Research On Slicing Domain Switching Techonogy Facing To Reconifgurable Service Carrying Network

As the diversification and large scale trend of network traffic, network technology has facedto flinty challenges. One size fit all is impossible. Based on sharing different substrate resources,many heterogeneous networks can coexist on the same reconfigurable flexible network by con-structing Reconfigurable Service Carrying Network (RSCN) as various network services. Innetwork level, optimal RSCNs are constructed considering factors such as heterogeneity andhomogeneity, cost and income, traffic load-balanced and so on. In node level, RSCN logical iso-lation is realized by platform-based supporting and component-based processing. Switching el-ement in reconfigurable routing switching node is one of key issues to guarantee Quality of Ser-vice of network business time delay, time jitter and packet loss rate, so it has important role forsupporting switching resource used by one RSCN itself.Combined with the research and development of the sub-project―Research and Develop-ment of Reconfigurable Router Components‖and―Engineering Implementation of Reconfigura-ble Flexible Testbed Equipments‖belong to863project―The New GenerationHigh Dependabil-ity Network‖. This thesis focuses on the research of switching technology facing to RSCN whichcontains crossbar switching unit slicing domain switching technology, Clos switching networkslicing domain switching technology, Banyan-type network slicing domain self-routing switch-ing technology without buffer and load-balanced two-stage slicing domain self-routing switchingtechnology using buffer scheduling. In short, its main work and contributions are outlined asfollows:1. Aiming at the instance of one switching port contains one RSCN, RSCN switching portrequirement mapping was analyzed first. Switching unit slicing domain idea was proposed. Slic-ing domain model was constituted. Input queued Slicing Domain Longest Queue First(SD-LQF)scheduling algorithm was proposed. Input and output switching ports mapping course was de-duced. Simulation result showed that: The comparative complexity of SD-LQF scheduling algo-rithm is less than previous proposed LQF scheduling algorithm; The time delay of SD-LQFscheduling algorithm access to output queued fair scheduling is less than previous proposed LQFscheduling algorithm based on different business source.2. Aiming at the instance of one switching port contains more than one RSCN, combinedinput-crosspoint queued Slicing Domain Carrying Group(SD-CG) scheduling algorithm wasproposed. Smoothed deficit round-robin group scheduling was deduced. Domain scheduling andoutput port scheduling based on timestamp algorithm were deduced. SD-CG scheduling algo-rithm with better time performance and worst-case fairness index was proved. Simulation resultshowed that: SD-CG scheduling algorithm has better time performance and fairness based onuniform Bernoulli business source; It has100%throughput based on Diagonal unbalance Ber-noulli business source.3. Aiming at the instance of switching network slicing domain in RSCN constructing course.Clos switching network with middle buffer slicing domain model was constructed. Slicing Do- main Middle-stage Load-balance(SD-ML) scheduling algorithm was proposed. Crossbar slicingmechanism in Clos switching network was deduced. Arbitrating mapping course in input stage,middle stage and output stage was deduced. One method guaranteeing packet sequence ofSD-ML scheduling algorithm was proposed. Stabilization and throughput performance ofSD-ML scheduling algorithm were proved. Simulation result showed that: SD-ML schedulingalgorithm has100%throughput based on uniform or unbalance Bernoulli business source; Thetime delay of SD-ML scheduling algorithm with4times iterative is less than previous proposedscheduling algorithm in Clos switching network.4. Aiming at the scalability instance of RSCN self-routing mechanism because increasingthe number of ports induces oversize the number of stages, a Slicing DomainSelf-Routing(SD-SR) switching model was proposed. The basic switching fabric was construct-ed by2×2sorters and2G-to-G group concentrators. The stage exchange was described bybit-permuting. The unbalance group concentrators were determined by destination transform.The blocking probability iterative course was deduced based on unbalance Bernoulli businesssource. Simulation result showed that: The blocking probability of SD-SR switching model andthe number of ports in group concentrator have inverse connection; The blocking probability ofSD-SR switching model and the number of stages or load-balanced coefficient have direct con-nection; The time delay of SD-SR switching model and the number of ports, the number of stag-es or load-balanced coefficient have direct connection; SD-SR switching model can meet timedelay QoS requirement because the value of time delay is less than hundred nanosecond levellimit.5. Aiming at the instance of higher complexity of scheduling software multicast and lowerthroughput of fanout copying mode in RSCN multicast mechanism, a Stage Partial Fanout Mul-ticast(SPFM) switching model was proposed. The self-routing course of the model was realizedby partial fanout copying mode and quaternary symbols cut-through coding. Unicast blockingprobability, multicast blocking probability and multicast fa nout ratio iterative course was de-duced based on Hybrid unicast and multicast business source. Simulation result showed that: Thevalue of multicast blocking probability is less than1%based on unicast load intensity limit;SPFM switching model can meet multicast time delay QoS requirement because the value oftime delay is less than hundred nanosecond level limit.6. Aiming at the instance of higher blocking probability of SD-SR switching model,Load-balanced Input Slicing Output Assembly(LB-ISOA) sliching domain switching model wasproposed. It realized packet aggregating and splitting in input part, packet assembling and dis-tributing in output part. Traffic load-balance was guaranteed in the model. The100%throughputobtained was proved based on admissible traffic condition. Compared with other previous pro-posed load-balanced schemes, LB-ISOA has optimal property such as time jitter nonexistence,lower hardware complexity, lower buffer complexity and middle buffers nonexistence. S imula-tion result showed that: The throughput can get95%upwards based on input buffer get25%upwards of total buffer requirement; The time delay of LB-ISOA is less than time delay of3-stage buffered load-balanced switching model with same complexity and switching fabric pa- rameter G, K.