A Novel Optmzaton of the Dstance Source Routng (DSR) Protocol for the Moble Ad Hoc Networs (MANET) Syed S. Rzv 1, Majd A. Jafr, and Khaled Ellethy Computer Scence and Engneerng Department Unversty of Brdgeport Brdgeport, CT 06601 {srzv, majdals, ellethy}@brdgeport.edu Aasa Rasat Department of Computer Scence Insttute of Busness Management Karach, Pastan 78100 aasa.rasat@obm.edu.p Abstract- Ths paper presents a new scheme for the Dstance Source Routng (DSR) protocol whch shows the mprovement over the two major metrcs of the DSR protocol: Route Dscovery and Route Mantenance. In addton, we present a mathematcal model that ncludes probablty densty functon for these two observed metrcs. Our smulaton results demonstrate a sgnfcant mprovement n the route dscovery, transmsson tme, and the overall networ utlzaton. As an nterestng sde result, our analyss also shows that the proposed model can be used to effectvely reduce the pacet losses. Keywords- DS-CDMA, bt error rate, data throughput, multuser communcatons I. INTRODUCTION The Dynamc Source Routng (DSR) protocol s dealt under On-Demand Routng (ODR) protocol whch s just an exact opposte to the Table-Drven Routng (TDR) [2, 3]. Generally, there are two man phases use n the DSR protocol. One s the Route Dscovery (RD) phase whch dscovers all the possble paths for the pacets to be transferred from a partcular source to a destnaton. It s essental to properly mantan the RD phase snce mantanng a separate table for storng routng detals nvolves cost ssues. The second phase of the DSP protocol s the Route Mantenance (RM) phase whch fxes all the possble paths from one partcular source to a destnaton [5]. In DSR, the pacets are transmtted only one tme for each node. If the node does not receve the pacet, the prevous node s responsble to mae attempts n order to transmt the pacet. On the other hand, f the destnaton node receves the pacet successfully, an acnowledgment s transmtted bac to the source node for the receved pacet. Snce the use of the DSR protocol does not requre the mantenance of a cache table, t allows us to avod unnecessary updatng wors whch results space and tme savng advantages. In the exstng DSR scheme, the malfunctonng of one or more lns along a certan route requres the retransmsson of all pacets bac to the orgnatng source node. Ths unnecessary amount of retransmsson results a sgnfcant transmsson overhead that can severely degrade the overall networ performance by ncreasng the average tme delay. In order to mnmze the transmsson overhead and maxmze the networ throughput, we present an alternatve scheme that can be used to optmze the performance of DSR protocol. Specfcally, our proposed scheme suggests mprovement n the RD and the RM metrcs of the DSR protocol. Based on the proposed optmzaton, we derve a mathematcal model whch proves the correctness of the proposed scheme. II. PROPOSED OPTIMIZATION FOR THE DSR PROTOCOL Our man goal s to mantan the orgnal underlyng archtecture of the DSR protocol. Therefore, we consder the DSR scheme as a blac box. The DSR protocol fals to mantan route consstency n the presence of broen lns. When one of the lns goes down, the DSR protocol locates an alternate route and transmts bac the pacet to the source node where the pacet was orgnated. On contrary to the actual scheme of the DSR protocol, our proposed scheme uses a reserve drecton search method. In our proposed scheme, the pacets would be transmtted to the mmedate pror node where the actual error was occurred. The proposed scheme then fnds one or more alternatve routes from the current locaton to the destnaton. Ths mples that the whole searchng procedure of the proposed scheme wll 1 Contact author: srzv@brdgeport.edu,
Fg.1. Fndng the alternate path n DSR protocol accordng to the proposed scheme be done n the opposte drecton startng from the destnaton node. Our smulaton results demonstrate that the proposed scheme consderably ncreases the chance of fndng a vald route for salvage pacets that are typcally stored n the send buffer. For nstance, consder an example for locatng a route based on the reverse drecton search scheme as shown n Fg. 1. It can be observed that the route fnds by the RD procedure from node A (source node) to L would be: A D E I L. Durng transmsson of the pacets, t s detected at run tme that the shortest ln between node E and I goes down. Consequently, the proposed scheme mmedately starts searchng the best avalable alternate routes. In order to reach the destnaton node, the proposed scheme locates the neghborng nodes (.e., node B, D, and H from node E). Ths process of fndng the alternate route from the locaton of error results an optmal alternate route: A D E I H L. Ths mples that our proposed scheme nether send any feedbac to the destnaton node A nor t ntates the route dscovery from the source pont. Therefore, repeatng ths search n the reverse drecton from the current locaton of error to the neghborng nodes results a sgnfcant ncrease n the chance of fndng a vald optmzed route. A. Proposed Reverse Drecton Search Scheme In order to formulate the proposed scheme, we present a model that shows smple steps that need to be mplemented for fndng a vald and optmze route n the presence of ln falures. The model s presented n Fg. 2. The model s typcally dvded nto two parts. The upper part of the model represents the RD procedure where as the lower part represents the RM procedure. The RD procedure s based on an exhaustve search of an nternal cache. Durng the transmsson of a pacet, f one of the lns goes down, the proposed scheme mentons that the pacet wll be Fg. 2. Flow chart showng proposed model of DSR algorthm mmedately forwarded to the next avalable node and starts transmttng from the new locaton. Unle the DSR protocol, the proposed scheme mnmzes the transmsson overhead by avodng the unnecessary transmsson of data to the source node n the presence of a faulty ln. In other words, the proposed scheme does not provde any feedbac to the source node that leads to a sgnfcant mprovement n the networ throughput. Snce the RD can be done on the current node, we do not need to focus on the source node. Ths mples that the proposed scheme suggests the best
delvery of the pacets even n the presence of ln falure. In addton, the repetton of the pacets due to the floodng wll be cut down. In the proposed model, we manly focus on the RD and the RM. Durng the RD process, f the entres are found n the nternal cache of the next node, the proposed scheme determnes the optmal path that wll be used to forward all the pacets to the next node. At that current node locaton, the same procedure for searchng the optmal path wll be repeated over the passage of tme n order to fnd the best path towards the destnaton. An empty entry n the nternal cache represents that there s no vald route exst for a partcular destnaton. In such a scenaro, the proposed scheme wll looup nto the next neghbor s cache and determne the best avalable route for the desred destnaton. Once the optmal route s dscovered, the pacet can then be transmtted. In the RM process, whenever there s a ln falure along the path, the pacet would not go further at the pont of error and there s no need to send any feedbac to the orgnal source node. Instead, the proposed scheme determnes and performs the RM process on the best avalable alternate path. B. Mathematcal Model We derve our mathematcal model based on the proposed reverse drecton scheme. In our mathematcal model, we show that the transmsson of pacets va an alternate route s more effcent as compared to transmttng pacets from the source node usng a prmary route. Ths s especally true n the presence of error. All system varables, along wth ther defnton, are lsted n Table I. The accuracy of the proposed scheme s essentally dependent on how effcently we can dscover the alternate routes n the presence of faulty lns. In general, the accuracy s partally related to a certan nterval by whch we perform the RD procedure for a specfc type of networ traffc such as a stream of pacets. In partcular, we frst need to derve an expresson for a random varable, x, that can be used to characterze the behavor of RD process wth respect to tme. Therefore, n order to mplement the proposed scheme, one must measure the frequency of route dscoveres. In order to determne the nterval between the route dscoveres, the followng mathematcal expresson can be derved for a random varable, x: + xf ( x ) dx (1) It should be noted that equaton (1) s based on the PDF whch s used to fnd the frequency of route dscovery for a partcular par of source and destnaton. Fgure 4 represents the proposed scheme wth the prmary and the secondary paths along wth ther correspondng lns. It can be seen n Fg. 3 that the node P represents the prmary route whereas the node S represents the secondary Parameters P S X P X S X R T E o f T(t) Z P S TABLE I SYSTEM PARAMETERS AND DEFINITIONS Descrpton Ths represents the th ln n a prmary path. Ths represents the th ln n a secondary path. Lfe tme of the th prmary route. Lfe tme of the th secondary route. Mnmum lfe tme for the collecton of all values n the prmary path lns Intervals for route dscovery An event that shows any of the gven ln fals Frequency of route dscovery Maxmum lfe tme among all avalable values. Represents the faulty prmary ln due to an event E Represents the faulty secondary ln due to an event E route. If an error occurs n the prmary route, the proposed scheme wll mmedately dscover an alternate route S 1 rather than gong bac to the source node A. In other words, n the presence of faulty lns, the proposed scheme searches the nternal cache and determnes the alternatve route S 1 whch s typcally stored n the local cache. For ths partcular scenaro, the success of the proposed scheme s heavly dependent on the rate at whch one may need to execute the RD procedure. In addton, the success of the proposed scheme s not only dependent on the rate at whch the RD procedure wll be performed but also dependent on the accuracy and the effcency by whch the alternate routes wll be determned. In order to fnd the frequency of an alternatve RD, we assume that an event E mght occur at a dscrete pont n tme n the networ whch causes an error n one of the two types of routes (.e., the prmary P and the secondary S routes). Thus the transmsson of an event can be mathematcally descrbed as: E = PS + ( P( P + S ) S ) + ( P( P( P + S ) S ) S +... (2) 1 1 2 1 2 1 3 1 2 3 2 1 Fg. 3: Proposed scheme wth prmary and secondary path and ther lns
where P represents the faulty prmary-ln where as represents the faulty secondary-ln whch caused due to the occurrence of an event E at dscrete pont n tme wthn a networ. Equaton (2) represents a generc equaton that shows how the occurrence of an event n the networ may cause an error n the alternate routes. Equaton (2) can be further extended for the maxmum K number of forwardng lns wthn the avalable prmary paths. It should be noted that the occurrence of an event E s representng a cause of malfunctonng n the currently used vald route. Tang these factors nto account, one may wrte the followng mathematcal expresson: E = P S + P S S + P S S S + + 1 1 2 2 1 3 3 2 1... P S S 1 S1 where P and S n (3) represent the faulty prmary and secondary lns, respectvely. Both of these faulty lns are caused due to the occurrence of an event E at a dscrete pont n tme wthn a networ. It should be noted that we only consder the values of the most forwardng lns that one may fnd wthn the prmary path lns from the generc equaton (2). One of our observatons about the two phases of the proposed scheme s the lfe tme of the prmary path whch we use to transmt the pacets to the desred destnaton n the presence of the faulty lns. In other words, n order to effectvely mplement the proposed scheme, we must determne the mnmum value of the lfe tme for prmary path lns. Ths calculaton s essental, snce the raton of determnng the accurate vald prmary lns s crtcally dependent on the nowledge of accurate values of lfetme. The mnmum lfe tme of prmary path lns s smply chosen from one of the prmary lns that has a smallest value for the lfe tme. In other words, f one of the th prmary routes has the smallest lfe tme value, ths wll be chosen as a mnmum lfe tme value for the prmary path lns. Ths hypothess can be changed nto a smple expresson: R p1 p2 p S (3) X = Mn X, X,..., X (4) where X R n (4) represents the mnmum lfe tme value for the collecton of all values n the prmary path lns. The rght hand sde expresson of (4) represents the lfe tme of each ndvdual prmary route startng from X p1 to X p. These values are consdered as a lfe tme of the sub lns n the prmary path. Smlar to (4), we can further extend our mathematcal model for computng the nterval of tme for the RD procedure: T = Max X, X,..., X (5) s1 s2 sn where T represents the ntervals of tme for the RD and X S represents the lfe tme of the th secondary route. Equaton (5) gves an estmate of the tme to be taen by the proposed scheme for the RD procedure. Ths value s evaluated from the maxmum values of the collected tme n the sub lns of the secondary path. The rght hand sde expresson of (5) represents the lfe tme of each ndvdual secondary route startng from X s1 to X sn. For the sae of the smulaton and the performance evaluaton, we assume that the value of T wll be measured n mllsecond. Combnng (4) wth (5), we can compute the value of the alternatve route dscovery as follows: T = Mn Max( X. X ), Max( X. X. X ) p1 s1 p2 s2 s1... Max( X, X, X,... X ) p s s 1 s1 Equaton (6) gves the value of the alternatve RD. Ths can be consdered as the optmum value whch s determned from all the avalable maxmum values for both the prmary and the secondary lns. Usng (6), we can compute the values for the RD metrcs whch s one of the subparts of the proposed scheme. Z = M a x X, X, X 1,..., X 1 (7) p s s s where Z represents the maxmum lfe tme among all avalable values for both prmary and the secondary paths. Recall (1), we can now derve an expresson for the frequency of RD usng equatons (2) to (6). N N λ t λ t f T ( t ) = λ e (1 e ) = 1 = 1 (6) (8) where the rght hand sde of (8) represents the frequency of RD. Equaton (7) also has a sgnfcant mpact on the RD for the alternate path. Implementng the results of (7) on (8), we can derve a new expresson for the frequency of the RD whch tae nto account the maxmum lfe tme among all avalable values for both prmary and the secondary paths. In addton, ths mplementaton descrbes the PDF n Z wth respect to the RD metrcs.
f Z ( t) = λ ( ) ( λ ( ) ) 1 j e j t + + 1 λ j= 1 = 1, j ( ) (1 e t) (9) where λ = = and for 1/l ( ) j / l for j 1, 2... j=+1. Equaton (9) descrbes the summaton of all the possble routes whch can lead us to the desred destnaton. Equaton (9) can be further extended for the followng gven expressons: T = Mn Max( X. X ), Max( X. X. X ) p1 s1 p2 s2 s1... Max( X, X, X,... X ) p s s 1 s1 Z = Max( X, X, X... X ) p s s 1 s1 T = Mn( Z1, Z2, Z3... Z ) Based on the above three expressons, we can approxmate the PDF of T for the frequency of RD as follows: f ( t) = lm p[ t T t + dt]/ dt (10) T dt 0 Equaton (10) gves the value for the frequency of the RD n terms of a PDF functon. Relatng (8) and (9) wth (10), we can derve the followng mathematcal expresson f ( t) = f ( t) p[ z > z ] T Z j = 1 j= 1, j 1 f ( t) = f ( t) (1 Fz ( t)) T Z = 1 j= 1, j 1 (11) where, F z(t) n (11) was ntroduced from (7) to mae Z as a functon of PDF. Equaton (11) shows that we derved the expected expresson whch can be used to compute the nterval between the rout dscoveres. In other words, one could use (11) to determne the frequency of the alternate RD process. The same frequency value can be used to measure the effcency of the networ. In addton, the fnal results show that the use of the proposed reverse drecton scheme wth the derved mathematcal model can effectvely mnmze the transmsson delay especally n the presence of collsons (lns error) or faulty lns due to the malfunctonng. Fg.4. Number of nodes versus RD III. SIMULATION RESULTS We smulate our model based on the predcted data from the exstng DSR model suggested n [1, 4]. For the sae of smulaton and the performance evaluaton, we consder two major metrcs for RD and RM. These metrcs are consdered for the evaluaton of the effcency of a networ. For the sae of the frst smulaton (see Fg. 4), we characterze the behavor of the RD phase of the proposed scheme wth respect to the number of nodes present n the networ. The purpose of ths experment s to show the performance of the RD phase for dscoverng the alternate prmary and the secondary path. Durng the smulaton, we consder that as the number of nodes ncreases n the networ, the more pacets wll be accumulated n the networ that could affect the performance of the RD phase. It can be clearly evdent n Fg. 4 that the RD phase of the proposed scheme performs better for the prmary paths dscoveres than for the secondary path. When we have small Fg. 5. Pacet loss n fractons versus number of nodes
expermentally verfed that both the RD and the RM metrcs perform well wth the proposed scheme than the exstng nfrastructure of the DSR protocol. Our performance evaluaton s also well supported by the smulaton results presented n ths paper. REFERENCES Fg. 6. Tme delay versus number of nodes number of nodes, t can be seen n Fg. 4 that the performance of the RD phase for both prmary and secondary path dscoveres s overlappng. However, as networ grows n terms of the number of nodes, the performance dfferences between the prmary and the secondary path s obvous. Fg. 5 shows the pacet losses (n the fracton value) wth respect to the number of nodes durng the transmsson usng both prmary and the secondary paths. In addton, Fg. 6 represents a comparson between the tme delay (represents n seconds) and the number of nodes. It can be seen n Fg. 6 that the tme requred to dscover the prmary paths usng the RD phase s very low as compared to the tme requred to dscover the secondary paths. Based on the smulaton results of Fg. 6, we can observe that the tme delay for prmary paths s not only small but also lnear wth respect to the number of nodes. In other words, when we ncrease the number of nodes n the networ, more pacets wll be accumulated that mae a lnear ncrease n the tme delay for dscoverng the secondary paths whch s not really desrable as far as the optmum performance of the DSR protocol s concerned. [1] P. Papadmtratos and Z. Haas, Secure Routng for Moble Ad hoc Networs, In Proceedngs of the SCS Communcaton Networs and Dstrbuted Systems Modelng and Smulaton Conference (CNDS 2002), San Antono, TX, January 27-31, 2002. [2] J. Raju and J. Garca-Luna-Aceves, A comparson of on demand and table drven routng for ad-hoc wreless networs, In Proc. IEEE Internatonal Conference on Communcatons (ICC 2000), June 2000. Volume 3, Issue 2000, pp. 1702 1706, 2000. [3] J. Raju and J. Garca-Luna-Aceves, Effcent On-Demand Routng Usng Source-Tracng n Wreless Networs, In Proc IEEE Global Telecommuncatons (GLOBECOM 2000), Vol. 1, Issue 2000, pp. 577 581, November 2000. [4] B. Johnson, A. Maltz, and Y. Chun, "The Dynamc Source Routng Protocol for Moble Ad Hoc Networs. (DSR)," IETF INTERNET DRAFT, 24 February 2003. [5] V. Par and M. Corson, "A Hghly Adaptve Dstrbuted Routng Algorthm for Moble Wreless Networs," Sxteenth Annual Jont Conference of the IEEE Computer and Communcatons Socetes. Drvng the Informaton Revoluton (INFOCOM '97), pp.1405, 1997. IV. CONCLUSION In ths paper, we presented a new scheme that mproves the retransmsson mechansm for the exstng DSR protocol. In order to support our hypothess, we provded a complete mathematcal model that shows the formulaton of the proposed scheme. In partcular, we nvestgated the RD and the RM phases wth respect to the proposed reverse drecton scheme. We also showed that how effectve the proposed scheme would be when we mplement t wth the reverse drecton search for dscoverng the prmary paths. Our analyss also suggested that the dscovery of alternate prmary paths from the current source of error sgnfcantly mproves the networ performance n terms of RD process, tme delay, and the pacet losses. Moreover, we have