TELK OMNIKA , V ol. 15, No . 3, September 2017, pp . 1061 1071 ISSN: 1693-6930, accredited A b y DIKTI, Decree No: 58/DIKTI/K ep/2013 DOI: 10.12928/telk omnika.v15.i3.5019 1061 P ac ket Loss Rate Diff erentiation in slotted Optical P ac ket Switc hing OCDM/WDM Omar Najah * , Kamaruzzaman Seman , and Khairi Abdulrahim F aculty of Science and T echnology , Univ ersiti Sains Islam Mala ysia, Bandar Bar u Nilai 7180 0, Nilai, Neger i Sembilan, Mala ysia * Corresponding author , e-mail: omar .najah@gmail.com Abstract W e propose a m ulti-class mechanism f or Opt ical Code Division Multiple xing (OCDM), W a v elength Division Multiple xing (WDM) Optical P ac k et Switch (OPS) architecture capab le of suppor ting Quality of Ser- vice (QoS) tr ansmission. O CDM/WDM has been propose d as a competitiv e h ybr id s witching technology to suppor t the ne xt gener ation optical Inter net. This paper addresses perf or mance issues in the slotted OPS netw or ks and proposed f our diff erentiation schemes to suppor t Quality of Ser vice . In addition, w e present a compar ison betw een the proposed schemes as w ell as , a sim ulation scheduler design which can be suitab le f or the core s witch node in OP S netw or ks . Using softw are sim ulations the perf or mance of our algor ithm in ter ms of losing probability , the pac k et dela y , and scalability is e v aluated. K e yw or ds: Quality of Ser vice (QoS), Slotted Optical P ac k et Switching, P ac k et Loss Probability (PLP), shared Fiber Dela y Line (FDL). Cop yright c 2017 Univer sitas Ahmad Dahlan. All rights reser ved. 1. Intr oduction Upcoming optical netw or ks are e xpected to suppor t man y n umbers of users with m ultiple bit r ate tr ansmissions , finer g r an ular ity capacity and quality of ser vice (QoS) demands in a fle x- ib le manner [1].The g ro wth of inter activ e application and m ultimedia, such as video-on-demand, teleconf erencing and digital video broadcasting depend on the Inter net to e xchange data tr affic. Consequently , Optical P ac k et Switching (OPS) is supposed to be among one of the g reatest can- didates to be implemented in f uture optical bac kbone netw or k [2], [3], it is fundamental to de v elop ne w QoS diff erentiation mechanisms that consider the OPS netw or k char acter istics . Ov er y ears , plenty of diff erent s witching technologies such as optical code division m ul- tiple xing (OCDM), and w a v elength division m ultiple xing (WDM) ha v e been in v estigated in optical fiber comm unication [4],[5]. These technologies , OCDM constitutes a potential candidate f or the ne xt gener ation optical netw or ks par ticular ly due to f eatures lik e synchronous and asynchronous oper ation simplified netw or k control the easy addition of ne w users and the possibility of diff eren- tiated quality of ser vice QoS at the ph ysical. No w ada ys , there has been an increasing demand f or ne w m ultimedia applications , f or instance , e-lear ning e-health high-data-capacity 3-D full-HD (high-definition) the video , m ultipla y er games , video-on-demand, etc., as a consequence , these ne w applications will require diff erentiated QoS and m ultiple r ates tr ansmission which are becom- ing a challenge f or future optical netw or ks . Ho w e v er , w a v elength division m ultiple xing, Optical code-division m ultiple xing (OCDM/WDM) is a technic , which can use f or the future gener ation of the Inter net [6]. In WDM technology , the optical media spectr um divided into a n umber of w a v elength bands , which arent o v er lapping and each w a v elength can pro vide a large amount of r a w bandwidth. Moreo v er , WDM can suppor t a m ulti-coded comm unication channels oper ating at high data r atio . Presently , WDM technology suppor ts the m ultiple xing of hundreds of spectr um into a single optical fiber , at a tr ansmission r ate of 10s Gb/s per spectr um channel [7]. In optical code-division, m ultiple xing (OCDM) can pro- vide fle xib le , heterogeneous , and synchronous m ultiple bit-r ate tr ansmissions at a sub-w a v elength le v el to connections with enhanced secur ity and netw or k scalability [8]. In addition, OCDM can Receiv ed March 8, 2017; Re vised J une 26, 2017; Accepted J uly 19, 2017 Evaluation Warning : The document was created with Spire.PDF for Python.
1062 ISSN: 1693-6930 pro vide QoS diff erentiation at the ph ysical la y er le v el. Moreo v er , optical code gener ation, process- ing, and decoding are perf or med entirely in the optical domain [1]. This ar ticle is orga niz ed as the f ollo wing sections . In Section II, w e descr ibe the related w or k. The diff erentiated contention resolution in photonic pac k et s witching net w or ks is discussed in Section III. In Section IV , w e discuss the sim ulation perf or mance . The conclusion has been e xplaining in Section V . 2. Related W ork Actually , high bandwidth does not solv e the QoS issue; theref ore , ne w schemes f or sup- por ting QoS in an optical bac kg round are needed. Currently , there is diff erent ser vice diff eren- tiation ha v e been proposed to suppor t m ulti-r ate ser vices pro vision in OCDM netw or ks such as emplo ying codes with diff erent lengths [9]. A diff erent approach is to use the m ulti-cored tech- nique , where the n umber of sim ultaneously assigned code w ords to each user is a function of the data r ate of the ser vice class , while QoS diff erentiation is realiz ed b y v ar iab le-w eight code w ords f or each ser vice class [10]. Adapting the n umber of allocated codes according to the requested tr ansmission r ate , so as to suppor t time-v ar iant data r ates and m ultiser vice tr ansmissions b y emplo ying a m ulti-code v ar iab le-w eight tw o-dimensional (2D) one coincidence frequency hopping code , optical or thog- onal code (OCFHC/OOC) as the signature code [1]. Ho w e v er , the author proposed dynamic changes in the req uested r ate b y means of allocating/de-allocating encoders/decoders to the e xisting connections . T w o m ulticode assignment methods , used, r andom m ulticode assignment (RMA) policy and unif or m m ulticode assignment (UMA) policy . Consequently , the result sho ws that b loc king probability obser v e w orse end-to-end b loc king probability . Fur ther more , RMA and UMA policies ha v e the same b loc king probability . On the other hand, utilizing m ultilength v ar iab le-w eight (ML VW) codes [9, 11], and m ul- ticode v ar iab le-w eight (MCVW) codes [10], b oth m ulti-r ate and m ulti-QoS tr ansmission are sup- por ted sim ultaneously , as a compar ison is carr ied out betw een using MCVW codes and ML VW codes in [10]. In the ar ticle [9], the idea w as based on OCDM scheme in gener aliz ed m ulti- protocol label s witching (GMPLS) netw or k using m ultilength v ar iab le-w eight optical, or thogonal codes (ML VW -OOC) as signature se quences . The w eight and th e length of codes w ere de- signed based on the ser vice char acter istics , with limitation the n umber of connections based on the activity n umber of the connected users to guar antee the desired ser vice . Moreo v er , m ulti- length v ar iab le-w eight, optical or thogonal codes (ML VW -OOC) is used as signature sequences of a h ybr id WDM/OCDM system, the code length and code w eight of ML VW -OOCs are designed based on the char acter istics of the demanded classes of ser vice [11]. In [12] an o v er all technique of constr ucting ML-OOCs , with arbitr ar y correlation v alue f or m ulti-r ate OCDM netw or ks ha v e been introduced to constr uct a v ar iab le-length mapping sequence with unpar alleled positions , with which map shor t-length OOCs into long-length OOCs to realiz e the gener al constr uction of ML-OOCs with correlation v alue with high efficiency . Fur ther more , the author in, [13, 14] ha v e proposed f or malisms can be successfully ap- plied to 1-D and 2-D MWM L-OOC OCDM netw or ks with an y n umber of user classes . In [13], the bit error r ate (BER) and pac k et correct probability e xpressions w ere der iv ed, consider ing the m ultiple-access interf erence as binomially distr ib uted. On the other hand, P ac k et through- put e xpressions w ere der iv ed consider ing P oisson, binomial, and Mar k o v chain approaches f or the composite pac k et arr iv als distr ib utions , with the latter defined as a benchmar k. Numer ical results sho w that the P oisson approach underestimates the throughput perf or mance in unaccept- ab le le v els and incorrectly predicts the n umber of successfully receiv ed pac k ets f or most off ered load v alues e v en in f a v or ab le condit ions , such as f or the 2-D MWML OOC OCDM netw or k with a consider ab ly large n umber of sim ultaneous users . The authors in [15] ha v e proposed to categor iz e the fiber w a v elength spectr um into a n umber of w a v ebands and allocate the spectr um of each w a v eband f or a par ticular class of ser- vice and the corresponding codes w ere designed based on the char acter istics of the cla s s of ser vice . The y ha v e suggested tw o scenar ios: path estab lishment with tr affic management and TELK OMNIKA V ol. 15, No . 3, September 2017 : 1061 1071 Evaluation Warning : The document was created with Spire.PDF for Python.
TELK OMNIKA ISSN: 1693-6930 1063 path estab lishment without tr affic man agement. In the first scenar io , a controlling mechanism manages the distr ib ution of the connected paths so that all w a v elengths ha v e the same n um- ber of connected paths; whereas , in the second scenar io , connected paths are divided among w a v elengths r andomly . Moreo v er , in [16] an or iented pa th length based w a v elength assignment str ategy f or w a v elength-routed WDM netw or ks w as pur posed which assign the w a v elength to the connection request according to the path length. In their scheme , the connection requests with shor ter light path w ere assigned the w a v elengths ha ving higher dispersion and the w a v elengths ha ving lesser dispersion w ere assigned to the light paths with longer distance . Additionally , w a v e- length assignment has been used in [17, 18] , and w a v elength reser v ation with shared b uff ers in [19]. In [19], the authors divided the w a v elengt hs to sets , theref ore regarding to the reser v ation scheme , the high pr ior ity pac k ets (le v el 1) can be shifted to an y sets of a v ailab le w a v elengths , in decreasing order from L1 to L4, where the rest pac k ets of other QoS le v els w ere consequently shifted to their o wn sets . 3. Diff erentiated Contention Resolution in Photonic P ac ket Switc hing Netw orks QoS diff erentiation fr ame w or k suppor ts OPS netw or ks to pro vide m ultiple ser vice classes . Efficient and f air resource categor ization betw een the classes of ser vice increases the total net- w or k efficiency and utilization [20]. In this ar ticle , the control and data planes of the proposed WDM/OCDM-based OPS scheme is designed consider ing QoS requirements . The measurement cr iter ia of QoS are PLP and, FDL dela y . Fur ther more , w e pro vide a s witch architecture and chan- nel allocation scheme implementation f or suppor ting diff erentiated b uff er i ng and diff erentiated w a v elength con v er ters in photonic pac k et-s witched netw or ks . Gener ally , absolute le v els can be specified f or each QoS metr ic [20, 21]. Ho w e v er , without loss of pr inciple , tw o classes are considered f or e v e r y metr ic , to be specific , high pr ior ity (HP) and lo w pr ior ity (LP) classes . Consider ing HP and LP classes f or the char acter iz ed f our channel allo- cation schemes , there are f easib le . In an y case , all possib le combinations of allocation schemes are sim ulated. W e then discuss the v ar ious par ameters f or pro viding diff erentiated ser vice in photonic pac k et-s witched netw or ks . 3.1. Node ar c hitecture In the recent w or k [22] , w e ha v e proposed h ybr id OCDM/WDM combined with shared FDL as it is demonstr ated in Figurer 1. The adv antages of the h ybr id system [22] w ere to impro v e the perf or mance of the node and, in [23] w as to optimiz e the ultimate n umber of FDL to reduce the cost and to k eep the perf or mance high. The proposed algor ithm, initially tr ies to eliminate the contention b y means of optical code b y assigning the pac k et to a v ailab le code and, if the re is no free channel code in that par ticular w a v elength, the algor ithm tr ies to shift the pac k et to another w a v elength in the same out fiber link using the w a v elength con v er ter from the shared w a v elength con v er ter pool, then, if there is no free w a v elength con v er t er , the algor ithm sends the pac k et to either FW b uff ers or FB b uff ers regarding to their a v ailability . The only cases where the pac k ets w ould be dropped are if there is no free FB b uff er is a v ailab le . In the s witch assumption, w e assumed that all s witch matr ix is function synchronously with constant length pac k ets receiv ed at N ing ress optical link in the fitted time slot. The matr ix of core s witch has (( N M F ) + K ) (( N M F ) + R + D + K ) s witching matr ix. It has N Input/output fiber por ts (IF/OF), K FB , R w a v elength con v er ter , D FW FDLs . Moreo v er , e v er y fiber suppor ts M w a v elengths , and e v er y w a v elength can hold up to F pac k ets coded along, b y using Optical Code . Means that, each input and output channel is reco gniz ed b y the three v ar iab le ( i; j ; O C k ) . The proposed scheduling algor ithm is e x ecuted at each slot time of the m ultiser vice OCDM/WDM optical pac k et s witch. The proposed scheduling algor ithm has been modified from [22, 23], in order to suppor t m ultiser vice tr ansmission [24]. In this paper , the m ultiser vice diff eren- tiation is assisted b y a n umber of technologies such as code con v ersion, w a v elength con v ersion, and Fiber Dela y Lines (FW and FB) b uff er ing [25]. The m ultiser vice OCDM/WDM pac k et s witch sho wn in Figure 1 perf or ms the f ollo wing oper ations: (1) incoming pa c k et s on each input fiber are w a v elength dem ultiple x ed and code decoded b y means of M WDM dem ul tiple x er and M :F P ac k et Loss Probability of h ybr id OCDM/WDM (Omar Najah) Evaluation Warning : The document was created with Spire.PDF for Python.
1064 ISSN: 1693-6930 Figure 1. Architecture of OCDM/WDM with shared FDL. code decoders; (2) the control unit processes and chec k f or all High pr ior ity pac k ets in entire input f or a time slot and giv e them the pr ior ity to use resource first, resolv es pac k et contentions , and directs pac k ets either to code w ord, w a v elength con v er ted or b uff ered based on the r ules of the scheduling algor ithm, f or HP pac k ets , when a pac k et arr iv es at ing ress first tr y to tr ansmit it im- mediately to desired out w a v elength through free code channel, if contention occurs , then, tr y to solv e the prob lem using w a v elength con v er ter , if there is no free con v er ter , then, send the pac k et to FW b uff ers , if there are no free FW b uff ers and no g uar antee of successful tr ansmit later on, at th at time , the HP pac k et will be dropped, this scenar io r uns up to finish all HP pac k ets , same scenar io goes with LP pac k ets accept that, if it is not successfully sent to FW b uff ers then, send it to FB b uff ers unless all FB b uff ers are occupied , at that moment, the LP pac k et will be dropped; (3) all successful pac k ets r outed either direct code w ord pac k ets , w a v elength con v er ted pac k ets and b uff ered pac k ets are directed to S F i s witches based on decisions tak en b y the control unit; (4) finally , pac k ets are routed to w ards to desired output channel after been coded and m ultiple x ed [24]. Once again, there are only three possib le places where the pac k et could be dropped; first place , if HP or LP pac k et d id not get free FW b uff er and no guar antee of output reser v ation; sec- ond place , if LP pac k et did not get free FB b uff er ; third place , if LP pac k et e xceed f our round in FB b uff er and no chance to lea v e the node . Moreo v er , the pac k et losses are the main concer n in the s witch consider ation, and pac k et dela y which is eff ected f actor b y both lac ks of w a v elength con v er ters and conte ntions at an output por t [26]. Hence , in this ar ticle , w e concentr ate on the P ac k et Loss Probability (PLP) and pac k et dela y as the main QoS par ameters , which means that the v ar iety ser vice classes wil l be diff eren- tiated from each other based on diff erent PLP and pac k et dela y . PLP is defined as the a v er age n umber of dropping pac k ets per unit time . Ho w e v er , f or the sak e of perf or mance e v aluation of the s witching node under diff erent tr affic load, PLP is defined as r atio of the total n umber of losing pac k ets to the tot al n umber of arr iving pac k ets; as w ell as throughput is defined as r atio of the total n umber of successful depar ture pac k ets to t he total n umber of arr iving pac k ets . The a v er age pac k et dela y is defined as the n umber of time slots a pac k et has w aited since its arr iv al into the input b uff er till it is tr ansmitted to its output por t. P LP = P Lossing pack ets P ar r iv ing pack ets (1) T hr oug h = P depar tur e pac k ets P ar r iv ing pack ets (2) D el ay = P w aitedsl ots P buf f er dpack ets (3) TELK OMNIKA V ol. 15, No . 3, September 2017 : 1061 1071 Evaluation Warning : The document was created with Spire.PDF for Python.
TELK OMNIKA ISSN: 1693-6930 1065 3.2. Channel Allocation Sc hemes Basically , por ts restr iction based QoS diff erentiation scheme , the optical code allocation algor ithm (OCAA), has been studied. In the optical code allocation, which pro vides QoS diff er- entiation in synchronous b uff ered OPS netw or ks wi th (full r ange and limited n umber) of shared w a v elength con v er ters , cer tain allocation str ategies ha v e been sim ulated in order to obtain QoS . Fur ther more , the algor ithm has been adapted from our pre vious w or k [22, 23] ho w e v er , f our mech- anisms ha v e been sim ulated f or the single node based QoS diff erentiation: shared inpu t, shared output (SSM); shared input, pr iv et output (SPM); pr iv et input, shared output (PSM); and pr iv et input, pr iv et output (PPM) [25]. In the OCAA, the total a v ailab le codes ( F ) at an input/output w a v elengths are divided into tw o pools according to pr ior ity , i.e . a class HP pool with FH codes and a class LP pool with FL codes . Incoming pac k ets can access these codes only if the y ha v e the necessar y pr ior ity le v el. This means that at share mechanism phase , both classes are shar ing the por ts with the pr ior ity ser vice (i.e . WC , FW) to HP pac k ets . On the other hand, at pr iv et mechanism phase , class HP tr affic can access only codes fro m the class FH pool, as w ell as class LP tr affic can access only codes from the class LP pool, where is the total of F H + F L = F codes on one w a v elength. By adjusting the v ar iab le FH and FL, w e achie v e an y desired le v el of the PLR f or class HP tr affic , which is impor tant in order to pro vide QoS to the core netw or k nodes [27]. F or incoming pac k ets which f ail to acquire an y ser vices (FW b uff er f or HP and FB b uff er f or LP) are immediately dropped at the node . In order to o btain sim ulation results f or the PLP f or both pr ior ity classes , w e ha v e b uilt the OCAA algor ithm. The first algor ithm is PPM, where the coded channels are divided into tw o sets FH and FL f or in/out por ts . All pac k ets arr iv ed at the in puts are ser viced regarding their pr ior ity class . Theref ore , the HP pac k ets are ser viced at first, and the y ha v e the pr ior ity to use nodes resource . Fur ther more , each class pac k ets are separ ated b y channels in use (i. e . each class has its o wn channels in/out). In this phase , the initial v a lue of HP set is (FH =2, 4, 6, and 8) channels , in the same time , the initial v alue of FL is (FL= 8, 6, 4, and 2) channels . The second algor ithm is PSM, where the input coded channels are divided into tw o sets FH and FL, at the same time the output por ts are shared among all pac k ets , depend on the channel a v ailability . F or instance , the HP pac k ets reach the inp ut node in pr iv et coded channels and lea ving the node using shar ing channels with FL pac k ets . The third algor ithm is SPM, where the input coded channels are shared b y all the diff erent classes , b ut the output is pr iv et each class has its o wn channel to use . The f our th algor ithm is SSM, where all por ts are shared in use f or all diff erent classes . 4. Sim ulation P erf ormance In this section, w e e v aluate the perf or mance of one node based QoS mechanisms f or OPS netw or ks with off ered (0.8) tr affic load, to this end, w e consider OCAA QoS diff erentiation mechanisms as e xplained ear ly section. The sim ulation results are obtained using Scilab . In the sim ulations , it is assumed that time slots are fitted to pac k et boundar y and all are equal length dur ation. The proposed diff erentiation schemes algor ithm has in v olv ed fix ed-lengt h pac k ets , sim- plifying the algor ithm and oper ation of the s witching node . In another w ord, it is assumed that all pac k ets entered the s witch are synchroniz ed with respect to their boundar ies , consequently that all pac k ets are fitted to their timeslot. The sim ulation en vironment that been used is r un f or a paired of 5000-time slots . In sim ulation e xper iment, the s witch node siz e is N = 32 , w a v elength M = 16 , optical code F = 10 , and with share d pool w a v elength con v er ter R = 24 . A sim ulation r uns enough time with the pur pose of diff erentiation mechanism. The pac k ets arr iv e according to Ber noulli distr ib ution ne w slot. T r affic is equally distr ib uted to all output of s witch 1 = N and it is considered that pac k et dur ation is fit to time cell. In the e xper iments , dur ing each time slot e x ecution, the HP pac k ets ha v e g iv en the pr ior ity to be ser viced first, f or instance , occupies free output channels , use a v ailab le w a v elength con v er ter s , and FW b uff ers o v er LP pac k ets . On another w ord, the HP pac k ets cant use FB b uff ers to reduce the dela y . In Figure 2 sho ws the result of an e xper imental sim ulation of PPM. The pac k et loss prob- P ac k et Loss Probability of h ybr id OCDM/WDM (Omar Najah) Evaluation Warning : The document was created with Spire.PDF for Python.
1066 ISSN: 1693-6930 Figure 2. P ac k et Loss Probability of h ybr id OCDM/WDM system at diff erent n umbers of code channel, pr iv et input pr iv et output mechanism. Figure 3. P ac k et Dela y of h ybr id OCDM/WDM system at diff erent n umbers of code channel, pr iv et input pr iv et output mechanism. ability and pac k et dela y of h ybr id OCDM/WDM system with diff erent n umbers of reser v ed code channels f or HP under (0.8) total tr affic load. The result of the algor ithm perf or mance sho ws that, when FH increases , the PLP decre ases , tak e note that F H + F L = 10 is the total n umb er of coded channels in the single w a v elength. Its ob vious that when the small n umber of FH used the probability of losing pac k ets is higher and reduced when increasing the n umber of FH, because of the contention is reduced b y increasing the n umber of out channels . On another hand, Figure 3 sho ws the perf or mance of the pac k et dela y f or HP is almost flat f or all n umber of the channels , as a consequence of high tr affic directed from the same n umber of inpu t channels to the same n umber of output channels . Figure 4 and Figure 5 i llustr ated the perf or mance of PSM. The n umber of input coded channels which been reser v ed to class HP is FL=2, 4, 6, 8, at each w a v elength, and where the output channels are shared among all classes . The algor ithm policy receiv es the HP pac k ets class from pr iv et channels and sends them through shared channels with t he pr ior ity ser vice f or HP class . The result of the algor ithm perf or mance displa ys that, PLP f or HP almost is z er o f or up to 6 channels , and f or 8 channels the PLP is 10 4 , the e xplanation is that the HP pac k ets are ser v ed first as the y ha v e more out channels than inputs . Note , e v en that the PLP of LH pac k ets has perf or med better than PPM and SPM. The dela y perf or mance of this algor ithm f or HP class is almost z er o when reser v ed channels less than 6 channels , then star ts to r ise up due to the hea vy tr affic flo wing from the increased n umber of reser v ed channels . The perf or mance of algor ithm SPM is presented in Figure 6 and Figure 7. The e xper i- ment tested all the probability of input tr affic f or a diff erent n umber of reser v ed output channels . TELK OMNIKA V ol. 15, No . 3, September 2017 : 1061 1071 Evaluation Warning : The document was created with Spire.PDF for Python.
TELK OMNIKA ISSN: 1693-6930 1067 Figure 4. P ac k et Loss Probability of h ybr id OCDM/WDM system at diff erent n umbers of code channel, pr iv et input shared output mechanism. Figure 5. P ac k et Dela y of h ybr id OCDM/WDM system at diff erent n umbers of code channel, pr iv et input shared output mechanism. P ac k et Loss Probability of h ybr id OCDM/WDM (Omar Najah) Evaluation Warning : The document was created with Spire.PDF for Python.
1068 ISSN: 1693-6930 Figure 6. P ac k et Loss Probability of HP h ybr id OCDM/WDM system v ersus a v ar iety of HP tr affic from total tr affic 0. 8, f or diff erent n umbers of channels reser v ed to HP , shared input pr iv et output mechanism. Figure 7. P ac k et Dela y of HP h ybr id OCDM/WDM system v ersus a v ar iety of HP tr affic fr om total tr affic 0.8, f or diff erent n umbers of channels reser v ed to HP , shared input pr iv et output mechanism. The result is dro wned with three v erbal , which are PLP , tr affic load, and a n umber of reser v ed channels , the char t sho ws the PLP only f or HP class . The SPM algor ithm perf or ms high pac k et loss probability compar ing with the other three proposed algor ithms , due to the limited a v ailab le output channels . Fur ther more , Figure 7 sho ws the pac k et dela y perf or mance , where it perf or ms v er y high dela y , due to v er y high tr affic flo ws to FW b uff ers . In gener al, this algor ithm did not perf or m w ell in both PLP and dela y , due to limited a v ailab le output channels . In Figure 8 and Figure 9 demonstr ate the results of an e xper imental sim ulation of SSM, where the input and output are shared among the classes with the respect of the pr ior ity . The Figure 8 plotted the PLP v ersus the HP tr affic load from total tr affic (0.8). F rom the plot dr a wn in Figure 8, ob vious that the SSM algor ithm perf or mance significance , with the consider ation, that the node is w or king in h ybr id netw or ks . The SSM algor ithm perf or ms v er y lo w pac k et loss probability compar ing with the other three proposed algor ithms , due to the v ar iety of a v ailab le channels . Note that the PLP of HP class at hea vy tr affic is star ting to r ise up as a result of HP pac k ets cant use FB b uff ers where dropping pac k ets are increased. On the other hand, the perf or mance of pac k et dela y is plotted in Figure 9. The HP class perf or med no dela y at lo w tr affic and toler ab le dela y at hea vy tr affic. The reason behind this is at high tr affic more contention occurs with limited w a v elength con v er ters . TELK OMNIKA V ol. 15, No . 3, September 2017 : 1061 1071 Evaluation Warning : The document was created with Spire.PDF for Python.
TELK OMNIKA ISSN: 1693-6930 1069 Figure 8. P ac k et Loss Probability of h ybr id OCDM/WDM system v ersus a v ar iety of HP tr affic from total tr affic 0.8, shared input shared output mechanism. Figure 9. P ac k et Dela y of h ybr id OCDM/WDM system v ersus a v ar iety of HP tr affic from total tr affic 0.8, shared input shared output mechanism. P ac k et Loss Probability of h ybr id OCDM/WDM (Omar Najah) Evaluation Warning : The document was created with Spire.PDF for Python.
1070 ISSN: 1693-6930 5. Conc lusion This research aims at reducing pac k et loss probability of HP h ybr id OCDM/WDM system in which a huge amount of data is being tr ansmitted throug h high-speed optical links . In order to do so , our objectiv e is to mak e QoS diff erentiation and resource utilization propor tional to each other . F our diff erentiation algor ithms called PPM, PSM, SPM, and SSM w as proposed in order to diff erentiate bet w een the classes . Sim ulation results indicate the f act that the proposed algor ithms PSM and SSM yield a significant pac k et loss probability with an acceptab le le v el of pac k et dela y compared with PPM and SPM algor ithms . Moreo v er , since that the s witch node is a par t of h ybr id netw or ks connected to each other and theref ore if w e use PSM algor ithm, the ne xt node m ust use pr iv et input mechanism, where the perf or mance of the netw or k ma y suff er from high contention. The sim ulations sho w that SSP method has v er y high perf or mance efficiency , it can approach the Raddo bound. Consequently , it is clear that the SSP algor ithm is a significant algor ithm f or such h ybr id netw or ks . Ref erences [1] A. E. F arghal, H. M. Shalab y , K. Kato , and R. K. P okhare l, “P erf or mance analysis of m ulticode ocdm netw or ks suppor ting elastic tr ansmission with qos diff erentiation, IEEE T r ansactions on Comm unications , v ol. 64, no . 2, pp . 741–752, 2016. [2] I. A. Ashour , S . Shaar i, P . S . Menon, and H. A. Bakar man, “Optical code-division m ultiple- access and w a v elength division m ultiple xing: Hybr id scheme re vie w , Jour nal of Computer Science , v ol. 8, no . 10, pp . 1718–1719, 2012. [3] V . Er amo , A. Ger moni, A. Cianfr ani, C . Raff aelli, and M. Sa vi, “Ev aluation of qos diff erenti- ation mechanism in shared-per-w a v elength optical pac k et s witches , in Optical Netw or k De- sign and Modeling, 2009. ONDM 2009. Inter national Conf erence on . IEEE, 2009, pp . 1–6. [4] H. Br ahmi, G. Giannoulis , M. Menif , V . Katopodis , D . Kala vrouziotis , C . K ouloumentas , P . Groumas , G. Kanakis , C . Stamatiadis , H. A vr amopoulos et al. , “On the fly all-optical pac k et s witching based on h ybr id wdm/ocdma labeling scheme , Optics Comm un ications , v o l. 312, pp . 175–184, 2014. [5] T . Iser nia, A. Massa, A. Mor abito , and P . Rocca, “On the optimal synthesis of phase-only reconfigur ab le antenna arr a ys , in Antennas and Propagation (EUCAP), Proceedings of the 5th European Conf erence on . IEEE, 2011, pp . 2074–2077. [6] H. S . Ab bas and M. A. Gregor y , “Ocdm netw or k implementation of 1-d ooc and passiv e correlation receiv er , in Ar tificial Intelligence with Applications in Engineer ing and T echnology (ICAIET), 2014 4th Inter national Conf erence on . IEEE, 2014, pp . 311–316. [7] B . P , M. Mathimanir angan, V . Vinodh, K umar , and M. Bar akathulla, “h ybr id optical s witching in optical code division m ultiple xing netw or ks , Inter national Jour nal of Research in Engineer ing and T echnology , v ol. 3, no . 5, pp . 663–667, 2014. [8] R. S . Fy ath and H. M. M. Ali, “T r ansmission perf or mance of optical code di v i s i on m ultiple ac- cess netw or k based on spectr al amplitude coding, Jour nal of Emerging T rends in Computing and Inf or mation Sciences , v ol. 3, no . 3, pp . 444–455, 2012. [9] H. Be yr an v and and J . A. Salehi, “All-optical m ultiser vice path s witching in optical code s witched gmpls core netw or ks , Jour nal of Lightw a v e T echnology , v ol. 27, no . 12, pp . 2001– 2012, 2009. [10] A. E. F arghal, H. M. Shalab y , and Z. Ka w asaki, “Multir ate m ultiser vice all-optical code s witched gmpls core netw or k utilizing m ulticode v ar iab le-w eight optical code-division m ul- tiple xing, Jour nal of Optical Comm unications and Netw or king , v ol. 6, no . 8, pp . 670–683, 2014. [11] H. Be yr an v and and J . A. Salehi, “Multir ate and m ulti-quality-of-ser vice passiv e optical net- w or k based on h ybr id wdm/ocdm system, Comm unications Magazine , IEEE , v ol. 49, no . 2 , pp . s39–s44, 2011. [12] X. Li and L. Lu, “Gener al constr uction method of m ultilength optical or thogonal codes with arbitr ar y cross-correlation constr aint f or ocdma m ultimedia netw or k, Jour nal of Optical Com- m unications and Netw or king , v ol. 7, no . 3, pp . 156–163, 2015. TELK OMNIKA V ol. 15, No . 3, September 2017 : 1061 1071 Evaluation Warning : The document was created with Spire.PDF for Python.