A Multipath Energy-Conserving Routing Protocol for Lifetime Improvement of Wireless Ad Hoc Networks

Transcription

1 A Mltpath Eergy-Coservg Rotg Protocol for Lfetme Improvemet of Wreless Ad Hoc Networks Omar Smal *, Berard Cos 2, Zolkha Mekkaka ad Rachda Mekk Abstract Ad hoc etworks are wreless moble etworks that ca operate wthot frastrctre ad wthot cetralzed etwork maagemet. Tradtoal techqes of rotg are ot well adapted. Ideed, ther lack of reactvty wth respect to the varablty of etwork chages makes them dffclt to se. Moreover, coservg eergy s a crtcal cocer the desg of rotg protocols for ad hoc etworks, becase most moble odes operate wth lmted battery capacty, ad the eergy depleto of a ode affects ot oly the ode tself bt also the overall etwork lfetme. I all proposed sgle-path rotg schemes a ew path-dscovery process s reqred oce a path falre s detected, ad ths process cases delay ad wastage of ode resorces. A mltpath rotg scheme s a alteratve to maxmze the etwork lfetme. I ths paper, we propose a eergy-effcet mltpath rotg protocol, called AOMR-LM (Ad hoc O-demad Mltpath Rotg wth Lfetme Maxmzato), whch preserves the resdal eergy of odes ad balaces the cosmed eergy to crease the etwork lfetme. To acheve ths goal, we sed the resdal eergy of odes for calclatg the ode eergy level. The mltpath selecto mechasm ses ths eergy level to classfy the paths. Two parameters are aalyzed: the eergy threshold β ad the coeffcet α. These parameters are reqred to classfy the odes ad to esre the preservato of ode eergy. Or protocol mproves the performace of moble ad hoc etworks by prologg the lfetme of the etwork. Ths ovel protocol has bee compared wth other protocols: AOMDV ad ZD-AOMDV. The protocol performace has bee evalated terms of etwork lfetme, eergy cosmpto, ad ed-to-ed delay. eywords: Ad hoc etwork, Mltpath rotg, Eergy effcecy, Network lfetme. Itrodcto A ad hoc etwork s characterzed by freqet chages the etwork topology, lmted badwdth avalablty, ad lmted power of odes. The ad hoc etwork topology chages freqetly as odes are able to move collectvely or dvdally ad ofte a predctable way. Whe oe ode moves ot of or to the trasmsso rage of aother ode, the wreless lk betwee the two becomes dow or p. Aother case of the topologcal chages s the stablty of the wreless lk qalty, whch mght become hgh or low de to the sgal fadg (obstacles betwee the two wreless odes), terferece from other sgals, or the chage the *Correspodece: omar.smal@v-sto.dz Faclty of Mathematcs ad Iformatcs. Compter Scece Departmet. Uversty of Sceces ad Techology, Mohamed Bodaf USTO-MB, Ora, Algera. 2 IRISA/Uversty of Rees, Frace. ode trasmsso power level. I the followg we refer to the dsappearace of the wreless lk for ay reaso as lk falre. Moble odes are battery powered; whe they r ot of battery power, they fal. All these types of falres crease the testy of chages the etwork topology. As the odes have a lmted commcato rage, the path from sorce to destato sally has mltple hops (the data packets are retrasmtted by some termedate odes); hece these characterstcs make rote dscovery complex. The rotg problem moble wreless etworks has attracted cosderable terest the research commty. Several research stdes have focsed o rotg protocols of ad hoc etworks [ 5]. Whle the proposed protocols have some relevat characterstcs, they have lmtatos, especally the case of hgh moblty of odes or hgh etwork load. The tradtoal approach of rotg moble ad hoc etworks adopts a sgle actve rote betwee a sorce ode ad a destato ode for a gve commcato. Ths sally ses proactve [6,7] or reactve (o-demad) [8,9] rotg

2 protocols. I [0], t s show that proactve protocols are very expesve terms of eergy cosmpto compared to reactve protocols, becase of the large rotg overhead crred the former. Bt reactve protocols sffer from hgh latecy drg the process of dscoverg fresh paths, especally large etworks ad dese etworks []. I recet years, the research commty has focsed o the mprovemet of ad hoc rotg, wth the developmet of several rotg mechasms. Mltpath rotg seems to be a effectve mechasm ad hoc etworks wth hgh moblty ad hgh load to gard agast the problem of freqet chages of the etwork topology cased maly by lk falres. The cocept of mltpath rotg s that the sorce ode s gve the choce betwee mltple paths to reach a gve destato. The mltple paths ca be sed alterately; the data traffc takes a sgle path at oe tme or several paths smltaeosly. A mltpath betwee a sorce ad destato mst be chose wsely so that a path falre does ot dstrb other paths as less as possble. There are two types of dsot paths: lk dsot paths ad ode dsot paths. Node dsot paths have o ode commo except the sorce ad destato. Lk dsot paths have o commo lks, whle they may share some odes. Ay path of a mltpath ca be sed to trasmt a data packet betwee a sorce ad a destato. Ths to maxmze the data flow ad to get a larger share of the etwork badwdth, the data packets of a flow betwee a sorce ad a destato ca be splt betwee the paths [2]. Mltpath rotg s hghly stable for mltmeda applcatos, to esre secre trasmsso ad t s proposed for dstral ad hoc etworks for mprovg relablty ad determacy of data trasmsso. Aother beeft of mltpath rotg protocols s the redcto of the rotg overhead, for whch several mltpath rotg protocols have bee developed [8,3 5]; these works propose a sgle path dscovery process able to bld several lk or ode dsot rotes towards the destato based o broadcast reqests. The approaches [6,7] are mprovemets of sgle path rotg protocols. They cotrbte redcg delays ad creasg throghpt becase they resst ode moblty comparso wth sgle path approaches. I ad hoc etworks, each ode has a power battery ad a lmted eergy spply. Over tme, odes deplete ther eergy spples ad are evetally removed from the etwork, whch costras the packet rotg. Some exhasted odes may be crtcal for packet trasmsso whe they are volved the oly path from the sorce to the destato ode. To solve ths effcecy, we propose a protocol whch exteds the lfetme of the etwork. We sed the eergy levels of odes to classfy the odes of paths betwee a sorce ad a destato. The ode eergy level s calclated based ts resdal eergy ad o odes eergy that cotrbtes the rotg; ths formato s collected the dscovery of mltple paths sg the same prcple as the dscovery of a sgle path. Two parameters are trodced, the eergy threshold β ad the coeffcet α, to defe the ode class order to preserve the cosmed eergy. Or protocol selects the odes wth smlar classes to costrct a gve path. At the ed of the rote dscovery process the sorce ode wll have a set of homogeeos paths terms of eergy. The best classfed paths are selected to balace the eergy cosmed betwee the dfferet paths of a mltpath. The proposed techqe dscards odes wth a crtcal eergy level ad makes sre they do ot partcpate rotg, otherwse we may have several lks falres cased by the depleto of odes. The paper s orgazed as follows. Secto 2 provdes a revew of related works for eergy-aware rotg protocols wreless ad hoc etworks. Secto 3 presets a overvew of the AOMDV protocol ad motvates ts explotato. Secto 4 gves the desg detals of or AOMR-LM protocol. Secto 5 provdes the smlato reslts of ts performace evalato. Secto 6 cocldes ths paper. 2 Eergy-Aware Rotg Protocols The ma obectve of ad hoc rotg protocols s to delver data packets amog etwork odes effcetly, whch depeds o the ode lfetme. The lfetme of a ode s drectly proportoal to ts battery's eergy. The ode battery's eergy s prmarly cosmed whle trasmttg ad recevg packets or comptg. As ad hoc etworks are mlthop etworks, oe ode may be volved retrasmsso of packets set betwee a sorce ode ad a destato ode. The odes partcpatg packet trasmsso, ca exhasted ther eerges ad removed from the etwork. Ths affects the relablty of the packet delvery servce of the etwork. So a mlthop commcato, the selecto of the approprate odes by the rotg protocol s very mportat. Ths, rotg algorthms play a mportat role savg the commcato eergy, ths exteds the lfe of the odes ad ths of the whole etwork. May works focsg o eergy-aware rotg protocols exst the lteratre [8-23]. These eergy-aware rotg protocols ca be sgle path or mltpath. I the followg, the ma proposed works abot eergy-aware sgle path rotg protocols are descrbed. I [24] the athors preseted several rotg protocols wth eergy measremets that track eergy effcetly. Mmm Total Trasmsso Power Rotg (MTPR) [25] was tally developed to mmze the power cosmpto of odes volved a path. However, sch protocols do ot take ode eergy capacty to accot. Ths, the eergy cosmpto s ot farly shared amog odes the etwork. M-Max Battery Cost Rotg (MMBCR) [26] cosders the remag power of odes as a metrc for the acqsto of paths order to exted the lfetme of the etwork. Bt, ths protocol caot garatee that the eergy cosmpto wll be mmzed. The ma advatage of o-demad rotg comes from the redcto of the rotg overhead, as a hgh rotg overhead has a sgfcat mpact o the performace of wreless etworks. A o-demad rotg protocol attempts to dscover a rote toward a destato whe a data packet s preseted to be trasmtted to that destato. Provdg mltple paths for trasmttg the data packets s beefcal ad hoc etwork commcatos: t ca be sefl for mprovg the effectve badwdth of commcato, redcg the rotg overhead, ad decreasg delvery delays. Recetly, several o-demad rotg mltpath protocols have bee proposed that preserve eergy order to avod etwork falres as log as possble.

3 I [27], L et al. propose a ew algorthm called Collso- Costraed mmm Eergy Node dsot mltpath rotg Algorthm for ad hoc etworks (ECCA). The algorthm s a trade-off betwee collso avodace ad eergy savg. It calclates a pper lmt of the correlato factor depedg o the servce reqred ad fds a dsot mltpath wth eergy whch satsfes the pper lmt. ECCA ca sgfcatly redce the packet loss rate ad the cosmed eergy. Hece, t attempts to provde eergy savgs at the odes. Bt t does ot take to accot the power stats of the odes ad ths the drato of avalablty of the etwork. The Mltpath Eergy-Effcet Rotg protocol (MEER) [28] ses a cotrol mechasm for ratoal power. The rote dscovery process whch the sorce tres to dscover rotes wth hgh eergy s based o SMR (Splt Mltpath Rotg) [29]. Ths protocol protects the odes from cosmg too mch eergy compared to the other odes the etwork. However the proposed approach s ot based o dsot rotes, ad ths t caot explot the beefts of path dsotess, whch s sefl to balace the eergy cosmpto amog etwork odes. Max- M Resdal Eergy (MMRE-AOMDV) [30] s a mltpath rotg protocol based o AOMDV (Ad hoc O-demad Mltpath Dstace Vector) []. Ths protocol fds the mmal odal resdal eergy of odes of each path ad the selects the path wth maxmal resdal eergy to forward the data packets. The MMRE-AOMDV protocol ses the rotg formato already avalable the derlyg AOMDV protocol. Ths lttle addtoal overhead s reqred for the comptato of the maxmal odal resdal eergy the rote. The MMRE-AOMDV protocol has two ma compoets: frst the comptato of the odal resdal eergy of each rote drg the rote dscovery process, ad secod the sortg of rotes by the descedg order of ther odal resdal eergy. It ses the rote wth maxmal odal resdal eergy to forward data packets. Smlato reslts showed that the proposed MMRE-AOMDV rotg protocol performed better tha AOMDV terms of packet delvery fracto, throghpt, ad etwork lfetme. Bt ths protocol does ot evalate the eergy cosmpto ad ts mpact o etwork performace, kowg that the etwork lfe depeds o the ode exprato, whch tr depeds po eergy cosmpto. I [3], L et al. proposed the Mltpath Rotg protocol for Network Lfetme Maxmzato (MRNLM), a protocol that defes a threshold to optmze the forwardg mechasm. It proposes a eergy-cost fcto ad ses the fcto as the crtero for mltple path selecto. Drg the trasmsso phase, they se a method called data trasmsso mltple paths oe by oe to balace the eergy cosmpto o the mltple paths. It s show that MRNLM cosmes less eergy tha AOMDV bt does ot mprove the ed-to-ed delay. Mltpath Mltmeda Dyamc Sorce Rotg (MMDSR) [32] s a mltpath rotg protocol able to self-cofgre dyamcally accordg to etwork states. The athors sed cross-layer techqes to mprove the ed-to-ed performace of vdeo-streamg servces over etworks sg IEEE 802.e. MMDSR ses a aalytcal model to estmate the path error probablty. Ths model s sed by the rotg scheme to estmate the lfetme of paths. I ths way, t s hoped that proper proactve decsos ca be take before the paths are broke. However, the comparso wth DSR (Dyamc Sorce Rotg) s ot far sce t s ot a mltpath rotg protocol. The Mltpath Eergy Effcet Rotg Protocol (MEERP) [33] s a exteso of the exstg rotg protocol AOMDV. Rote dscovery s modfed MEERP, whereby each termedate ode s prohbted from geeratg a rote reply message. The proposed protocol selects eergyeffcet mltple ode dsot paths based o the resdal eergy ad sccessfl trasmsso rate. I ths algorthm, mltple rotg paths are selected. However, oly oe path s sed for data trasmsso at a gve tme. Drg the pathdscovery process, each termedate ode calclates the cmlatve ode cost tl the destato, ad the path wth the hghest cost s selected. Ths cost depeds po two measremets: the sccessfl trasmsso rate ad the resdal eergy of the ode. The smlato reslts show that the proposed scheme ca acheve a great mprovemet of the etwork lfetme by redcg ed-to-ed delay ad overhead. I [34], the athors propose a mltpath rotg protocol based o AODV rotg algorthm (ZD-AOMDV). The represeted protocol ths stdy tres to dscover the dstct paths betwee sorce ad destato odes wth sg Om drectoal ateas, to sed formato throgh these smltaeosly. Ths protocol cots the mber of actve eghbors for each path, ad fally t chooses some paths for sedg formato whch each ode has lower mber of actve eghbors all together. Here, actve eghbors of a ode are defed as odes that have prevosly receved the RREQ (Rote Reqest).The am of ths work s to try to mprove the eergy effcecy of ad hoc etworks. All of the above stdes solve the problem of eergy coservato, bt the maorty of power-savg mechasms are based oly o the remag power caot be sed to establsh the best rote betwee sorce ad destato odes. O oe had, f a ode s wllg to accept all rote reqests oly becase t crretly has eogh resdal battery capacty, too mch traffc load may be roted throgh that ode. O the other had, excessve eergy savgs eglects the power cosmpto at the dvdal odes, t reslts etwork parttog de to odes battery exhasto. Ideed, t redces etwork performace. Hece, shared ad balaced eergy cosmpto s a remedy for those types of problems. Fally, the maorty of these protocols have bee compared oly wth orgal protocols (AODV, AOMDV, DSR, SMR,...), whch do ot explctly cosder eergy cosmpto, ad ths these performace evalatos are ot far. We wll se ZD- AOMDV [34] as a referece for or performace evalato becase t ams to mprove the lfetme of the ad hoc etwork ad has the same characterstcs as or protocol, amely ts reactvty, mltpath character, ad se of AOMDV as ts basc protocol. 3 AOMDV Overvew The exstg rotg protocols ad hoc etworks sch as proactve ad reactve protocols ca be modfed to corporate a power cotrol fcto whch prologs the etwork lfetme ad optmzes eergy cosmpto. We chose

4 AOMDV (Ad hoc O-demad Mltpath Dstace Vector) [] as the bass for or protocol. AOMDV has bee prove to be a good protocol that ses mltpath rotes; all of ts paths are dsot paths ad t ca garatee a loop-free path becase t allows oly alterate rotes wth lower hop cots. The AOMDV protocol performs a rote dscovery process betwee the sorce ode ad the desred destato ode whe the sorce eeds to sed a data packet ad the path to the destato s ot kow. I ths process, rote qery (RREQ) messages are broadcasted by every ode the etwork. The destato seds a rote reply (RREP) message for all of the receved RREQ packets. A termedate ode forwards a receved RREP packet to the eghbor that s alog the path to the sorce. Ths dscovery process ca be exploted to collect fresh ode formato, sch as resdal eergy, load level, ad so o. Several stdes [35 37] have show that the AOMDV protocol s more robst ad performs better most of the smlated scearos. So we selected ths protocol, stead of ay other reactve protocol (sch as SMR [29]), as the referece for performace evalato of or protocol. 4 The Ad hoc O-demad Mltpath Rotg wth Lfetme Maxmzato protocol I ths secto, a mproved rotg protocol, amed Ad hoc O-demad Mltpath Rotg wth Lfetme Maxmzato (AOMR-LM), s preseted. AOMR-LM s a mltpath rotg protocol based o AOMDV protocol, wth a ew path classfcato mechasm accordg to the eergy level of the odes formg these paths, whch ca be hgh, average, or low. The dea s to bld homogeeos paths terms of eergy, ths ca balace the cosmpto eergy of odes ad avod lk falres de to odes eergy depleto. The protocol sets a eergy threshold ad a coeffcet, sed to defe the class of each ode. These parameters are reqred to forward the reply packet decso. Ths dea helps at prologg the etwork lfetme ad mproves the eergy performace moble ad hoc etworks. I the sgle path rotg AODV protocol, mateace of paths occrs by sedg perodc short messages (called HELLO message). If three cosectve HELLO messages are ot receved by a ode from a eghborg ode, the ode cosders the lk to be broke. Whe the ode receves a data packet whch shold be roted va ths broke lk, a rote error message (RERR, Rote ERRor) s set back to the packet sorce dcatg the broke lk. I ths case a ew path dscovery process eeded, whch adds a extra cost terms of delay, throghpt, cotrol messages, ad cosderable eergy cosmpto for the addtoal trasfer of cotrol messages. Amog the reasos of falre of a ode s the lmtato of ts battery eergy. De to these ode falres, lks a path may become temporarly avalable ad make the path vald. The rotg algorthm decdes whch of the etwork odes eed to be selected a partclar path. Selecto of odes wthot takg to accot the ode eergy whe determg the paths leads to a mbalaced eergy level the etwork. Nodes o the mmm eergy paths cold qckly become draed whle other odes rema tact. Ths wll reslt the early death of some odes. For ths prpose, mltpath rotg has bee show to be effectve sce t dstrbtes the traffc load amog more odes ad proporto to ther resdal eerges. Whe the eergy cosmpto amog odes s more balaced, the mea tme to ode falre s prologed ad so the etwork lfetme. The proposed AOMR-LM protocol s a reactve protocol for mltpath rotg. Or selecto mechasm preserves the resdal eergy of odes ad balaces the cosmed eergy. Ths prologs the etwork lfetme ad mproves etwork performace. I the ext sectos, we frst trodce defe some assmptos ad the provde the ma detals of mltpath dscovery, selecto, data trasmsso, ad mateace procedres. 4. Problem defto A ad hoc wreless etwork s represeted by a drected graph, G = (V, E), where V s the set of etwork odes ad E s the set of etwork bdrectoal lks. Let ), V, represet the resdal eergy at ode. Let c(, v), (, v) E, be the eergy reqred to trasmt a packet from ode to ode v. We assme that c(, v) = c (v, ) for all (, v) E. If P s ay set of paths of G; the we cosder P( 0, ) to be the set of paths betwee odes 0 ad ode. Let P ( 0, ) = 0,,..., m,...,, be the th path P( 0, ), the sorce ode s oted by 0 or 0 ad the destato ode s oted by or. We defe P, ) ( 0 as the mber of odes o the path P ( 0, ) ad P, ) ( 0 as the mber of paths the mltpath dscovery process. The sm of the resdal eergy of a path P ( 0, ), deoted by e sm (P ( 0, )) s gve by: e sm ( P(, )) = ) 0 () 0 We cosder e sm (P( 0, )) to be the sm of the resdal eerges of all paths belogg to the set P( 0, ): sm ( P (, )) = e ( P(, )) 0 e (2) sm P( 0, ) Let e average (P ( 0, )) be the average resdal eergy of oe ode belogg to a path, e average (P ( 0, )) s gve by: e average ( P (, )) e 0 ( P ( 0, )) (, ) sm 0 = (3) P 0 The average resdal eergy of a set of paths betwee a same par of odes s deoted by e averagenet (P( 0, )), whch s the average resdal eergy of odes that partcpated the mltpath dscovery process betwee oe sorce ode 0 ad oe destato ode, e averagenet (P( 0, )) s gve by: e averagenet ( P(, )) e = ( P( 0, )) P(, ) sm 0 Where P P (4) 0

5 We defe 0, e ( ), the eergy level of ode drg a level dscovery process betwee a sorce ode 0 ad a destato ode, gve by: ( ) 0, w elevel ( ) = (5) e ( P(, )) averagenet Gve a sorce s, a destato d, ad a data packet to be roted, the sorce tates the mltpath dscovery. Ths dscovery provdes mltple dsot paths wth eergy effcecy propertes. The path selecto s based o the resdal eergy level of ts odes to preserve the ode eergy; ad to balace the eergy cosmed over a large set of paths we choose to classfy paths ad trasmt the data by choosg oe path of the avalable hghest class. 4.2 Mltpath dscovery Based o the o-demad rotg scheme, the sorce ode starts the mltple paths dscovery process to create a set of paths able to forward data towards the destato ode from the sorce ode. Whe a sorce reqres a rote toward a destato, the sorce checks ts rotg table for ay avalable path toward ths destato. If a path s ot preset or s vald, the sorce performs rote dscovery: t broadcasts a RREQ (Rote Reqest) message to all of ts eghbors. Whe a ode receves a RREQ, t esres that the receved RREQ s ot a dplcate RREQ by comparg the RREQs detfers, order to prevet loopg paths. Otherwse, the RREQ-recevg odes verfy whether they have ay vald path toward the destato ther rotg tables. If they have, they forward the RREQ to those vald path eghbors. I ths case, for or protocol, retrg back a RREP (Rote Reply message) to the sorce s ot desrable becase we wold lke to collect fresh eergy formato. Otherwse, the recevg ode retrasmts the RREQ message to all of ts eghborg odes to fd the paths toward the destato. Whe the destato receves the frst RREQ, t wats for a certa tme ad collects all other RREQs arrvg drg ths tme terval. Or protocol ses the destato seqece mber the same way as AOMDV to dcate the freshess of the rote, whch esres loop-freedom. Moreover, we se the oto of a advertsed hop cot to mata the mltple paths for the same seqece mber. The advertsed hop cot cotas the hop cot of the logest path allowed. Oce the seqece mber chages, the advertsed hop cot s reset ad remas chaged for ths seqece mber. Ideed a ode blds a alterate path to a certa destato ode va a eghborg ode oly f ths alterate path has a smaller advertsed hop cot. Several chages are eeded the AOMDV rote-dscovery procedre to eable comptato of the sm ad average odal resdal eergy of the etwork. Each RREQ ow carres a addtoal feld, called e sm (P ( 0, )), whch represets the sm of resdal eergy from the sorce to the crret ode. Aother feld ) s added, so that a ode kows the resdal eergy of ts eghbors. Whe a termedate ode receves o a RREQ, t creases the feld e sm (P ( 0, )) by the vale of ts resdal eergy. The mltpath dscovery mst take to accot the message seqece mber order to esre the freshess of paths ad the maxmm hop cot for all the paths, deoted respectvely by seqm ad advertsed_hopcot. Ay RREQ message receved wth a seqece mber lower tha the largest seqece mber receved ay prevos RREQ message from that sorce towards that destato s dscarded. The same process s repeated tl the RREQ message reaches ts fal destato; see Algorthm. f (seqm d <seqm d ) the seqm d :=seqm d ; f ( d ) the e sm (P( 0, )):= e sm (P( 0, ))+ ); advertsed_hopcot d := ; rote_lst d := NULL; sert (, advertsed_hopcot d +, ) ) to rote_lst d ; else advertsed_hopcot d :=0; edf elsef (seqm d =seqm d ) ad ((advertsed_hopcot d,)>(advertsed_hopcot d,)) the e sm (P( 0, )):= e sm (P( 0, ))+ ); sert (, advertsed_hopcot d +, )) to rote_lst d ; edf Algorthm : AOMR-LM path dscovery process Fgre shows the strctre of a etry of the rotg table of a ode. For each destato kow by the ode there s a etry. Rote_lst cotas all kow eghborg odes of a ode whch leads to that destato. Each eghbor for that destato s detfed ts exthop address, ad the hopcot feld s the mber of hops reqred to reach that destato sg ths eghbor. We add two ew felds, ) ad marked_ode, the rote_lst. The feld ) deotes the resdal eergy of a ode ad the feld marked_ode dcates whether or ot a ode has bee selected by a reverse path (explaed later). Destato Seqece_mber Advertsed_hopcot Rote_lst {(exthop, hopcot, ), marked_ode), (exthop2, hopcot2, 2 ), marked_ode), } Exprato tmeot Fgre : Strctre of a rotg table etry for AOMR-LM. 4.3 Mltpath selecto After recepto of the frst RREQ packet, the destato ode wats for a certa perod of tme RREQ_Wat_tme before startg the rote selecto procedre. Ths perod of watg geerates a addtoal delay for the mltpath rotg selecto. However the reslts of the evalatos (Secto 5) demostrates ts low mpact o ed-to-ed delay. Whe ths

6 perod of tme expres, the destato ode geerates a rote reply (RREP) message ad seds t back to the sorce. I the covetoal AOMDV, the propagato of rote reqest messages from the sorce towards the destato establshes mltple paths. However, t does ot cosder the resdal eergy of odes. I or protocol, the eergy of a ode ) s cosdered whe the RREP packet s forwarded back. So the selecto of the ext ode toward whch the RREP packet s forwarded depeds o the ode class. To determe the belogg of a ode to oe of the three classes, we defe two thresholds, deoted α ad β. α beg the lower threshold whch separates the low class from the mddle class, β the pper threshold whch separates the mddle class from the pper class. The eergy threshold β s compted for all etwork odes. So for or solto, we assme that the threshold β s the average resdal eergy of the etwork. If the eergy of ode (.e. )) s above the threshold ( ) β), the ode has a hgh probablty of beg able to trasmt data packets, ad ths t s classfed as hgh class. Whe ) < β, the defto of the ode class depeds o the threshold β ad the coeffcet α, whch s ecessary for the forward decso. The coeffcet α s trodced to determe the ode's capacty to spport data traffc terms of eergy. Ths coeffcet decdes the ode partcpato rotg. ) Whe ) < β the <. Let α < ; we obta: β ) α * < (6) β To aalyze the vales of the coeffcet α, we defe T Net (P( 0, )), whch descrbes the effectve partcpato of etwork odes the trasfer of data o paths of P( 0, ). The rest of the odes do ot affect the data trasfer, so they wll ot be cosdered the formlato of or solto. We have: T Net ( P(, )) = 0 = ) α α * β β * = ) Applyg the eqalty of arthmetc ad geometrc meas αx + α 2x α x α α α 2 α x x... x α where α α + α α 0, = 2 > f we take, = we obta α, ad x = w ) = The we have = ) = 2 ( ( )) α β * = whch proves that T Net α β α ) β ( P( 0, )) = * ( = )) = *( ) α * β )) = To smplfy the aalyss, we pose = ) e averagenet ( P, )) ( 0, sce the sm of the resdal eergy of odes dvded by the mber of odes gves the same vale calclated by Eqato (4). It s cosdered that f the resdal eergy of a ode s greater tha the average resdal eergy of the etwork, the ths ode has sffcet eergy ad has a hgh probablty of trasmttg more data packets before beg exhasted. Ths case correspods to ( ) β), so: averagenet ( P, )) β = e (7) Therefore: T ( T Net ) Net = ( 0 = ) α * β k α, ad we obta: α (8) Followg (6), we obta β α < (9) ) Wth (8) ad (9), we have β α ( T ) Net, (0) ) After recevg a RREP message, a termedate ode checks whether the RREP message has already bee receved (the termedate ode detfes each RREP message by the IP address of the orgatg ode, ad the destato seqece mber order to detect a dplcate message). If the RREP message has already bee receved (or s older tha a already receved RREP message), the ode dscards t drectly; otherwse t calclates the eergy levels of all eghborg odes ad determes ther classes. Or dea s to classfy odes accordg to ther eergy levels. Whe the vale of ) of ode s hgh ad Eqato (7) s sed, ths case s redced to e level ( ). Three classes of odes are defed:

7 Low: The eergy level of the ode s below α. Average: The eergy level of the ode s betwee α ad. Hgh: The eergy level of the ode s greater or eqal to. I the AOMDV protocol, the termedate ode establshes a reverse path by selectg the frst eghborg ode from the rote_lst feld of the rotg table (see Fgre ). I or protocol the termedate ode determes the class of each eghborg ode stored the table ad forwards the RREP message to the eghborg ode of the same class. If the classes of all eghborg odes are dfferet from the class of the termedate ode, the eghborg ode wth the lowest class above the class of the termedate ode s selected; f o hgher class exsts, the eghborg ode wth the hghest class der the class of the termedate ode s selected. Ths process s repeated at every termedate ode tl the sorce ode. At the ed of the process all dscovered paths have same or ear odes classes, ths reslts homogeeos paths. The sorce ode wll have all paths eergy formato, ths facltates the paths selecto for rotg. To esre lk dsot paths, a termedate ode marks each eghborg ode selected by aother reverse path; ths property s esred by the marked_ode feld of the rotg table (see Fgre ); addto or solto evalates the eergy level of each ode, so the exhasto of a ode s cosdered ad does ot affect the lks of paths. The traectores of each RREP message may tersect at a termedate ode, bt each takes a dfferet reverse path to the sorce to esre lk-dsotess. Whe the ode sorce receves the frst RREP message for a destato, t wats a specfed tme, called RREP_Wat_tme, before selectg the best path; ths tme s a addtoal delay for mltpath rotg protocols. 4.4 Data trasmsso phase The choce of the best path betwee a sorce ode s ad destato ode d depeds o the eergy levels of odes. AOMDV trasmts data packets sg the frst path from the lst of avalable paths whe mltple paths are establshed. I AOMR-LM protocol, paths are groped by class (hgh, average, ad low) accordg to the classes of odes that bld these paths. The paths of hgh class are selected frst to forward the data packets. I ths case, all packets gog to the same destato follow the same path as log as there s o lk falre ths path. Oce the hgh class s empty, the algorthm selects ay path of the average class. The same process s repeated wth the low class. Ths process balaces the odes eergy cosmpto whch exteds the etwork lfetme. Oe path s selected at a tme for a data packet trasmsso betwee a sorce ad a destato. If the selected path fals, the sorce ode receves a RERR; ths case, a path s selected from the hghest avalable class. Path dscovery s tated aga whe there are o more paths; ths dscovery process s descrbed Secto Rote mateace Rote error detecto AOMR-LM s smlar to rote error detecto AOMDV. It s lached whe a lk fals betwee two odes alog a path from a sorce to a destato. Whe a eghborg ode does ot respod to three sccessve HELLO messages set by a ode, the lk s cosdered to have faled. If a ode detects a falre of a lk a actve path, t erases the rote from ts table ad the seds a RERR message to the sorce ode of the path. Each termedate ode retrs ths RERR message alog the reverse path to the sorce ode. Whe a sorce ode receves a RERR message, t erases the path from ts table ad looks for a alterate path towards the destato ode, f oe s avalable; otherwse t tates a path dscovery process to resme the data trasmsso. A alteratve path s selected as descrbed Secto Performace Evalato of AOMR-LM I ths secto, we preset smlato reslts to demostrate the effcecy of or proposed protocol. Frst we preset the metrcs sed for performace evalato ad the we aalyze the vales of coeffcet α to select the most approprate vale for the rest of the smlatos; ths vale allows s to classfy odes accordg to ther eergy levels. Fally we evalate or protocol by comparg t wth two protocols the lteratre, amely AOMDV ad ZD-AOMDV. Ths evalato s accompaed wth a aalyss ad dscsso of reslts. 5. Performace parameters We evalate three key performace metrcs. The etwork lfetme ca be defed three ways [38]: the tme take to exhast the battery of the frst etwork ode, the tme take to exhast the battery for N etwork odes, ad the tme whe the battery of the last etwork ode s exhasted. We choose the secod way. Eergy cosmpto s the average of the eergy cosmed by odes partcpatg packets trasfer from the sorce ode to the destato ode drg the whole smlato. Ed-to-ed delay s the average trasmsso delay of data packets that are delvered sccessflly over the total drato of the smlato. 5.2 Aalyss of the coeffcet α I or protocol the choce of the path sed by a sorce ode to trasmt data packets toward a destato ode s based o the path class, whch depeds o the classes of odes belogg to ths path. I or protocol, the class defto of a ode s maly affected by the vale of α. I order to fd approprate vales of α, we apply Eqatos (8) ad (0) to fd the mmm vales to defe maly two classes (low ad average). The maxmm vale of α s determed by the eqato (9), we have ot stded ths case, becase or work the thrd class (hgh) s determed by the vale of β. We assme that f the mber of odes partcpatg the trasfer of data s ear to, the the vale of T Net approaches zero. Table shows α m vales measred by varyg from 0 to 00 odes. Table : Measrg α m relatve to the mber α m α m

8 I Fgre 2 we ca see that α m creases wth, ad teds toward zero as becomes small, therefore α m s a fcto of whe T Net teds to zero. For a etwork sze of 90 odes, the mber of odes sed for trasferrg data packets from a sorce ode to the destato ode s betwee 30 ad 40 (correspodg to of Table )[39]. Fgre 2: α m verss. For the smlato of or protocol AOMR-LM, we chose 90 as the etwork mber odes; hece or choce of α m = Performace evalato We carred ot smlatos to determe the effectveess of or protocol. The prcpal goal of these smlatos s to aalyze or protocol by comparg t wth other protocols, maly AOMDV [] ad ZD-AOMDV [34]. The vales of smlato parameters are smmarzed Table 2. The Radom Waypot model s sed to smlate ode movemet; each ode moves wth a speed radomly chose from 0 to 5 m/s. The rado model ses characterstcs smlar to a commercal rado terface, Lcet s Wave LAN. Wave LAN [4] s a shared-meda rado wth a omal bt-rate of 2 Mbt/s ad a omal rado rage of 250 m, whch s compatble wth the IEEE 802. stadard. Each smlato s carred ot der a dfferet mber of etwork odes ad the performace metrcs are obtaed by averagg over 20 smlato rs from oe sorce to oe destato radomly selected. We assme that a ode cosmes 28.8 mw whle recevg ad 28.8 mw whle trasmttg [42]. It was show [43] that o real ode eergy optmzato ca be acheved the presece of overheatg or dle state. For ths reaso, the eergy cosmpto drg dle or overheatg tme s ot cosdered ths model. I or smlatos, we talzed the eerges of the odes radomly betwee 0 ad 60 Joles (form dstrbto), whch correspods to the average capacty of a battery. We evalate the performace of AOMR-LM by comparg t wth the AOMDV ad ZD-AOMDV rotg protocols. The etwork lfetme metrc s show Fgre 3 wth the mber of etwork odes eqal to 90. The etwork lfetme of AOMR-LM s loger tha those of AOMDV ad ZD- AOMDV. Or protocol exhasts fewer odes compared to ZD-AOMDV ad AOMDV protocols, whch creases the lfetme of the etwork. Table 2: Smlato parameters. Commcato Model Costat Bt Rate (CBR) MAC type IEEE 802. Moblty model Radom Waypot Terra rage Trasmsso rage 840 m 840 m 250 m Nmber of moble odes 30 to 90 Data payload 52 bytes RREQ_Wat_Tme.0 s RREP_Wat_Tme.0 s To evalate AOMR-LM, we se the etwork smlator s-2 [40]. Each smlato r has a drato of 300 secods. Drg each smlato, costat bt rate (CBR) coectos are geerated; each of them prodces for packets per secod wth a packet sze of 52 bytes. The vales of RREQ_Wat_Tme ad RREP_Wat_Tme are set to.0 secods, the same vale as that sed for the protocol AOMDV []. We vary the mber of etwork odes from 30 to 90 to obta dfferet scearos a 840 m 840 m evromet. Fgre 3: Network lfetme verss mber of exhasted odes. Ths AOMR-LM balaces the eergy amog all the odes ad prologs the dvdal ode lfetme ad hece the etre etwork lfetme. Fgre 4 shows the eergy cosmed dfferet scearos by the AOMR-LM, ZD-AOMDV, ad AOMDV protocols. AOMR-LM does ot perform too well at the begg of the smlato, bt t mproves later. Itally, t s ot better tha ay other protocol, becase tally the maorty of data packets are ot yet trasmtted, so the total eergy of sedg ad recevg packets s ot mportat. Bt the last stage, as tme creases, there s some mbalace of eergy that comes to play ad the the mpact of or algorthm comes to play. We ca see that the eergy

9 cosmed AOMR-LM s less tha those cosmed by ZD- AOMDV or AOMDV. Fgre.5: Ed-to-ed delay verss mber of odes. Fgre 4: Eergy cosmed verss tme. AOMR-LM cosmes less eergy tha ZD-AOMDV or AOMDV, frstly becase AOMR-LM s able to balace the eergy betwee paths. Ths eergy s balaced ot across the etwork, redcg eve eergy cosmpto. Secodly, AOMR-LM s able to avod odes wth low eergy the costrcto of the mltpath. Ths meas that paths wth hgher eergy are detfed ad selected for trasmsso. Fgre 5 shows the average ed-to-ed delay of the compared protocols. The average ed-to-ed delay for all tested protocols creases whe creasg the etwork sze, bt the average ed-to-ed delay of AOMR-LM s lower tha those of ZD-AOMDV ad AOMDV, for dfferet odes speeds, see Fgres (a) ad (b). Whe the mber of odes s less tha 70 Fgre 5 (a) ad (b), the protocol ZD-AOMDV protocol has a delay eqal or slghtly better tha or protocol, de to the smltaeos se of paths. Oce the sze of the etwork creases, or protocol prodces a better delay, ths for ay odes speed. The reaso s that or AOMR-LM protocol favors odes havg a hgh eergy level ad prevets the crtcal odes from partcpatg the data packet trasmsso. Ths prodces fewer broke lks ad greatly redces the ed-to-ed delay. 6 Coclso I ths artcle we have provded a solto to the problems of rotg a ad hoc etwork. Moble ad hoc etworks are characterzed by ther lack of frastrctre ad ther dyamcty: lk falres ad rote breaks occr freqetly. Moreover, the freqet chages of topology exhast the batteres of the odes, whch decreases the etwork performace. A ew mltpath rotg protocol, AOMR-LM, has bee proposed ths paper, performg eergy aware rotg moble ad hoc etworks. We has show that AOMR-LM coserves the resdal eergy of odes ad balaces the cosmed eergy over mltple paths. AOMR- LM rotg protocol s a exteso of the exstg mltpath rotg protocol AOMDV. It ses a eergy-aware mechasm, whch explots the resdal eergy of odes to select ad classfy the paths accordg to the eergy level of ther odes. Ths cocept exteds the etwork lfetme ad mproves eergy cosmpto whe compared wth other soltos kow the lteratre. The coeffcet α s aalyzed order to fd approprate vales, whch are reqred to defe the class of a ode drg the reply-forwardg process ad to preserve the ode resdal eergy. Comparg the performace of AOMR-LM wth those of the AOMDV ad ZD-AOMDV protocols, AOMR-LM s able to balace the eergy cosmed. It creases the lfetme, cosmes less eergy ad has a lower average ed-to-ed delay tha the other smlated protocols, becase paths are compted depedg o the eergy level of ther odes, ad the oe of the best paths s selected. I the ftre we pla to stdy the cooperato of or rotg protocol wth a MAC layer power-cotrol techqe to see how they ca cooperate to decrease the eergy cosmpto of ad hoc etworks.

10 Refereces. D Mrthy, J J Garca-La-Aceves, A rotg protocol for packet rado etworks. I Proc. of the st ACM Iteratoal Coferece o Moble Comptg ad Networkg (Mobcom 95), pp , T H Dye, W Beapolakl, D P Mh, Performace evalato ad comparso of dfferet ad hoc rotg protocols. Elsever Compter Commcatos. Vol. 30, No. -2, pp , Sep G Jayakmar, G Gopath, Performace comparso of two o demad rotg protocols for ad-hoc etworks based o radom way pot moblty model. Amerca Joral of Appled Sceces, Vol. 5, No. 6, pp , Je H Rvao, F Theoleyre, F Valos, A Framework for the Capacty Evalato of Mlthops Wreless Networks. Ad Hoc & Sesor Wreless Networks (AHSWN), Vol. 9, No. 3, pp , Jaary A Radwa, T M Mahmod, E H Hosse, Performace measremet of some moble ad hoc etwork rotg protocols. Iteratoal Joral Compter Scece Isses. Vol. 8, No., pp. 07-2, Jaary E M Beldg-Royer, Rotg approaches moble ad hoc etworks. Moble Ad Hoc Networkg, S. Basag, M. Cot, S. Gordao, I. Stomeovc, Eds., IEEE/Wley, pp , C E Perks, P Bhagwat, Hghly dyamc destato seqeced dstace vector rotg (DSDV) for Moble Compters. I Proc. of ACM SIGCOMM 94, pp , Sept D G S Meller, Mltpath rotg moble ad hoc etworks: sses ad challeges. Performace Tools ad Applcato to Networked Systems, Lectre Notes Compter Scece, M. C. Calzarossa, E. Gelebe, Eds., Vol. 2965, pp , Apr C E Perks, E M Royer, S Das, Ad hoc o-demad dstace vector (AODV) rotg. RFC 356, Jly T X Brow, S Dosh, Q Zhag, Optmal power aware rotg a wreless ad hoc etwork. I Proc. of IEEE LANMAN, pp , M Mara, S R Das, O-demad mltpath dstace vector rotg ad hoc etworks. I Proc. of the Iteratoal Coferece for Network Protocols, S Che, Nahrstedt, Dstrbted qalty-of servce rotg ad hoc etworks. IEEE Joral o Selected Areas Commcatos, Vol. 7, No. 8, pp , M Tarqe, E Tepe, S Adb, S Erfa, Srvey of mltpath rotg protocols for moble ad hoc etworks. Joral of Network ad Compter Applcatos, Elsever, Vol. 32, No. 6, pp , Nov P Peryasamy, E arthkeya, Srvey of Crret Mltpath Rotg Protocols for Moble AD Hoc Networks. I. J. Compter Network ad Iformato Secrty, vol. 2, pp , Y Zo, Z Lg, Y Ya, A hybrd mlt-path rotg algorthm for dstral wreless mesh etworks. EURASIP Joral o Wreless Commcatos ad Networkg, Vol. 203, No. 82, M Foto et al., Evalatg eergy-aware behavor of proactve ad reactve rotg protocols for moble ad hoc etworks. I Proc. of SPECTS 07, Jly C Che, W W, Z L, Mltpath rotg modellg ad hoc etwork. I Proc. of IEEE Iteratoal Coferece o Commcatos, May V Rodopl, T H Meg, Mmm eergy moble wreless etworks. IEEE Joral o Selected Areas Commcatos, Vol. 7, No. 8, pp , Agst X Wag, L L, C Ra, A eergy-aware probablty rotg maets. I IEEE Workshop o IPOperatos ad Maagemet, pp. 46 5, October Q Zhao, L Tog, Eergy effcecy of large-scale wreless etworks: proactve verss reactve etworkg. IEEE Joral o Selected Areas Commcatos, Vol. 23, No. 5, pp. 00 3, May L Bao, J J Garca-La-Aceves, Stable eergy-aware topology maagemet ad hoc etworks. I Proc. of Ad Hoc Networks, Vol. 8, No. 3, pp , O Smal, Z Mekkaka, B Messabh, R Mekk, B Cos, Eergy Coservato for Ad Hoc O-Demad Dstace Vector Mltpath Rotg Protocol. Iteratoal Joral of Compter Network ad Iformato Secrty (IJCNIS), Vol. 6, No. 6, pp. -8, F De Rago, F Gerrero, P Fazo, Lk-stablty ad eergy aware rotg protocol dstrbted wreless etworks. IEEE Trasactos o Parallel ad Dstrbted Systems, Vol. 23, No. 4, pp , Apr S Sgh, M Woo, C S Raghavedra, Power-aware rotg moble ad hoc etworks. I Proc. of MobCom 98 Coferece, October Scott, N Bambos, Rotg ad chael assgmet for low power trasmsso PCS. I Proc. of Iteratoal Coferece o Uversal Persoal Commcatos, pp , C-. Toh, Maxmm battery lfe rotg to spport bqtos moble comptg wreless ad hoc etworks. IEEE Commcatos Magaze, pp , Je M L, Z X et al, Collso-costraed mmm eergy ode dsot mltpath rotg ad hoc etwork. I Proc. of Wreless Commcatos, Networkg ad Moble Comptg, pp. 5, P Ya, Y Ba, H Wag, A mltpath eergy-effcet rotg protocol for ad hoc etworks. I Proc. of Iteratoal Coferece o Commcatos, Crcts ad Systems, Vol. 3, pp , S J Lee, M Gerla, Splt mltpath rotg wth maxmally dsot paths ad hoc etworks. I Proc. of IEEE Iteratoal Coferece o Commcatos (ICC), pp , Y L, L Go, H Ma, T Jag, Eergy effcet o demad mltpath rotg protocol for mlthop ad hoc etworks. I Proc. of IEEE 0th Iteratoal Symposm o Spread Spectrm ad Applcatos, pp , Ag J L, J Che, Y o, Mltpath rotg protocol for etworks lfetme maxmzato ad-hoc etworks. I Proc. of the 5th Iteratoal Coferece o Wreless Commcatos, Networkg ad Moble Comptg (WCom '09), M Aglar Igarta, V Carrascal Frías, Self-cofgred mltpath rotg sg path lfetme for vdeo-streamg servces over ad hoc etworks. Compter Commcatos, Vol. 33, pp , S V Gole, S V Mallapr, Mltpath eergy effcet rotg protocol. Iteratoal Joral of Research ad Revews Compter Scece (IJRRCS), Vol. 2, No. 4, pp , Ag H. Naseh, N. T. Java, A. B. Aghababa, Y. G. Brga, Improvg eergy effcecy maets by mlt-path rotg. Iteratoal Joral of Wreless & Moble Networks (IJWMN), Vol. 5, No., Feb G Parssds, V Leders, M May, B Platter, Mltpath rotg protocols wreless moble ad hoc etworks: a qattatve comparso. I Proc. of 6th NEW2AN, pp , S Mal, E aa, A Valarmath, C Dael Sdar Raa, Traffc based eergy cosmpto aalyss of AOMDV protocol a moble ad hoc etwork. Iteratoal Joral of Compter Applcatos, Vol. 4, No. 6, pp , Feb P Peryasamy, E arthkeya, Performace evalato of AOMDV protocol based o varos scearo ad traffc patters. Iteratoal Joral of Compter Scece, Egeerg ad Applcatos (IJCSEA), Vol., No. 6, pp , M Bheemalgaah, M M Nad, D Sreevasa Rao, G Varaprasad, Eergy aware ode dsot mltpath rotg moble ad hoc etwork. Joral of Theoretcal ad Appled Iformato Techology, Vol. 2, No.2, pp , N Meghaatha, A ode dsot mlt-path rotg protocol based o locato predcto for moble ad hoc etworks. Iteratoal Joral of Egeerg, Scece ad Techology, Vol. 2, No. 5, pp , The Network Smlator s-2, avalable ole at 4. LM Feeey, M Nlsso, Ivestgatg the eergy cosmpto of a wreless etwork terface a ad hoc etworkg evromet. IEEE INFOCOM, pp , Apr E S Jg, N H Vadya, A power cotrol MAC protocol for ad hoc etworks. I Proc. of the ACM Iteratoal Coferece o Moble Comptg ad Networkg (MOBICOM), pp , Sept D m, J J Garca-La-Aceves, Obraczka, J-C Cao, P Mazo, Rotg mechasms for moble ad hoc etworks based o the eergy dra rate. IEEE Trasactos o Moble Comptg, Vol. 2, No. 2, pp. 6 73, 2003.