Optimizing Route Length in Reactive Protocols for Ad Hoc Networks

Transcription

1 Optimizig Route Legth i Reactive Protocols for Ad Hoc Networks Thomas Heide Clause *,PhilippeJacquetadLauretVieot INRIA Rocquecourt, Projet Hipercom, Domaie de Voluceau, B.P. 5, 7853 Le Chesay cedex, Frace Telephoe: Fax: Abstract May protocols for Mobile Ad-hoc Networks propose costructio of routes reactively usig floodig. The advatage hereof is that o prior assumptio of the etwork topology is required i order to provide routig betwee ay pair of odes i the etwork. I mobile etworks, where the topology may be subject to frequet chages, this is a particularly attractive property. I this paper, we ivestigate the effect of usig floodig for acquirig routes. We show that floodig may lead to o-optimal routes i terms of umber of hops. This implies that more retrasmissios are eeded to sed a packet alog a route. We proceed by providig a qualitative aalysis of the route legths. Fially, we propose alterative floodig schemes ad evaluate these schemes through simulatios. We fid that usig these schemes, it is ideed possible to provide shorter routes. I. INTRODUCTION Growig iterest has bee give to the area of Mobile ad-hoc etworkig sice the apparitio of powerful radio devices allowig coectio of mobile odes. A keypoit i coectig a group of mobile odes is the desig of a routig protocol that allows distat odes to commuicate through relayig of their traffic by itermediate odes. Developmet ad stadardizatio is the subject of the IETF workig group MANet [3], [7], i which several protocols have bee proposed. A subset of those protocols are those costructig routes from a source ode to adestiatioodeodemad,i.e.whethesourceode has data traffic to trasmit to the destiatio ode. These are usually idetified as reactive protocols [4], [], [6], [], []. These protocols basically costruct routes usig the followig mechaism: ) The source ode emits a request packet, which is retrasmitted oce by all odes which receive it, icremetig a hop couter i the packet; * Thomas Heide Clause may also be cotacted at MidPass Ceter for Distributed Systems, Departmet of Computer Sciece, Aalborg Uiversity, Fredrik Bajers Vej 7E 922 Aalborg Ø, Demark 2) The destiatio ode ackowledges the request packets by sedig back reply packets via the reverse path to the source ode; 3) Amog the paths ackowledged by the destiatio, the shortest oe is used for the data trasmissio. The protocols differ i how they store the path followed by the request (e.g. the path is stored i the itermediate odes i AODV, ad i the packet i DSR), however without alterig the basic route discovery priciple. There are also other techiques i order to reduce the floodig cost. For example, may protocols attempt to reduce the cost of floodig by allowig itermediate odes to omit request retrasmissio if they already kow a route to the destiatio. However, whe a destiatio is requested for the first time, full floodig is still required. The work preseted i this paper is focused o idepth aalysis of two problems related to route discovery, amely: ) Floodig geerates a large amout of cotrol traffic ad thus itroduces a large overhead; 2) Routes, discovered through floodig may be suboptimal. Agoodpartoftheliteratureaboutmaetroutigprotocols has bee devoted to the aalysis of their cotrol overhead (i.e. overhead due to cotrol traffic), while little has bee devoted to the overhead due to route legth suboptimality. However, route legth sub-optimality may also be a o-egligible source of protocol overhead sice it is proportioal to the actual data traffic. Ideed, i situatios where the etwork is close to overloaded, route sub-optimality may be the mai source of overhead ad thus of etwork limitatio. The first result i this paper cocers the aalysis of route legth as obtaied by floodig. Simulatios ad aalytical meas are employed to show that routes created by floodig are ofte suboptimal. This result correspods with the results preseted i [2], however we preset a more qualitative isight o how much ad whe

2 2 sub-optimality is observed. The secod result i this paper cocers two complemetary solutios, proposed as ways of obtaiig shorter routes through floodig. The first solutio, called Super Floodig is a expasio of a full floodig scheme. The purpose is to get optimal routes at the expese of more emissios. The secod solutio, called MPR floodig, relies o the usage of multipoit relays [5] to both reduce the legth of the obtaied routes as well as the umber of emissios required to complete the floodig. Combiig this two solutios may be a promisig way for optimizig route legth i reactive protocols. A. Paper Outlie The remaider of this paper will be orgaized as follows: sectio II presets the proposed modificatios of floodig rules which potetially yield better performace with regard to route legth. These modificatios ca easily be adapted to be employed i most reactive protocols. The, i sectio III, we provide a explaatio of why ad uder which coditios suboptimal routes are obtaied by floodig. We utilize a ideal physical layer ad preset both aalytical results (i sectio III) ad simulatios (i sectio IV). Fially, i sectio V, we the validate these results uder more realistic settigs usig the etwork simulator s2 [5]. II. SCHEMES FOR OPTIMIZING ROUTE LENGTH DISCOVERED BY FLOODING I this sectio, we propose two modified schemes for floodig. Both schemes rely o modifyig the rules a ode obeys whe decidig if a give flooded packet is to be retrasmitted or ot. Reactive protocols use sequece umbers to prevet a ode from relayig a flooded message more tha oce. I.e. by default a ode obeys the followig rule, i the followig deoted basic floodig: amessageisforwardedifitisthefirsttimeitisecoutered by that ode. The two schemes we propose are complemetary i the sese that the first scheme makes this rule looser while the secod scheme makes it stricter. Both schemes ca easily be combied. Before describig the schemes i detail, it is importat to otice that reactive protocols ofte employ more complex mechaisms tha simple floodig i order to save cotrol traffic. For example, utilizig expadig rig floodig will prevet distat odes from retrasmittig a flooded message through utilizig a time-to-live (TTL), which is associated with the flooded message. Also, a itermediate ode may omit a retrasmissio if it is able to provide a valid route to the required destiatio (e.g. from a local cache). These are classical restrictive retrasmissio rules (used e.g. i AODV ad DSR). It is importat to otice that our schemes are compatible with such additioal rules. A. Super Floodig Scheme Our first floodig scheme is called Super Floodig. We loose the retrasmissio rule by allowig a ode to forward a flooded cotrol message more tha oce. Specifically, whe a ode receives a flooded cotrol message, it obeys the followig rule: amessageisretrasmittedif: the message has ot bee received by the ode before, OR the hop-cout of the message is smaller tha the hop-cout of the previously retrasmitted istace of the message. Notice that this requires that a ode is able to get the hop-cout for a received message. I the case of e.g. AODV ad DSR this iformatio is already available. This scheme has the advatage of providig optimal routes whe o collisios occur, but at the cost of more cotrol traffic. B. MPR Floodig Our secod floodig scheme is called multipoit-relay floodig or MPR floodig for short. It is ispired by the broadcastig scheme of the proactive protocol OLSR [8]. This scheme requires a eighbor discovery mechaism which allows a ode to acquire iformatio about the odes i its eighborhood as well as the eighborhood of these eighbor-odes (i.e. a ode s 2-hop eighborhood). The multipoit-relay priciple is the followig: a ode selects amog its eighbors a set of odes, called multipoit-relays, such that ay ode i the 2-hop eighborhood is reachable through at least oe multipoit-relay. A ode should try to get the smallest umber of multipoit-relays possible. The rule for retrasmissio of a flooded cotrol message is restricted as follows: amessageisretrasmittedif: the message has ot bee received by the ode before, AND the ode is selected as multipoit relay by the ode from which it received the message (the previous hop of the message). Notice that if a ode has received a message oce (from ay eighbor), it will ot retrasmit ay other istaces of the message, regardless of whether the first istace was retrasmitted or ot ad regardless of ay followig istaces arrivig from odes which have selected it as multipoit relay. Also otice that the defiitio of multipoit-relay isures that the packet is propagated to the etire etwork

3 3 (if o other restrictive rule applies which prevets that). Alargeamoutofcotroltrafficcabesavedwiththis scheme, especially i dese etworks (see [8], [9] for more details ad for a heuristic for computig multipoitrelays). This scheme implies some additioal cotrol traffic i form of a eighbor discovery mechaism. The beefit of this scheme is the that it allows a reductio of the overhead from floodig cotrol messages. Ituitively, usig this scheme may yield shorter routes: the miimizatio of the umber of multipoit-relays ecourages that a ode select MPRs that each cover a large fractio of the 2-hop eighborhood. This icreases the probability of usig log rage liks. III. IDEAL PHYSICAL LAYER ANALYSIS I this sectio, we will explore the details of floodig with the purpose of discoverig how suboptimal routes may occur. We will do so i the cotext of a ideal physical layer. Thus, we will cotiue by defiig the model for this ideal physical layer, followed by a aalysis of the optimality of routes obtaied through basic floodig. This ideal physical layer will also be used for the simulatios described i sectio IV. A. Model We assume a etwork of odes coected through wireless liks, the radio iterfaces (trasmitters ad ateas) beig idetical for all odes i the etwork. We assume that the radio iterfaces have a fixed rage, ad that a collisio avoidace scheme is employed. I this sectio, we further suppose that o collisios occur ad that access to the radio media is fair (i.e. amog the odes competig for the right to trasmit i a regio, all odes have the same probability of succeedig). Notice, that real lik layers such as IEEE 82. usually strive to reach these goals of collisio avoidace ad fairess which we assume. The coverig area of a ode is a circular disk, cetered at the ode. A ode ca commuicate directly to ay ode i its coverig area. The odes i our etwork are radomly placed i a rectagular field. Fially, the aalysis i this sectio assumes that o mobility is preset (i.e. odes do ot move). This radio model does ot take shadowig effects ito accout While this free space rectagular field model is highly improveable, it provides a approximatio to the real world i which it is possible to coduct a aalysis. I sectio V, we provide simulatios usig a more comprehesive model of the etwork (ad, ideed, the physical layer), iteded to validate the results from the aalysis i this approximated model. Fig Y Propagatio of a floodig message i D etwork. This model is related to the uit graph model [5] where odes are distributed i a square ad a valid lik exists betwee ay pair of odes as log the distace betwee the odes is shorter tha the uit (the radio rage). B. Uidimesioal Aalysis I this sectio, we aalyze the legth of routes as costructed by floodig i a uidimesioal (D) etwork. This model ca be see as a dese strip (i.e. high desity ad very arrow field). As a practical example of a oe dimesioal etwork, cosider a highway with cars, equipped with radio commuicatio uits. We are goig to show that the ratio of the route legth over optimal distace asymptotically teds to. This basically meas that the farther away a ode is, the less optimal the route to that ode will be. Obviously, the path which a message will take whe beig flooded i a D etwork will describe a straight lie. I a 2D-etwork, ituitively, a flooded packet has the possibility to follow ay curved path available. Thus, it ca be assumed that, though this aalysis cocers oly a D etwork, the average floodig distace i a 2D etwork may be greater tha that of the D etwork. We will ow preset a formal proof of the above assertio. To simplify the aalysis, we restrict the defiitio of floodig distace by the legth of the first path reachig the destiatio ad ot the shortest path reachig the destiatio. However, sice the shortest path always comes from a eighbor ode for which it was a first path, the legth of the shortest path will at least be the estimated legth mius oe. Thus, our asymptotic estimatios are valid for the floodig scheme, described i sectio II. I priciple, a sigle floodig creates a route from the source of the floodig to ay ode receivig the floodig packet. We call the legth of the route obtaied for a ode, the floodig distace of that ode. Suppose that the odes are desely placed o a lie (a D etwork), as illustrated i figure. We ow cosider propagatio of oe flooded message. We cosider the propagatio of the floodig as a message, origiated i the left-had side of the etwork, prop-

4 4 (a) Fig floodig distace x x x x x 2 3 odes reached by the floodig (b) (c) limit y x + y + ew limit x + y + uiformed odes distace Floodig distace versus cartesia distace i D etwork. agatig to the right. The uit idicated is the coverig area of a ode (a emissio at positio covers ). Nodes i the iterval will already have received the flooded message, ad will hece ot cosider the message agai, accordig to the forwardig rule described i sectio II. Nodes i the iterval will receive this message for the first time ad will hece attempt to re-emit the message. At a give time, the odes, willig to re-emit the message, form a dese set i the iterval. Ayode i that set may succeed i emittig (due to radomizatio i the chael access protocol). Let, be the positio of the first emissio of the flooded message i the iterval,..., ad for let be the positio of the first reemissio i. I (the iterval of odes, which are reached by the retrasmissio from ad which have ot received the trasmissio before), let be the limit betwee the odes that have same floodig distace as ad those havig floodig distace oe more tha.the floodig distace is clearly a icreasig fuctio, ad as the odes i have received the emissio of,theirfloodigdistaceisotgreaterthathefloodig distace of plus oe. This is illustrated i figure ad i figure 2. I the figure 2 the floodig distace, sice iteger, have discotiuities which correspods to the limits above described. (a) presets the limits before icludig the curret limit at.(b)correspodstothecasewhere the ext emissio o the right of is o the left of the limit.(c)correspodstothecasewheretheext emissio o the right of is o the right of the limit. We ca compute the probability distributio of.let d be the probability that.noticethat depeds oly o.asillustratedbyfigure2,thereare maily two cases, depedig o the positio of with regard to the limit. Ifitisotheleft, oegets,thelimitfor is the same as for.ifitisotheright,oegets,thelimitfor is the right boud of.weimmediatelydeduce the followig recurrece (remember that is uiform i ): Clearly.Wethusdeduce.Wethecompute. is thus statioary for with. The mea value of is thus d.theprobabilitythataewfloodigdistace limit is created is thus.theaveragedistaceof is clearly.itsfloodigdistaceistheumberofdistict floodig limits i, i.e..theratiofloodig distace over distace i this model is thus asymptotically whe icreases. IV. IDEAL PHYSICAL LAYER SIMULATIONS We coduct a umber of simulatios, usig the ideal physical layer. The purpose of these simulatios is to expose the problem of floodig distace usig basic floodig, as well as to ivestigate the impact o the floodig legth of usig super-floodig ad MPR floodig. We coduct simulatios, usig two basic scearios: strip etwork Afield-sizeof5x3.Varyigodedesity (5 odes, odes ad 222 odes) square etwork Afield-sizeofx. Varyig ode desity ( odes ad 222 odes) The odes are radomly distributed throughout the field. The coverig area of each ode is a disk with radius 25m. Nodes are fixed (i.e. o mobility) ad placed radomly i the field. I the followig subsectios, we preset our simulatio results, comparig the various floodig schemes. A. Distributio of Floodig Distaces Figure 3 shows the distributio of floodig distaces of odes i the strip etwork with a total of odes. Figure 4 shows the distributios of floodig distace with odes but i the square etwork. We preset results d d

5 Basic floodig MPR floodig Super floodig MPR super floodig.4.2. Basic floodig 5 odes MPR floodig 5 odes Basic floodig odes MPR floodig odes Basic floodig 222 odes MPR floodig 222 odes Number of odes at fd Number of odes at fd Floodig distace fd Floodig distace fd Fig. 3. Number of odes observed at a certai floodig distace for various schemes i the strip etwork with odes. Fig. 5. Fractio of odes observed at a certai floodig distaces i the strip etwork. Oly odes at optimal distace 4 where cosidered Basic floodig MPR floodig Super floodig MPR super floodig.6.4 Basic floodig odes MPR floodig odes Basic floodig 222 odes MPR floodig 222 odes 2.2 Number of odes at fd Number of odes at fd Floodig distace fd Floodig distace fd Fig. 4. Numberumber of odes observed at a certai floodig distace for various schemes i the square etwork with odes. Fig. 6. Fractio of odes observed at a certai floodig distaces i the square square. Oly odes at optimal distace 4 where cosidered. correspodig to basic floodig, MPR floodig ad Super Floodig. Notice that MPR ad Super Floodig provides sharper distributio, deotig a shorter floodig distace tha basic floodig. MPR floodig distace ad Super Floodig distaces are very close, provig that MPR floodig distaces are close to optimal distace which, by defiitio, is give by Super Floodig. Oe should ot cofuse the distace obtaied via MPR floodig ad the distace obtaied by the computatio of routes via MPR as provided i OLSR [8], [9]. The latter routes are optimal but eeds the proactive advertisemet of MPR liks throughout the etwork via Topology Cotrol (TC) messages. B. Compariso Floodig versus MPR Floodig I figure 5 we display the floodig distace distributio whe the optimal distace is fixed (for istace a optimal distace of 4 hops). As expected the MPR floodig distaces are shorter ad have less variatios. I figure 6 we display the floodig distace distributio i a square etwork for odes at optimal distace 4. The differece with MPR floodig is ot as sigificat as i the strip etwork. C. Ratio Floodig Distaces with Optimal Distace Figure 7 displays the average ratio betwee floodig distace ad optimal distace versus respectively obtaied obtaied with 5,, 222 ad 3 odes. The simulatio is doe i a 5x3 strip which correspods more or less to the D model. Figure 8 displays the same quatity but with MPR floodigs. Figure 9 show the same quatities as the two previous figures but i the square etwork.

6 6 Average floodig distace / Optimal distace odes odes 222 odes 3 odes Average floodig distace / Optimal distace basic floodig with odes basic floodig with 222 odes MPR floodig with odes MPR floodig with 222 odes Optimal distace Optimal distace Fig. 7. Ratio floodig distace over optimal distace i the strip etwork with basic floodig. Fig. 9. Ratio floodig distace over optimal distace i the square etwork with basic floodig ad multipoit-relay floodig. Average floodig distace / Optimal distace basic floodig with 5 odes basic floodig with odes MPR floodig with 5 odes MPR floodig with odes Mea umber of emissios Basic floodig MPR floodig Super floodig MPR super floodig Optimal distace Number of odes Fig. 8. Ratio floodig distace over optimal distace i the strip etwork with basic floodig ad multipoit-relay floodig. Fig.. Average umber of emissios i the strip etwork for various umber of odes. D. Cotrol Traffic Overhead Figure displays the umber of request packet retrasmissio usig the differet floodig mechaisms i the strip etwork. It is oticed that MPR floodig causes atremedousreductioipacketretrasmissioadthat the extra trasmissios caused by Super Floodig, as compared to basic floodig, is still reasoable. V. NS-2 SIMULATIONS We coduct simulatios usig the etwork simulator s2 [5] with the purpose of validatig that our results from usig the ideal physical layer ad the free space rectagular field are idicative for real-world etworks. As a real-world etwork, we pick a MANET, ruig the reactive routig protocol AODV [3]. The sceario we use are the same as those used i sectio IV, except that we itroduce a umber of (low-itesity) CBR data traffic streams. We do this sice AODV is a reactive routig protocol, ad hece does t attempt to set up routes before data trasmissio is eeded. We coduct simulatios usig basic floodig as well super-floodig, ad measure the average route legth, take by the data traffic. The data traffic patter we employ have the followig characteristic: 5 cocurret streams (source, destiatio pairs), each stream with a duratio of secods (after which the sources ad destiatios chage), ad each stream carries 64bytes/s. Furthermore, we coduct our simulatios both with ad without ode mobility. With ode mobility, the odes move at a idividually radomly chose speed betwee ad. For each of the described scearios, we radomly geerate differet sceario files for s2. Thus, with traffic streams of secods, for each sceario we get 25

7 7 2 Super floodig, square Super floodig, strip Basic floodig, square Basic floodig, strip Super floodig, square Super floodig, strip Basic floodig, square Basic floodig, strip bytes/secod 6 Hops Nodes Nodes Fig.. Cotrol traffic overhead i a static etwork. Fig. 2. Path legth i a static etwork. samples of the route legth, provided. We use the exact same sceario files for the simulatios of AODV as well as AODV + super-floodig. We compute ad compare the average route legths ad the cotrol traffic overhead of the two approaches i the differet scearios Super floodig, square Super floodig, strip Basic floodig, square Basic floodig, strip A. Cotrol Traffic Overhead Our first observatio cocers the amout of cotrol traffic geerated by the two floodig schemes. We recall that, from the defiitio of the floodig schemes i sectio II, a ode may potetially forward the same message whe receivig it multiple times, depedig o the TTL of the message. We otice, that assumig the first message received by the ode has traveled the shortest possible path, the exact same amout of cotrol traffic overhead should occur with both basic floodig ad Super Floodig. Our simulatio results describig the cotrol traffic overhead are icluded i figure. We observe that the amout of cotrol traffic geerated through usig Super Floodig is orders of magitude larger tha that of usig basic floodig. Give that we have used idetical simulatio scearios ad traffic patters, this leads to the coclusio that basic floodig does, ideed, ot provide optimal routes. The observed path legths are icluded i figure 2. We observe that, cosistetly, the path legth as obtaied by Super Floodig is shorter tha that of basic floodig. Relatig the path legth obtaied by basic floodig to the path legth obtaied by Super Floodig (i.e. optimal paths), we further observe that the path legths provided by basic floodig cosistetly are 5-% loger. This makes Super Floodig attractive as soo the traffic payload exceeds te times the floodig route request overhead. Hops Fig Nodes Path legth i a etwork with mobility. Figure 3 shows the path legths i a etwork with mobile odes. We observe, that the differece betwee the path legths, obtaied through Fuper Floodig ad through basic floodig are substatial - ad i favor of Super Floodig. We otice that the paths are sigificatly loger tha the paths obtaied by the simulatios o the ideal physical layer. This is caused by two reasos: (i) the simulated physical layer o s2 is ot ideal ad hece, sice it is more realistic, collisios occur that may kill propagatio o more optimal path, (ii) special AODV feature such as local route repair, oe-hop route request, or expadig rig search, are likely to icrease the path legth, with or without Super Floodig. Figure 4 shows the amout of cotrol traffic, geerated by the routig protocols. It is evidet that, as i figure the cost of usig Super Floodig to obtai optimal routes is a substatial icrease i cotrol traffic. Ayhow, this cost is likely to be absorbed by the spare i packet retrasmissios due to reduced path legth.

8 8 bytes/secod Fig. 4. Super floodig, square Super floodig, strip Basic floodig, square Basic floodig, strip Nodes Cotrol traffic overhead i a etwork with mobility. VI. CONCLUSION AND FURTHER WORKS The total overhead, icurred by a routig protocol, cosists of two elemets: overhead i form of cotrol traffic geerated by the protocol, as well as overhead from data traffic, forwarded through routes of o-optimal legth. Such o-optimal routes brig a o- egligible overhead that is proportioal to the data load of the etwork. We have show, through a simple aalysis, that basic floodig does, ideed, yield o-optimal routes. We have the proposed two simple complimetary floodig schemes, which aim at reducig route legth overhead: MPR floodig ad super-floodig. MPR floodig reduces both the route discovery floodig overhead as well as provides shorter routes. The drawback is the requiremet of a eighbor sesig mechaism. Super-floodig, likewise, provides shorter path - however at the cost of a icreased route discovery overhead. We have preseted simulatios, substatiatig our aalytical results. Sice the MPR floodig scheme ad the super-floodig scheme are complimetary, we aticipate that combiig the two could yield some beefits. Ivestigatig this hypothesis will costitute parts of our future efforts. REFERENCES [] G. Aggelou ad R. Tafazolli. Rdmar: A badwidth-efficiet routig protocol for mobile ad hoc etworks. I Proceedigs of ACM MobiCom99/WoWMoM99, August999.Washigto,USA. [2] J. Broch, D.A. Maltz, D.B. Johso, Y.C. Hu, ad J. Jetcheva. Aperformacecomparisoofmulti-hopwirelessadhocetwork routig protocols. I Proceedigs of the Fourth Aual ACM/IEEE Iteratioal Coferece o Mobile Computig ad Networkig, October 998. Dallas. [3] M. S. Corso ad J. Macker. Mobile ad hoc etworkig (MANET): Routig protocol performace issues ad evaluatio cosideratios. RFC 25, Jauary 999. [4] J.J. Garcia-Lua-Aceves ad M. Spoh. Efficiet routig i packet-radio etworks usig lik-state iformatio. I Proc. IEEE WCNC 99,August999. [5] P. Jacquet, A. Laouiti, P. Miet, ad L. Vieot. Performace aalysis of olsr multipoit relay floodig i two ad hoc wireless etwork models. Techical Report 426, INRIA, 2. [6] David B. Johso ad David A. Maltz. Dyamic source routig i ad hoc wireless etworks. Mobile Computig, 5:53 8,996. Kluwer Academic Publishers. [7] Scott Corso Joseph Macker. Mobile adhoc etworkig ad the ietf. ACM Mobile Computig ad Commuicatios Review, 998. [8] A. Laouiti, A. Qayyum, ad L. Vieot. Multipoit relayig: A efficiet techique for floodig i mobile wireless etworks. Techical Report RR-3898, INRIA, 2. [9] A. Laouiti, A. Qayyum, ad L. Vieot. Multipoit relayig: A efficiet techique for floodig i mobile wireless etworks. I 35th Aual Hawaii Iteratioal Coferece o System Scieces (HICSS 2). IEEEComputerSociety,2. [] S.-J. Lee, M. Gerla, ad C.-C. Chiag. O-demad multicast routig protocol. I Proceedigs of IEEE WCNC 99, pages , September 999. New Orleas, LA. [] V. Park ad M. S. Corso. A highly adaptive distributed routig algorithm for mobile wireless etworks. I Proc. IEEE INFOCOM 97. Kobe,Japa,997. [2] Guagyu Pei, Mario Gerla, ad Tsu-Wei Che. Fisheye state routig: A routig scheme for ad hoc wireless etworks. I Proceedigs of the IEEE Iteratioal Coferece o Commuicatios, 2. [3] C. E. Perkis, E. M. Royer, ad S. R. Das. Ad hoc o-demad distace vector (AODV) routig. Iteret Draft,draft-ietf-maetaodv-9.txt, November 9 2, Work i progress. [4] Charles E. Perkis ad Elizabeth M. Royer. Ad-hoc o-demad distace vector routig. I Proceedigs of the 2d IEEE Workshop o Mobile Computig Systems ad Applicatios, pages9, February 999. New Orleas, LA. [5] Network Simulator. s-2. Available at