traffic RTS/CTS adaptation, priority queueing, timeouts and rate control provided significant gain in improving the loss rate since latency and jitter remained in the acceptable range. However, the loss rate highly depends on the number of hops and none of these QoS mechanisms can satisfy alone the strict demands of real-time applications. Thus, the investigation of their combined effects is desirable what we have already done in our companion paper [2]. As future work, we plan to investigate the discussed QoS extensions in a real test-bed environment.

[1] J. Broch, D. A. Maltz, D. B. Johnson, Y.-C. Hu, and J. Jetcheva. A performance comparison of multi-hop wireless ad hoc network routing protocols. In Proceedings of ACM/IEEE International Conference on Mobile Computing and Networking (MobiCom), pages 85–97, October 1998. [2] D. Budke, K. Farkas, O. Wellnitz, B. Plattner, and L. Wolf. Real-Time Multiplayer Game Support Using QoS Mechanisms in Mobile Ad Hoc Networks. In Proceedings of The Third Annual Conference on Wireless On demand Network Systems and Services (WONS 2006), Les M´enuires, France, January 2006. [3] Erik Nordstr¨om. AODV-UU implementation. http://core.it.uu.se/AdHoc/AodvUUImpl. [4] Gahng-Seop Ahn and Andrew T. Campbell and Andras Veres and Li-Hsiang Sun. SWAN: Service Differentiation in Stateless Wireless Ad Hoc Networks. In Proc. IEEE INFOCOM 2002, pages 457–466, June 2002. [5] IEEE-SA Standards Boards. Part11-Wireless LAN Medium Access Control (MAC) and Physical Layer (PHY) Specifications. 12 June 2003. http://standards.ieee.org/getieee802/802.11.html. [6] Information Sciences Institute ISI. The Network Simulator NS-2, February 2005. http://www.isi.edu/nsnam/ns. [7] Mobile Entertainment Industry and Culture (mGain). Mobile Entertainment in Europe: Current State of the Art. http://www.mgain.org/MGAIN-wp3-d311-revisedfinal.pdf, April 2003. [8] C. E. Perkins, editor. Ad Hoc Networking. 2001. [9] Prasun Sinha and Raghupathy Sivakumar and Vaduvur Bharghavan. CEDAR: a core-extraction distributed ad hoc routing algorithm. In Proc. IEEE INFOCOM 1999, pages 202–209, Mar. 1999. [10] T. S. Rappaport. Wireless Communications, Principles and Practice. PRENTICE HALL INTERNATIONAL, 1996. ISBN: 0-13-042232-0. [11] P. Ruiz and F. Ros. UM-OLSR Webpage. http://ants.dif.um.es/masimum/. [12] R. Steinmetz. Human Perception of Jitter and Media Synchronization. IEEE Journal on Selected Areas in Communications, Vol. 14(1):61–72, January 1999. [13] Xiao Hannan and Chua Kee Chaing and Seah Khoon Guan Winston. Quality of service models for ad hoc wireless networks. CRC Press, Inc., Dec. 2002. [14] Xiaowei Zhang and Seoung-Bum Lee and Gahn-Seop Ahn and Andrew T. Campbell. INSIGNIA: An IP-based Quality of Service Framework for Mobile Ad Hoc Networks. Journal of Parallel and Distributed Computing, Special Issue on Wireless and Mobile Computing and Communications, Vol. 60(4):374–406, April 2000.