A Reliable and Fault Tolerant Routing for Optical WDM Networks

A Reliabl e and Fault- Tolerant Routing for Optical WDM Networ ks G.Ramesh Depart ment of Infor mation Technolog y KLN College o f Engineeri ng, Mad urai, India grameshphd @gmail.co m S.SundaraVadivelu Depart ment of Electro nics and Communicat ion Engineer ing SSN College o f Engineeri ng, Kalava kkam, Chennai, Ind ia. Abstract — In optical W DM networks, since each lightpath can carry a huge mount of traffic, failures may seriously damage the end-user applications. Hence fault-tole rance becomes an important issue on these netw orks . The light path w hich carries traffic during normal operation i s called as primary path. T he traffic is re routed o n a b ackup path in case of a failure. In t his paper we propose to d esign a reliable a nd fault-tolerant routing algorithm for establishing primary and bac kup paths. In order to establish the primary path, this algorithm uses load b alancing in which link cost metrics are estimated based on the cur rent load of the links. In backup p ath setup, the source calculates the blocki ng probability through th e r eceived fee dback f rom the desti nation by sending a small fraction of probe packets along the existing paths. It then selects the opti mal light path w ith the lowest blocking probability. Based on the si mulation results, w e show that the relia ble an d fault toler ant ro uting algorit hm re duces the blocking probability and latenc y w hile increasing the through put and channel utilization. Keywords- Relia ble; faul t-tolerance; blocking probability; load balancing; feedba ck. I. I NTRODUCTIO N A. Waveleng th-Division -Multiplex ing (WDM) Net works The Simultaneous transmissio n of multiple strea ms of d ata [1] with the help of the exclusive propertie s of fiber optics is called as wavelength divisio n multiple xing (WDM). The users ha ve the ca pabilit y to transfer huge amou nt o f data at high speed s over la rge d istances which i s offered by t he WDM networks. For the backbone of future next-ge neration inter net, wavelength divisio n multip lexing (W DM) is consid ered as the most capable te chnology [ 2]. Data’ s a re routed thro ugh optic al channels called lig ht pat hs, in WDM all-optical net works. T he establish ment of light path req uires the sa me wavelengt h to be used along entire ro ute of t he light p ath wit hout the wavelength conver sion capabilit y. T his is co mmonly r eferred as the wavele ngth contin uity co nstraint. By allowin g many ind epend ent signals with differen t wavelength to be tra nsmitte d simultaneo usly on o ne fiber, WDM enables the emplo yment o f a s ubstantial portion o f the available fiber bandwidth [3]. Sin ce the wa velength determines the co mmunicatio n path b y actin g as t he signat ure address of the origin, d estinatio n or routing, the routin g and detection of these si gnals can be achie ved independent ly. Therefor e the wavele ngth selective co mponents are req uired, allowing for the trans mission, reco very or ro uting of spe cific wavelength s. B. Fau lt Tolerant in WDM Networks Since each lightpath can carry a huge mount of traffic, failures in such network s ma y serio usly dama ge e nd-user applicatio ns. According to the scale o f their e ffect, failures in all-optical WDM networks can be classified i nto two categories [4]. One ca tegor y is a wavelength -level failure which i mpacts the qualit y of transmis sion o f each indi vidual lightpath. T he ot her categor y is a fiber-le vel failure wh ich affects all the lightpat hs on an individ ual fiber. Since each lightpath i s e xpected to oper ate at a rate of severa l gigab ytes per seco nd, a failure can le ad to a severe d ata loss. The ability o f networ k to with -stand fa ilures is ca lled a s fault-tolera nce. Fail ures arise due to the node failure or link failure. When a link fails al l its cons tituent fibers al so fa ils. All the con nections which use these fibers are to be rerouted and a wavelength will be a ssigned. The lig ht path which carries traffic during normal oper ation is called as primary path. The traffic is re routed on a backup path in case of a failure. Optica l net works whi ch u se the wavele ngth divisi on multiplexin g (WDM) and wavelen gth routi ng are subjected to failures. Fa ult to lerance bec omes a n i mportant i ssue because of the large a mount of traffic o n these net works in contradictio n to the con ventional cop per links. Fault tolera nce scheme s can be broad ly classified into • Path Pr otection • Restoration 1) P ath protectio n: In path protection, bac kup reso urces are reserved d uring connectio n set up and bo th primary a nd bac kup lightpath are co mputed before a failure occurs. T here are two types of pro tection sche mes: de dicated a nd shared pr otection. Dedicated- path pro tection: In dedicated -path p rotection (also called 1:1 p rotectio n), the re sources along a bac kup pa th are de dicated for only one co nnectio n and are not shared with the backup paths for o ther connection s. Shared-pat h protect ion: In shared -path protec tion, the resources alo ng a back up path may b e shared with other backup paths. As a result, backup channels are multiplexed among differe nt failure scena rios, and there fore, shared-pa th (IJCSIS) International Journal of Computer Science and Information Security, Vol. 6, No. 2, 2009 48 http://sites.google.com/site/ijcsis/ ISSN 1947-5500 protectio n is more capacit y efficient when co mpared wit h dedicate d-path prote ction. 2) Path Restoratio n: In path restoration, t he source and destination node s o f e ach connectio n traversin g t he fa iled link participa te in a distributed al gorithm to dynamica lly d iscover an e nd-to-end b ackup route. If no ro utes are availa ble for a broken connect ion, then t he co nnection is dr opped . II. R ELATED W ORK Michael T. Fre derick and Arun K. Somani [6] have presented an L+1 fault toler ance which is used for the recovery o f o ptical networks from si ngle link fail ures without the allocatio n of valuable system reso urces. W hile t he appro ach in its s implest form performs well a gainst the existing sche mes, t he flexibili ty of L+1 lea ve many optio ns to examine possible ways to further incr ease perfor mance. Muriel M ’edard [7] has d escribed that the pro tection r outes are pre-co mputed at a single locatio n a nd thus it is ce ntralized . Before the restor ation of the traffic, so me d istributed reconfiguratio n of op tical s witches is essential. O n the other hand, restoratio n te chnique s depend upo n d istributed signaling between nod es or o n the allo cation of a ne w p ath by a centr al manager. Hongsik Choi, S uresh Subra maniam and H yeong-Ah Choi [8] have co nsidered the network survi vability which is a critical require ment in the high-speed op tical networks. A failure model is consid ered so that any t wo lin ks i n the network may fai l in an random or der. They have pre sented three loop back method s of recoverin g fro m double -link failures. Only the first two method s re quire the identificati on of the failed li nks. But pr e-computing the b ackup paths for the third method i s mor e co mplex than the f irst t wo methods. The double link f ailures are tolerated by the heuristic algorith m which pre -computes the backup paths for lin ks. Yufeng Xin, Jing Teng, Gigi Karmous-Ed wards, GeorgeN.Rou skas and Da niel [9 ] have studied the i mportant fault management issue which concentra tes on the fast restoratio n mechanisms for Op tical Burst Switched (OB S) networks. T he OBS network o perates under the J IT signali ng proto col. The basic routing mechanis m is similar to the IP networks, where ever y OBS node maintains a local for wardi ng table. The entries in the for warding tab le consist of the ne xt hop information for the burs ts p er destination a nd per F EC (For ward Equi valent Class). Based on loo king up t he ne xt-hop information in their forwarding tables, OBS nodes forwards the comin g b urst co ntrol pa ckets and set up the co nnections. The connection set up process is si gnified by the burst forwarding or burst routi ng. Jian Wang, Laxman Sa hasrabudd he and Biswanath Mukherjee [10] have considered t he fault -monitoring functions w hich are usually provided by the optic al- transmission systems. In o rder to m easure the bit erro r ra te in the wavelength channels u sing SONE T framing, t he B1 b it in the SONE T overhead can b e used. Mo reover, to d etect cer tain failures like fiber cut i n other for matted optical channel s, the optical power lo ss ca n b e used. Optic al-Electric al-Opti cal (OEO) conversio n is used befo re each OX C po rt b ecause most of the OX Cs u se electro nic switc hing fabric. T herefore, fa ults can be d etected o n lin k-by-link b asis. B oth the e nd nod es of the failed lin k can detect the fi ber cut for all-optical s witches. Lei Guo [ 11] has studied the pr oblem of mu ltiple fail ures in WDM network s. In order to improve the surviva ble performance he pro posed a heuristic algorith m called Shared Multi-sub backup paths Reprovision ing (SMR). The survivable p erformance o f SMR in multiple failure s was considerab ly improved when co mpared with the pr evious algorithm. Guido Maier, Achil le P attavina, Luigi Barb ato, Francesca Cecini and Mario Martinelli [ 12 ] have investi gated t he issue of dynamic con nections in WDM net works. I t is a lso loaded with the hig h-prior ity pr otected static connectio ns. T hey have compared various r outing strategie s b y discrete event simulation in ter ms o f bloc king pro bability. B ased on the occupanc y cost function the y ha ve prop osed a heuristic algorithm which take s se veral p ossible ca uses of bloc king into account. T he be havior o f their algorithm was tested in well known case study o f mesh networks, wit h and with out wavelength co nversio n. A. Rajkumar and N.S.Murt hy Sharma [13] have propo sed a distributed prior ity based routing algorit hm. In o rder to establish the p rimar y and backup light paths they have proposed a varie ty of traffic classes whic h u ses t he co ncept of load balancing. Ba sed on the lo ad o n the links, their al gorithm estimates the cost metric. T he routing o f high p riorit y traf fic was p erformed over the light ly lo aded links. Ther efore while routing t he pri mary and bac kup p aths, the l ightly lo aded links are c hosen i nstead o f c hoosing the links w ith heavier loads. The load balancing will not reflect the d ynamic load c hanges because it is used in the r outing metric. Dong–won shin, Ed win K.P .Chon g a nd Ho ward Jay S iegel [14] have develop ed t wo heu ristic multipath routi ng sc hemes for survivable mu ltipath pro blem ca lled CPMR ( Condition al Penalization Multipat h Ro uting) and SPMR (Successive Penalization Multipath Rout ing). T heir schemes use “lin k penalization” methods to co ntrol (but not prohib it) link- sharing to deal with the dif ficulties c aused b y the link s haring. When compared with the ro uting scheme t hat searches fo r disjoint p aths, their method s have considera bly higher routing success rates which are shown through the si mulation res ults. All the above existin g works, did not p rovide the solution s based on the cha nging traffic loa d and the bloc king prob abilities of the pa ths. III. P ROBL EM D EFINITION AND S OLUTI ON Accepting as many de mands a s possible und er the networ k resource constraint is the main obj ective of the dynamic routing algor ithm. T his goal can be achieved by ce ntralize d algorithm ( CA) [5] through finding a pri mary/backup lightpath pair which u ses the mi nimum number of free chan nels for the current de mand. Therefor e more net work resources are left for the future d emands. On the other hand, CA is not scalable to large networ ks b ecause of its ce ntralized nature a nd the failure of the Networ k Manage ment Syste m (NM S) can b ring do wn the en tire net work. I n c ontrast to this, Simple Distrib uted Algorithm ( SD A) does not have the scalab ility p roble m and (IJCSIS) International Journal of Computer Science and Information Security, Vol. 6, No. 2, 2009 49 http://sites.google.com/site/ijcsis/ ISSN 1947-5500 the single po int of failure problem of CA. Moreover its performance in terms of the number of de mand blocking is much worse t han CA. Increasing tota l band width from so urce to destination i s the traditional use of multiple paths to di stribute data. One might anticipate t hat t he opti mal solutio n is caused b y balanci ng the load a mong multiple paths and these a pproa ches are p roposed in IP networks. I n such case s, the load for a gi ven so urc e- destination pa ir is distrib uted on each path in p ropo rtion to the available bottlenec k bandwidth of that path. But unpredictabl y, i n W DM netwo rks, the ab ove strate gies lea d to the worst perfor mance. It will be b etter if we deli ver the entir e load on a single opti mal pa th d ependi ng upo n t he c urrent network-wide load status. If the traf fic load for each sourc e and d estination pa ir remains static, then the single static light path which is based on the load gives best p erformance. But it will not be optimum for the d ynamic traffic loa d. For this case, in ord er to p rovide the choice o f selecti ng t he be st light pa th, multiple light pa ths needs to be mai ntained base d on the chan ging traffic lo ad conditions. In this paper we propose to design a reliable and fa ult tolerant routing ( RFTR) algorithm for p rimary a nd b ackup paths. In orde r to e stablish t he pri mary path, t his algo rithm use s the co ncept o f lo ad balancing . G iven a physic al network with the link costs and the traffic require ments bet ween e very source-destinat ion pair, then finding a route of the light paths for the net work with le ast congestion, is called as .l oad balancing. In this algorithm, based o n the load o f the links t he cost metric i s esti mated. T he traffic is route d over the light ly loaded link s. Therefore when routing the primary path, the links with the li ghter loads a re selected instead of links with the heavier lo ads. Using path restoration back up paths are established. In backup path set up, the source sends a small fractio n o f p robe packets alon g the e xisting pat hs. For a h igher burst ar rival r ate, the fraction o f traf fic pro bing will be lower. For a slow changing traffic, t he period of upd ate will be higher res ulting in an even s maller fractio n. The source edge can monitor and identify the requests that are rej ected at the netwo rk based on receiv ing the PACKS/NA CKS from the destina tions. Thus the source c an easily calc ulate the b loc king pro bability throu gh the monitore d results fro m t he probe pac kets. The ingress edge node sele cts the optimal light path with the lowest bloc king probab ility based on the measured blocki ng pr obabilitie s a nd forwards the data through this opti mal light path. On the other hand, it keeps pro bing the sub-opti mal path for their current blocki ng prob ability. IV. R ELIA BLE F AULT -T O L ERANT R OUTING P ROTOCOL A. Comp uting P rimary Path The link cost function f or primary pat h computatio n is designed b ased on the follo wing step s: 1) For each lin k L j , j = 1,2,3,… calculate t he load index of the link L j as Load (LI) = C f / C n (1) Where C f gives the n umber of free channel s in that link and C n is the to tal no. of c hannels in t hat link. 2) The link cost function Cost (L j ) is then de fined as Cost (L j ) = 1- Load (LI), if Load (LI) > LT = 1+ Loa d (LI), if Load (LI) > 0 and Loa d (LI) < = LT = ∞ , if Load (LI) = 0 (2) where LT is the load thres hold. 3) After we assi gn eac h li nk a c ost usin g t he ab ove formula, Dijkstra ’s shortest p ath algorithm is then used to compute the least-cost path as the primary path. If t he least -cost pat h has a cost of infinity, t hen the d emand is bloc ked; o ther wise a backup p ath is computed usin g the metho d given in the ne xt subsection. B. Compu ting Back up Path Let t he nu mber of path s bet ween source S and destination D are n. We p ropose a n adaptive multipath protocol to select the optimum pat h based on t he b locking pro babilities of the p aths. In our ap proach, small fractio ns o f pro be packets are sent by the non-opti mal paths suc h that these paths ar e selected ver y rarely. The prob e packets co ntain seq uence n umbers to identify the p ackets. (i) Let P j , j =1,2….k be the set of pro be p ackets sent on the paths R j , j=1 ,2…k. (ii) On receiving the prob es packets, the destination D for the path R j , send an PACK packet to the source, for each packet cor rectly received . T he missing or dropp ed pac kets can be identified using t he sequ ence number s of the rec eived packets. For e ach dr opped p acket, it sends a NACK packet to the source. (iii) For the path R j , the source calculates the blocking prob ability BP j such that BP j = P lost / P sent (3) Where P lost is the number of packets dro pped and P sent is the number of p ackets se nt. (iv) Similarl y for all the p aths R j , the so urce S calc ulates their b locking probab ilities BP j b ased on the P ACK and NACK feedb ack from the destination D. (v) No w sort the p aths { R j , j=1 ,2.,…k} in asce nding orde r of BP j values. (vi) T he paths which are hav ing less blockin g pro babilities BP 1 , BP 2 , BP 3 … are selected as bac kup pat hs. If t here i s an y (IJCSIS) International Journal of Computer Science and Information Security, Vol. 6, No. 2, 2009 50 http://sites.google.com/site/ijcsis/ ISSN 1947-5500 sudden or unexpecte d failure occurs in the pri mary path, traffic can be rerouted through these b ackup p aths. At t he sa me time, for their bloc king p robab ility it keep s searching the sub o ptimal paths. Beca use of thi s, we ca n a ble to jump quickl y to a new pa th when blo cking p robab ility of the current pa th increase s. T his o ccurrence is quite obvious in IP net works where the tr affic p atterns may vary significa ntly. By sending a small fraction of traffic for probing, the aggregated t hroughput is red uced. Ho wever b y findin g a new optimal path quickl y thi s red uction is compe nsated. The value of s mall fraction depends upon the sample size for accurate ly calculatin g the block ing prob ability. For a hig her burst ar rival rate the fractio n of traf fic for p robing is ver y low. V. S IMULATION R ESULTS A. Simu lation Mod el and Pa rameters In this section, we exa mine the per formance of our reliable and fault to lerant routi ng algorith m ( RFTR) with an e xtensive simulation stud y ba sed upon the ns-2 network simula tor [15]. We use the O ptical WDM networ k si mulator ( OWNs) patch in ns2. In o ur simula tion, we simulate a n 8-Node topolo gy as an example (Fi gure.1) which can be extended to any nu mber of nodes. Var ious simulatio n para meters are give n in table 1. T ABLE I . S IMULATION P ARAMETERS Topo logy Mesh Tota l no. of nodes 8 Link Wavelen gth Nu mber 8 Link Dela y 10ms Wavelengt h Conversio n Factor 1 Wavelengt h Conversio n Distance 8 Wavelengt h Conversio n Time 0.024 Link Utilizatio n sample Interval 0.5 Traffic Arri val Rate 0.5 Traffic Ho lding Ti me 0.2 Packet Size 200 No. of Session -traffics 4 Max Request s Number 50 In our experiments, we use a dynamic traf fic model in which c onnection request s arr ive at the network acco rding to an exponential process with an arrival rate r ( call/seco nds). The sessio n holdi ng time is exponentia lly distrib uted with mean holding ti me s (seco nds). The co nnection r equests are distributed rando mly o n a ll the network nodes. In all the experiments, we compare the results of RFT R with DP BR [1 3] scheme. Figure: 1 8-Node Topol ogy B. Performa nce Metrics We measure the following metrics in all the simulation experiments:  Bloc king Prob ability  Thro ughput in terms o f Packets Rec eived  End-to -End Delay  Channel Utiliz ation C. Results A. Based On Rate In the initial experiment, we var y the traffic rate as 2Mb, 4Mb….8 Mb a nd measure the bloc king pro babilit y, e nd-to-end delay and cha nnel utilizatio n. R ate V s B lo ck ing Pr ob ab ility 0 0.2 0.4 0.6 0.8 1 2 4 6 8 R ate (M b ) B lo ck in g P r o b a b i li t y D P BR R F T R Figure: 2 Rate Vs Blo cking Probability R ate V s D e lay 0 100 200 300 400 500 600 2 4 6 8 R ate ( M b ) D e l a y D P BR R F T R Figure: 3 Rate Vs Delay (IJCSIS) International Journal of Computer Science and Information Security, Vol. 6, No. 2, 2009 51 http://sites.google.com/site/ijcsis/ ISSN 1947-5500 R ate V s U t iliz atio n 0 0.01 0.02 0.03 0.04 0.05 0.06 2 4 6 8 R ate ( M b ) U t il i z a t io n D P B R R F T R Figure: 4 Rate Vs Utilizatio n Figure.2 shows the bloc king prob ability obta ined with our RFTR algorithm co mpared with DPBR scheme. It sho ws t hat the block ing prob ability is signi ficantly les s than t he DP BR, as rate increases. Figure.3 sho ws the e nd-to-en d delay occ urred for various rates. It sho ws that the de lay o f RFTR is significa ntly les s than the DPBR. Figure.4 shows the channel utilization obtained for vario us rate. I t shows t hat RT FR has better utilizat ion tha n the DP B R scheme. B. Ba sed On Time In the second exper iment, we vary the time interval as 2 , 4, 6….40 seco nds and mea sure the blocki ng p robab ility, e nd-to- end dela y, throughput a nd channel utilization. Tim e V s B lock ing Pr o bab ility 0 0.1 0.2 0.3 0.4 0.5 0.6 2 8 1 4 20 26 32 38 Tim e ( s e c) B l oc ki ng P ro ba bi li ty D P B R R F T R Figure: 5 Time Vs Blocking Pro bability Tim e V s D e lay 0 100 200 300 400 500 600 2 8 14 20 26 32 38 Tim e (s e c) De la y D P BR R F T R Figure: 6 Time Vs Del ay Ti m e V s Th r ou g hp ut 0 5000 100 00 150 00 200 00 2 8 14 20 26 3 2 38 Tim e (s e c ) Thr oughput D P BR R F TR Figure: 7 Time Vs Thro ughput Tim e V s U t iliz ation 0 0.01 0.02 0.03 0.04 0.05 0.06 2 8 14 20 2 6 3 2 38 Tim e ( s e c) U til iza t io n D P BR R F T R Figure: 8 Time Vs Utilization Figure.5 shows the bloc king prob ability obta ined with our RFTR algorithm co mpared with DPBR scheme. It shows that the block ing prob ability is signi ficantly les s than t he DP BR, as time increases. Figure.6 sho ws the e nd-to-en d delay occ urred for various time. It shows that the d elay of RFTR is signi ficantly less than the DPBR. Figure.7 shows the t hroug hput occurr ed for vario us ti me. As we can see fro m the fi gure, the thro ughput i s more in t he case of RFTR when co mpared to DPB R. (IJCSIS) International Journal of Computer Science and Information Security, Vol. 6, No. 2, 2009 52 http://sites.google.com/site/ijcsis/ ISSN 1947-5500 Figure.8 shows the channel utilization obtained for vario us time. RFT R shows better utilizatio n than the DP BR scheme. C. Based on Tr affic Sources In t his e xperiment, we vary the number of tra ffic sources as 1, 2, 3, and 4 a nd measure the blockin g pro bability, e nd-to- end dela y and throughput. N o. Of Tra ffic V s Blo ck ing P roba bi l i ty 0 0.2 0.4 0.6 0.8 1 1 2 3 4 s ou r ce s Blo ckin g Prob ab ilit y D P BR R F T R Figure: 9 Traffic Vs Blo cking Probabil ity N o. Of T ra ff i c V s De l a y 0 20 40 60 80 100 1 2 3 4 s ou r ce s De l a y D P B R R F T R Figure: 10 Traffic Vs Del ay N o. Of Tra ffic V s T hroughput 0 1000 2000 3000 4000 1 2 3 4 s ou r ce s Through put D P B R R F T R Figure: 11 Traffic Vs Through put Figure.9 shows the bloc king prob ability obta ined with our RFTR algorithm co mpared with DPBR scheme. It sho ws t hat the b locking prob ability of RFT R is significa ntly less than the DPBR, as the number of tr affic sources increa ses Figure.10 shows t he end -to-end d elay o ccurred when varying the numb er of traffic sources. It shows that t he dela y of RFTR is s ignificantl y less than the DPBR. Figure.11 shows the thro ughput occurr ed when varying the number of traffic sources. As we can see fro m t he figure, the throughput is mor e in t he case of RFT R when c ompared to DPBR. VI. C ONCLUSION In this paper we have designe d a reliab le and fault-tolerant routing a lgorithm for establis hing primary and backup pat hs in optical W DM networks. In ord er to establish the pri mary path, this algo rithm uses load bala ncing i n whic h li nk cos t metrics are estimated based on the current load of the links. The traffic i s routed o ver the lightly load ed li nks. T herefore the links with the li ghter loads a re selected instead of links with the heavier loads. I n bac kup path setup, the source se nds a small fractio n o f prob e packets along the exi sting pat hs. It c an monitor a nd identify the re quests that are rej ected at the network based on the re ceived positive and negat ive feedb ack from the d estinations. The so urce then c alculates t he block ing prob ability from the recei ved feedback and sele cts the opti mal light p ath with the lowe st blo cking prob ability. Based on t he simulation result s, we ha ve sho wn that the reliab le a nd fault tolerant ro uting al gorithm red uces the bloc king pro bability and latency while increasing the t hroughp ut a nd chan nel utilization. As a future work, we will concentrate on designin g an efficient fault de tection and lo calization technique i n W DM networks. R EFERENCES [1] Canhui (Sam) Ou Hui Zang ,Narendra K. Singha l, Keyao Zhu, Laxman H. Saha srabuddhe, Robert A. Macdonald, And Biswanath Mukherjee, “Sub p ath Protection For Scalability And Fast Recovery In Op tical WDM M esh Netw orks”, I EEE Journ al On Sel ected A reas In Communications, Vol. 22, No . 9, November 2004. [2] Vinh Trong Le, Son Hong Ngo, Xiao Hong Jiang, Susumu Horiguchi and Yasushi I noguchi, “A Hybrid A lgorithm for Dy namic L ightpath Protection in Survivable WDM Optical Netwo rks”, IEEE,2005 [3] Sateesh Chandra S hekhar, “ Survivable Mu lticasting in WDM Optical Networks”, August,2004 [4] S.Ramamurthy, Laxman Sahasrabuddhe, and Biswanath Mukherjee, "Survivable WDM M esh Networks", Journ al of light wave technology , vol. 21, no. 4, April 2 003. [5] Lu Ru an, Haibo Luo, and Chang L iu, “A Dynamic Routing Alg orithm with Load Balancing Heuristics for Restorable Connections in WDM Networks”, IEEE Journa l on Selected Areas in Communications, Vol. 22, no. 9, November 2 004. [6] Michael T. Frederick a nd A run K . Somani, “ A single-fault recove ry strategy for optical Networks using sub graph routing”, The International Journal of Computer and Telecommunications Networking, Volume 50, I ssu e 2 (February 2006) [7] MurielM´edard, “Netw ork Reliability an d Fault Tolerance”, I EEE Transactions on Volume 50, Issue 1, Page(s) :85 – 91, Mar 2001 [8] Hongsik Choi, Suresh Subramaniam, and Hyeong-Ah Choi, “On Double-Link Failure Recovery in WDM Optical Networks”, INFOCOM Volume: 2, On page(s): 808- 816 vol.2, 2002 [9] Yufeng Xin, Jing Teng, Gigi Karmous-Edwards,GeorgeN.Rouskas and Daniel St evenson, “ Fault Management with Fast Restoration for Opti cal Burst Switched Netw orks”, IEEE,2004 (IJCSIS) International Journal of Computer Science and Information Security, Vol. 6, No. 2, 2009 53 http://sites.google.com/site/ijcsis/ ISSN 1947-5500 [10] Jian Wang,Lax man Sahasrabuddhe a nd Bisw anath Mukherje e, “ F ault Monitoring and Restoration in Optical WDM Networks”, National Fiber Optic Engineers Confer ence NFOEC 2002 [11] Lei Guo, "Heuristic Survivable Routing Algorithm for Multiple Failures in WDM Networ ks", in p roc. of 2nd IEEE/IFI P International Workshop on Broadband Convergence Networks, pp : 1-5, 21 May 2 007, Doi: 10.1109/BCN.2007.372 750. [12] Guido Maier, Achille Pattavina, L u igi Barbato, Francesca C ecini and Mario M artinelli, "Routing Algorithms i n WDM Networks under Mixed Static and Dynamic Lambda-T raffic", Journal on Photonic Network Communications, vol. 8, no. 1, pp: 69- 87, June 2004, Doi: 10.1023/B:PNET.0000 031619.18955.b4. [13] Ra jkumar and D r.N.S.Murthy Sha rma, "A D istributed Priority B ased Routing Algorithm for Dynamic Traffic in Survivable WDM Networks", International Journal of Computer Science and Network Security, vol.8, no.11, November 2008. [14] Dong-won Shin, Edwin K.P.Chong and Howard Jay Siegel, “Survivable Multipath Routing Usi ng Link Pe nalization”, Computing and Communications, 2004 I EEE International Co nference. [15] Network Simulator – www.isi.edu/ns nam A BOUT A UTHORS G.Ramesh is working as Assistant Profe ssor in the Department o f Information Technology a t KLN College o f Engineering, Madurai, India. He has completed the graduation in Electrical and Electronics En gineer ing and Post Graduation in Computer Science a nd Engineering. He is doing research work in optical netwo rking Dr.S.SundaraVadivelu is working as Professor in Department of Electronics and Communication Enginee ring at SSN Colleg e of Engineering, Kalavakkam, Chennai, India. His research in terest is in Optical Communication and Networks. (IJCSIS) International Journal of Computer Science and Information Security, Vol. 6, No. 2, 2009 54 http://sites.google.com/site/ijcsis/ ISSN 1947-5500