Integrating fast mobility in the OLSR routing protocol

Transcription

1 Integrating fast obility in the OLSR routing protocol Mounir BENZAID 1,2, Pascale MINET 1 and Khaldoun AL AGHA 1,2 1 INRIA, Doaine de Voluceau - B.P.105, Le Chesnay Cedex, FRANCE ounir.benzaid, pascale.inet@inria.fr 2 LRI, Bât 490 Université Paris Sud, Orsay Cedex, FRANCE alagha@lri.fr Abstract - With the current increase in ad-hoc obile networks in public doains (e.g. airports, cities, etc.), and the widespread use of IEEE wireless LAN, there is a growing need to handle and anage fast obility. Extending the coverage area of ad-hoc networks and taking into account fast obility in routing protocols could offer a copleentary solution to the UMTS for fourth generation (4G) obile networks. In this paper we present an extension of the Optiized Link State Routing protocol (OLSR), denoted Fast-OLSR, which is designed to eet the need for fast obility in Mobile Ad-hoc NETworks (MANETs). Perforance evaluation of Fast-OLSR is done by siulation, and the results show that the loss rate can be iniized while aintaining a reasonable overhead traffic. Keywords: obile ad-hoc networks, wireless networks, fast obility, routing protocol, proactive protocol, OLSR. I. INTRODUCTION With recent technological advances in laptop coputers and wireless data counication devices, wireless counications represent one of the fastest growing segents of the counications industry today. Wireless networks bring a new diension to obility; indeed they enable a user to access the Internet anywhere and at any tie. At present, obile wireless networks can be classified according to two ain types: infrastructure and infrastructureless obile networks; this last type of network is called an ad-hoc network [1], [2], [3]. A Mobile Ad-Hoc NETwork (MANET) [4] is a collection of obile nodes that counicate using a wireless ediu, foring an autonoous network. There is no centralized access point or pre-existing infrastructure. Such networks have dynaic, rando, soeties rapidly changing topologies, liited bandwidth, variable throughput links, and liited power (e.g. battery operated devices). When a node needs to counicate with another node, it uses either a direct wireless link or a ulti-hop route to reach the destination. This eans that all the nodes ust incorporate routing capability to ensure that packets are delivered to the designated destination. Moreover ad-hoc routing protocols ust iniize the induced control traffic. With the increasing appearance of ad-hoc obile networks in public doains (e.g. highways, cities, etc.), there is an increasing need to handle and anage fast obility. Extending the coverage area of ad-hoc networks and taking into account fast obility in routing protocols could offer a copleentary solution to the UMTS in fourth generation (4G) obile networks. The design of a fast and efficient routing protocol is necessary for the perforance of an ad-hoc network in particular in the case of large and dense networks with fast oving nodes. This paper focuses on how to deal with fast oving nodes in the OLSR routing protocol. OLSR [1] is one of the protocols discussed in the MANET working group. The protocol as described in this paper inherits the stability of a Link-State routing protocol and the availability of routes when needed due to its proactive nature. However, when a node is oving fast, the links with its neighbors are valid only during a short tie interval. If packets are forwarded on an invalid link, not yet detected as broken, they are lost. Hence to iniize packet loss, broken links between the node and its neighbors ust be quickly detected. In this paper, we propose the Fast-OLSR extension to account for fast nodes in routing while keeping the routing overhead as low as possible. This extension is based on the initial study [5]. A tradeoff is found between the routing overhead and the loss rate. The reainder of this paper is organized as follows: in Section 2, we describe the two failies of routing protocols (i.e. reactive and proactive protocols) discussed in the MANET working group. In Section 3, we briefly present OLSR,a proactive protocol. In Section 4, we show how this protocol can be extended to take into account fast obility. In Section 5, we evaluate the perforance of the Fast-OLSR extension in ters of essage loss and induced overhead. II. MANET ROUTING PROTOCOLS Different routing protocols are proposed in the MANET working group of the IETF [4]. They address the proble of unicast routing, while taking into account the features of wireless, ulti-hop, obile ad-hoc networks. All these protocols generally deal with low obility environent conditions. Such protocols can be divided into two categories: proactive and reactive, depending on the route discovery echanis that is used. With reactive protocols, a node discovers routes on-deand and aintains only active routes. Thus, a route is discovered

2 whenever a source node needs to counicate with a destination node for which a route is not available. This discovery is based on pure flooding in the network. The source node broadcasts a route request essage to all its neighbors. The neighbors in turn rebroadcast the route request to their neighbors if they do not have a route to the destination. When the route request reaches either the destination or a node that has a valid route to the destination, a route reply essage is generated and transitted back to the source. Therefore as soon as the source receives the route reply, a route is created fro the source to the destination. The advantage of reactive protocols is that no control essage is needed for non-active routes. The drawback is the latency when establishing a route. Exaples of reactive protocols include AODV [6] and DSR [7]. With proactive protocols, each node continuously aintains the routes to all other nodes in the network by the periodic exchange of control essages. When a node needs to send a packet to any other node in the network, the route is iediately available. The ain advantage of proactive protocols is that they do not introduce a delay before sending data. Furtherore, these types of protocols are useful for traffic patterns where a large subset of nodes is counicating with another large subset of nodes, and where the [source, destination] pairs are changing over tie. Exaples of proactive protocols include DSDV [8] (an adaptation of Routing Inforation Protocol [9]), OLSR [1] (an optiization of the Link-State algorith OSPF [10]) and TBRPF [11]. III. OPTIMIZED LINK STATE ROUTING PROTOCOL (OLSR) OLSR [1] is a proactive routing protocol, providing the advantage of having routes iediately available in each node for all destinations in the network. It is an optiization of a pure Link State routing protocol. This optiization is based on the concept of ultipoint relays (MPRs) [12]. First, using ultipoint relays reduces the size of the control essages: rather than declaring all links, a node declares only the set of links with its neighbors that are its ultipoint relays. The use of ultipoint relays also iniizes flooding of control traffic. Indeed only MPRs forward control essages. This technique significantly reduces the nuber of retransissions of broadcast control essages [13]. OLSR is characterized by two types of control essages: neighborhood and topology essages, called respectively Hello essages and Topology Control (TC) essages. OLSR provides two ain functionalities: the first is Neighbor Discovery, and the second is Topology Disseination. These will be detailed in the following. A. Neighbor Discovery Each node ust detect the neighbor nodes with which it has a direct link. Due to the uncertainties in radio propagation, a link between neighboring nodes ay enable the transission of data in either one or both directions over the link. For this, each node periodically broadcasts Hello essages, containing the list of neighbors known to the node and their link status. The link status can be either syetric (if counication is possible in both directions), asyetric (if counication is only possible in one direction), ultipoint relay (if the link is syetric and the sender node of the Hello essage has selected this node as a ultipoint relay), or lost (if the link has been lost). The Hello essages are received by all one-hop neighbors, but are not forwarded. They are broadcast at a low frequency deterined by the refreshing period Hello-interval (the default value is 2 seconds). Thus, Hello essages enable each node to discover its onehop neighbors, as well as it two-hop neighbors (the neighbors of its neighbors). This neighborhood and two-hop neighborhood inforation has an associated holding tie Neighborhold-tie, after which it is no longer valid. On the basis of this inforation, each node of the network independently selects its own set of MPRs aong its one-hop neighbors. The ultipoint relay set of, denoted MPR() is coputed as follows: it is a iniu subset of one-hop neighbors with a syetric link, suchthatalltwo-hopneighborsof have syetric links with MPR(). This eans that the MPRs cover (in ters of radio range) all the two-hop neighbors. Figure 1 shows the MPRs selection by node. A possible algorith to select these MPRs is described in [12]. The MPR set is coputed whenever a change in the one-hop neighborhood or twohop neighborhood is detected. Fig. 1. Multipoint relays of node. ultipoint relays of node Each node aintains the set of its ultipoint relay selectors (MPR selectors). This set contains the nodes which have selected as a ultipoint relay. Node only forwards broadcast essages received fro one of its MPR selectors. B. Topology Disseination Each node of the network aintains topological inforation about the network obtained by eans of TC essages. Each node selected as a MPR, broadcasts a TC essage at least every TC-interval (the default value is 6 seconds). If a change occurs in the MPR selector set, the next TC can be sent earlier (e.g. after soe pre-specified iniu interval). The TC

3 essages are flooded to all nodes in the network and take advantage of MPRs to reduce the nuber of retransissions. The TC essage originated fro node declares the MPR selectors of. Thus, a node is reachable either directly or via its MPRs. This topological inforation collected in each node has an associated holding tie Top-hold-tie, after which it is no longer valid. The neighbor inforation and the topology inforation are refreshed periodically, and they enable each node to copute the routes to all known destinations. These routes are coputed with Dijkstra s shortest path algorith. Hence, they are optial as concerns the nuber of hops. Moreover, for any route, any interediate node on this route is a MPR of the next node. The routing table is coputed whenever there is a change in neighborhood inforation or a change in topology inforation. IV. FAST-OLSR In an OLSR ad-hoc network, when a node is oving fast, its neighborhood changes quickly and the default Hello frequency in OLSR is not sufficient to track the nodes otion. Therefore, routes to this node becoe inactive and essages sent to this node ay be lost. A higher Hello frequency would overcoe this proble, but at the cost of an additional control overhead. We propose an extension of OLSR to address the issues for fast obility nodes. This Fast-OLSR extension has two ain objectives: first, The induced control traffic is tuned to node obility, in such a way that it allows a fast node s otion to be tracked, i.e. there is very low overhead when there is no obility and an appropriate overhead as obility increases. Second, the bandwidth consued ust reain reasonable (i.e. aintained below a certain threshold). Fast-OLSR has been designed as an extension of the OLSR protocol such that: (i) not all nodes are required to ipleent Fast-OLSR; OLSR and Fast-OLSR can coexist in the sae obile ad-hoc network. (ii) Fast-OLSR is based on the sae basic principles as OLSR. It only differs in the Neighbor Discovery functionality, which is adapted to deal with fast obility. Moreover, to stay general and copletely independent of the underlying link layer (e.g. IEEE , Bluetooth, etc.), Fast- OLSR does not ake any assuptions concerning the inforation provided by the chosen link layer. This eans for instance that Fast-OLSR ust work even when neither inforation on signal attenuation nor link layer notification of broken links are available. In such conditions, the only way that Fast-OLSR can detect node obility is to observe the neighborhood changes. The basic idea of neighbor discovery in Fast-OLSR is to enable a fast oving node,, to quickly discover a sall nuber of neighbors. Aong these, a sall nuber of ultipoint relays are selected to aintain connectivity with other nodes in the network. To achieve this, a fast oving node establishes a sall nuber of syetric links refreshed at a high frequency by eans of Fast-Hellos. Such links are called Fast links and this high frequency is deterined by the refreshing period Fast-hello-interval. There are three ain echaniss in Fast-OLSR: Switching to the Fast-Moving/Default ode: when a node detects that it is oving fast (e.g. a high nuber of changes in its neighborhood), it switches to the Fast-Moving ode and starts sending Fast-Hellos. On the other hand, when a node in Fast-Moving ode detects that it is no longer oving fast (e.g. a sall nuber of changes in its neighborhood), it switches back to the Default ode, which is the initial one. Establishing Fast links: a node in Fast-Moving ode sends Fast-Hello essages at high frequency. A Fast-Hello is siilar to a Hello, however its size is saller because it contains a reduced nuber of neighbor addresses. Fast-Hello essages are used to establish Fast links. When a node in Default ode receives a Fast-Hello fro a node in Fast-Moving Mode, it replies with a Fast-Hello. Aong the received replies, node selects a sall nuber of MPRs. These are declared in the next Fast-Hello. The declared nodes will then broadcast TC essages to all the nodes in the network declaring that they are MPRs of. Only nodes in Default ode can be selected as MPRs. This siplified selection of MPRs is used because the node, which is oving fast, only partially knows its two-hop neighborhood. Refreshing Fast links and Detecting new/broken links: a node in Fast-Moving ode sends Fast-Hellos containing the addresses of its MPRs and its MPRs reply with epty Fast-Hellos. Epty Fast-Hello savebandwidthusage, and they are sufficient to enable node to know that it can always be reached by these MPRs. If a MPR of has not received the Fast-Hello of, it sends a TC essage to infor all nodes in the network that it is no longer a MPR for. By eans of Fast-Hellos, Fast-OLSR enables a broken link to be detected quickly. Hence the coputation of a new route is ade earlier, and essage loss is reduced. Moreover, as the nuber of Fast links refreshed by a node in Fast-Moving ode is reduced, the additional overhead reains reasonable. As soon as a node in Default ode no longer has any Fast links, it stops sending Fast-Hellos and returns to noral OLSR behavior, by sending only Hello and TC essages. V. SIMULATION In this section, we study the behavior of Fast-OLSR in worstcase conditions. We conduct a perforance evaluation by siulation. The velocity of obile nodes ranges fro a cyclist s speed up to a car s speed on a highway. Two paraeters are easured: the packet loss and the overhead produced by Fast- OLSR. A. Siulation odel The considered siulation odel is depicted in Figure 2. A fixed node C is located in a central position. It counicates by syetric links with 6 nodes denoted 1 to 6. These nodes are fixed. One node, oving at high speed, is also

4 considered. It oves continually around 1 to 6 in a circular way. The velocity of is constant, and its value depends on the siulation. This odel can, for instance, represent the effective architecture where C is a gateway and i, i [1, 6] are base stations that connect a fast oving wireless node to another wired/wireless network via the gateway. We consider that there is no overlapping between the radio coverage areas of two adjacent nodes i and i+1. This eans that it is ipossible for to be in a soft handoff situation where it can receive or send packets fro both i and i+1. Thus, node loses connectivity with i before having connectivity with i+1. We ake this assuption in order to study the behavior of Fast-OLSR in worst-case conditions. Indeed, overlapping areas would enable a obile node to aintain connectivity while a new route is being established. Moreover, we assue that neither link layer notification, nor inforation on signal attenuation are available. In each siulation, there is an initialization tie in which there is an exchange of Hello and TC essages to establish the links between node C and nodes 1 to 6. Then, switches to Fast-Moving ode. Fast-OLSR perforances are evaluated while is oving 9 ties around 1 to 6. The signal attenuation distance is fixed to 60 eters (ean value of the attenuation distance in IEEE802.11). This odel can be applied with IEEE In such a case, collisions could occur, they result fro siultaneous transissions of C and i.asinall considered scenarios, they are reduced to a iniu, and are not considered in our odel C 1 Fig. 2. Siulation odel. B. Siulation Measureents and Paraeters Let us assue that a Constant Bit Rate (CBR) strea of packets is being transitted fro node C to obile. With our odel, C is always two hops fro but the next hop to reach changes when oves. We are interested in evaluating the nuber of lost packets due to hard handoffs, assuing the worst conditions (no buffering and no retransission). - Packet Loss: This nuber is coputed as follows. We first copute the nuber of lost packets during a handoff between nodes i and i+1. Packets are lost during the tie interval [t 1, t 2 ) where t 1 is the tie when loses connectivity with i and t 2 is the tie when can be reached through i+1. The total nuber of lost packets is obtained by adding the nuber of packets lost in each handoff generated during the siulation. 2 - Induced overhead: We copute the overhead induced by Fast-OLSR due to fast obility of node. This overhead is equal to the total nuber of (i) Fast-Hello essages generated in the network and (ii) the TC essages sent each tie a Fast link is established or broken. - Paraeters: We have easured the packet loss and the overhead induced in various siulations, each siulation being characterized by two paraeters: first, the refreshing period of Fast links (i.e., Fast-hello-interval), secondly the velocity of in kiloeters per hour. C. Siulation results Figures 3 and 4 highlight the perforance of Fast-OLSR in ters of loss rate and overhead, when node C sends CBR traffic to obile. Mobility ranges fro 20 k/h to 150 k/h. Figure 3 depicts the packet loss rate versus obility. Several curves are drawn with regard to the value of the Fast-hellointerval (fro 100 s to 300 s). As Figure 3 shows, the greater the Fast-hello-interval is, the greater the loss rate is, and, of course, the packet loss becoes greater as the speed increases. Our siulation results show that the packet loss rate can be aintained saller than 10.6% for a velocity up to 90 k/h. Recall that these results are obtained in a worst-case where is no overlapping radio coverage area of two adjacent nodes i, i+1 (see Figure 2) and no buffering or retransission are considered. The routing overhead produced for the sallest Fast-hello-interval (i.e. 100 s) is kept below 7.7 kbit/s. To establish a coparison, this interval is equivalent to the 120 s Slow Associated Control CHannel (SACCH) in GSM radio interface. The SACCH channel is used to report received signal strength and thus triggers the handoff between cells. For a given axiu acceptable loss rate and the axiu reachable speed, we can deterine the largest Fast-hellointerval that produces the least routing overhead. First we draw the vertical line indicating the axiu speed, and the horizontal line representing the acceptable loss rate. The acceptable region is then deliited by the x-axis, the y-axis, the speed line and the loss rate line. The ost appropriate Fast-hello-interval is the one given by the intersection of the highest curve in the acceptable region with the speed line. For exaple, for a axiu speed of 125 k/h and a loss rate of 15%, the only possible Fast-hello-interval is 100 s. However, for a speed liited to 40 k/h with a loss rate of 10%, there are several possibilities for the Fast-hello-interval: 100 s, 140 s, 200 s. We select the highest curve, which gives us a Fast-hello-interval of 200 s. Indeed this value incurs the sallest overhead in the acceptable region. Figure 4 depicts the routing overhead induced by Fast-OLSR versus obility. Several curves are drawn with regard to the value of the Fast-hello-interval (fro 100 s to 300 s). As Figure 4 shows, the saller the Fast-hello-interval is, the greater the routing overhead is. The routing overhead becoes slightly greater as the speed increases.

5 Loss rate (%) Fast_hello_interval 100s 140s 200s 240s 300s Mobility K/h Fig. 3. Loss rate versus obility. Overhead traffic byte/sec Fast_hello_interval 100s 140s 200s 240s 300s Mobility K/h Fig. 4. Overhead traffic versus obility. Table I. Mobility classes and associated results. Class Speed Loss rate 10% 10%< Loss rate 15% 15%< Loss rate 20% Fast-hello Overhead Fast-hello Overhead Fast-hello Overhead k per hour s kbps s kbps s kbps Cyclist 20 S< > Urban 45 S< Road 90 S< Highway 120 S< Figure 4 also shows the additional overhead resulting fro a decrease in the Fast-hello-interval. The choice of a saller Fast-hello-interval is justified only when the resulting gain in loss rate is significant. For exaple, for a axiu speed of 50 k/h and a loss rate of 10%, a Fast-hello-interval of 140 s is chosen. A Fast-hello-interval of 100 s would iprove the loss rate (it would be 7%) but at the cost of an additional overhead of 1 kbyte/s. On the other hand, for a axiu speed of 90 k/h and a loss rate of 25%, a Fast-hello-interval of 240 s is chosen. This iproves the loss rate (it would be 30% with a Fast-hello-interval of 300 s) at the cost of an additional overhead of only 30 Byte/s. Results of our siulations are suarized in Table I. Four classes of obile node are defined. Each class is defined by its axiu speed. For each class, we deterine the best Fast-hello-interval and associated axiu routing overhead in ters of kbit per second, copliant with the acceptable loss rate. VI. CONCLUSION With the deployent of obile ad-hoc networks in public doains (e.g. highways, cities,etc.), routing protocols ust support fast obility. In this paper, we have proposed a Fast- OLSR, an extension of OLSR dealing with fast obility. Fast- OLSR aintains connectivity with fast oving nodes, while aintaining a reasonable overhead. The perforances of Fast- OLSR are evaluated by siulation in worst-case conditions (no buffering, no retransission, no overlap in coverage areas and no link layer notification). Siulation results show how to tune the value of the refreshing period (Fast-hello-interval) for different classes of obile and different acceptable loss rates. Thus, in the considered configuration, ad-hoc network nodes can ove as fast as cellular network nodes with a velocity that can reach 150 k/h and the loss rate is aintained below 15% with an acceptable overhead (the refreshing period being equal to 100 s). In further work, we will study how to dynaically adjust the value of the refreshing period as a function of the obile speed and the acceptable loss rate. REFERENCES [1] P. Jacquet, P. Muhlethaler, A. Qayyu, A. Laouiti, L. Viennot, T. Clausen, Optiized Link State Routing Protocol, draft-ietf-anet-olsr- 06.txt, IETF, Septeber [2] J. Broch, D. Maltz, D. Johnson, Y. Hu, J. Jetcheva, A perforance coparison of ulti-hop wireless ad hoc network routing protocols, in ACM Mobico 98, (Dallas, USA), October [3] C. Perkins, Ad Hoc Networking. Addison Wesley, [4] Mobile Ad-hoc Networks (MANET), anet-charter [5] P. Jacquet, P. Muhlethaler, P. Minet, A. Qayyu, A. Laouiti, L. Viennot, T. Clausen, OLSR Extensions, draft-ietf-anet-olsr-extensions-00.txt, IETF, August [6] C. Perkins, E. Royer, S. Das, Ad hoc On-Deand Distance Vector Routing, draft-ietf-anet-aodv-08.txt, IETF, March [7] J. Broch, D. Johnson, D. Maltz, The Dynaic Source Routing Protocol for Mobile Ad Hoc Networks, draft-ietf-anet-dsr-01.txt, Dec [8] C. Perkins, P. Bhagwat, Highly Dynaic Destination-Sequenced Distance-Vector Routing for Mobile Coputers, in ACM SIG- COMM 94, (London, UK), August [9] J. Malkin, RIP Version 2, RFC 1388, IETF, Jan [10] J. Moy, OSPF version 2, RFC 2328, IETF, Jan [11] R. G. Ogier, F. L. Teplin, B. Bellur, M. G. Lewis, Topology Broadcast Based on Reverse-Path Forwarding, draft-ietf-anet-tbrpf-05.txt, IETF, March [12] A. Qayyu, L. Viennot, A. Laouiti, Multipoint Relaying: An Efficient Technique for flooding in Mobile Wireless Networks, Tech. Rep. RR- 3898, INRIA, Feb [13] P. Jacquet, P. Muhletaler, A. Qayyu, A. Laouiti, T. Clausen, L. Viennot, Optiized Link State Routing Protocol, in IEEE INMIC, (Pakistan), Dec 2001.