Design And Performance Analysis Of Optical Packet Switching Structure

Optical packet switching can provide powerful switching ability and flexiblepacket routing capability, making it really important in the future development of thenext generation all optical network. Aiming to cope with packet competition in outputports of optical packet switching structure, a solution combined with fiber delayline（FDL） and tunable wavelength converter（TWC） is proposed to schedule thecompeting packets in order to reduce the packet loss rate. The study is based oncomputer simulation, using Simulink platform of MATLAB software. The selectedstructure is asynchronous blocking core switching node.Firstly, two problems which are brought by FDL, namely excess load and virtualoccupation, are discussed. Then an optical buffering structure focused on these twoproblems is proposed, including buffering queue, inhomogeneous feedback FDL groupand shared tunable FDL group. The first two parts are designed for reducing excess loadwhile the last one is for reducing virtual occupation. Then three wavelength-convertingstructures are presented for contrast, such as SPC（shared-per-channel）, SPL（shared-per-link） and SPN（shared-per-node）. Beside the above contents, the traditionalfist-come-first-served （FCFS） strategy of wavelength-converting structure and thewavelength backup techonology are also discussed.Secondly, OPS structures with FDLs, with TWCs and with the both are tested onthe simulation platform. The adoptive condition, also called the conventional condition,is showed as follows: the quantity of switching structure’s ports N=8/16，travel loadρ=0.8, wavlength number per port W=8, the coming packet which follows Poissondistribution with an average length T=100and the distribution parameter λ=0.5. Thesimulation result indicate that, when the referential packe loss rate（PLR） is set as10^-5,for the buffering structre, the upper limit of buffering queue H=300, the lower limitL=250, the inhomogeneous FDL groups’ arrangement is two FDLs per group, itsbuffering granularity D=T=100, the number of FDLs M=6when N=8, and M=7whenN=16; shared tunable FDL groups’ maximum buffering depth B_max=255when N=8, andB_max=511when N=16; while for the wavelength-converting structure, the optimal model is SPN with a backup wavelength number ΔW=1when N=8and ΔW=2when N=16.The relative packe loss rate is PLR=5.62×10^-6when N=8, while PLR=9.33×10^-6whenN=16. By contrast with previous research, the result reaches a low level of resourceoccupation.Finally, some improvements about the solution are presented, including adjustableFDLs, FDL scheduling strategy and TWC scheduling strategy. At first, the previousfirm-length feedback FDLs are replaced by adjustable counterparts in order to expandthe buffering ability. Moreover, the FDL scheduling strategy combined withconservative mode and greedy mode as well as the TWC scheduling strategy combinedwith FCFS and LPFS are employed. The simulation result suggest that after theadoption of these improvements, the number of TWCs used in both structures can bereduced to m=5, bringing about an effective spending cut.