CONFERENCE ON SIGNAL PROCESSING AND REAL TIME OPERATING SYSTEM (SPRTOS) MARCH 26-27 2011 COMO1O1-1 Comparison of LFSR and CA as Pseudorandom Number Generator Ashutosh Mishra*, Dr. Harsh Vikram Singh**, S.P. Gangwar** *Student (M.Tech), ** Astt. Prof. Dept. of Electronics Engineering, KNIT SULTANPUR Abstract Pseudorandom number generator (PRNG) is aKeyelementinstreamcipher.Thispaper comparestwotechniques:LinearFeedback ShiftRegister(LFSR)andCellular Automata(CA),usedforpseudorandom numbergeneration.BothLFSRandCAare analyzedbasedontheirconstructionand characteristics.AcomparisonofLFSRand CAispresentedtodemonstratetheir shortfallsandsuitabilitytocertain applications. Introduction Incryptography,astreamciphercombines theplaintextbitswithapseudorandom cipherbitstream(keystream),typicallyby anexclusive-or(XOR)operation.Ina streamciphertheplaintextdigitsare encryptedoneatatime,andthe transformationofsuccessivedigitsvaries during theencryption.In practice, the digits are typically single bits or bytes. A stream cipher makes use of much smaller andmoreconvenientkey128bits,for example.Basedonthiskey,itgeneratesa pseudorandomkeystreamwhichcanbe combinedwiththeplaintextdigitsina similarfashiontotheone-timepad. However,thiscomesatacost:becausethe keystreamisnowpseudorandom,andnot trulyrandom,theproofofsecurity associatedwiththeone-timepadnolonger holds: it is quite possible for a stream cipher tobecompletelyinsecure.Binarystream ciphersareoftenconstructedusingPRNG such as linear feedback shift register (LFSR) oradditivecellularautomat(ACA)in todays cryptographic scenario.LFSRscanbeimplementedinhardware, andthismakesthemusefulinapplications that require very fast generation of a pseudo-randomsequence,suchasdirect-sequence spreadspectrumradio.LFSRshavealso beenusedforgeneratinganapproximation ofwhitenoiseinvariousprogrammable sound generators. An ACA is a cellular automaton whose rule iscompatiblewithanadditionofstates. Typically,thisadditionisderivedfrom modular arithmetic. Additive rules allow the evolutionfordifferentinitialconditionsto becomputedindependently,thentheresults combinedbysimplyadding.Theresultsfor arbitrary starting conditions can therefore be computed very efficiently by convolving the evolution of a single cell with an appropriate convolutionkernel(which,inthecaseof two-color automata, would correspond to the set of initially "active" cells). Linear feedback shift register A linear feedback shift register (LFSR) is a shiftregisterwhoseinputbitisalinear function of its previous state. Theonlylinearfunctionofsinglebitsis XOR,thusitisashiftregisterwhoseinput CONFERENCE ON SIGNAL PROCESSING AND REAL TIME OPERATING SYSTEM (SPRTOS) MARCH 26-27 2011 COMO1O1-2 bitisdrivenbytheexclusive-or(XOR)of some bits of the overall shift register value. TheinitialvalueoftheLFSRiscalledthe seed,andbecausetheoperationofthe register is deterministic, the stream of values producedbytheregisteriscompletely determined by its current (or previous) state. Likewise,becausetheregisterhasafinite number of possible states, it must eventually enterarepeatingcycle.However,anLFSR withawell-chosenfeedbackfunctioncan produceasequenceofbitswhichappears random and which has a very long cycle. Additive cellular automata Cellularautomataevolveinstepandthe valueofnodedependsonthevalueof neighbors.AnadditiveAutomata(ACA) consistsofacollectionofcells/nodes formedbyflip-flopswhicharelogically relatedtotheirnearestneighborsusing XOR/XNORgates[5].Whenthevalueofa nodeisdeter-minedonlybyneighboring cells theACA is knownas one-dimensional linear CA. The logical relations which relate anodetoitsneighborsareknownasrules andtheydefinethecharacteristicsofan ACA.Therearemanyruleswhichcanbe used to constructan ACA register, the most popular being rules 90 and 150 illustrated in figure . =(Rule 90)

(Rule 150) Thenextstate(t+1)ofthenodeXiis determinedby thecurrentstateX(t)of neighboringnodesXi-1andX i+1forrule90 andnodesXi,Xi-1andXi+1forrule150.All the nodes of a CA register do not have to be implementedwiththesamerule,different nodescanemploydifferentrules.Thefirst and the last nodes of a CA register have only oneneighborunlikeallothernodeswhich havetwo,hencenormalrulescannotbe appliedhere[5].Onesolutionistoassume that the missing neighbor is fixed at logic 0 (nullboundarycondition).Theother solutionassumesthelastandfirstnodesto be neighbors and is connected using normal rules(cycliccondition).Connection between the end nodes (first and last nodes) introducesafeedbackloopinthecyclic boundarycondition;thismakesnull boundary condition a better choice. Comparison Building on the results of Serra, et.al [3] the consequencesofthesimilarity transformationbetweencellularautomata and LFSRs has been explored. By definition theLFSRobtainedbyasimilarity transformationofthetransitionmatrixofa CAhasthesamecharacteristicpolynomial astheoriginalCA.Ithasbeenshownthat thecharacteristicequationdeterminesthe recursionrelationamongthebitsinthe D Q Xi-1 D Q Xi D Q Xi+1 D Q Xi-1 D Q Xi D Q Xi+1 CONFERENCE ON SIGNAL PROCESSING AND REAL TIME OPERATING SYSTEM (SPRTOS) MARCH 26-27 2011 COMO1O1-3 outputbitsequenceofaCA.Thisimplies thatthesamelineardependenciesexistin theoutputbitstreamofaCAasinthe output of the similar LFSR. LFSRandCAarecharacterizedbytheir transitionmatrices,theanalysisofthese matricesalongwithsimulationsgivethe measureoftherandomnessinthepatterns generated;thesemeasuresshowthehigher randomnessofpatternsproducedbyCAs. Parallel patternsgenerated byLFSRs(using outputsfromdifferentnodesofanLFSR) have a strong correlation between each other duetotheshiftingofdata.Pattern generationinCAsdoesnotinvolveshifting of data. Thereisgreaterprobabilityofanerrorin LFSRbyaliasingcomparedtoACAdueto shiftingofdatainLFSR[1].IncaseofCA eachnodevalueisafunctionofthe neighboringnodesresultinginalower probabilityofanerror.Thepresenceof XORgatesinthefeedbackpathofan ExternalFeedbackLFSRandlackofa feedbackpathinanullboundarycondition results in higher operating speed for CAs. LFSR have a feedback from their end nodes; this means a redesign of the LFSR is needed if the pattern length has to be changed. This isnotthecasewithACA.ACAislogically connectedtotheironlytotheirneighbors andthereisnofeed-backforanACA employingthenullboundarycondition[5]. Therefore,thepatternlengthgeneratedby CAs can be easily changed by cascading the nodes. The regular structure of the nodes for CAmakesthemidealforCADtoolsby providingthemuchneededflexibilityin design. However, it is difficult to construct a maximum length sequence CAas compared toanLFSRwhichcanbeconstructedusing theprimitivepolynomialswhicharevery well documented.AnLFSRcanbeimplementedusingonlya fewXORgateswhereasaCArequiresat least one XOR gate for each node. This fact bringsupanobviousdraw-backofCA: Higherareaoverheadinvolvedin implementationofCAcomparedtoan LFSR. So, the designer has to pay a penalty ontheareaover-headbychoosingCAover LFSR. InACAthecommunicationisgenerally local,beingrestrictedtothenearest neighboringcellsandcellsareregularand topologicallyequivalenttooneanotherbut in the case of LFSR these property does not exist [2]. Followingtableshowsthesummaryof comparison CharacteristicsLFSRACA Performance Verygood incaseof internal feedback, poorfor external feedback Good-no feedback pathand maximum onXOR gate between nodeRandomnessof generated pattern Low- shiftingof bitcauses correlation between pattern Highno shiftingof bit CAD FriendlyNo- requires redesignfor changein pattern length Yes-node canbe cascaded easily Speed Lowerthan CA Higher than LFSR Error Probability Greater-by aliasingdue toshifting Lower-no shiftingof bit CONFERENCE ON SIGNAL PROCESSING AND REAL TIME OPERATING SYSTEM (SPRTOS) MARCH 26-27 2011 COMO1O1-4 of bit Conclusion LFSR has been well researched and provide a compact and simpler circuit for application using polynomial division and generation of patterns.Theirpopularityisowedtosimple andcompactdesignforcryptographic applicationtheyfinduseinpseudorandom numbergeneration.Butresearchhasshown someoftheshortfallsoftheLFSRcanbe overcomebytheuseofCellularAutomata. ACAuselargernodesascomparedtothe LFSRbutprovidemuchmorerandom patternsandcanbeeasilycascadedfor design flexibility.

References [1] M. Serra, T. Slater, J. C. Muzio & D. M. Miller,TheAnalysisofOneDimensional LinearCellularAutomataandTheir AliasingProperties,IEEEtrans.OnCAD, pp. 767-778, 1990. [2] Ph. Tsalides, T.A. York, A. Thanailakis, Pseudorandom number generators for VLSI systemsbasedonlinearcellularautomata, IEEProceedings-e,Vol.138,no.4,July 1991. [3]M.Serra,D.M.Miller,J.C.Muzio, LinearcellularautomataandLFSRsare isomorphic,Proc.Thirdtech.Workshop on new dir. For IC Testing, Halifax N.S. Pp. 213-223, oct. 1988. [4]PinoCaballero-Gil,AmparoFster-Sabater,OscarDelgado-Mohatar,Linear CellularAutomataasDiscreteModelsfor GeneratingCryptographicSequences, JournalOfResearchAndPracticeIn InformationTechnology,Vol.40,No.4, November 2008. [5]S.Nandi,B.K.Kar,andP.Pal Chaudhuri,TheoryandApplicationsof CellularAutomatainCryptography,IEEE Transactions on Computers, vol. 43, no. 12, December 1994. [6]S.Zhangt,R.Byrnel,J.C.Mutiot,D.M. Miller,WhyCellularAutomataAreBetter ThanLFSRsasBuilt-InSelf-Test GeneratorsforSequential-TypeFaults, IEEEInternationalSymposiumonCircuits and Systems, Vol. 1, 1994. [7]N.M.Thamrin,G.Witjaksono,A. Nuruddin,M.S.Abdullah,AnEnhanced Hardware-basedHybridRandomNumber GeneratorforCryptosystem,International ConferenceonInformationManagement and Engineering, 2009. [8]PaulH.Bardell,AnalysisofCellular AutomataUsedasPseudorandomPattern Generators,IEEEInternationalTest Conference 1990. CONFERENCE ON SIGNAL PROCESSING AND REAL TIME OPERATING SYSTEM (SPRTOS) MARCH 26-27 2011 COMO1O2-1

COST-EFFECTIVE MARKOV BASED SEGMENTATION S.Robinson Sobitha Raj Mrs.V.S.Bindhu M.E., Applied Electronics, Asst. Professor, EEE dept Noorul Islam University, Kumaracoil.Noorul Islam University, Kumaracoilrobins.singh@yahoo.co.inbindhuvs2003@gmail.com. AbstractImagesegmentationisanimportant preprocessingstepinasophisticatedandcomplex imageprocessing algorithm. In thispaper we are going tointroduceaproceduretominimizethe misclassificationcostwithclass-dependentcost.This procedureassumesthehiddenMarkovmodel(HMM) whichhasbeenpopularlyusedforimagesegmentation inrecentyears.WerepresentallfeasibleHMM-based segmenters(orclassifiers)asasetofpointsinthe receiveroperatingcharacteristic(ROC)space.Then, theoptimalsegmenter(orclassifier)isfoundby computing the tangential point between the iso-cost line withgivenslopeandtheconvexhullofthefeasibleset intheROCspace.Weillustratetheprocedureby segmentingaerialimageswithdifferentselectionof misclassification costs. Indexterms---Convexhull,hiddenMarkovmodels, image segmentation, iso-cost line, ROC convex analysis, ROC curve. I.INTRODUCTION Imagesegmentationextractsexplicitinformation about content, and it allows human observers to understand imagesclearly byfocusing specific regions of interest. For thisreason,itisoftenusedasaninitialprocedureto simplifyasophisticatedandcompleximageprocessing system.Segmentationoftenrequiressmoothboundary betweenregionsfordifferentclasses,andhiddenMarkov model(HMM)is possiblyoneofthemostpopularmodels for it [4], [8], [11]. The HMM assumes that the true hidden classMarkoviandependencyand,thus,hassmooth boundarybetweensegmentedregions.Thepopularlyused Markovmodelshavetwoparameters,whichwedenoteby and , where indicates the popularity of each class over imagesandtheotherindicatesthestrengthofspatial coherence.Theparametersareestimatedinsegmentation procedures or pre-decided by an expert. In real examples, the cost of misclassification can dependontheclasses.Forexample,incancerdiagnosis, misclassifyingtumorcellspaysmuchhighercostthan doingnormalcellstotumorcells;or,segmentingand detectingmilitarytargets,mistakenlydetectingtargetsto nontargetsmaycostmorethantheothertypeof misclassification.However,allexistingsegmentation proceduresdonottakeintoaccountthecostof misclassification, particularly the unequal cost that depends on the classes. Themainthemeofthispaperistodesigna classifier to minimize the expected cost of misclassification withclass-dependentmisclassificationcost.Wedenoteit asacost-effectiveclassifier.Thecost-effectiveclassifiers studiedmuchwiththenameROCconvexanalysisin machine learning literature [2], [3], [5]-[7] [9], [10], where targets to be classified are often independent to each other. However,ithasrarelybeenstudiedintheframeworkof imagesegmentation,inwhichtargetsaredependentfrom the Markovian model. In this paper, we introduce the ROC convexanalysisprocedureforHMM-basedimage segmentationtoconsiderclass-dependentmisclassification cost. TheROCconvexanalysisdrawsgreatattention inthemachinelearningsociety.Inatwo-classproblem (positive and negative class), the ROC curve (or set) is the plotoftheprobabilityoffalsepositivedecision(false positiverate,FPR)andthatoftruepositivedecision(true positive rate, TPR). The ROC convex hull analysis finds an optimumpointinanROCspacetominimizethe misclassification cost of classifier , which is defined as (1) Where impliesfalsenegativeratewhichisequal to.Oncemisclassificationcostisgiven,we have a family of iso-cost lines with slope On which costs of classifiers arethe same. Then, the ROC convex analysis finds the optimal classifier as the classifier whose(FPR,TPR)pairisthetangentialpointbetween these iso - cost line and the convex hull of the ROC curve. The ROC convex analysis requires the entire set of feasible classifiers. In the HMM based segmentation, the model has twounknownparameters whichareestimatedin segmentingtheimage.Fixingthesetwoparameters,not estimating, provides the set of classifier C (, ) s which are indexedby .TheROCset isthesetoferrorrates sofclassifier .Wedenoteitsconvex hull as . Theset shouldbecharacterizedtofindthe mostcosteffectiveclassifier,butitsnotcomputationally feasible. In this paper, we provide an alternative way which doesnotcomputetheentireset ,butweapproximateits boundary. We find that the tangential point between the iso costlinewithgivenslopeandexistsontheboundary of ,whichwedenoteas .Also,wefindthatisthe convex hull of the boundary of. Hence, we suggest tocomputeorapproximate .Toapproximatethe boundaryof ,wefixtheparameterbutestimatein segmentingthetargetimages.Wecompute the of classifiers ins. CONFERENCE ON SIGNAL PROCESSING AND REAL TIME OPERATING SYSTEM (SPRTOS) MARCH 26-27 2011 COMO1O2-2 Weapplytheproposedproceduretosegment aerialimageswithtwoclasses,man-maderegionand natural region, which could be used for target recognition andtracking.Here,thecostofmisclassifyingtargets(or man-maderegions)ishigherthanthatofmisclassifying non targets (or natural regions) to targets. The results show that,asthemisclassificationcostofman-maderegions increases, the HMM based segmenter tends to classify both man-made and natural regions to man-made ones. Thus, the rate of falsely classifying into man-made regions increases. Theremainderofthispaperisorganizedasfollows.We introduceROCconvexanalysisinsectionII.Westudyits applicationtoHMMbasedsegmentationprocedurein sectionIII.InsectionIV,Weapplytheproposed procedure to segment an aerial image. Section V concludes this paper. II. ROC CONVEX ANALYSIS Inabinaryclassifier,theROCcurve(orspace) plots two accuracymeasures of a classifier, FPRand TPR. SupposeweuseacontinuousclassificationscoreXto diagnose a certain disease, and we classifya subject into a disease(positive)groupifhis/herscoreXishigherthana givencriticalpoint;otherwise,weclassifyhim/herinto non disease group (negative) group. The ROC curve plots a series of (FPR, TPR) pairs produced from different choices of . TheoptimalROCcurveistheoneproducedby theclassifierswhichhasthemaximumTPRgivenFPR. The optimal ROC curve has several geometrical properties includingconvexity.Supposeitisnotconvexonan interval , where a and b correspond to critical values andinthewaythattheFPRatandisa andb,respectively.Then,forthediagnosiswithcritical values ,wefindabetterdiagnosticsystem (classifier)whichhasthesameFPRbuthigherTPRby randomlychoosingbetweentwodiagnoseswithcritical values and.Thus,theconvexhulloftheobserved ROCcurvesrepresentstheROCcurveofthesetof potentiallyoptimalclassifiers.WeletinROCspacebe thesetof sofallclassifiersweconsider,and bethesetoftheirrandommixtures.Then,isthe smallest convex region which contains .Thegoalofthispaperistofindtheclassifier whichminimizeshecost(1)amongclassifiersin . Since , (2) Werepresentthecostfunctions sin the ROCspace.Then,theinterceptoftheiso-costline(2)is and,asinFig.2.1,thepointtominimizethe cost is the tangential one between the line with slopeand the convex area . In summary, proposed procedure to find the cost-effective classifier has the following steps: S1)wedefineasetofclassifiers,sayC,tobeconsidered. In case of HMM-based segmentation, this becomes a set of classifierscorrespondingtoeachoftwoparametersIsing modelprior.WewillseedetailsonHMM-based segmentation in the next section. S2) we compute FPRs and TPRs of classifiers in C and plot them. This defines the region in the ROC space. Further, we get the ROC convex hull region of. S3) we find the tangential point between the line with slope in(2)andtheROCconvexhullregion .Theclassifier correspondingtothefoundtangentialpointisthemost cost-effective classifier. Fig.2.1. Illustration of ROC convex hull analysis III. COST MINIMIZATION IN HMM-BASED IMAGE SEGMENTATION In this section, we find the HMM-based classifier tominimizethe(expected)costofmisclassification(1) (ECM).Togetbetterunderstandingoftheproblem,we beginwithHMMwithoutspatialcoherencesuchasGMM [1]. The model assumes that the testing image is composed ofmanyindependentsub-blocks,sayXsfork=1,2,, K. The model has unknown parameter which indicates the prevalenceofeachclass.Weletbetheratioofprior probabilityofpositiveclass(classP)tothatofnegative class(classN).Then,given,theoptimalclassifierto minimize the ECM among all classifiers is the maximum a posteriori(MAP)classifierthatassignsxtoclassN,for each k, if (3)

Wherefn(x)andfp(x)istheprobabilitydensity function of class N and class P, respectively. In GMM, both fn(x)andfp(x)aredensityfunctionsoftheGMM. Suppose we denote the classifier in (3) given as C (), and theircollectionasF.Wefurther findthatthiscollectionD isinvarianttothecost.Inotherwords,wealwayshave same collection regardless of what we choose. Inpractice,tofindtheoptimalclassifier,wesettobean arbitraryfixednumber,andweconsiderthesetof conditionally optimal classifiers which minimize the cost of misclassificationgiven>0.WeletDbethesetoftheir FPRsandTPRsintheROCspace.Followingthestepsin theprevioussection,wegetDandfindthetangentpoint and between the line (2) and D. Now,wemovetotheHMM-basedsegmenter. TheHMMbasedsegmenterassumesthatthehidden processisfromtheMarkovrandomfieldhavingtwo CONFERENCE ON SIGNAL PROCESSING AND REAL TIME OPERATING SYSTEM (SPRTOS) MARCH 26-27 2011 COMO1O2-3 unknownparametersand;theparameteristhe parameterthatrepresentsthemagnitudeofmagnetization of the random field which implies the dominance of class P againstclassNincommonwords;itisalsorelatedtothe ratio of the prior probability of P to N; is the parameter to measure the strength of spatial coherence. For example, the HMGMMmodel[11]usesthegeneralizedbond percolation (BP) model. Let Z = {} with Z = 1 or 1: Z = 1 andZ=1impliestheclassPandN,respectively.The generalized BP assumes that the probability of Z is (4) Where (Z) ( (Z)) is the number of concordant (dis- concordant)adjacentpairswhichareneighborstoeach other. Here,) is the partition function that is Fig.3.1. ROC convex Hull for HMGMM Classifiers Asinthespatiallyuncorrelatedmodel,weconsiderthe collection F of MAP classifier C , which, given and ,assignstheobservedimage i,j=1,2,,n}to .WeletCbetheMAP classifier,givenand,andFbethecollection of .WefurtherdefineDand similarly.Tofind theoptimalclassifier,again,wecomputethetangential point between the line (2) and , the convex hull of D in the ROC space. InHMM-basedsegmenter,evaluationofDis computationallyquiteintensiveinpractice.Here,we introduceasuboptimalbutfastalgorithmforit.Many algorithmsarestudiedfromdeterministicannealingto simulatedannealingtoMarkovchain MonteCarlomethod tofindtheMAP.Theyoftenassume=0andestimate, the parameter of spatial coherence in finding the MAP. For example,in[11],theparameterisestimatedusingthe maximum likelihoodestimate (MLE) along with the Gibbs sampler.TheMLEis(Z)|X}/totalnoof edges),andweapproximatetherighthandsideofthe equation using Monte Carlo method to get the estimate.Welet betheestimateofgiven,andwe approximate the boundary points of with FPRs and TPRs of bymoving.Asstatedbefore,knowing boundary issufficienttofindthetangentialpoint betweeniso-costlinesand .Theapproximatedboundary, denoted by ,isacurvefrom(0,0)to(1, 1). Finally,is approximated by the convex hull of, and the optimal classifier is found using the procedure in Section II. IV. EXPERIMENTAL RESULTS In this section, we apply the procedure in Section IIItosegmenttheaerialimagewithHMGMMwith generalized BP model in (4). The aerial image is composed ofmanysub-blockswhichareclassifiedintonatural regions or man-made regions. We call the natural sub-blockasnegativeandman-madeoneaspositive. TheHMGMMmodelhastwoparametersandwhich reflectstheoverallportionofman-madeandspatial coherencebetweenadjacentsub-blocks,respectively.Each introducesaclassifier,say ,andapointof (FPR,TPR)intheROCspace.WeletbeDthecollection ofalltheses.Theinputimagetakenis illustrated below. Fig.4.1. Input image Intheexperiment,we varyfrom0.35to 2.85 by 0.1, and we get the empirical FPRs and TPRs from the segmented results . We then compute the convex hull of ( s ofs to approximate the convex hullofD.Wedenotetheconvexenvelopeas .Fig.2 plotssandtheirconvexenvelopeintheROC space.Wefindtheoptimalclassifiertominimizethe misclassification cost (1) is the one that corresponds to the tangentialpointbetweenthe(2)and .Theis piecewise linear function with 11 supporting points (points thathavechangesinslope).Eachpointonisthe optimal classifier to minimize the cost for a specific choice of . Fig.4.2. Segmented image Now,wereportsomedetailsoftheanalysisfor=0.5,1, and 2. The noise mixed image is given by, segmented imageCONFERENCE ON SIGNAL PROCESSING AND REAL TIME OPERATING SYSTEM (SPRTOS) MARCH 26-27 2011 COMO1O2-4 . Fig.4.3. Noised image Here,thehighvalueofimplieshighcostof misclassificationofmanmadesub-blocks(tonatural ones). Thus, as.increases, the segmenter tends to classify bothnaturalandman-madeblockstoman-made ones. The enhanced image is given by, Fig.4.4. Enhanced image The analysis graph is given below: Fig.4.5. Analysis graph Finally,theoriginalHMGMMin[11]istheclassifier with .ItsFPRandTPRis0.22and0.97, respectively, which is close to the (approximated) boundary of. Some computation shows that the HMGMM is close to the optimal forbetween 0.16 and 0.22. V. CONCLUSION Weproposeaproceduretofindtheoptimal classifiertominimizethemisclassificationcostbasedon HMM.Thecostfunctiontominimizeallowsunequalcost thatdependsonthetrueclass.Theoptimalclassifieris given as a classifier whose error rate pair (FPR, TPR) is the tangentialpointbetweentheiso-costline(2)andthe convexhulloffeasiblesetoferrorratesbyHMM-based segmenters.Weapplytheproceduretosegmentanaerial imagewithdifferentselectionofunequalmisclassification cost. The procedure in this paper does not depend on a specifictypeofimagebutisapplicabletoawiderclassof imagesandHMM-basedclassifiersforthem.Thesame procedurecanbeextendedtootherapplicationssuchas cancerdiagnosisbasedonMRIimagingor3-Dtemporal images.Apossibledifficultyintheseextensionisfrom obtaining , the boundary of the ROC convex space. The currentsuboptimalalgorithmcouldbemoreexhaustiveas thenumberofHMMparametersincreases,anditmight provide rough approximation if the size of training data sets is not enough. This would be our next step to move. VI.REFERENCES [1] A. Aiyer, K. Pyun, Y. Huang, D. B. OBrien, and R. M. Gray, Lloyd clustering of Gauss mixture models for image compressionandclassification,SignalProcess.:Image Commun., vol. 20, pp. 459485, Jun.2005. [2]H.BlockeelandJ.Struyf,Derivingbiasedclassifiers for improved ROC performance,Informatica, vol. 26, pp. 7784, 2002. [3]R.CaruanaandA.Niculesen-Mizil,Anempirical comparisonofsupervisedlearningalgorithms,inProc. 23rd Int. Conf. Machine Learning, 2006, pp. 161168. [4]P.A.Devijver,Segmentationofbinaryimagesusing thirdorderMarkovmeshimagemodels,inProc.8thInt. Conf. Pattern Recognition, 1986, pp. 259261. [6]P.A.FlachandS.Wu,Repairingconcavitiesinroc curves,inProc.19thInt.JointConf.Artificial Intelligence, 2005, pp. 702707. [7] D. M. Gavrila, J. Giebel, and S. Munder, Vision-based pedestrian detection: The protector system, inProc. IEEE Intelligent Vehicles Symp., 2004, vol. 1318. [8]J.Li,A.Najmi,andR.M.Gray,Imageclassification by a two dimensional hidden Markov model, inProc. Int. Conf.Acoustics,Speech,andSignalProcessing,1999,pp. 33133316. [9]R.C. Pratiand P.A.Flach,Roccer:Analgorithmfor rulelearningbasedonROCanalysis,inProc.19thInt. Joint Conf. Artificial Interlligence, 2005, pp. 823828. [10]F.ProvostandT.Fawcett,Robustclassificationfor impreciseenvironments,Mach.Learn.,vol.42,pp.203231, 2001. [11]K. Pyun,J.Lim,C.S.Won,andR. M.Gray,Image segmentationusinghiddenMarkovGaussmixture models,IEEETrans.ImageProcess.,vol.16,no.7,pp. 19021911, Jul. 2007. noised imageenhanced imagePerformance Comparison of VANET Routing ProtocolsNITIN SHARMADepartment of Computer Science and EngineeringMNIT, Allahabad, India nsharma283@gmail.comMobile:+91 8861 6365 11POOJA RANIDepartment of Information TechnologyITM, Sector-23A, Gurgaon, India me_pooja@msn.comMobile: +91 9899 7470 33ABSTRACTVehicular Ad-hoc Network (VANET) is a collectionof communicationvehicles, moving in different directions. The vehicles forma communication group to disseminate desired information. Various routing protocols are implemented in a VANET, each having benefitsand shortcoming in the domain ofimplementation. Inthispaperperformanceofthree routing protocols, namely Ad hoc On-Demand Distance Vector Routing (AODV),Destination Sequenced Distance Vector (DSDV)and Dynamic Source Routing (DSR) iscompared for different parameters. The protocols are simulated on Network Simulator-2(ns-2).Index TermsRouting protocols, Simulation, VANET, Vehicle-to-Vehicle Communication, WAVE. I - INTRODUCTIONVANETisanad-hocnetworkformedbetween vehicles as per their need of communication.In order todevelopaVANETeveryparticipating vehicle must be capable of transmitting and receivingwireless signals uptorangeof three hundred meters. VANET communication range is restricteduptoonethousandmetersinvarious implementation [1]. The performance of a VANETremains optimumwithinonethousand meters and beyond that it is not feasible to communicate among vehicles because of high packet loss rate [2],[3]. Finding optimum path is typical task for dynamic protocols as management of vehicle movement is quite complex. There is a need to update entries in the route finding node [4]. VANET is not restricted up to Vehicle-to-Vehicle communication, it takes benefits of road side infrastructure that can also participate in communication between vehicle [5], but in this paper the main focus is on Vehicle-to-Vehicle communication.There are various challenges for VANET suchas highspeedof vehicle, dynamic route finding, building, reflecting objects, roadside objects, other obstacles in path of radio communication, different direction of vehicles, concernabout privacy, authorizationofvehicle, security of data and sharing of multimedia services.High speedofvehiclerequiresregular updateof routingtablewhereas dynamicroute findingwouldresult intohightimelossbefore static communication [6].Various user groupamongVANETare getting popular, mostly it is used in traffic management agencies, highway safety agencies, law enforcement agencies and emergency services.The main service provided through VANETareGPSnavigationsystem, electronic payment of toll tax, authenticity of vehicle without human intervention, traffic message. Internet access, broadcastingof trafficscenario and multimedia streaming.As most of the properties of a Mobile Ad-hocNetwork(MANET) are commonwiththe VANET,various MANETroutingprotocolsare usedinVANET[7]. Basicdifferencebetween MANET and VANET is that under VANET PERFORMANCE COMPARISON OF VANET ROUTING PROTOCOLS1movement of vehicleisat highspeedandless random as compared to MANET, existing MANETrouting protocols are not compatible with VANET. VANET routing protocols are broadly divided into two categories Table Driven protocols and Source Initiated on Demand protocol.In TableDriven protocoleach vehicle maintains a table of neighborhood vehicles within its communication range and any change in vehicle position is updated regularly. In Source Initiated on Demand protocol, firstly source vehiclebroadcasts aquerytofindrouteupto node gets a route up to the destination node. The destinationrouteis repliedbacktothe source node via same path. In this paper, the performance of following three routing protocols are compared:Ad Hoc on-Demand Distance Vector Routing (AODV)It is a Source Initiated on Demand routing protocols used in VANET. In this protocol every vehiclemaintains arouteinformationof every vehicle. It uses sequence number concept to acknowledge the entry update time and time stampbasedconcept for tableentry. If atable entryisnot usedwithinacertaintimelimit, it will bedeletedfromtableandif thereis any breakage in linking with a vehicle toanother vehicle, route error (RERR) packet is forwarded so that vehicle route is effectively updated in the routing table.Destination Sequenced Distance Vector (DSDV)It is Table Driven routing protocol which is used inVANETandis basedonclassical Bellman-Ford algorithm. Initially every vehicle broadcasts its own route tables to its neighbor vehicles. The neighbor vehiclesupdateroutingtablewiththe help oftwotypeof packets- FullDumpPacket and Incremental Packet. Full Dump Packet contains information about every participating vehicle in the VANET. These packets are transmitted periodically after a long time interval. Incremental Packet contains updatedchangein vehicle position since last Full Dump Packet. These packets are transmitted periodically in short interval of time and are stored in additional table. Routes are selected with the latest entry in the table. DSDVis goodfor networks where location of nodes are less dynamic. If position of a vehicle changes very often, its performance goes down because more Full Dump Packets are needed to sent in the network, resulting into wastage of bandwidth.Dynamic Source Routing (DSR)It is Source Initiated on Demand routing protocol used in VANET and is based on link state routing algorithm. When a vehicle wants to communicate data to another vehicle, firstly it finds route up to that vehicle. For route discovery,source vehicle initiates arouterequest (RREQ) packet inthe networkandother nodes forward theRREQby changing their name as sender. Finally when RREQ packetreachestothe destinationvehicle or toavehicle havingpathtothe destination vehicle, a route reply (RREP) packet is unicasted tothe sendernode.Ifthe reply isnot received, thesourcevehiclerestartsaggressivediscovery of route up to the destination vehicle.The simulation setup is discussedinthe next sectionandresultsarepresentedinsectionIII. Section IV describes conclusion and future scope.II - SIMULATION SETUPTheProtocol stackof VANETconsists of two combination of standards Dedicated Short Range Communication (DSRC) and Wireless Access in Vehicular Environment (WAVE). DSRC contains three layer Physical Layer, Media Access Control (MAC) Layer, and Logical Link Control (LLC) Layer. The layers make communication possible in wireless environment. WAVE lies on top of the DSRC layer. WAVE is alsoknownas IEEEP1609andis usedas a standard for communication, which is further classified as follows: IEEEP1609.1:It is standard for resource manager, which defines the key resources for data-flow and other resources as well as the command messages.PERFORMANCE COMPARISON OF VANET ROUTING PROTOCOLS2 IEEEP1609.2:It is standard for security services in application and controlling messages, encoding and decoding of messages. IEEE P1609.3: It is the standard for network services and involves the key features of network and transport layer such as IP addressing,informationpassing andsecurity while switching services. IEEEP1609.4: It is thestandardfor MAC layer support andincreases communication capacity of node. Also it provides multichannel operation to maintain Quality of Service (QoS) with high speed.The ns-2 is used to evaluate performance of routing protocols in VANET. It is a network simulator andis alsowidelyusedfor VANET related simulation work [8]. Ns-2 provides a standard for IEEE 802.11p simulation in the form of tcl scripts. Also various types of communication patterns, traffic scenarios and resources are available with ns-2.The simulation grid is 500m x 500m. in which the initial position of every vehicle is specified at thestartingtimeof simulationbycallingC++ object and the destination of every vehicle is set after a certain time interval. The vehicle approach their destination with variable speed. In simulation, threeinput parametersareprovided namely, speed of vehicle vary between 10MPH to 100MPH, number of vehicles vary between 20 to 100 and distance between vehicles very between 20mto100m. forthetrafficpurpose, Constant Bit Rate (CBR) traffic with a fixed packet size is used. The vehicle antenna is omni directional and its transmission range is 150m. The channel data rate is 2Mbps. IEEE 802.11p standard is used for vehicle-to-vehicle communication, which is available is ns-2 are wireless support [9]. It follows most of the WAVE standard, Nakagami model is used for radio propagation, which is the best model for WAVE environment[10]. While implementationaroutingprotocol inns-2the behavior of queue and its adaptability criteria are carefully considered.In the simulation work, the available standard and data units are followed. Ns-2 produces output in the form of trace file which is furtherprocessedbyshell scriptingtocalculate thedesiredparameter. Shell scriptingandawk are widely used to process trace files. III - SIMULATION RESULTThe comparison of routing protocols are done on the basis of following parameters:Average Number of Hop Count:It is the number of vehicles running between source and destination and it signifies error in the network.Timeto Live(TTL) is decided onthe basis of Hop Count, which helps in avoiding the congestion in the network.Average Jitter Rate:It isthedelaybetweentwoconsecutivepacket delivery atanode.QualityofService (QoS)of the network is measured by Average Jitter Rate.Packet Delivery Ration:The fractional of total packets received at a node to the packets sent by the node. It is associated withtheQoSandbandwidthutilizationinthe network.Routing Overhead:ItistherationofnumberofMACpacketsand total number of packets sent, Routing Overhead increases with increases in vehicle density.Throughput:Throughput is the sum of data to all the nodes in the system during a period. In a time interval the throughput reflects the bandwidth utilization.For each of the above parameters, various traffic scenarios are simulated by changing the number of vehicles, distance between vehicles and vehicle speed. The effect of these changes in the routing protocols are analyzed andresults are shown below.PERFORMANCE COMPARISON OF VANET ROUTING PROTOCOLS3 Fig 1:Jitter Rate with various vehicle distances Fig 2:Packet Delivery Ratio with various vehicle distancesFigure 1 shows the Jitter Rate variation at different distances between vehicles. It is observed that DSRJitter Rate remains lowest throughput the interval of observation, while DSDVandAODVshowcomparativelyhigher Jitter rate.Figure 2 shows the Packet Delivery Ratio at various distances between vehicles. It is observed that DSRPacket DeliveryRatio remains high during entire duration of observation. AODV Packet Delivery Ratio in between other two and DSDVPacket DeliveryRatiois lowest among three.Fig 3:Routing Overhead with various vehicle distancesFig 4:Throughput with various vehicle distances Figure 3 shows the Routing Overhead at various distances betweenvehicles. It is observedthat DSRRoutingOverheadis lowest, AODVand DSDVRoutingOverheadremain similar from 20mto100mbut after 100mAODVRouting Overhead is going down.Figure 4 shows the Throughput with various vehicle distances is shown. Clear inference from graph DSR Throughput is highest. AODV Throughput remains inbetweenother twoand DSDV Throughput is lowest among all three.PERFORMANCE COMPARISON OF VANET ROUTING PROTOCOLS4In the simulation work, we compared the performance of AODV, DSDVand DSRon fifteen criteria as shown in Table I. It is observed that no protocol performs good in all the conditions. One protocol performs well in specific domain area and others in different domain area. DSR had better performance than other two protocols becauseof its dynamicroutefinding nature that decreases Hop Count as every time it increases portability of finding optimum route. It also increases Packet Delivery Ratio. Routing Overheadis alsodecreasebecause indynamic network it is found that most of the maintain any routingtable that benefits inHopCount, it is increase Throughput of network by reducing regular update packet.TABLE IPARAMETER COMPARISION OF ROUTING PROTCOLSParameters Variables Routing ProtocolsAODVDSDV DSRHOP COUNTDist. b/w Vehicles L M HHOP COUNTNo of Vehicles L H MHOP COUNTSpeed of Vehicles L M HJitter Rate Dist. b/w Vehicles M H LJitter Rate No of Vehicles M H LJitter Rate Speed of Vehicles M H LPacket Del RatioDist. b/w Vehicles M L HPacket Del RatioNo of Vehicles H L MPacket Del RatioSpeed of Vehicles H L MRouting Overhead Dist. b/w Vehicles H M LRouting Overhead No of Vehicles H M LRouting Overhead Speed of Vehicles H M LThroughput Dist. b/w Vehicles M L HThroughput No of Vehicles H L MThroughput Speed of Vehicles H L MAbbreviation:L- Low Value, M- Medium Value & H- High ValueAODV performs better than DSDV in most of the parameters because of its time limit in table entry usages. That benefits in Throughput, Packet Delivery Ration and some times that also benefits in Throughput Routing Overhead. DSDV Performanceisworst amongthreeprotocols, it uses updation of table and depends on table the unnecessary decreases Packet Delivery Ratio, increasesnumberofHopCountsanddecreases Throughput.IV - CONCLUSTION AND FUTURE WORKOnthebasis of simulationresults presentedin Table I for different traffic scenario and for three protocols, performance of DSR has been found to bebetter thanthat of AODVandDSDV. The performance results are summarized for Highway, Urban and Freeway traffic scenarios in Table II.Fromthe parameter values characterizing the three traffic scenarios. DSR is found suitable for Highway and Freeway traffics, whereas AODV is suitable for Urban traffic scenario.Thusit canbeconcludedthat asingleprotocol doesnt give best performance in all traffic scenarios. Since traffic scenarios change throughput the day, some part of hybrid adaptive protocol would give better performance.TABLE IIDIFFERENT TRAFFIC SCENARIOTraffic Scenario Freeway Highway UrbanDistance b/w VehiclesSmall Large smallDensity of VehiclesLow Low HighHigh Speed of Vehicles No Yes NoSuggested Protocol DSR DSR AODVPERFORMANCE COMPARISON OF VANET ROUTING PROTOCOLS5REFERENCES[1] Yi wang. Akram Ahmad, Bhaskar Krishnamachari and Konstantinos Posunis, IEEE 802.11p Performance Evaluation and Protocol Enhancement.IEEE International Conference on Vehicular Electronics and Safety, 2008. [2] Bo Xu, Aris Ouksel and Ouri Wolfson, Opportunistic Resouce Exchange in Inter-Vehicle Ad-hoc Networks,International Conference on Mobile Data Management, 2004.[3] Lin Yand,Jindua Guo and Ying Wu,Piggyback CooperativeRepetitionforReliableBroadcastingofSafety Message in VANETsIEEE Internatitional Conference on Consumer Communication and Networking, 2009, pp-1-5[4] Katrin Bilstrup, Elisabeth Uhlemann, Erik G.Strom, andUrbanBilstrup,OntheAbilityofthe 802.11pMACMethodandSTDMAtosupport Real-Time Vehicle-to-Vehicle Communication.EURASIP Journel onWirelessCommunicationandNetworking 2009.[5] RainerBaumann, SimonHeimilicherandMartin May, Towards Realistic Mobility Models for VehicularAd-hoc Networks.IEEEInternational Conference of Mobility Networking for Vehicular Environment, 2007.[6] ValeryNaumov, Rainder BaumannandThomas Gross,An Evaluation of Inter-Vehicle Ad-hoc Networking Based on Realistic Vehicular races,ACM International Symposium on Mobile AdHoc Networking and Computing, 2006.[7] D. Rajini Girinath and Dr. S. Selvan,Data Dissemination to regulate vehicular traffic using HVRPin urbanmobilitymodel,International Journel of Recent Trends in Engineering, 2009[8] Victor Cabrea, FranciscoJ. Ros andPedroM. Ruiz,SimulationbasedStudyofcommonIssueson VANET Routing Protocols,IEEE Vehicular Technology Conference, 2009.[9] Djamel Djenouri, WassimSoualhi and Elmalik Nekka,VANETs Mobility and Overtaking: An Overview,International Conference onInformation and Communication Technologies, 2008, pp1-6[10] Christoph Sommer, Isbel Dietrich and Falko Dressler, Realistic Simulation of Network Protocols in VANET Scenarios. International Conference on Mobile Networking for Vehicular Environments, 2007PERFORMANCE COMPARISON OF VANET ROUTING PROTOCOLS6Greedy Grid Scheduling Algorithm in Static EnvironmentPOOJA RANIDepartmentof Information TechnologyITM, Sector-23 A, Gurgaon, IndiaE-mail: me_pooja@msn.comMobile: +91 9899 7470 33NITIN SHARMADepartment of Computer Science & EngineeringMNNIT, Allahbad, IndiaE-mail: cs0802@mnnit.ac.inMobile: +91 8861 6365 11Abstract Gridschedulingisatechniquebywhichthe userdemands aremet andtheresources areefficiently utilized. The scheduling algorithms are used to minimize thejobswaitingtimeandcompletiontime. Most of the minimization algorithms are implemented in homogeneous resource environment. In this paper the presented algorithm minimize average turnaround time in heterogeneous resource environment. This algorithmis basedongreedy approachwhichis usedinstatic job submission environment where all the jobs are submitted at same time. Taken all jobs independent the turnaround time of eachjobis minimizedtominimizetheaverage turnaround time of all submitted jobs. Keywords- greedy, grid, heterogeneous, high performance computing, scheduling. I. INTRODUCTIONUsing the distributed resources to solve the applications involvinglargevolumeofdataisknownasgridcomputing [1], [2]. There exists many tools to submit jobs on the resources which have different computational power and are connected via Local Area Network (LAN) or Virtual Private Network(VPN). Themainchallengeingridcomputingis efficient resource utilization and minimization of turnaround time. Theexistingsystemmodel consistsofthewebbased gridnetworkplatformwithdifferent management policies, formingaheterogeneous systemwherethecomputingcost and computing performance become significant at each node [3], [4].In grid computing environment, applications are submitted for use of grid resources byusers fromtheir terminals. The resources include computingpower, communicationpower and storage. An application consists of number of jobs; users want to execute these jobs in an efficient manner [5]. There are two possibilities of submission of jobs/data on resources; in one of them, job is submitted on the resources where the input data is available and in the other, on the basis of specific criteria, resource is selected on which both job and input data are transferred. This paper uses second approach, wherein the job is submitted on a scheduler and data on a resource identified by the scheduler. A resource in existing algorithms is selected randomly, sequentially or according to its processing power [2], [6], [7]. In this paper the proposed algorithmchooses a resource on the basis of processing power, job requirement and time to start at that resource.The next section describes details about system model. Section 3 describes the proposed scheduling algorithmin static job submission environment. In section 4 the experimental details and the results ofexperiments are presentedwithcomparisonof someexistingalgorithms. In section 5 conclusions and suggestions for future improvements are proposed. II. SYSTEM MODELA grid is considered as the combination of multiple layers. In our model the whole systemis composedof three layers (Fig.1). The first layer is the user application layer in which the user authentication is done and jobs are submitted to the scheduler bytheuser. Thesecondlayer containsscheduler and GIS. The scheduler schedules jobs among various resources after taking resource statusinformationfrom GIS. The second layer is connected through a VPN to user. VPN provides additional securityandonlyauthorizedusers can access services. All the resources reside in third layer where user's jobs areexecutedwhicharealsoconnectedthrough VPN. Fig. 1: Layered architecture.III. PROPOSED ALGORITHM The existinggrid schedulingalgorithms are basedonthe speed of resources [6], [7]. Each resource of layer 3 (Figure.1) has different processingpower andall theresources of layer 3 are connected via homogeneous communication environment in which the communication delay between 1scheduler and resources is assumed constant, also the jobs are assumed to submitted on layer 1 having different job requirement.An algorithm is proposed in this paper which is suitable for static job submission in heterogeneous resource environment connected to the scheduler through homogeneous communication environment. Greedy approach is used to solvethejobschedulingproblem. Accordingtothegreedy approach A greedyalgorithm always makes the choice that looks best at that moment. That is, it makes a locally optimalchoiceinthehopethat this choicewill leadtoaglobally optimal solution" [8]. The proposed algorithm uses the similar approach; it takes every job as independent of each other and eachofthemisscheduledonaresourcetogiveminimum turnaround time for that job.The overall turnaround time of all the jobs is thus minimized. The parameters used in this algorithm are as follows: A set of resources, R = {R1, R2, R3,........, Rn}.Ji = The submitted ith job.Arr_timei = Arrival time of job Ji.Proc_powerj = Processing power of resource Rj.Strt_timej = Estimated time at which a job starts execution at resource Rj.Job_reqi = Length of job Ji.Schd_valueij=Expected turnaround time of ithjob at jth resource.Min = The minimum of Schd_valueij among all resources.Res_id= Current selected resource id having optimum turnaround time.The algorithm used to schedule a job is given as follows: GREEDY_SCHEDULE/*The users submit their jobs on the scheduler.*/For all resource Rj /*Initialize the start time at resources.*/ Strt_timej=0.0 End For /*The jobs are stored in a queue Q.*/ Insert all the jobs Ji in Q While Q is not empty do Delete the job Ji SUBMIT_NEW_JOB UPDATE_STATUS Advance the Q pointer End WhileEnd GREEDY_SCHEDULE The scheduler uses SUBMIT_NEW_JOB algorithm to find the best suited resource that minimizes the turnaround time. The turnaround time is calculated on the basis of expected completion time of a job. The detailed SUBMIT_NEW_JOB algorithm is as follows:SUBMIT_JOB Min = For every resource Rj/* Calculating the expected turnaround time*/Schd_valueij = Strt_timej + (Job_reqi/Proc_powerj)If Min is greater than Schd_valueij ThenMin = Schd_valueijRes_id = RjEnd IfEnd ForSubmit the job Ji to Res_id resourceSubmit the input data of Ji job to Res_id resourceEnd SUBMIT_NEW_JOBOnce the scheduler submits a job to a resource, the resource will remains for sometimeinprocessingof that job. The UPDATE_STATUSalgorithmisusedtofindout whenthe resource will be available to process a new job. The UPDATE_STATUS algorithm is given below: UPDATE_STATUS/* Res_id is the resource on which the job Ji is submitted. j is the index of resource on which the job Ji is submitted and Rj = Res_id*/Strt_timej = Schd_valueijEnd UPDATE_STATUS Theabovepresentedalgorithmhasthetimecomplexityof O(n) foreach job,where n isthe numberofresources. The above algorithm required additional space to store the resorces current status for availability.IV. EXPERIMENTAL RESULTSThe GridSim simulator [6] is used to simulate the algorithms. The GridSim toolkit is used to simulate heterogeneous resourceenvironment andthecommunicationenvironment. Theexperiments areperformedwiththreealgorithms. The algorithms are RandomResource SelectionandEqual Job Distribution and Proposed Algorithm. The input data is taken to be the same for all the three algorithms. The simulation is conducted with three resources which are shown in Table 1. TABLE 1: Resources with their architecture and processing power.Resource R0 R1 R2Architecture Sun Ultra Sun Ultra Sun UltraOS Unix AIX UnixProc_power(in MIPS)48000 43000 54000Theschedulersubmitsthesejobsonresourcesaccordingto thesealgorithms. Thealgorithms arepresentedonebyone with their simulation results. A. Random Resource SelectionInthis algorithmthescheduler contacts GIStoobtainthe resource information and then it chooses a resource randomly [7]. The job is submitted on this chosen resource. This algorithm is very simple to implement and has less overhead on the scheduler. The bar chart (Fig. 2) shows the turnaround time of different jobs. The completion time is a time at which 2the result of a job is available.After simulation the average turnaround time is found to be 20105.65seconds and all the jobs are completed at the 64420.25th second. Fig. 2: Jobs and turnaround time using Random Resource Selection.B. Equal Job DistributionIn Equal Job Distribution we firstly calculate the total length of all the jobs andthendistributetheselengthequallyoneveryresource. The main notations which are used in the formula are as follows:L = Total length of all the jobs taken together.Proc poweri = Processing power of resource Rj.tProc power = Total processing power of all resources.Loadi = Load assigned on resource Rj.Theformulausedtocalculatethejobdistributionisgiven bellow:Loadi = L* (Proc poweri /tProc power)The turnaround time of each job is shown by the bar chart in Fig. 3. Experimental results show that the average turnaround timeis17968.55secondsandthelast result isoutputedat 39000.22th second. Equal Job Distribution reduces the average turnaround time by10.62%and it takes less time in comparison to the Random Resource Selection to give all the results. Fig3: Jobs and turnaround time using Equal Job Distribution.C. Proposed Algorithm In Proposed Algorithm, the scheduler finds the resource information with the help of GIS and calculates the approximatecompletiontimeofthisjoboneveryresource. Using these values the scheduler chooses a resource which has the minimum of completion time and submits that job on this resource. Theturnaroundtimeofeachjobisshowninbar chart in Fig. 4. Through this algorithm the average turnaround timeofthesejobsis17208.77secondsandall thejobsare completed at41840.88thsecond. The Proposed Algorithm further reduces the average turnaroundtime by4.22%as compared with Equal JobDistribution. The completiontime of all jobs takes some more time than Equal Job Distribution algorithm. Fig.4: Jobs and turnaround time using Proposed Algorithm.V. CONCLUSION AND FUTURE WORKThe proposed scheduling algorithm reduces the average turnaround time of all submitted jobs. The considered environment executedthejobsondifferent resourceswhich are geographically distributed. It is observed that the Proposed Algorithmreduces the average turnaround by4.22%with Equal Job Distribution (as shown in Table 2). Thealgorithm uses meta-scheduler where resource failure is not considered. TABLE 2: Algorithms with their average turnaround time and completion time.Algorithms Average Turnaround Time(In Seconds)Completion Time(In Seconds)Random Resource Selection20105.65 64420.25Equal Job Distribution 17968.55 39000.22Proposed Algorithm17208.77 41840.88 3REFERENCES[1] Ammar H. Alhusaini, Viktor K. Prasanna,C.S. Raghavendra, UnifiedResourceSchedulingFrameworkfor Heterogeneous Computing Environments", in Proceedings of theEighthHeterogeneousComputingWorkshop, SanJuan, Puerto Rico, pp. 156-165, 1999.[2] N. Muthuvelu, J. Liu, N. L. Soe, S.r Venugopal, A. Sulistio and R. Buyya, ADynamic Job Grouping-Based Scheduling for Deploying Applications with Fine-Grained Tasks on Global Grids", Proceedings of the 3rd Australasian Workshop on Grid Computing and e-Research (AusGrid 2005), Newcastle, Australia, 41-48, January 30 - February 4, 2005. [3] I.Foster, C Kesselman, The Grid: Blueprint for a new computing infrastructure", Morgan Kaufmann Publishers, San Francisco, USA, 1999.[4] R. Buyya, D. Abramson, J.Giddy, Nimrod/G: An Architecture for a Resource Management and Scheduling System in a Global Computation Grid", International Conference on High Performance Computing in Asia-Pacific Region(HPCAsia2000), Beijing, China. IEEEComputer Society Press, USA, 2000.[5] Cong Liu, Sanjeev Baskiyar and Shuang Li, A General Distributed Scalable Peer to Peer for Mixed Tasks in Grids", High Performance Computing HiPC 2007,ISBN:978-3-540-77219-4, 320-330, 2007. [6] Rajkumar Buyya,Manzur Murshed, GridSim: a toolkit for the modeling and simulation of distributed resource management and scheduling for Grid computing",Technical Report, MonashUniversity, Nov. 2001. Toappear inthe Journal of Concurrency and Computation: Practice and Experience (CCPE), pp. 1-32, Wiley Press, May 2002.[7] Volker Hamscher, Uwe Schewiegelshohn, Achim Streit, Ramin Yahyapour,Evaluation of Job-SchedulingStrategies for Grid Computing", in 1st IEEE/ACM International WorkshoponGridComputing(Grid2000), Berlin, Lecture Notes in Computer Science (LNCS), Springer, Berlin, Heidelberg,NewYork,pp. 191-202,2000.[8]Cormen TH, LeisersonCE, Rivest RL, Introductiontoalgorithms 2nd edition", MIT and McGraw-Hill Book Company, Boston Massachusetts, cp. 16, 370-403, 2001.4CONFERENCE ON SIGNAL PROCESSING AND REAL TIME OPERATING SYSTEM (SPRTOS) MARCH 26-27 2011 COMO1O6-1 A Comparative Study of IP and MPLS Based Networks Sandeep singhSchool of Information and Communication Technology, Gautam Buddha University Greater Noida, U.P. er.sandeep_vhdl@yahoo.comM.Altamash Sheikh School of Information and Communication Technology, Gautam Buddha University Greater Noida, U.P. altsheikh@gmail.com

Abstract IntodaysworldofnetworkingIProuters buildup and uses the routing tables and these packets of data are routed on the IP prefixes which are already stored in routing tables of routers. These routers fallows the lookups to find out the next destination of packets. This processgivesseveraloverheadslikedelay, jitterandtrafficdrops.simpledatacan acceptmuchdelaybuttherealtimetraffic needs guarantied QoS with low jitter , delay andpacketdropetc.optimumperformance ofIP based network can not beachieved by justapplyingtheQoSbutMPLScan providebettersupporttoQoS.Herethe performancefactorofthenetworkcanbe improvedwithMPLSinabetterwayas compared to just IP routing. Keywords:QOS,TOS,MPLS,IPBased routing , Traffic Engineering. IntroductionWheneverthereistrafficto flownormalroutingfacesseveralissues relatedtocongestionanddelay.Routerneedstolookupthereroutingtableevery timesothismakesrouterverybusyduring theprocessing.Routersdecidesthepath basedonitsIPprefixesandthispathis alwaystheshortestpathbasedonshortest pathfirstalgorithm(SPF).Thisresultsthe unnecessaryloadonaparticularlinkand gives traffic drop. When we are dealing with therealtimetrafficitneedsguaranteed service[4] , [5]. IngeneralMPLScanprovidesacombined and better environment for the network with capabilityofsupportingQOSandtrafficengineering.Intheemerginganddevelopingtrendoftechnologysmall,mid sizedandlargecompaniesarefrequently movingforchangingthereinfrastructure. Thesecompaniesaremigratingtowards Traffic engineering[5] , [6] , [7]. Related work :BackgroundofQOS:Attheinitialstage whenIPheaderwascreatedtherewasthe spaceinthatheadercalledTypeOfService (TOS)bytewhichwasbasicallyforfuture perspective.Itwaswellknownthat technologiesweregrowingveryrapidlyas noonecouldthinkofit.ThisTypeof service(TOS)fieldwasreservedforfuture purpose of improving QOS[17].In the early development of internet the applicationsrunningoveritdidnottake muchcareaboutTOSbecauseduring forwardingofIPdatagramthedeviceslike CONFERENCE ON SIGNAL PROCESSING AND REAL TIME OPERATING SYSTEM (SPRTOS) MARCH 26-27 2011 COMO1O6-2 routerdidnotinterpretthisfielduntilit gives any kind of service parameters. In this TOSbytethreefieldswerereservedforIP precedenceandotherthreefortypeof service[12], [13].HighertheIPprecedence datawouldtreatedmorepriorityandlower theIPprecedencedatawouldbedealtwith low priority . During the initial stage of internet it was notneededinthenetworkbecause commercialization was not so much but as it grownuptheneedofdifferenttypeof service were also grown up[16]. ClassificationoftrafficbasedonQOSClassificationofdifferentflowsoftraffic intoclassesmaydependonseveral parameters.Basicallysimilardatapackets areconsideredinthesameclass[18].The most common way to classify the flows may dependonheaderfieldssuchasIP precedence and DSCP fields[15]. One of the header of TCP is also used for classification of traffic by identifying the length of coming packets or by checking the MAC address of both senders and receivers address[14]. When the traffic is classified three main classes comes out (i) High priority for sensitive traffic : The trafficlikerealtimerequiressomespecial treatmentlikenodelayandlessjitter.This comesinHighPrioritytrafficandmost commonexampleofsuchtrafficis VoIP[10]. (ii)Bestefforttraffic:Whentrafficdoes notneedanydelayrelatedguarantees.In this case packets are send after high priority trafficandthebestexampleofthiskindof traffic is email[9] , [10].(iii)Lowprioritytraffic:Thiskindof traffic does not need any guarantee , priority orsignificanceanditisnotrequiredto deliverthetrafficatitsappropriate destination . This kind of traffic includes the codes generated by a hacker or spam kind of mails[18]. MPLS: MultiProtocolLabelSwitchingisa new technology for getting the fast and rapid transferofdata.Highspeedconvergenceof scalabilityispossibleinMPLSbased network[1],[2],[3].MultiProtocolLabel Switchinghasbecomethefirstchoicefor packettransportation,whichfulfillsthe several requirements of next generation[19].Several service providers aregoing to deploytheMPLSonacommonplatformto achieveconvergenceofexisting technologies like X.25 ; ATM / Frame relay(FR) and best effort services. It provides the transmission resources by providing the Diff Servetoanetwork.MPLSprovidesthe intelligentroutingandgivestheimportant improvementsintheswitchingperformance ofthenetworkinsteadofwhatnetworks architectureis.MPLSalsoprovides scalabilitylikeadvantagesforVirtual PrivateNetwork(VPN)andmaintainend-to-end QOS [1] , [3] , [5]. InnormalIPbasednetworksIP routersperformsdestinationbasedrouting. when they have to sent trafficrouter always usesthesimpleshortestpathfirstalgorithm tocomputetheshortestpathbetween CONFERENCE ON SIGNAL PROCESSING AND REAL TIME OPERATING SYSTEM (SPRTOS) MARCH 26-27 2011 COMO1O6-3 themselves and destination[8] . This distance canbehopcountforprotocolssuchas RoutinginformationProtocol(RIP)orleast totalmetrics.Theydontcarewhether alternatepathexistinthenetworkornot. Thiscreatesaproblemwhiledeliveringthe trafficbecausewhenthetrafficexceedsthe limitoflinkitgivestrafficdrop[11].Butin MPLSTrafficengineeringitenablesthe operatortobuildatrafficengineeringtunnel for safely delivery of data. MPLS Architecture MPLSisanIETFstandardwhichprovides Label-swapping-basedforwardinginthe presenceofroutinginformation[20].It basicallyconsistofsomeprincipal components : (1)Control:Controlcomponentusesa Label distribution protocol to maintain label forwarding information for all destination in MPLS networks. (2)Forwarding:Theforwarding componentswitchespacketsbyswapping labelsusingthelabelinformationcarriedin the packet. Figure 1. MPLS Architecture (3)Ingresslabeledgerouter:Thisis basicallyanedgerouterofMPLSnetwork which classify packets and add MPLS labels to the packets. (4)Labelswitchrouter:Itisanetworks coreMPLSenabledrouterwhichforward thepacketsbasedonlayer2MPLSlabels and later removes the label before sending it to the egress label edge router. (5)Egresslabeledgerouter:Egress label edgerouterswitchestheunlabeledpackets based on its destination IP address. ProblemStatementandProposed Solution : Theprimarygoalofthispaperistoexpose howtheMPLSsupportsQOSneeds.Italso highlighttheseveralissuesandindepth knowledge regarding it. (i)HowMPLShelpstoprovideQOS demanding traffic . (ii)ComparisonofMPLSwithtraditional technologies like IP routing. (iii) Practical implementation of MPLS with IP based routing in simulation environment. CONFERENCE ON SIGNAL PROCESSING AND REAL TIME OPERATING SYSTEM (SPRTOS) MARCH 26-27 2011 COMO1O6-4 Figure 2. Network Diagram of MPLS Domain Thepacketdroprateisincreasedinthe besteffortservicebecausepacketarrivalis nottreatedwithanydifferentiation.Dueto fillthebuffer,networkdeviceslikerouter andswitchesarefilled[24].Thepackets weregoingtodropsoreceiverrequestsfor retransmissionwhichgivesunwanteddelay. Realtimetrafficlikevoiceandvideo conferencingcannotstayinaqueuefor longtime.Errorrecoveryisalsoneededto check the data to ensure whether the content ofpacketissameasitwassentoriginally during the transmission [11]. QualityofserviceforMPLSbased networkisnotverymuchdifferfromthe wayasanyotherIPnetwork.Inthisway [22]theQualityofservicestrategymay divided into four stages :- (I)Classificationoftrafficisconsideredat theinitialstageinprovisionofQOSatthe edgeofMPLSdomain.Thetrafficis classifiedonthebasisofpreference[25]. Edgeroutersaregenerallyresponsiblefor such task but switches can also install at the edge for classifying traffic. (II)Polarizationoftraffichelpsthe important traffic to be forward first based on itssensitivityandimportance.Thisis achievedbyclassofserviceforagiven flow.MPLSprovidesdifferentcategories basedonprioritylikegold,silver,and bronze[26]. (III)Managementofbandwidthisanother big issue for accurate delivery of traffic. It is assumedthatanamountofbandwidthis neededtopassthelowprioritytraffic[25], [26].Realtimetrafficisalwaysgiven priorityandsomebandwidthisreservedfor itstransmissioninsuchawaythatwhenit exceedsthethresholditwouldbedeniedin order to pass the rest of the traffic. (IV)Trafficcanbeshapedinsuchaway thatitremainsinthelimitsofavailable bandwidthpool[22].Forthispurposetraffic maybecompressedordecompressed. Routeritselfperformthistaskof compressionanddecompression.Resources areallocatedinthenetworkbyapplying admissioncontrolonalllabelswitch paths[21]. Simulation and Results : Thesimulationenvironmentiscreated inlaboratorywheretwoscenariosarebeing consideredinordertotesttheworkingof MPLS.Infirstscenariothesimpleand generalnetworkisbuildwithatraditional IProutingthenthesamenetworkis configuredwithMPLSandresultswere obtained.Theobtainedresultsshowsthat theMPLSnetworkgivesbetteroutputin termsofperformanceandthroughputas compared to traditional IP routing. Policies for IP based Networks Class Matching CriteriaDSCP PrecedenceCritical RTP 46 07Web SMTP , HTTP 18 03Interactive ICMP , OSPF , EIGRP26 5Default ClassAny00 00 CONFERENCE ON SIGNAL PROCESSING AND REAL TIME OPERATING SYSTEM (SPRTOS) MARCH 26-27 2011 COMO1O6-5 Table 1. Policies for IP Network Table 2. Precedence Policies of IP Network TheQualityofServiceisintroducedin bothscenarios.Normallywhenordinary trafficistobetransferitdoesntneedof anyguaranteeofservice.Factorsof performancedegradationlikeJitter,delay anddroppingofpacketsdidnotharmthis data[23].Butrealtimetrafficistransferred in consideration of more sensitively for such factors. Policies for MPLS based networks Class Matching Criteria MPLS EXPCritical Precedence 7 5Web Precedence 3 3Interactive Precedence 5 4Default Class Precedence 0 0 Table 3. Policies for MPLS Network Table 4. Policies for MPLS Network (Critical Class) Results : The above two scenarios produces the results related to the performance of MPLS and IP routing. following statistics were obtained by issuing pkt-seq-drop-stats, show delaystats,show jitter-stats commands. Here we use a PAGENT router. PAGENT router contains the Cisco IOS which is able to generate the automatic traffic. Statistics of IP Networks Sent Receive DroppedDrop 688 688 000Minimum Maximum AverageJitter 0.000002 0.030636 0.000342Delay 0.017585 0.048495 0.017951 Table 5.O/P Statics of IP Network Statistics of MPLS based networks Sent Received DroppedDrop 755 755 000Minimum Maximum AverageJitter 0.000005 0.004931 0.000222Delay 0.017692 0.022794 0.017928 Policies for IP based Networks Class Matching CriteriaBandwidth %Average RatePrec7 Precedence 750 240000/120000Prec5 Precedence 535 160000/80000Prec3 Precedence 315 60000/30000Default ClassPrecedence 0-- 800000/400000Policies for MPLS based networks Class Matching Criteria Bandwidth %MPLS EXP 5 60 %CONFERENCE ON SIGNAL PROCESSING AND REAL TIME OPERATING SYSTEM (SPRTOS) MARCH 26-27 2011 COMO1O6-6 Table 6. O/P Statics of MPLS Network (1)PacketforwardinginMPLSisdepends onlabelsratherthenIPprefixes,which reduces overhead of processing of routers. (2) It has been observed that features of load balancingofMPLSremovestheunwanted congestion from the network. (3)PracticalscenarioprovedthatQOS works better in MPLS than IP routing. (4)Largeroutingtableswerebeing generatedbytherouters(IP)whichmakes IP router slower. (5)InIProuting,theinteriorgateway protocol(IGP)usedtheshortestpathto forwardthepacketsbetweeneachhop,but inMPLStunnelingremovessuchoverhead of load balancing (6)InMPLSdroppingofpackets,delay related problemsand issues of jitter are very nominal in comparison to IP based routing. ConclusionThis research focus on practical comparison ofMPLSandIPbasednetworkingwhich helpstounderstandtheconceptsand workingofMPLS.Guaranteedqualityof serviceisneededtotransfertherealtime trafficoverthenetwork.InMPLSbased networkQOSgivesbetterresultsas compared to IP networks regarding surety of datatransfer,delayandjitteretc.Delay, jitter and results related to packet drops were obtainedbyissuingcommands.Thenthe networkwasreconfiguredforMPLS networkandagainsomeparameterswere calculated for MPLS based networks. Thestatisticscollectedfromboththe networks and MPLSgives better result than normalIProutednetworks.Packetofdata faces longer delays in IP based networks and otherfactorswerealsoimprovedinMPLS basednetworks.Allthenecessary configurations of routers have been recorded andtwodifferentscenarioswerecompared. ThenroutingtablesofIPbasednetworks and Cisco Express Forwarding (CEF) tables of MPLS networks have been obtained. AcknowledgementThisresearchworkwassupportedby Prof(Dr.)SanjayJasolaundertheSchoolof InformationandCommunication TechnologyinGautamBuddhaUniversity, Greater Noida. References[1].LawrenceJ,Designingmultiprotocol label switching networks, Communications Magazine,IEEE,Volume39,Issue7,July 2001. [2].Kuribayashi,Tsumura,OptimalLSP selection method in MPLS Networks, Communications,ComputersandSignal Processing, IEEE acific Rim Conference on 22-24 Aug. 2007. [3].MaheshK.P,YadavS.V,Charhate M.E,TrafficAnalysisofMPLSandNon MPLSNetworkincludingMPLSSignaling ProtocolsandTrafficdistributioninOSPF andMPLS,FirstInternationalConference on Emerging Trends in Engineering and Technology 2008. [4].DavidApplegate,MikkelThorup, LoadoptimalMPLSroutingwithN+M CONFERENCE ON SIGNAL PROCESSING AND REAL TIME OPERATING SYSTEM (SPRTOS) MARCH 26-27 2011 COMO1O6-7 labels,AT&TLabsResearchShannon Laboratory180ParkAvenueFlorhamPark, 2003. [5].LucDeGhein,MPLSFundamentals, Cisco Systems, Cisco Press 800 East 96th StreetIndianapolis,ISBN:1-58705-197-4, 2007. [6].BenjaminTang,AhmetA.Akyamac, Chi-HungKelvinChuandRamesh Nagarajan,MPLSNetworkRequirements andDesignforCarriers:Wirelineand WirelessCaseStudies,TelecommunicationsNetworkstrategyand planning symposium, pages 1-6, Nov 2006. [7].MuzammilAhmadKhan,Quantitative Analysis of Multiprotocol Label Switching (MPLS),StudentConfrence,Volume1, Issue 16-17, page(s): 56-65, Aug 2002. [8].LoaAndersson,StewartBryant,"The IETFMultiprotocolLabelSwitching Standard:TheMPLSTransportProfile Case,"IEEEInternetComputing,vol.12, page(s):69-73, Aug, 2008. [9].MaheshK.P,NjulataYadavS.V, CharhateM.E,TrafficAnalysisofMPLS andNonMPLSNetworkincludingMPLS SignalingProtocolsandTrafficdistribution in OSPF and MPLS, 2008. [10].SrinivasVegesnaQualityofService By Published by Cisco Press, 2001 ISBN 1578701163, 9781578701162 [11].B.E.Nichols,Differentiatedservices intheInternetCarpenter,Proceedingsof theIEEE,Volume90,Issue9, page(s):14791494, Sept 2002. [12].LeonardoBalliache,NetworkQoS UsingCiscoHOWTO,April2003v0.1 PublishedonMay1,2003,v0.2May15, 2003, v0.3 September 10, 2003 [13].Trafficconditioningfactors http://www.cisco.com/en/US/tech/tk543/tk757/technologies_tech_note09186a00800949f2.shtmlhttp://en.wikipedia.org/wiki/ IP_traceback#Packet_marking, http://en.wikipedia.org/wiki/Traffic_policing ,http://en.wikipedia.org/wiki/Traffic_shaping#Traffic_Classification(AccessDate January 12, 2009) [14].CarterHorneyforNuntiusSystems, QualityofServiceandMPLS(White Paper),Inc.13700AltonPkwy.,Suite154-266 Irvine, CA 92612 949.295.0475 voice www.nuntius.com [15].P.Prabagaran&Joseph B,Experiences with Class of Service (CoS) TranslationsinIP/MPLSNetwork, Proceedingsofthe26thAnnualIEEE ConferenceonLocalComputerNetworks, ISSN:0742-1303, page(s):243, 2001. [16].Ji-FengChiu,*Zuo-PoHuang,*Chi-WenLo,*Wen-ShyangHwangandCe-KuenShieh,AnApproachofEnd-to-End DiffServ/MPLS QoS Context Transfer in HMIPv6Net,AutonomousDecentralized Systems, Eighth International Symposium, ISBN:0-7695-2804,page(s):245-254,Mar 2007. [17]. Sundeep .B.Singh and Girish.P.Saraph, DiffServ over MPLS: Tuning QOS parametersforConvergedTrafficusing Linux Traffic Control, Indian Institute of Technology Bombay, Powai, Mumbai India. [18].HongyunMan,LinyingXu,ZijianLi, LianfangZhang, END-TO-END QOS IMPLEMENTBYDIFFSERVAND MPLS,ElectricalandComputer CONFERENCE ON SIGNAL PROCESSING AND REAL TIME OPERATING SYSTEM (SPRTOS) MARCH 26-27 2011 COMO1O6-8 Engineering,volume2,page(s):641-644,May 2004. [19].InternationalWorkshop NGNT,DiffServandMPLSTmea Dreilinger,BudapestUniversityof TechnologyandEconomics,Departmentof TelecommunicationandTelematics,High Speed Networks Laboratory. [20].F.LeFaucheur,L.Wu,S.Davari,P. Vaananen,R.Krishnan,P.Cheval, J.heinanen, Multi-Protocol Label Switching (MPLS) Support of Differentiated Services, RFC 3270, May 2002. [21].JohnBartlettandRebeccaWetzel BCR,QOSoverMPLStheComplete Story,P2PforCommunications:beyond filesharing,Volume36,Number2, February 2006. [22].VictoriaFineberg,QoSSupportin MPLSNetworksMPLS/FrameRelay Alliance, White Paper May 2003. [23].LIMing-huiandXizJing-bo, ResearchandSimulationonVPN NetworkingBasedonMPLS,Wireless Communication,Networkingandmobile Computing ,page(s): 1- 4, Oct 2008. [24].HiroshiYamada,End-to-End PerformanceDesignFrameworkofMPLS VirtualPrivateNetworkServiceacross AutonomousSystemBoundaries, TelecommunicationsNetworkstrategyand planningSymposium,page(s):1-6,Nov 2006 [25].Labeldistribution: http://www.cisco.com/en/US/docs/ios/mpls/configuration/guide/mp_mpls_overview_ps6350_TSD_Products_Configuration_Guide_Chapter.html(lastaccesseddate10April 2009). [26].JingWu,DelfinY.Montuno,Hussein T.Mouftah*, Improving reliability of label distributionprotocol,Proceedingsofthe 26th Annual IEEE Conference on Local ComputerNetworks,ISSN:0742-1303, page(s) 236, 2001. SandeepSinghhasreceivedhis B.Tech.degreeinElectronicsand CommunicationEngineeringfrom Chaudhary Charan Singh University, Meerut in 2008. He is a M.Tech. student at Gautam BuddhaUniversity,GreaterNoida.Hehas beenworkedasafacultyofDr.K.N.Modi InstituteofEngineeringandTechnology, Modinagar, Ghaziabad.He is working as an AssistantProfessoratSobhasaria Engineering College, Sikar, Rajasthan. He is alifetimememberofIACSIT,Singapore andmemberofCSI.Hehaspublisheda paperinnationalconferenceon communicationnetworkandsecurity, Alwar,Rajasthan.Hisresearchinterest includesAd-hocwirelessnetworks,Mobile IP,QualityofService,MPLSbased networks and Traffic Engineering. CONFERENCE ON SIGNAL PROCESSING AND REAL TIME OPERATING SYSTEM (SPRTOS) MARCH 26-27 2011 COMO1O6-9 M.AltamashSheikhhas receivedhisB.Scdegreein ElectronicsEngineeringin 2007fromAnnamalai University,TamilNadu.He receivedhisM.Sc.degreeinElectronics CommunicationEngineeringfromJamia MiliaIslamia,NewDelhiin2009.Heisa M.Tech.studentatGautamBuddha University,GreaterNoida.Hisresearch interestincludes,mobileIP,Sensor Networks,Ad-HocNetworks,&Bluetooth Technology. CONFERENCE ON SIGNAL PROCESSING AND REAL TIME OPERATING SYSTEM (SPRTOS) MARCH 26-27 2011 COMO1O7-1 INTEGRATED SERVICES DIGITAL NETWORK AJAY BHATIA (Dept. Electronics & Communication) Punjab College of Engineering and Technology, Mohali ABSTRACT IntegratedServicesDigitalNetwork(ISDN) isadigitalcommunicationstechnologythat enablesasmallbusinessoranindividualto connectdirectlytoboththeInternetand otherusers.ISDNprovidesastandard interfaceforvoice,fax,video,graphicsand data- all on a single telephone line.ISDN is asetofcommunicationsstandardsfor simultaneous digital transmission ofvoice, video, data, and other network services over the traditional circuits of the public switched telephonenetwork.Itwasfirstdefinedin 1988inthe CCITT redbook.ISDNenables thetelephonenetwork,whichwasbuiltfor traditional analogcalls, to carry information digitallyathigherspeedsandwithoutthe errorsofthetraditionalanalogsystemor POTs (Plain Old Telephone Services) lines.As a result, ISDN has the ability to deliver a widerangeofdesirableapplicationsin education,healthcare,business,and multimedia.Inlightofthissubstantially increasedfunctionality,ISDNcanbethe basisforanewtypeofusefulmediumfor Internetaccessthatis:reasonablypriced, readilyavailable,andalsoallowsforthe deliveryofinformationathighspeeds,and providesformanytypesofnewand previouslyunavailableapplications.The technicaldesignofISDNismotivatedby theprimaryaimofmodernizingtraditional telephonenetworks.Theuseofdigital switchingandtransmissiontechnology withintheISDNfacilitatesthetransmission ofdigitaldatabesidesthedigitizedvoice.KEYWORDS ISDN, ISDN traffic, voice/data networks, value added services, video conferencing. INTRODUCTION Integrated ServicesDigital Network(ISDN) iscomprisedofdigitaltelephonyanddata-transportservicesofferedbyregional telephonecarriers.ISDNinvolvesthe digitalizationofthetelephonenetwork, whichpermitsvoice,data,text,graphics, music, video, and other source material to be transmittedoverexistingtelephone.The emergenceofISDNrepresentsaneffortto standardizethesubscriberservices, user/networkinterfaces,andnetworkand internetwork capabilities. ISDN applications includehigh-speedimageapplications, additional telephone lines in homes to serve thetelecommutingindustry,high-speedfile transfer,andvideoconferencing.Voice service is also an application for ISDN. This papersummarizestheunderlying technologiesandtheservicesassociated withISDN.Also,therearetwosignificant CONFERENCE ON SIGNAL PROCESSING AND REAL TIME OPERATING SYSTEM (SPRTOS) MARCH 26-27 2011 COMO1O7-2 developmentsdrivethecommercialization ofISDN:digitalswitchinganddigital transmission. It is estimated that nearly all of theaccesslinesandtrunkroutesinthe UnitedStateswillbedigitalwithinafew years and a third of these will be fiber optic cable.PulseCodeModulation(PCM)and TimeDivisionMultiplexing(TDM)are importantrelatedtechnicaldevelopments enablingdigitalswitchingandtransmission. ISDN STANDARDS TheISDNBasicRateInterface(BRI) serviceofferstwoBchannelsandoneD channel(2B+D).BRIB-channelservice operatesat64kbpsandismeanttocarry user data; BRI D-channel service operates at 16kbpsandismeanttocarrycontroland signalinginformation,althoughitcan support user data transmission under certain circumstances.TheDchannelsignaling protocolcomprisesLayers1through3of theOSIreferencemodel.BRIalsoprovides forframingcontrolandotheroverhead, bringingitstotalbitrateto192kbps.The BRIphysical-layerspecificationis InternationalTelecommunicationUnion TelecommunicationStandardizationSector (ITU-T)(formerlytheConsultative CommitteeforInternationalTelegraphand Telephone[CCITT])I.430.ISDNPrimary RateInterface(PRI)serviceoffers23B channelsandoneDchannelinNorth AmericaandJapan,yieldingatotalbitrate of 1.544 Mbps (the PRI D channel runs at 64 Kbps).ISDNPRIinEurope,Australia,and otherpartsoftheworldprovides30B channels plus one 64-Kbps D channel and a totalinterfacerateof2.048Mbps.ThePRI physical-layer specification is ITU-T I.431.

ISDN SIGNALING Therearetwodifferenttypesofsignaling usedinISDN.Forcommunicatingwiththe local phone company, ISDN uses the Digital SubscriberSignalingSystem#1(DSS1). DSS1defineswhatformatthedatagoesin on the D-channel, how it is addressed, etc. It also defines message formats for a variety of messages used for establishing, maintaining, CONFERENCE ON SIGNAL PROCESSING AND REAL TIME OPERATING SYSTEM (SPRTOS) MARCH 26-27 2011 COMO1O7-3 anddroppingcalls,forinstanceSETUP messages,SUSPENDandRESUME messages,andDISCONNECTmessages. OnceyourDSS1signalmakesitthethe phonecompany,theirownsignalingsystem takesovertopassthecallinformation withintheirsystem,andbetweenother phonecompanies.SignalingSystem#7 (SS7)issupposedtobeusedforthis.SS7 definesacommunicationsprotocol,and formatssimilartoDSS1,howeverSS7is designedinabroader,moregeneralway. DSS1 is specific to ISDN, however SS7 will handlethesignalingneedsofISDNaswell asotheroldersignalingsystemsand (hopefully)willadaptwelltofutureneeds. OneimportantfeatureofSS7isproviding CCS.Thismakesitharderformalicious usersofthephonenetworktoputoneover on the phonecompany.It also improves the service,forinstancebyofferingfaster connectionestablishment.Olderequipment stilllooksforthesignalinginformationin thesamechannelasthevoice,intheeighth bit of each piece of voice data. ISDN SWITCHING WithpureISDN,switchingisthe departmentofthephonecompany. TraditionalphoneservicesisCircuit Switched Voice (CSV). The voice data goes throughseveralswitchesbeforereachingits finaldestination.Forpoint-to-pointdata connections,youneedCircuitSwitched Data (CSD) - the exact same thing with data insteadofvoice.IfCSVareused,theyare freetorouteyourcallthroughanytypeof switch,eventheoldanalogswitches.The digitalchannelmayalsobesharedwith otherchannels,inthemomentswhenthere issilenceonthephoneline.Andthedigital partsofaCSVcallcangothroughnoisy switchesthatmightcreateanundetected error.ThereareactuallyastandardsetofcombinationsdefinedforsettingupBRIs. ThesearecalledNationalISDNInterface Groups(NIIGs),sotherewillbealimited menuofofferingsavailable.Onecanget bothB-channelsfordata,oroneforvoice andtheotherfordata,oroneforvoiceand the other for either voice or data. In order to facilitatethis,NorthAmericanphone companies use an optional part of theISDN standardtoidentifyeachTE1orTAyou use.ThephonecompanyassignsaService ProfileIdentifier(SPID)toeachofthese devices, and one has to manually enter them intoeachdeviceuse.Thephonecompany thenstoresthisdatasomewhere,andwhen oneconnectthemachinetothenetwork,it sendsitsSPIDtothenearestphone companyswitchwhichidentifieswhattype ofconnectionthedeviceneedsand (therefore)howtorouteitscalls. Presumably,theSPIDshavetorefertoa configuration that matches one of the two B-channelsyouhave.Bytheway,theSPIDS arearbitrarynumbersthatrefertodata storedbythephonecompany.Thephone Companyoftenincludesthephonenumber in the SPID for their own convenience. One oldertypeofphonecompanyswitch,a DMS-100,wasimproperlydesignedwith respecttothestandardsrelatingtoSPIDs. ThisswitchmisguidedlyassignsoneSPID toeachB-channelthatisuse,ratherthanto each device. Therefore ifthe nearest switch is a DMS-100, one will only be able to hook up two devices to the CPI, rather than eight. Ifoneisonlygoingtobehookingupa singledevicetotheISDN(i.e.settingitup inapoint-to-pointconfiguration,onemight notneedaSPIDatall,asthephone companycanidentifytheISDNlineasone particulartype,fulltime.Thisdependson whatequipmenttheyhave-theoldDMS-100 switch will still requireto have a SPID. CONFERENCE ON SIGNAL PROCESSING AND REAL TIME OPERATING SYSTEM (SPRTOS) MARCH 26-27 2011 COMO1O7-4 ISDN FRAME Protocol discriminator Theprotocolusedtoencodethe remainder of the encoder. Length of call refrence value Definesthelengthofthenextfield.the callrefrencemaybeoneortwooctets longdependinguponthesizeofthe being encoded. Flag Set to zero for the messages sent by the partythatallocatedthecallrefrence value,otherwise set to one. Call refrence value Anarbitraryvaluethatisallocatedfor thedurationofthespecificsession whichidentifiesthecallbetweenthe devicemaintainingthecallantthe ISDN switch. Message type Definestheprimarypurposeofthe frame.themessagetypemaybeone octetortwooctets.whenthereismore than one octet ,the first octet is coded as eight zeroes. ISDN AND VALUE ADDED SERVICES TheusualreferencetoanISDNimpliesan integrationofservicesonthesenseofthe transportofdigitalizedvoiceanddata,the formerwithtelephonicsignalingandthe lattereithergainingaccessviatelephonic signaling and then employing data signaling orviadatasignalingonly.Circuitand message,synchronousandasynchronous, packet-alltraffictypesaretobe accommodated.Packetswitchingis consideredbysomeasavalue-added service,butbyothersasatransport mechanism, which can be combined within acommunicationsnodewithcircuittraffic. Thepotentialforcombiningthe telecommunicationsandthecomputer technologiesbecomesapparentwhenthe deliveredsignalisnotjustareplicaofthe inputsignal,buthasbeenprocessedina usefulmanner.Itcanbedebatedon operational,economical,technicaland politicalgroundswhethersuchprocessing shouldbeincludedwithinthepublicand private network itself, treated and offered as anetworkservice,orwhetherthese functions should be external to the network, providedatitsperiphery. CONFERENCE ON SIGNAL PROCESSING AND REAL TIME OPERATING SYSTEM (SPRTOS) MARCH 26-27 2011 COMO1O7-5 ISDN VIDEOCONFERENCING Today,amodernorganizationisconnected viaanIPnetworkresemblingahub betweendedicatedconnectionsinmultiple remoteoffices.EmployeesusetheIP networkfortext-basedcommunications, meetregularlyviatelephoneandfrequently traveltoremoteofficesforface-to-face meetings.Whenattheirdesks,employees have both Intranet services and access to the publicInternet.Videoconferencingisbeing exploredasatechnologythatcanimprove groupandindividualcommunicationsand therebyaffectboththesestrategic objectives. CONNECTIVITY OPTIONS Inthepast,networkoptionsfor videoconferencingweresimpler.Heavy usersplacedendpointsonacommon privatenetworkofswitchedcircuits leasedfromtelecommunicationscarrier. Mostcompaniespreferredtoconnect videoconferencingendpointstoapublic networkviaIntegratedSwitchedDigital Network (ISDN) connectionsobtained from the local exchange carrier. As a result of the improvedabilitytoguaranteequalityof servicesinpacketnetworks,theubiquityof switchedlocalareaEthernetnetworks,and thereleaseofnumerousproductsfor businessqualityH.323videoconferencing, IP-based networks are now firmly on the list ofvideoconferencingconnectivityoptions. ISDN VIDEOCNFERENCING SECURITY Inordertoproperlysecurevideo conferencemeeting,onemustbeconcerned aboutthreekeyareas:datastorage,data radiation, and data encryption. Data Storage:-Therearetwotypesofvideoconferencing systems,orcodecs,commerciallyavailable today; PC-based and appliance.The PC-BASEDVIDEOCONFERENCING SYSTEMSthatrunPCoperatingsystems andutilize,atleastinpart,standardCOTS PChardware.Sincethesesystemsinclude internalharddrivesandplug-and-play connectivityforexternalstoragedevices, onemusttakeadditionalstepstosecure these systems. Specifically, the system must beequippedwitharemovablehard-drive setup.Inaddition,twodifferentharddrives mustbeused;oneforsecurecallsandone fornon-securecalls.BecausesecuringaPC-basedvideoconferencingsystemrequires additionalcostandwork,thesesystemsare notverywellsuitedforsecure videoconferencing.APPLIANCEVIDEOCONFERENCING SYSTEMSarenotbasedonaPCplatform. Thesedeviceshavebeencustomdesigned andmanufacturedtoprovideonlyspecific functionalityandtypicallydonotutilize standard PC- based hardware or software. In addition,thestoragecapabilitiesofthese systemsareusuallylimitedtostoring addressbookinformation,usagedata,and configurationsettings.Sincethesedevices do not provide data storage capabilities, they areagoodchoiceforsecure videoconferencing.CONFERENCE ON SIGNAL PROCESSING AND REAL TIME OPERATING SYSTEM (SPRTOS) MARCH 26-27 2011 COMO1O7-6 Data Radiation and Encryption:- Anotherareaofconcernisthatconfidential informationmaybeelectricallyradiatedby theconferencingequipment.Whenthis occurs, data is susceptible to monitoring and eavesdroppingbyunauthorizedpersonnel. AccordingtotheF.A.S.(Federationof AmericanScientists),thiswouldallowa person to view all of the contents of another personscomputer monitor from over half a mileawaywithoutbeingdetected. Therefore,toprovideasecure videoconference environment, data radiation mustbecontrolledoreliminated. BENEFITS OF VIDEOCONFERENCING Cost savings. Time savings.Improved quality of life. Gaining ad-hoc access to contacts. Bandwidth on demand Efficient and Flexible Reliable Communication ISDN BACKUP BROADBANDISDN (B-ISDN) B-ISDNisnotacomputernetworkbutthe telephonenetwork.Specifically,itisthe worldwidedigitaltelephonesystem currentlybeinginstalledinvirtuallyevery developedcountryoftheworld.Once completed,B-ISDNwillpermititsusersto communicateoverhigh-quality,high-speed digitalcommunicationchannels.The supportedmediaincludetelex,fax,voice telephone,videotelephone,audio,high definitionTV(HDTV),and,ofcourse, computer networking. CONFERENCE ON SIGNAL PROCESSING AND REAL TIME OPERATING SYSTEM (SPRTOS) MARCH 26-27 2011 COMO1O7-7 B-ISDN WITH ATM B-ISDNistheso-calledInformation Superhighway.AlongwithATM, AsynchronousTransferMode,isoften confusedwithB-ISDN.Infact,ATMis simplyaservicethatcanrunoverB-ISDN. Itisliterallylittlemorethanthe specificationfora48-bytepacketorcellof informationwithafive-byteheaderwhich tells the telephone system where that packet isgoing.Itcanrunoveranumberof differentphysicalmedia,rangingfromUTP (unshieldedtwistedpair),tosomething calledTAXI,totheB-ISDNsuperhighway. APPLICATIONS OF ISDN Studioqualityaudiotransmissionfor broadcast. Imagearchives(Realestate,medical images, photographic image banks etc.). Preparation of printed materials. Electronic manuals online. Stock quote for brokers. Credit card authorization. CONCLUSION AsISDNisdeployed,fewpeopleare currently replacing their home phone system with an ISDN phone network. The trend for nowseemstobeprovidinganentireISDN networkinasinglebox,withtheNT1,TA, and TE1 equipment all built in. An example wouldbethePipeline25,fromAscend, whichprovidesISDNtoEthernet connections, using IP. It has an NT1 built in, andprovidestwophonejacksforstandard POTS telephones. REFERENCES www.radvision.com Computer networks ISDN systems. AnISDNapproachtointegrated corporate networks. www.perey.com www.searchnetworking.com CCITTRedBook, RecommendationsonISDN, Geneva, 1985. M.HuetandP.Put,TRANSPAC andEmergingISDN,P.J.Kuehn, ed., New Communication Services: A ChallengetoComputerTechnology (North-Holland,Amsterdam,1986) 119-124. M.Intorella,M.BenedettiandG. Gasparrone,Italy'sPilotISDN Service,in:P.J.Kuehn,ed.,New CommunicationServices:A Challenge to Computer Technology. www.data .com CONFERENCE ON SIGNAL PROCESSING AND REAL TIME OPERATING SYSTEM (SPRTOS) MARCH 26-27 2011 COMO1O8-1 PERFORMANCE ANALYSIS OF WIRED NETWORKS UNDER VARYING NETWORK ATTRIBUTES Vaibhav Nijhawan Department of Electronics & Communication Engineering Allenhouse Institute of Technology, Kanpur {vaibhav.nijhawan@gmail.com} Abstract:Wirednetworkisaninterconnection ofremotenodesthroughphysical infrastructureofwiresandcables.Itprovides advantagesoffast,reliable,secureandlong haulcommunication.Themodelingand analysisofanetworkprovidesunderstanding ofbehaviorofvariousnetworkparameters whichcanbefurtherusedforcomputationof desiredresultsandefficiency.Inthispaper analysisofwirednetworkshasbeendoneto evaluatetheperformanceofanetworkunder varyingbandwidthoftransmissionlinksand number of workstations (nodes) attached to the network. Keywords:OPNET,WiredNetworks, Transmission links I.INTRODUCTION Data communication and networking is the fastest growingtechnologyinthemodernera.Advanced researchprojectsagencynetwork(ARPANET) addedfueltothegrowthofwirednetwork industryandsincethenwirednetworksare emergingassocialnetworkswhicharelinking people,organizationsandknowledgeworldwide. Thisradicalchangeinthewirednetworking domainoverthelastdecadesistheresultof convergenceofinternetcoupledwithengineering advancesinthefieldofinformationand communicationtechnologies.Wiredcomputer networkshavebridgedthegapbetweencountries andcontinentsandbroughtupaneraof globalization. II.WIRED NETWORKS Wirednetworkisaninterconnectionofremote nodes through physical infrastructure of wires and cables.Itprovidesadvantagesoffast,reliable, secureandlonghaulcommunications.Thewired networksaremostsuitableforenvironmentswith fixedentitieslikehomeandofficenetworksand highdataratesenvironmentswherededicated linksarenecessaryasalinkbetweenkeyboard andcentralprocessingunit(CPU).Theprimary partsofawirednetworkarenetworkcables, networkadapters,hubs,switches,routersand processgoverningsoftwaresliketransmissi