144 JOURNAL OF EMERGING TECHNOLOGIES IN WEB INTELLIGENCE, VOL. 2, NO. 2, MAY 2010transmission distance is significantly greater than theelectronic ones, signal need not to be regenerated <strong>in</strong>optical networks. Optical fiber has very less weight <strong>in</strong>comparison to electronic MINs. Thus Optical networksgive the comb<strong>in</strong>ation <strong>of</strong> high bandwidth and low latency.TABLE IICOMARISON OF ELECTRONIC AND OPTICAL NETWORKSCharacteristicsElectronicMultistageNetwotksOpticalMultistageNetworksSpeed Less HighEnergyElectricity LightTransmittedBandwidth Used for lessbandwidthapplicationsUsed for highbandwidthapplicationsLatency High LessError Probability High LessWeight More LessCost Less MoreSwitch<strong>in</strong>g Packet Switch<strong>in</strong>g CircuitSwitch<strong>in</strong>gPathProvide Multi path Provide s<strong>in</strong>glefrom source to path fromdest<strong>in</strong>ation. source todest<strong>in</strong>ationComplexity More Complex Less ComplexStructureconsidered2-dimensional 3-dimensionalIV. PROBLEMS IN OPTICAL NETWORKSDue to the difference <strong>in</strong> speeds <strong>of</strong> the electronic andoptical switch<strong>in</strong>g elements and the nature <strong>of</strong> opticalsignals, optical MINs also hold their own challenges.A. Path Dependent LossPath dependent loss means that optical signalsbecome weak after pass<strong>in</strong>g through an optical path. In alarge MIN, a big part <strong>of</strong> the path-dependent loss isdirectly proportional to the number <strong>of</strong> couplers that theoptical path passes through [16]. Hence, it depends on thearchitecture used and its network size. Hence, if theoptical signal has to pass through more no <strong>of</strong> stages orswitches the path dependent loss will be more.B. Optical CrosstalkOptical crosstalk occurs when two signal channels<strong>in</strong>teract with each other. There are two ways <strong>in</strong> whichoptical paths can <strong>in</strong>teract <strong>in</strong> a switch<strong>in</strong>g network. Thechannels carry<strong>in</strong>g the signals could cross each other.Alternatively; two paths shar<strong>in</strong>g a switch couldexperience some undesired coupl<strong>in</strong>g from one path toanother with<strong>in</strong> a switch. For example, assume that thetwo <strong>in</strong>puts are y and z, respectively, the two outputs willhave ly+lxz and lz+lxy, respectively, where l is path lossand x is signal crosstalk <strong>in</strong> a switch. Us<strong>in</strong>g the best devicex=35 dB and l=0.25 dB. For more practically availabledevices, it is more likely that x=20 dB and l=1 dB [5].Hence, when a signal passes many switches, the <strong>in</strong>putsignal will be distorted at the output due to the loss andcrosstalk <strong>in</strong>troduced on the path.Crosstalk problem is more dangerous than the pathdependentloss problem with current optical technology.Thus, switch crosstalk is the most significant factor thatreduces the signal-to-noise ratio and limits the size <strong>of</strong> anetwork. Luckily, ensur<strong>in</strong>g that a switch is not used bytwo <strong>in</strong>put signals simultaneously can elim<strong>in</strong>ate first-ordercrosstalk. Once the major source <strong>of</strong> crosstalk disappears,crosstalk <strong>in</strong> an optical MIN will have a very small effecton the signal-to-noise ratio and thus a large optical MINcan be built and effectively used <strong>in</strong> parallel comput<strong>in</strong>gsystems.V. APPROACHES TO SOLVE CROSSTALKA. Space Doma<strong>in</strong> ApproachOne way to solve crosstalk problem is a spacedoma<strong>in</strong> approach, where a MIN is duplicated andcomb<strong>in</strong>ed to avoid crosstalk [8]. The number <strong>of</strong> switchesrequired for the same connectivity <strong>in</strong> networks with spacedoma<strong>in</strong> approach is slightly larger than twice that for theregular network. This approach uses more than doublethe orig<strong>in</strong>al network hardware to achieve the same. Thusfor the same permutation the hardware or we can say theno <strong>of</strong> switches will be double. Thus cost will be morewith the networks us<strong>in</strong>g space doma<strong>in</strong> approach. In allthe four cases only one <strong>in</strong>put and only one output isactive at a given time so that no cross talk occurs. Withthe space doma<strong>in</strong> approach, extra switch<strong>in</strong>g elements(SEs) and l<strong>in</strong>ks are used to ensure that at most one <strong>in</strong>putand one output <strong>of</strong> every SE will be used at any giventime.Figure 3. Crosstalk avoidance us<strong>in</strong>g space doma<strong>in</strong> approachB. Time Doma<strong>in</strong> ApproachAnother way to solve the problem <strong>of</strong> crosstalk is thetime doma<strong>in</strong> approach [3]. With the time doma<strong>in</strong>approach, the same objective is achieved by treat<strong>in</strong>gcrosstalk as a conflict; that is, two connections will beestablished at different times if they use the same SE.Whereas we want to distribute the messages to be sent tothe network <strong>in</strong>to several groups, a method is used to f<strong>in</strong>d© 2010 ACADEMY PUBLISHER
JOURNAL OF EMERGING TECHNOLOGIES IN WEB INTELLIGENCE, VOL. 2, NO. 2, MAY 2010 145out which messages should not be <strong>in</strong> the same groupbecause they will cause crosstalk <strong>in</strong> the network. A set <strong>of</strong>connections is partitioned <strong>in</strong>to several subsets such thatthe connections <strong>in</strong> each subset can be establishedsimultaneously <strong>in</strong> a network. There is no crosstalk <strong>in</strong>these subsections. This approach makes importance <strong>in</strong>optical MINs for various reasons. First, most <strong>of</strong> themultiprocessors use electronic processors and opticalMINs. There is a big mismatch between the slowprocess<strong>in</strong>g speed <strong>in</strong> processors and the highcommunication speed <strong>in</strong> networks carry<strong>in</strong>g opticalsignals [15]. Second, there is a mismatch between therout<strong>in</strong>g control and the fast signal transmission speed. Toavoid crosstalk, the TDM approach is used, where the set<strong>of</strong> messages are partitioned <strong>in</strong>to several groups such thatthe messages <strong>in</strong> each group can be sent simultaneouslythrough the network without any crosstalk.VI. METHODS FOR MESSAGE PARTITIONING INTDM APPROACHA. W<strong>in</strong>dow methodW<strong>in</strong>dow method is the method that is used to f<strong>in</strong>d themessages that are not <strong>in</strong> the same group because it causescrosstalk <strong>in</strong> the network. If we consider the network <strong>of</strong>size N x N, there are N source and N dest<strong>in</strong>ation address.Comb<strong>in</strong><strong>in</strong>g source and its dest<strong>in</strong>ation address formscomb<strong>in</strong>ation matrix. From this, optical w<strong>in</strong>dow size is M-1 where M = log 2 N and N is size <strong>of</strong> network. In w<strong>in</strong>dowmethod, the number <strong>of</strong> w<strong>in</strong>dows is equal to the number <strong>of</strong>stages [11].After f<strong>in</strong>d<strong>in</strong>g conflicts us<strong>in</strong>g w<strong>in</strong>dow method,conflict graph is generated shown <strong>in</strong> figure. The number<strong>of</strong> nodes is the size <strong>of</strong> the network. The nodes that arehav<strong>in</strong>g conflict are connected through edge. Degree <strong>of</strong>each message is the number <strong>of</strong> conflicts to the othermessage. Conflict graph is shown <strong>in</strong> figure 4.Figure 4. Conflict graphThe conflict matrix is a square matrix with N x Nentry, it consists <strong>of</strong> the output <strong>of</strong> the w<strong>in</strong>dow method, asshown <strong>in</strong> figure 5. The def<strong>in</strong>ition <strong>of</strong> Conflict Matrix is thematrix M ij with size N x N. N is the size <strong>of</strong> the network.B. Improved w<strong>in</strong>dow methodIn this method the first w<strong>in</strong>dow is elim<strong>in</strong>ated for thiswe make the conflict matrix <strong>in</strong>itialized to 0, here Number<strong>of</strong> w<strong>in</strong>dows is M-1. It takes less time to f<strong>in</strong>d conflictsthan the w<strong>in</strong>dows method. Therefore, it is calledimproved w<strong>in</strong>dows method [11,12].Figure 5. Conflict matrix© 2010 ACADEMY PUBLISHER