Deployment of VoIP in the IEEE s Wireless Mesh Networks

Transcription

1 Deployment of VoIP in the IEEE s Wireless Mesh Networks Master Research Report June 6, 2008 Submitted By: Abbasi Ubaid Supervised By:Adlen Ksentini Team: DIONYSOS (IRISA)

2 Abstract Wireless mesh networks (WMNs) are shaping up into a critical part of internet protocol (IP) networks. The integration of WMNs with other networks such as the Internet, cellular, IEEE , IEEE , IEEE , sensor networks, etc., can be accomplished through the gateway and bridging functions in mesh routers. Voice over IP (VoIP) service meanwhile, known to be a great concern from last decade, however before deploying such application in WMNs there is a need of investigating the WMNs capacity to support quality of service (QoS). When using VoIP, factors like delay, packet loss and overhead require special consideration. Traditional routing protocol like dynamic source routing (DSR) fail to show satisfactory performance due to very high delay. Hybrid Wireless Mesh Protocol (HWMP) is designed to provide better QOS in WMNs. In this paper we present a performance comparison of different routing protocols for VoIP traffic. On the basis of result obtained, we design a new scheme for routing in WMNs. This scheme has been implemented and then compared with existing solutions. Keywords: Wireless Mesh Network, Voice over IP, Quality of Service ii

3 Acknowledgements I would like to thanks my supervisor Adlen ksentini for suggesting the topic for internship and supporting me whenever I face some problem. Without his help and cooperation this work was not possible. I am really grateful for his kind support. I would like to thanks my parents who were the source of encouragement for me. Without there support it would not possible to study here in France. iii

4 Table of Content 1 Introduction Background Proposed Goals Document Outline Background Wireless Mesh Networks System Architecture Physical Layer MAC Layer Routing Protocol WMN Application Scenarios Voice over IP Overview Characteristics of VoIP Traffic Delay Packet Loss Jitter VoIP Performance Comparison in IEEE802.11s QualNet Components of QualNet Developer Performance Evaluation Results Performance Comparison of HWMP Problem Definition Proposed Solution Optimization & Maintenance Preliminary Results Average End to End Delay Routing Overhead Conclusion...28 References...29 iv

5 List of Figures Figure 2-1Infrastructure WMNs...4 Figure 2-2. Client WMNs...4 Figure 2-3. Hybrid WMNs...5 Figure 2-4 Distributed MAC Protocols...7 Figure 2-5. Cost Function based on Airtime Link Metrics...11 Figure 2-6.AODV Route Discovery...12 Figure 2-7. HWMP Route Discovery...13 Figure 4-1. Mesh Point uniformly placed in QualNet...18 Figure 4-2. End to End delay comparison...19 Figure 4-3. Packet Loss Due to No Route...20 Figure 4-4. Routing Overhead comparison...21 Figure 4-5. Data Throughput...22 Figure 5-1. Average End to End delay Comparison...22 Figure 5-2. HWMP Packet Loss...23 Figure 5-3. Data Throughput for HWMP...24 Figure 5-4. HWMP Route Comparison...25 Figure 6-1. End to End delay comparison...27 Figure 6-2. Routing Overhead Comparison...28 v

6 1 Introduction 1.1 Background Wireless Mesh Network (WMN) is a communication network which provides multi-hop communication over wireless links, thus increasing the effective coverage area. Today, WMNs are widely known fore-runner of wireless multihop networking and network deployment [1]. A lot of research has been carried out in WMNs during last few years. An IEEE draft for WMNs has been accepted in April 2008 [2]. Several emerging and commercially interesting applications for commodity networks based on WMN architecture have been deployed recently [3]. These include intelligent transportation system and residential broadband access for hard to reach and scarcely populated areas. Nowadays data network are widely used for carrying different type of traffic, including voice, video and so on. In a VoIP system, voice is digitized and sent over an IP network in small packets. Therefore a single network can be used for carrying both voice and data traffic. This makes VoIP cost efficient and flexible. VoIP software phones like Skype has also increase the popularity of VoIP among home users. Using WMNs for transporting VoIP traffic is a cost effective solution. However, providing VoIP service over WMN is a challenging task. A practical framework for providing VoIP in WMN must include VoIP packet aggregation, efficient path selection technique to route VoIP traffic and efficient QOS mechanism. The path selection mechanism in WMN refers to the forwarding of data frames on a single path or multiple paths. The shortest path will ensure minimum delay, thus enhancing the performance of VoIP. 1.2 Proposed Goals The objectives of this master s thesis are: To provide an overview of the existing routing protocols in WMN. To provide a performance comparison of VoIP for different routing protocols in WMN. To suggest a solution, which enhance the performance of existing routing protocol in WMN for VoIP and other delay sensitive traffic. To implement this solution in the simulator 1

7 To evaluate the performance of new protocol with existing protocols using VoIP on application layer. 1.3 Document Outline The rest of this paper is organized as follows: Section 2 provides the background knowledge on WMNs, Medium Access Control (MAC) layer and routing in WMNs. Section 3 discusses the characteristics of VoIP. Section 4 presents VoIP performance comparison for different routing protocols used in wireless networks such as dynamic source routing and Adhoc on demand vector. Section 5 proposes the new algorithm for improving VoIP characteristics in WMNs. Implementation of the suggested solution and comparison is described in section 6. Section 7 sums up the report. Future work is also discussed in this section. 2

8 2 Background This chapter provides the necessary background information about WMNs and VoIP in order to understand next chapters. 2.1 Wireless Mesh Networks System Architecture Wireless local area networks (WLANs) have become very popular in recent years. IEEE is the standard for WLAN. This specification defines a physical layer and an Ethernet like MAC layer for wireless links [3]. IEEE consists of mobile stations (STAs) and access points (APs). A mobile station is a network device with a wireless network interface card. APs are acting as bridge by providing connectivity to stations. APs are connected to each other through wired links. IEEE provides a cost effective and simple way for wireless networking. However the problem is the wired connection between the APs. The wired links increases complexity and deployment cost in many situations [4]. Therefore it is desirable to connect the APs via wireless links as well and create a WLAN Mesh. In WMNs, APs turn into mesh access points (MAPs). Mobile stations are sometimes referred as mesh clients. The new IEEE s standard for WMNs introduces a third class of nodes called mesh points (MPs) [5]. MPs and MAPs support WLAN mesh services, allowing them to forward packets on behalf of other nodes to extend the wireless transmission range. Mesh clients can associate with MAPs but not with MPs. Mesh portals (MPPs) are MAPs which provides connectivity to other network thus acting as a gateway for Mesh Network. According to [1] there are three different types of WMNs Infrastructure WMNs: Infrastructure WMNs uses MPs and MAPs as backbone for the clients. Availability of gateway like MPPs provides connectivity to external networks such as Internet. The clients connect to the MAPs via standard , but do not forward packets. This is the most used architecture nowadays. APs are also used for increasing coverage area. Figure 2-1 shows typical structure of this type of network. 3

9 Figure 2-1Infrastructure WMNs Client WMNs: In this type of architecture, client nodes constitute the actual network to perform routing and configuration functionalities as well as providing end user applications to customers. As an example let us consider a collection of MPs connected to each other, they can communicate to each other within network and also forward data on behalf of other. In Client WMNs, a packet destined to a node in the network, hops through multiple nodes to reach the destination. Client WMNs are usually formed using one type of radios on devices. Moreover, the requirements on end-user devices is increased when compared to infrastructure meshing, since, in Client WMNs, the end-users must perform additional functions such as routing and selfconfiguration. Figure 2-2 shows the topology of client WMNs. Figure 2-2. Client WMNs Hybrid WMNs: This architecture is the combination of infrastructure and client meshing. Mesh clients can access the network through MPs as well as directly meshing with other mesh clients. While the infrastructure provides connectivity to other networks such as the Internet, Wireless 4

10 Fidelity (Wi-Fi), Worldwide Interoperability for Microwave Access, Inc. (WiMAX), cellular, and sensor networks; the routing capabilities of clients provide improved connectivity and coverage inside the WMN. The hybrid architecture will be the most applicable case in our opinion. IEEE s is the typical scenario of Hybrid WMNs, we have clients (Ordinary stations), which uses MAPs for accessing network and we have independent nodes (MPs) which can directly access the network. All these entities are based on based radio technology. The typical structure of hybrid WMN is shown in figure 2-3. Figure 2-3. Hybrid WMNs Due to different system architecture mesh networks have different requirement to the physical layer, the MAC mechanism and the routing protocol than legacy IEEE LANs. The changes made in above layer for WMNs are discussed in next section Physical Layer The original standard IEEE supports two physical layers which are based on radio transmission and infrared. Infrared physical layer has not been used today; however IrDA is the dominant standard for infrared communications [4]. For radio transmission, either frequency hopping spread spectrum or direct sequence spread spectrum are used. These two techniques aim to maximize interference resistance and signal robustness. IEEE uses either Gaussian frequency shift keying (GFSK) or phase shift keying (PSK). Data rates of 1 & 2Mbit/s are possible whereas IEEE n [6] uses orthogonal frequency 5

11 division multiplexing (OFDM) with multiple antennas, thus increasing the data rates up to 100 Mbit/s. In WMNs new antenna system has been proposed for increasing physical layer performance [1]. The use of multiple antennas at the sender and the receiver to create a multiple-input multipleoutput (MIMO) system can triple the performance. IEEE s doesnot include a physical layer specification MAC Layer MAC is required to provide fair access to resources & efficient utilization of Bandwidth. MAC layer is also responsible for providing system authentication, association with an access point, encryption and data delivery [7]. In the next section we will discuss MAC protocols for WLANs and the challenges and solutions for MAC protocols in WMN MAC Protocol for WLANs IEEE has specified two MAC protocols for single hop WLANs. These protocols are Point Coordination Function and Distributed Coordination Function. A. Point Coordination Function The optional priority-based point coordination function (PCF) provides contention-free frame transfer for processing time-critical information transfers. With this operating mode, a point coordinator resides in the access point to control the transmission of frames from stations. When a station wants to send data, it waits for short inter-frame spacing (SIFS), and then start transmission. If there is no data to send the point coordinator waits for PCF inter-frame spacing (PIFS) and poll the next station. PCF cannot provide quality of service (QOS) mainly due to delayed contention period. B. Distributed Coordination Function The basic access mechanism, called Distributed Coordination Function (DCF) is basically a Carrier Sense Multiple Access with Collision Avoidance mechanism (CSMA/CA). In CSMA/CA When a station wants to send, it senses the medium. If the radio channel is idle for at least the duration of DCF inter-frame spacing (DIFS), the station sends. If the medium is busy it waits until it is idle for DIFS. Then it enters a contention phase choosing randomly a multiple slot-time within a contention window (CW) as a back off timer. If the medium is idle when the timer expires, the station transmits. If the medium is busy before the timer expires, the station stops the timer and starts it again after the channel is idle for DIFS [8]. After each unsuccessful transmission attempt the size of the CW size is doubled. As a consequence, the waiting time increases and the 6

12 probability of a concurrent transmission decrease. As a result of this scheme the delay is increased when a high network load is present. CSMA/CA does not overcome the hidden terminal problem, since collisions can still occur at the receiver. Therefore an extension to DCF was made, introducing two new control packets and a virtual channel reservation scheme [8, 9]. It is called Distributed Coordination Function with RTS/CTS extension. Using this scheme a station waits for the duration of DIFS and then sends a request to send message (RTS). This includes a duration field specifying the expected time needed for the transmission of data and acknowledgement. Every node overhearing the RTS stores the medium allocation in his net allocation vector (NAV). The receiver waits for SIFS and sends a clear to send (CTS), also including the duration field. Again, all nodes overhearing the CTS set their NAV. From now the channel is reserved for the sender. He waits for SIFS and transmits the data. The receiver responds with ACK after waiting SIFS MAC Protocols for WMNs Distributed nature of WMN restrict the usage of MAC protocols like PCF therefore distributed MAC protocol (for e.g. DCF) can be used but these protocols shows poor performance in WMN and doesn t provide fairness [10]. It is also observed that protocols for multi-hop ad hoc WLANs do not meet all requirements for WMNs MAC protocols because WMNs have different properties than ad hoc networks [11]. WMNs have low mobility (sometime zero mobility) without any consideration for energy constraints. Therefore specialized MAC protocols should be designed for WMNs. Several ways of implementing MAC protocols in WMNs are shown in figure 2-4. Nodes either use single channel or multiple channels. Figure 2-4 Distributed MAC Protocols 7

13 I. Single Channel MAC Protocols There are different ways to adapt MAC protocols for WMNs. One solution is to change parameters like contention window size or back off procedures, but this didn t result in major performance gain [1]. Other solution is to combine physical layer enhancements such as directional antennas with new MAC protocols. New WMN MAC protocols focus on QOS mechanism and fairness [12]. Traffic prioritization along with resource reservation enhances QOS. II. Multi Channel MAC A multi-channel MAC can be implemented on several different hardware platforms, which also impacts the design of the MAC. There are three different ways to implement multi-channel operations. The first option is multi-channel single transceiver MAC. In this mechanism nodes have a single transceiver, which utilize multiple channels but it can send on only one channel at a time. MAC protocol selects the transmission channel. Multi-channel Multi-Transceiver MAC allows the nodes to select multiple channel simultaneously however there is a single MAC protocol to coordinate transmission. The third variant Multi-radio MAC uses an own MAC protocol for each of its independent radios and puts a virtual MAC protocol on top. Multi-channel single-transceiver MAC doesn t require additional hardware for multiple transceivers so it is easier to implement. Wireless mesh network standard IEEE s will support an optional Multi-channel singletransceiver MAC protocol called Common Channel Framework (CCF) [5]. The sender transmits a Request-to-Switch (RTX) message (instead of Request to send), which includes a suggested channel number, on a common channel to the receiver. The receiver responds with a Clear to Switch (CTX) (instead of Clear to send), accepting or declining the channel selection. After a successful RTX/CTX handshake sender and receiver switch to the channel, which causes the channel switching delay. Then the sender transmits the data and the receiver responds with an ACK. Afterwards both tune back to the common channel. To increase the utilization of the common channel a channel coordination window (CCW) is defined, in which the common channel is solely used for RTX/CTX. Outside the CCW the common channel can also be used for data transfers. The CCW is repeated after a defined period Routing Protocol Introduction Wireless mesh networks are multi-hop networks. Therefore a mechanism for finding a path between source and destination is needed. Static routing means that the path is set up manually, while dynamic routing requires a routing protocol which sets up routing tables. A router forwards 8

14 packets to a next hop neighbour, which is chosen upon a routing metric. This process is called routing [13, 14]. Wired networks either uses distance vector or link state routing protocol. With distance-vector protocols each node keeps a vector of its neighbours and their distance. The router periodically broadcasts the distance-vector to its one-hop neighbours. Distance is usually the hop count or weighted hop count. Routing Information Protocol (RIP) and Border Gateway Protocol (BGP) are two famous distance vector protocols. On other hand link state routing protocols construct a map of the connectivity of the network, in the form of a graph showing which nodes are connected to which other nodes. Finally using some shortest path algorithm like dijkstra s, best path to the destination is calculated. WMNs require a different classification of routing protocols. In WMNs routing protocols are classified as either proactive or reactive. As the name suggests proactive routing protocol keep the routes before one is needed. It tries to keep up to date information. Reactive protocol finds a route only when a node wants to communicate with another. Hybrid routing protocols are combination of the above two. Some routes to the destination are maintained proactively and other are created on demand. Two major attributes of routing protocols are efficiency and convergence time. Efficiency is the share of routing traffic in overall traffic. Proactive routing protocols are efficient in high load scenarios while reactive protocols work well in low traffic settings [5]. Convergence time refers to the time span needed to have correct routing tables after a topology change. Link state protocols usually have a better convergence time than distance vector protocols. Routing Protocols in WMNs has few differences as compared to conventional protocols. Hybrid Wireless Mesh Protocol (HWMP) which is the default protocol for IEEE s operates on MAC layer. This is because MPs and MAPs don t support IP addresses. Other reason may be that multi-radio routing for capacity improvement and multi-path routing for load balancing and fault tolerance can be desirable in WMNs. Routing Metrics in WMNs Routing metrics decides if one route is better than other. Selection of an appropriate routing metrics affects the network performance significantly. Conventional routing metrics used in wired networks like hop count, link state e.t.c are not suitable in WMNs. Manual assignment of link cost is not possible in self organizing WMNs. Hop count metrics may not provide the best path as a path having more hops can provide better links as compared to a single hop path. In order to ensure that the efficient utilization of resources, minimum weighted path selected by routing protocols must have good performance on term of high throughput and low packet delay.therfore a routing 9

15 metrics is required that can capture the characteristics of WMNs [15]. IEEE s WMNs uses Airtime Link metrics for calculation of link cost Air Time Link Metrics In order to compute the forwarding table for individually addressed frames from the cached link state information generated by each MP, the MP shall first calculate the link cost for each pair wise link in the Mesh. This sub clause defines a default link metric that may be used by a path selection protocol to identify an efficient radio-aware path. The extensibility framework of IEEE802.11s allows this metric to be overridden by any path selection metric. The cost function for establishment of the radio-aware paths is based on airtime cost. Airtime cost reflects the amount of channel resources consumed by transmitting the frame over a particular link. This measure is approximate and designed for ease of implementation and interoperability [2]. Airtime cost for each link can be calculated as: c a O ca O p Bt 1 r 1 e pt Where O ca, O p and B t are constants, and represent channel access overhead, protocol overhead and number of bits in test frame respectively. The input parameters r and ef are the bit rate in Mb/s and the frame error rate for the test frame size B t respectively. The rate r represents the rate at which the MP would transmit a frame of standard size (B t ) based on current conditions. The frame error rate ef is the probability that when a frame of standard size (B t ) is transmitted at the current transmission bit rate (r), the frame is corrupted due to transmission error, and its estimation is a local implementation choice. Fig 2-5 shows the link cost percentage calculated by Airtime link metrics. Lower value metrics shows better link. 10

16 Figure 2-5. Cost Function based on Airtime Link Metrics Dynamic Source Routing DSR is an efficient routing protocol designed for Adhoc networks. Intermediate nodes are used to forward packets over multiple hops between nodes not directly within transmission range of each other. The protocol is composed of two mechanisms of Route Discovery and Route Maintenance, which work together to allow nodes to discover and maintain source routes to arbitrary destinations in the Adhoc network [16]. In DSR if the source node wants to communicate with a certain destination, having the route to destination in its cache, it will insert the route in data packet headers. If source nodes don t have the route information it will start a route discovery process. In this process route request (RREQ) packets are flooded in the network until these packets reach the destination. Destination node on receiving the first route request will send the route reply (RREP) message to the source node. In case of link breakage route error (RERR) message is sent to the destination and destination will then test another route. Therefore, only after making error in current route, destination seeks another route. This mechanism causes delay in packet delivery [17]. The protocol allows multiple routes to any destination and allows each sender to select and control the routes used in routing its packets. Adhoc On-demand Distance Vector Routing (AODV) AODV is a hop by hop routing protocol developed for wireless Adhoc networks [18]. It offers quick adaptation to dynamic link conditions, low processing and memory overhead. When a host wants to find a route to a destination it broadcast a RREQ message. The RREQ contains addresses (source and destination), sequence number and a broadcast identifier. 11

17 Nodes other than destination receiving RREQ message either re-broadcast or respond with RREP, depending on flags setting in RREQ message. When forwarding a RREQ node stores broadcast identifier, source address and maintains a reverse route. In order to avoid loop, RREQ are re-broadcasted only when a request with the same source address and broadcast identifier has not been processed before. Concept of sequence number is used for route updation. Thus an intermediate host replies with a RREP when it has a fresh enough route to the destination. Figure 2-6 shows a typical example of route discovery using AODV protocol. RREQ message was broadcasted by source node. Intermediate node creates and maintains a reverse route to the source node. Destination node, on receiving RREQ sends a unicast RREP to the source node on the same path that was created during RREQ., Figure 2-6.AODV Route Discovery Hybrid Wireless Mesh Protocol HWMP is the default protocol for IEEE s WMN. This protocol utilize layer 2 addressing therefore IEEE s denotes it as path selection protocol instead of routing protocol. The word Hybrid in this protocol refers to the fact that it supports both Reactive and Proactive routing. HWMP utilizes the features of reactive AODV routing protocol called Radio-Metric AODV (RM- AODV). HWMP supports two modes of operation depending on the configuration. These modes are: On demand: this mode allows MPs to communicate using peer-to-peer paths. The mode is used in situations where there is no root MP configured. If no root portal is configured, RM-AODV is used 12

18 for path selection. For destinations within the mesh the route discovery works like normal AODV. If the destination is outside the mesh, the source receives no RREP upon a RREQ. Therefore it sends the messages to the route portal after a timeout. The portal forwards them to the connected network [19]. Proactive: Route is discovered before any request or demand and as a result when request arrived for a particular destination node it is fulfilled (if route to destination exists).root Portals (also called Mesh Portal) are configured to send announcement called root announcement (RANN) periodically. The root MP periodically floods a RANN message into the network. The information contained in the RANN is used to disseminate path metrics to the root MP. Upon reception of a RANN, each MP that has to create or refresh a path to the root MP sends a unicast path request (PREQ) to the root MP via the MP from which it received the RANN. The unicast PREQ follows the same processing rules defined in the on demand mode. The root MP sends path reply PREP in response to each PREQ. The unicast PREQ creates the reverse path from the root MP to the originating MP, while the PREP creates the forward path from the MP to the root MP. When the path from an MP to a root MP changes, it may send a PREP with the addresses of the MPs that have established a path to the root MP through the current MP. A mesh portal connects mesh networks to outside network like internet. A designated mesh portal (MPP) is selected as designated root MPP. This selection is done either by configuration or by selection process. As a result we have a tree structure with a root and thus it allows proactive routing towards MPP. Figure 2-7 shows the process of HWMP route discovery. Figure 2-7. HWMP Route Discovery 13

19 Summary Traditional routing protocols are not feasible for IEEE s WMN. AODV are not very effective due to the reason that it use hop-count metrics. Layer 2 routing is effective for the scenario of IEEE802.11s, thus HWMP address the short coming to some extent WMN Application Scenarios The ability to cover large areas with cheap hardware and a self-organizing nature creates a number of interesting application scenarios for WMNs, which are not possible with legacy IEEE LANs or cellular networks. Some of the applications suggested in [1] are: Community/public access networking: WMNs can be used to share internet access and provide location based services. The WMN can cover limited areas like university campuses or even whole cities. These Projects aim to build up non-commercial roof top networks for providing free internet access. Broadband home networking: WMNs eliminate the difficulties of adding new APs which need wired network access. WMNs can serve as a distribution system for internet access and multimedia streams. Enterprise Networking: WMNs can be used where no wired Ethernet is available or where an installation is too expensive. 3 Voice over IP 3.1 Overview The International Telecom Union (ITU) defines VoIP as the transmission of voice, fax and related services over packet-switched IP-based networks. Services that were traditionally provided by public switched telephone networks systems (PSTN) are run over IP based networks such as LANs or the Internet. The terminals, i.e. the phones, are IP enabled devices such as computers or IP phones. There are a number of standards used for signalling, speech coding and the transport of the voice packets over IP. This modularization makes VoIP flexible and the standards interchangeable when new requirements and applications emerge. 14

20 3.2 Characteristics of VoIP Traffic Delay Delay or latency is the time needed to process a voice sample from input to output by the system. The ITU calls this time end-to-end one-way delay [20]. The end-to-end delay is the sum of delays added by the sender, by the network transmission and by the receiver. When transmitting a packet over the network latency is added by queuing it in routers, serializing it and by forwarding it to the next hop. The delay from sender to receiver might not be the same as on the way back, since packets in packet-switched networks can take different paths in both directions [21]. At the receiver side delay is generated by the jitter buffer, depacketization and decoding. Depackatization and decoding increase the delay by approximately the same amount as packetization and encoding do at the sender side Packet Loss When packets are transmitted over the network, they can get lost or corrupted. Typical sources of packet loss are network congestion and transmission errors [7]. Sometimes, packets are dropped deliberately [23] to avoid congestion in the network. In correctly installed wired Ethernets packet loss is primarily caused by network congestion. In networks, packet loss is also caused by channel noise and colliding transmissions. Also, in packet loss cannot be detected by the MAC layer in all cases, whereas the CSMA/CD mechanism in wired Ethernet detects packet loss reliably. The packet loss ratio is the average number of packets lost divided by the number of sent packets. A packet is also counted as lost, when it arrives at the destination but cannot be processed because of too high jitter or too high delay. The acceptable loss ratio is dependent on the codec and packet size Jitter Jitter is the variation of delay of consecutive packets. It is a result of different queue lengths on the network, varying paths and congestion [7, 20]. If, due to jitter, the difference of arrival times of two succeeding packets is greater than the processing time of the first packet, the decoder cannot process the second packet in time. To smoothen the jitter, all incoming packets are stored in a jitter buffer. There they are held for a certain time and afterwards released to the decoder. A fixed jitter buffer holds back incoming packets for a fixed time, whereas an adaptive jitter buffer adjusts the 15

21 delay dynamically to the network condition [7]. A larger jitter buffer can compensate a larger jitter, but also introduces more delay. 4 VoIP Performance Comparison in IEEE802.11s VoIP has generally low performance in WMNs. In multihop WLANs the performance degrades exponentially with the number of hops, resulting in only one supported call over a five hop string topology [22]. In order to study the problems of VoIP in IEEE802.11s WMN a performance comparison is required. So far there is no such comparison focusing on routing protocols. First of all we will discuss performance comparison of different routing protocols for VoIP in WMNs and then try to find solution to improve VoIP quality. We did a performance comparison of three routing protocols in IEEE802.11s WMN, with VoIP at application layer. These protocols are AODV, DSR and HWMP. This performance comparison has been done using Qualnet QualNet QualNet is a fast, scalable and hi-fidelity network modelling software. It enables very efficient and cost-effective development of new network technologies. QualNet is network modeling software that predicts performance of networking protocols and networks through simulation and emulation. Using emulation and simulation allows to reproduce the unfavourable conditions of networks in a controllable and repeatable lab setting. QualNet provides the following key benefits: Speed. QualNet can support real-time and faster than real-time simulation speed, which enables software-in-the-loop, network emulation, hardware-in-the-loop, and human-in-theloop exercises. Scalability. QualNet supports thousands of nodes. It can also take advantage of parallel computing architectures to support more network nodes and faster modeling. Speed and scalability are not mutually exclusive with QualNet. Model Fidelity. QualNet offers highly detailed models for all aspects of networking. This ensures accurate modeling results and enables detailed analysis of protocol and network performance. Portability. QualNet runs on a vast array of platforms, including Linux, Solaris, Windows XP, and Mac OS X operating systems, distributed and cluster parallel architectures, and both 32- and 64-bit computing. 16

22 Extensibility. QualNet connects to other hardware & software applications, such as OTB, real networks, and STK, greatly enhancing the value of the network model Components of QualNet Developer QualNet Scenario Designer is a model setup tool that allows users to set up geographical distribution, physical connections, and the functional parameters of the network nodes. Using intuitive click and drag operations, the user can also define network layer protocols and traffic characteristics down to each node. QualNet Animator offers in-depth visualization and analysis. As simulations are running, users can watch traffic flow through the network and view dynamic graphs of critical performance metrics. Users can also assign jobs to run in batch mode on a faster server and view the animated data later. QualNet 3D Visualizer is a QT-based tool for rich animations of network simulations. Users set up QualNet scenarios in QualNet Scenario Designer and then send the simulation to the 3D Visualizer for animation. QualNet Analyzer is a statistical graphing tool that displays hundreds of metrics. Users can choose to see pre-designed reports or customize graphs with their own statistics. Real-time statistics are also an option, where users can view metrics as they are generated while a simulation is running. Multi-experiment reports are also available. All graphs are exportable to spreadsheets. QualNet Packet Tracer is a packet-level visualization tool for viewing the contents of a packet as it goes up and down the network stack. This is a valuable debugging tool. 4.2 Performance Evaluation We have created a simulation scenario for performance evaluation. There are 50 nodes (MPs) which are uniformly placed at a distance from each other. These nodes are establishing VoIP connections with each other. There is no other background traffic. Each node has a 2Mbps link. Airtime link metrics is used as the default metrics. The total simulation time is 300 seconds. Time Period for each VoIP call is 20 seconds. H323 [23] is used as the signalling protocol for VoIP. Figure 4-1 shows MPs uniformly placed and connected to a single subnet. 17

23 Figure 4-1. Mesh Point uniformly placed in QualNet Results Average End to End Delay Average end to end delay is the average elapsed time to deliver a packet from the source to the destination. This time includes all delays that occur from source to destination. Figure. 4-2 shows the End to End delay comparison. DSR has greater delay than AODV and HWMP. Delay increases as the number of nodes increases. The reason is that DSR doesn t have explicit mechanism to expire stale routes in the cache or prefer fresher routes. Also DSR uses route caching aggressively and replies to all requests reaching a destination from a single request cycle. Hence DSR is not useful for VoIP in IEEE802.11s. AODV and HWMP has approximately same delay graph. Both are Hop-by-Hop protocols. Here we have used the Reactive version of HWMP protocol which is based on the concept of AODV. Both protocols are more conservative, the fresher route is always chosen. Both uses the concept of 18

24 sequence number for maintaining up to date routes to the destination. The route deletion using RERR is also conservative End to End Delay Comparison HWMP AODV DSR Delay in secs No of active users Figure 4-2. End to End delay comparison Increase delay in HWMP as compared to AODV is due to the reason that HWMP also maintains a proactive path. For that purpose MPP broadcast route announcement (RANN) messages to all nodes Packet Loss due to No Route Both AODV and DSR utilize the intermediate nodes for route discovery. DSR uses source routing and routes are stored in a route cache, data packets carry the source route in the packet header. Figure 4-3 shows the packet loss due to no route. Both AODV and HWMP face the problem of packet loss due to no route and it increases as number of nodes increases. 19

25 Packet Loss due to No Route AODV HWMP DSR 1.4 Packets Lost (%) No of active users Figure 4-3. Packet Loss Due to No Route Routing Overhead Routing overhead describes how many routing packets for route discovery and route maintenance need to be sent in order to deliver the data packets from source to destination. Figure.4-4 shows the routing overhead in IEEE802.11s WMN. Routing Overheads are calculated by summing up different types of control packets. These include Number of RREQ packets initiated Number of RREQ packets retrieved Number of RREQ forwarded Number of RREP packets initiated as destination Number of RREP packets initiated as intermediate node Number of RREP packets forwarded. DSR has less routing overhead. Here we have not included DSR s routing header extension in each data packet (Required for source routing). DSR sent fewer overhead packets than AODV and HWMP due to aggressive caching; therefore it maintains more than one route per destination. In case of AODV and HWMP overhead increases as the route discovery process spread to almost every node. In single query-reply cycle of DSR, source learns route to each intermediate node in the route in addition to the destination. Each intermediate node also learns route to other nodes on 20

26 the route. Promiscuous listening is also used to learn the route to every node on the route. AODV and HWMP don t have source routing or Promiscuous listening, thus relying on RREQ broadcast for route discovery causing more overhead. Initially HWMP has greater overhead than AODV because control packets like RANN are broadcast for construction of proactive path Overhead Comparison AODV DSR HWMP Overhead (%) No of active users Figure 4-4. Routing Overhead comparison Data Throughput Figure.4-5 shows the variation of data throughput with the number of active users. Data Throughput is defined as the average total number of data bytes received by the destination over the total simulation time. AODV has better throughput in IEEE802.11s WMNs. It is also observed that, for 50 active users all the three protocols didn t saturated. HWMP shows poor data throughput for greater number of nodes. 21

27 AODV HWMP DSR Data Throughput Data Throughput (Mbps) No of active users Figure 4-5. Data Throughput 5 Performance Comparison of HWMP Before discussing the problem in HWMP it is necessary to study the behaviour of Proactive and Reactive protocol in HWMP so that we can analyze the problem and suggest an appropriate solution. Average End to End Delay in secs HWMP Proactive and Reactive Comparison Reactive Proactive No of active users Figure 5-1. Average End to End delay Comparison 22

28 Figure 5-1 shows the End to End delay comparison of HWMP. As the number of nodes increases delay increases and finally for 50 nodes Proactive path has greater delay. The reason is that Root has complete path information to all nodes and therefore when a request for route is sent to root it is sent to that node but as number of request to root increases delay also increases. On the other hand reactive path uses the mechanism of broadcasting packets for discovering route to destination, which will constantly increase with increased number of nodes. Performance of both protocols degrades with increased number of users. Packet loss is certainly greater in proactive routing because all the requests are generated for root node. Figure 5-2 shows the packet loss for proactive and reactive version of HWMP. One path (from source to root) gets more congested. Reactive routing follow different route for different node, using more updated link HWMP Packet Loss Reactive Proactive Packet Loss due to No Route (%) No of active users Figure 5-2. HWMP Packet Loss Figure 5-3 shows the data throughput for proactive and reactive HWMP. Reactive HWMP shows greater data throughput as compared to proactive version. For Proactive HWMP throughput tends to decrease for larger group of users. In proactive version of HWMP, Root node is suffered from large number of route request. As a result packet delivery to destination is suffered. 23

29 0.025 Data Throughput 0.02 Data Throughput (Mbps) Proactive Reactive No of active users Figure 5-3. Data Throughput for HWMP 5.1 Problem Definition In the last chapter we have seen HWMP has better performance considering VOIP application. However it has been observed that within Mesh network HWMP has more overhead, which increase with number of nodes. End to End delay is nearly same as AODV. HWMP doesn t select the best path in every case. Figure 5-4 suggests that the green path is selected by HWMP due to Hop by Hop nature. RREQ packets are broadcasted in network to find the best route to destination and finally the next best having low metrics value is selected. The cumulative sum of metrics value for green path is 12%, while on the other hand the cumulative sum of metrics value for red path is 10%. Thus HWMP doesn t always guarantee a low metrics value path. Here a more centralized approach is required as MPs in IEEE802.11s have very low or no mobility. Centralized approach means that path should be selected having complete knowledge of the topology. The reason for this approach is that IEEE s WMN has approximately no mobility, so distribution of topology information will be easy. Important factor to be considered in this regard is bandwidth. HWMP floods the network with RREQ broadcast and as the number of nodes increased more overhead will be experienced. A centralized approach not only assures low delay due to best path selection but also reduce overhead for greater number of nodes. One solution to the problem is that all nodes in the mesh network provide topology information to root node. This topology information includes neighbour information and metrics value for these neighbours. Root node will then create a global topology table. Whenever a source node wants to 24

30 communicate with destination it sends request to root node. Root node will apply some shortest path algorithm to the global topology table and then provide best path between source and destination. This will always provide a best path. The above solution sounds good but it will not work for large WMNs. First, for large number of nodes and all sending route request to root node causes congestion on the specified link to root. Secondly all nodes are dependent on root for communication with destination. 5.2 Proposed Solution Figure 5-4. HWMP Route Comparison Our proposed solution is an extension of existing HWMP. Our proposed protocol uses Root as the central node. In this protocol Root broadcast RANN messages (As in Proactive HWMP). All the nodes will send RREP message to Root node. This RREP message will contain the neighbour information of nodes along with their metrics value. On receiving all these messages Root node will build a table which depict the whole network topology. This table is broadcasted to all the nodes in the network. Whenever a node wants to communicate with destination node, it just has to use any shortest path algorithm like Dijkstra s in our case to compute the best path to destination. For route updation RERR messages are broadcasted in network so that every node can update its topology table. Following are the steps of our proposed solution. 25

31 I. Root node will broadcast RANN message (as in HWMP). When every node receives the RANN message, it will piggyback the neighbour information consisting of neighbour addresses and corresponding metrics value in RREP message. II. On receiving RREP from all nodes route will create a topology information table consisting of nodes, their neighbours and metrics value. III. This table is broadcasted to all the nodes in the network. IV. All nodes apply shortest path algorithm to calculate the best path to all destination. V. Whenever source nodes request a path to destination it will be forwarded to nest node according to table information. VI. A change in neighbour (Exit or Arrival) or change in metrics value (Link condition) is communicated to all nodes by broadcasting this information to all nodes. VII. Sequence number are used for avoid looping and updation of table. 5.3 Optimization & Maintenance Every source node can calculate a shortest path to destination without flooding the network. There is overhead initially in distributing topology information but after that all computation is done locally. We also restrict the nodes so that after providing neighbour information to Root node they will not send neighbour information unless and until some change occur (new neighbour arrived or existing neighbour quits). This will reduce the control overhead. Also when there is a change in metrics value it is also send in RREP message. This change will be broadcasted to every node. As a result the frequent change in topology is communicated to all nodes. This will ensure that every time source node discovers a path to destination, it is an updated path. Our proposed protocol is more centralized, having a global view of network topology. This algorithm doesn t rest solely on Root node as in proactive HWMP where traffic to all destinations goes through Root, neither it use RREQ messages which may cause congestion in network. 6 Preliminary Results We investigate the performance of our proposed solution with AODV. Performance comparison is done by increasing the number of users. We have implemented our protocol and observe its performance on same topology (same parameters) used for performance comparison earlier. MatLab is used for extracting the statistics and creation of graph. For that purpose all user are treated as MPs and one Root MP is selected manually. All MPs are uniformly placed at a distance from each other. Simulation time was 300 seconds. Topology information is 26

32 communicated to all nodes in 4.06 seconds for 50 nodes. Traffic is generated after 5 second. Statistics are collected every time when number of communicating nodes is increased. 6.1 Average End to End Delay The average time to send data from source to destination is calculated for both protocols. Once all the nodes have topology information, average End to End delay is low as compared to AODV. For both protocol the average delay is less than 0.5 seconds. In our proposed solution the increase in delay as compared to number of node is more moderate. AODV send more broadcast (RREQ) for discovering route with increasing number of users. Figure 6-1 shows the average end to end delay Proposed Solution AODV Delay in secs No of active users Figure 6-1. End to End delay comparison 6.2 Routing Overhead Routing Overheads are calculated for AODV and proposed solution. Figure 6-2 shows the performance results of routing overhead. Since the proposed solution has constant control overhead so it has less routing overhead as the number of user become larger. On the other routing overheads for AODV is much larger then proposed solution. The routing overhead for proposed solution are nearly 10% for 50 active users. Under the same condition the routing overhead for AODV are 27%. The reason for less overhead of proposed solution is that, once complete network topology has been communicated to all nodes there are very few control packets transferred. These control packets are for updation and error reporting. On the other hand AODV broadcast whole network when a new 27

33 request for route has been received. Also AODV delete the previous route information after a fixed time interval Proposed Solution AODV Routing Overhead (%) No of active users Figure 6-2. Routing Overhead Comparison 7 Conclusion WMNs are foreseen to lead to a disruptive change in wireless communications. However, to be successful, a set of criteria has to be fulfilled: a critical mass of subscribers/users needs to be present, applications need to be adequately supported, and dependable operation has to be guaranteed. However, one crucial factors possibly the tipping point in making WMNs a success story is the support for QoS. In contrast to the wired Internet bandwidth is scarce and overprovisioning cannot be applied to WMNs; thus, without adequate QoS mechanisms a lot of promising applications is likely to fail. Our algorithm improves the QoS by improving the performance of those factor which effects VoIP and other QoS sensitive traffic. It also shows better performance for large group of user. Delay has been decreased for small and large number of users. Routing Overheads are greater for smaller group of user but it gradually decrease as the number of users tends to increase. We are in process of investigating other results for the proposed solution. The work on the master internship raised a lot of interesting questions and issues. For improving VoIP quality in WMNs more concentration is required to decrease overhead and packet loss. HWMP also required a lot of improvement. Our comparison suggests that this protocol doesn t have satisfactory performance related to above mention factors. 28