Data Dissemination Based on Ant Swarms for Wireless Sensor Networks

Transcription

1 Data Dssemato Based o At Swarms for Wreless Sesor Networks S. Selvakeedy Scool of IT Madse Bldg F9 Uversty of Sydey, NSW 26, Australa. skeedy@t.usyd.edu.au S. Sappa Graduate Scool of Busess & Professoal Developmet, Uversty of Wollogog, NSW 2522 Australa suku@uow.edu.au Y Sag Dept. of Computer Scece, Uversty of Mssour-Columba, Columba, MO 652, USA. sagy@mssour.edu Abstract Tere are may dffcult calleges aead te desg of a eergy-effcet commucato stack for wreless sesor etworks. Due to te severe odes' costrats, protocols ave to be smple yet scalable. To ts ed, collectve socal sects beavor could be adopted to gude te desg of tese protocols. We explot te smple beavors of at swarms foragg ad brood sortg to desg a erarccal ad scalable data dssemato strategy. I order to allow a realstc evaluato, a compreesve smulator volvg crtcal compoets of te commucato stack s corporated. Te proposed sceme promotes a uform dstrbuto of clustereads, wc subsequetly eables substatal eergy savgs over oter clusterg algortms. Keywords-clusterg; data dssemato; eergy effcet; swarm tellgece. I. INTRODUCTION Wreless sesor tecology s garerg a lot of terests due to ts promses, eablg t to evolve rater rapdly. For stace, terms of te sesor ode ardware, te Mca2 mote as rougly egt tmes te memory ad commucato badwdt as ts predecessor, te Ree mote, developed 999 for te same power budget []. Tese sesor odes ave foud use may applcatos suc as eartquake motorg, target trackg ad survellace, ad structural motorg. Te odes are typcally less moble due to ter uque applcato eeds, substatally more resource costraed ad more desely deployed ta moble ad oc etworks (MANETs). Eve toug, tere ave bee sgfcat advaces recet years, more eergy-effcet solutos are requred wt te commucato stack for te coservato of te battery power. A approac tat s lkely to succeed s te use of a erarccal structure [2], wc also promotes scalablty of wreless sesor etworks (WSNs). Clusterg wt data aggregato s a mportat tecque ts drecto, ad t makes te tradeoff betwee eergy effcecy ad data resoluto. Most clusterg algortms am at geeratg te mmum umber of clusters ad trasmsso dstace. Tese algortms also dstgus temselves by ow te clustereads (CHs) are elected. Te LEACH algortm [3] ad ts related exteso [4] use probablstc self-electo, were eac sesor ode as a probablty p of becomg a CH eac roud. I [4], te autors proposed te tme-cotrolled clusterg algortm () tat allows te formato of multop clusters dyamcally. By cosderg bot tra- ad ter-cluster traffc, t was demostrated tat for s optmal operato, small multop clusters of two or tree ops seem to be te most approprate sze. Aoter crucal desg ssue to cosder s te etwork relablty. To ts ed, socal sect swarm beavor may provde a deal model for te desg of suc less cotrollable systems. To our kowledge, very few researcers ave cosdered or adopted suc ature-spred approaces for WSN desg. However, a umber of recet works as bee based o dfferet swarm beavors te desg of routg protocols for MANETs. As tere are may mportat smlartes betwee tese two ad oc tecologes, we beleve buldg o tese kowledge may be useful for WSNs. Most of tese swarm-based routg algortms are smple yet robust as well as adaptve to topologcal cages. However, suc algortms caot establs te best pats before suffcet umber of agets s flooded [5]. Ts mples tat as te umber of odes te etwork creases, te umber of agets requred to establs te routg frastructure may explode. A way to overcome te overead exploso ad atta scalablty s by usg erarccal routg approac. I a poeerg work [6], te autors ad used swarm tellgece to mplemet a dstrbuted etwork of moble sesors ad cotrolled te odes pyscal movemets. It was demostrated tat te swarm beavor could be used to esure safe separato betwee te agets to esure coverage effcecy wle eforcg a level of coeso tat matas a level of coectvty betwee te moble agets. I ts paper, we propose a ew clustered data dssemato strategy based o te at swarm beavor. Ts sceme s realzed based o our tal clusterg algortm proposal, [4]. Ulke, a at swarm dyamcally cotrols te CH electo process stead. Ts ew algortm corporatg te clusterg wt ts ANT electo sceme s termed as te algortm. aceves better performace ta tat of a flat dssemato strategy, LEACH ad pla. Te algortm aceves te

2 objectves by explotg two swarm beavors, amely foragg ad brood sortg. II. RELATED WORK Te approac tat s lkely to succeed to provde a scalable ad eergy-effcet soluto s of a erarccal structure. Itally, te clusterg algortms focused o te coectvty problem [7] but later eergy-effcecy was more of terest [3, 4, 8]. However, almost all focuses o reducg te umber of clusters formed, wc may ot ecessarly etal mmum eergy dsspato. Geerally, clusterg algortms segmet a etwork to o-overlappg clusters comprsg a CH eac. No-CHs trasmt sesed data to CHs, were te sesed data could be aggregated ad trasmtted to te sk. LEACH [3] requres posto kowledge to perform a precse trascever power cotrol, ad self-elects CHs usg a omated probablty p. Te algortm esures tat every ode wll be omated as a CH oly oce /p rouds for a certa fxed durato. However, t assumes tat all odes are eac oter s rado rage. I [9], a fxed clusterg algortm tat performs eergy-balacg to mprove etwork lfetme was proposed. It also takes to cosderato te teracto betwee clusterg ad routg. Aoter crucal desg aspect of WSNs to cosder s te etwork relablty ad fault-tolerace. It as bee demostrated dfferet cotext tat te collectve beavor of socal sects as may attractve features, ot te least robustess ad relablty. However, oly very lmted WSN proposals ave bee spred by ature or bologcally. I socal sects, sopstcated commuty beavor emerges from te teracto of dvduals were eac sect carres out smple tasks. Some kow collectve beavors are foragg, est costructo, termoregulato, brood sortg ad cemetery formato []. We ats de, tey are removed from te est by workers ad deposted ples outsde te est. Te cause to ts acto appears to be a attracto betwee dead bodes, but remas to be cofrmed. Tese collectve beavors of socal sects ave spred computer scetsts to replcate tem as tey exbts may attractve features, suc as robustess ad relablty troug redudacy [5, ]. Te frst MANET routg algortm based o at coloy prcples s ARA []. It exploted te at peromoe layg beavor. Peromoe s a qualty metrc dcatg te goodess of a pat. Altoug peromoe evaporates over tme, subsequet ats leave addtoal peromoe ad tus reforce te pat. Ats gradually establs te sortest pat betwee food ad ter est a fully dstrbuted ad autoomous maer. Te fact of te gradual decay of peromoe troduces a form of a egatve feedback to prevet old routes from remag te forwardg tables we routes fall out of favor wt ats. Routg scemes based o suc coloy beavor s bot robust ad adaptable. We te sortest route s lost due to some evet, te loger routes provde alteratve optos. Tere are tree basc cotrollg beavors tat gover movemets of agets wt te swarm. Kadrovac ad Lamot [6] ave summarzed tese beavors as sow: Separato: Avod collsos wt earby agets. Algmet: Attempt to matc velocty wt earby agets. Coeso: Attempt to stay close to earby agets. Swarm beavor s solely based o locally observable peomea, ad s reflected above by te adjectve earby. Te tegrato of tese beavors results a stable swarm formato, were every aget s at least some mmum dstace from oters. We captalze o te frst two beavors troug peromoe cotrol to aceve a ear uform dstrbuto of te CHs odes. Moreover, te algortm coverges faster to a optmal or ear optmal soluto we peromoe s also reduced drastcally from tose elemets tat make up te worst soluto eac terato. Tus, subsequet geeratos of ats are dscouraged from returg to poorer solutos see te past. Ts costtutes te smplest way to mplemet at-peromoe were te removal of peromoe s smulated by a reducto exstg peromoe levels. III. THE CLUSTERING ALGORITHM adopts two-pase clusterg process volvg te cluster setup ad steady state pases. To gude te CH electo, we cose to use a swarm of ats. Troug te use of a swarm of ats, we would guaratee tat te etwork always matas a optmal umber of clusters. Durg te ode talzato, te sk releases a umber of ats (.e. cotrol messages). Ramos ad Merelo [2] suggest tat te rato of te umber of ats to te umber of objects (.e. sesor odes) sould equal.. We te sk releases a at, t cooses oe of ts egbors at radom. Te at could travel to te etwork as deep as restrcted by ts tme-to-lve (TTL) feld. We a at arrves at a ode, te ext ode s radomly cose (excludg te seder) for ts subsequet stop f TTL as ot expred. If TTL expres, te at remas at ts ode. However, f te fal at locato overlaps wt aoter at, te former at must fd aoter locato. Te cluster setup (CS) pase s cotrolled troug a CS tmer. We ts tmer expres, a ode cecks to see weter t possesses a at. If te ode as a at, t becomes a CH. We a ode becomes a CH, t advertses to ts egbors by broadcastg a ADV message wt ts ode d ad a TTL feld to costra te ADV propagato. Upo recevg a ADV message, a regular ode records te CH d, te seder s d as ts paret, te op dstace to ts CH, te umber of ADV messages receved so far ad total op dstace to all see CHs, ad te rebroadcasts t f TTL permts. A ode decdes to jo a cluster we ts jo-tmer expres. It te computes ts peromoe level based o ts total op dstace () to CHs, te umber of CHs () ts egborood, ad ts ormalzed resdual eergy. Te peromoe expresso s based o te forwardg probablty formula used te at routg algortm [5], but expaded as: p + p p = () + p

3 were p s gve by: k p = 2 E E s te ode s op dstace to te selected CH, E res s te resdual eergy, E max s te referece maxmum battery eergy ad k s te learg rate of te algortm ( =.). Ts expresso esures tat p s ger we te ode s oly reacable by fewer CH odes (smaller ), far from CHs ( ), as ger resdual eergy (E res ) or s earer to ts selected CH ( ). A regular ode cooses te best cluster to jo based o ts op dstace to te CH, wc would esure mmal eergy dsspato durg te data dssemato rouds. Te ode jos a cluster by sedg a JOIN message wt ts d, te selected CH d ad ts peromoe level. If te CH s rage, te message s trasmtted drectly; oterwse forwarded troug ts paret to te CH. We a CH receves JOIN messages, t fds te member wt te gest peromoe level to attract ts at for te ext roud. Before te ext CS tmer expres, te ats wader to te odes wt te gest peromoe level amog ter egbors, ad tese odes wll be te future CH. Before a at leaves ts curret ode, a amout of at-peromoe s lad to mmc a rapd decay of peromoe level. Te peromoe removal s computed wt te at-peromoe rate (β). Te gve peromoe expresso gudes te evoluto of te swarm to aceve te separato beavor betwee ats te swarm as dscussed earler. It s foud emprcally tat separato s attaed rater quckly wt 3-5 rouds as a optmal swarm sze s used. Aoter useful swarm beavor s algmet. I our cotext, te area served by eac at represets te algmet beavor. It s reflected by te umber of members a cluster. We te swarm evolves to aceve separato, algmet s also aceved as a sde-beeft. Te peomea due to bot beavors are captured by te followg ftess fuctos, respectvely. Te CH electo ftess fucto S to capture te separato beavor s: S res max = c j j= were c s te umber of CH odes, s te umber of ADVs see by CH ad j s CH s op dstace to CH j. Te clusterg ftess fucto A to represet te algmet beavor s as follows: (2) (3) were r s te umber of regular odes ad s ode s op dstace to ts CH. I te steady state pase, eac regular ode seds or forwards ts sesory data to ts CH. It s possble tat te foragg ats may de due to te evrometal ucertaty or ode falure. To avod a reducg umber of ats te etwork over tme, ats ave a fte lfetme. We ats de, te sk re-releases te same optmal umber of ats to restart te process. IV. RESULTS AND DISCUSSIONS We vestgate performace agast LEACH, pla ad a flat strategy. Sce LEACH ca t be appled drectly to a multop etwork, we modfed ts algortm to use a routg protocol to forward messages weever te destato s ot wt te rado rage. We termed ts modfed algortm as multop-leach (or m-leach). For tese smulato expermets, we assumed tat tere are sesor odes dstrbuted radomly a square M M rego wt M = 5 m. Te trascever eergy parameters are set as: E elec = 5 J/bt ad ε fs = pj/bt/m 2. Te eergy for data aggregato s set to E DA = 5 J/bt per sgal [3]. Te cotrol ad data message szes are fxed at 3 bytes, ad sesory data s geerated at 2-secod terval. Eac CH ode retas ts CH status for 2 secods. Te umber of ats s fxed at ad te at-peromoe rate s.. Te performace metrcs beg vestgated are: Clusterg ftess: It represets te goodess of te cluster formato volvg all regular odes; CH electo ftess: It represets te goodess of all te elected CH odes; Average eergy per roud: It represets te average eergy usage by te odes per roud; ad Network lfetme: It represets te perod from te stat te etwork s deployed to te momet we te frst sesor ode rus out of eergy. Fgure depcts te clusterg ftess value at dfferet smulato tme. For, te tal value s g dcatg tat te swarm as ot yet aceved te algmet beavor as te ats are radomly released to te etwork. However, as peromoe s lad ad at-peromoe takes effect durg CS pases, te swarm algmet mproves. Wt te trd evoluto, te swarm s able to alg. As for te oter scemes, te ftess value vares rater wldly. Ulke, mostly operates sub-optmal faso. Also for m-leach, te ftess value s always smaller ta te oter scemes due to te ADV messages beg lmted to frst-op odes. Ay ucovered odes would ave to resort to drect trasmsso to te sk. Sce m-leach ad ave probablstc CH electo, t s possble tat te CHs may eve be clumped. We te CHs are clumped, te dsparty amog clusters s large terms of ter umber of members, as eac CH coteds for te same regular odes pool. A = r (4)

4 25 2 m-lea CH members eac cluster. After reacg good separato, t s stll possble for ats to be closer aga at later rouds order to promote load balacg. Clusterg Ftess Fgure. Clusterg ftess at dfferet smulato tme for, m- LEACH ad. I Fg. 2, te CH electo ftess s depcted for te same tree algortms. Aga, cosstet beavor as above s obtaed. For, t as a ger fucto value tally, but t quckly coverged somewat. Te ats move to a better ode based o te calculated peromoe level, ad wt fve rouds, te swarm s able to aceve te separato beavor. Ts beavor esures te elected CHs are dstrbuted as uformly as possble. Eve after te uformty s aceved, te ats keep movg at eac roud to esure tat te CH role s sared amog odes, ad eergy-load balacg s attaed. As for te oter scemes, te topology barely settles ad mostly as a lower value ta. A lower value dcates tat te CHs tese scemes are mostly too close to eac oter. I m-leach, te ftess fucto qute ofte assumes a zero value compared to. Ts s maly due to ts restrcted ADV propagato, were a CH s uable to recogze aoter CH located oly two ops away. (a) CH Electo Ftess m-leach Fgure 2. CH electo ftess at dfferet smulato tme for, m-leach ad. Te followg results demostrates ow promotes uform CH dstrbuto te etwork compared to (or m-leach). Fg. 3 depcts te clustered topology formato at dfferet CH rouds for cluster sze of two. We cose to use ats. I te followg fgures, uflled crcles represet CHs, flled crcles represet regular odes, ad te le segmets lk members to ter CH or troug ter paret to CH. Ulked regular odes ave to perform drect trasmsso to te sk. It s vsually evdet tat we te sk radomly releases te ats, tey could stll be forced to egborg odes, wc results some regular odes beg ucovered, as sow Fg. 3(a). However, as te swarm evolves, by roud tree te ats are able to aceve separato. At later evolutos, te ats are more uformly dstrbuted resultg smlar umber Fgure 3. Logcal topology of at roud umber: (a) ad (b) 3. Sce cluster sze was sow to ave a sgfcat mpact o clusterg algortms [4], we vared ADV s TTL value ad compared ad. I Fg. 4, bot algortms exbt te presece of a optmal cluster sze, ad ts optmal sze (.e. two) s te same for bot. However, aceves sgfcatly more eergy savgs ta for cluster szes up to four. We cluster sze s two, dsspates 27% lesser eergy compared to. Ts observato s cosstet wt ftess values reported for bot approaces Fgs. ad 2. Sce maly operates wt sub-optmally formed topology, ts eergy dsspato s worse off. However, at larger cluster szes, bot scemes exbt smlar eergy usage. Ts s maly caused by te eergy expeded durg te cluster setup pase tat s sgfcatly larger as ADV messages are flooded furter, ad te JOIN messages ave to (b)

5 P o ly.(t -A N T ) P o ly.(t C C A ) be forwarded may ops before reacg ter CHs. Smlarly, durg te steady state pase, sgfcat tra-traffc s geerated. Note tat te performace of m-leach s te same as wt cluster sze oe, ad as sgfcatly ger eergy dsspato experece. Average Eergy per Roud (J) Cluster Sze (ops) Fgure 4. Average eergy usage per roud agast te cluster sze of ad (ad m-leach). I Fg. 5, te mprovemet gaed troug s furter exemplfed by te etwork lfetme grap. For ts vestgato, we ave fxed E max at.j as a referece level. It s evdet tat exbts te logest lfetme wt all odes remag fully fuctoal. It s foud tat aceves almost 3.5 tmes te lfetme of m-leach ad almost 5 tmes of te flat approac. It also supports 5% loger lfetme ta. Number of Nodes Alve m-lea CH Flat Fgure 5. Network lfetme agast smulato tme of,, m-leach ad te flat strategy. V. CONCLUSIONS To our kowledge, te clusterg algortm s te frst ature-spred approac for data dssemato a quasstatoary wreless sesor etwork. Te algortm uses a swarm of ats to cotrol te clusteread electo a totally dstrbuted maer. It s evdet tat s able to aceve two desrable swarm beavors, amely separato ad algmet. Due to tese, a uform dstrbuto of clusteread s almost guarateed eablg te etwork to operate a optmal maer trougout ts lfetme. Eve toug t could also be aceved a cetralzed approac as LEACH-C [3], our algortm s dstrbuted, robust, does ot requre posto kowledge ad promses scalablty. also stores less ta % of state overead memory compared to LEACH or. ACKNOWLEDGMENT Ts work was supported by USyd grat L2844 U323. REFERENCES [] J. Hll, M. Horto, R. Klg, ad L. Krsamurty, "Te platforms eablg wreless sesor etworks," Comms. of te ACM, vol. 47, pp. 4-46, Ju 24. [2] K. Akkaya ad M. Yous, "A Survey o Routg Protocols for Wreless Sesor Networks," Elsever Ad Hoc Networks, vol. 3, pp , May 25. [3] W. B. Hezelma, A. P. Cadrakasa, ad H. Balakrsa, "A applcato-specfc protocol arctecture for wreless mcrosesor etworks," IEEE Trasactos o Wreless Commucatos, vol., pp , Oct 22. [4] S. Selvakeedy ad S. Sappa, "A Cofgurable Tme- Cotrolled Clusterg Algortm for Wreless Sesor Networks," Proc. of t It. Cof. O Parallel ad Dst. Sys., Fukuoka, Japa, Jul 25, pp [5] Y. Otak, N. Wakamya, M. Murata, ad M. Imase, "Scalable ANT-based Routg Algortm for Ad-Hoc Networks," Proc. of 3rd IASTED It. Cof. o Comms., Iteret, ad Iformato Tecology, St. Tomas, US Vrg Islads, Nov 24. [6] B. A. Kadrovac ad G. B. Lamot, "A partcle swarm model for swarm-based etworked sesor systems," Proc. of ACM symposum o Appled computg, Madrd, Spa, Mar 22, pp [7] D. Baker ad A. Epremdes, "Te Arctectural Orgazato of a Moble Rado Network va a Dstrbuted Algortm," IEEE Tras. o Commucatos, vol. 29, pp , Nov 98. [8] C.-F. Casser, I. Clamtac, P. Mot, ad A. Nucc, "A eergy-effcet metod for odes assgmet cluster-based ad oc etworks," Wreless Networks, vol., pp , May 24. [9] T. Su, M. Kruz, ad S. Vrudula, "Power balaced coveragetme optmzato for clustered wreless sesor etworks," Proc. of 6t ACM teratoal symposum o Moble ad oc etworkg ad computg, Urbaa-Campag, IL, USA, May 25, pp. -2. [] V. Hartma, "Evolvg aget swarms for clusterg ad sortg," Proc. of coferece o Geetc ad evolutoary computato, coferece o Geetc ad evolutoary computato, Ju 25, pp [] M. Gees, U. Sorges, ad I. Bouazz, "ARA - te at-coloy based routg algortm for maets," Proc. of ICPP Worksop o Ad Hoc Networks, Vacouver, BC, Caada, Aug 22, pp [2] V. Ramos ad J. J. Merelo, "Self-orgazed stgmergc documet maps: Evromet as mecasm for cotext learg.," Proc. of Spas Coferece o Evolutoary ad BoIspred Algortms, 22, pp