Study On The Key Technologies Of Data Transmissiaon Using Multi-path In IP Network

As a foundation of building unified IP carrying and operational platform, the routing andswitching equipment must meet the rapid expansion of information capacity and the diversiformrequirements of network traffic to establish a stable and fundamental basis for NationalInformation Infrastructure (NII). However, with the network scales rapidly and new networkapplications emerge frequently, bandwidth supply for today’s Internet could not catch up with therapid increasing requirements. Unfortunately, irrational using of network sources makes thingsworse. Actual network deploys single-next-hop optimization paths for data transmission, butsuch “best effort” model leads to the imbalance use of network resources and usually leads tolocal congestion. On the other hand Multi-path routing can use the aggregation bandwidth ofmulti paths efficiently and improve the robustness of network, security, load balancing andquality of service. As a result, multi-path has attracted much attention in the routing andswitching research fields and many important ideas and solutions have been proposed.Combined with the research on “Fast Self-Recovery Routing Protocol and ExperimentSystem”,“a Multi-Next Hop Routing Mechanism Based on Node Potential” and “A Multi-Next-hop Based Dynamical Load Balanced Routing and Switching”, this dissertation focuses onimplementing the parallel transmission of multi next-hop data,balancing the network traffic andreducing the congestion. In short, the contents of this dissertation can be included as follows:1. Multi-routing generation is the basis and premise of the multi-path data transmission innetwork. This paper proposes a framework of multi-routing generation algorithm based on nodepotential, which pointed out that the multi-route generation is equal to convert the networkdiagram to a directed acyclic graph. Under the framework of this algorithm, two algorithms areproposed, that are NPMA-Dijkstra which based on Dijkstra algorithm and NPMA-MCSalgorithm which based on minimum cost sum algorithm. It is easy for NPMA-Dijkstra algorithmto deploy.And NPMA-MCS algorithm can improve the performance of data transmissionbetween nodes on the routing level. Compared with commercial mature MDVA algorithm, thealgorithms proposed in this paper improves the number of multi next hops and paralleltransmission capacity.So, the algorithms can make fully use of the network link resources.2. Based on multi-routing generation, the multi-path network traffic engineering model isinvestigated. According to the characteristics of multi next hops transmission, this paperproposes a linear programming model MM-TE with the main idea of minimizing maximum linkutilization and analyzed the performance. As there is no traffic division for the model on someindividual nodes, improved traffic engineering model MM-TE+is proposed. MM-TE+modelconstrains the selection of the paths with excessive number of hops. Simulation on the actualnetwork topology shows that the MM-TE+model can cooperate with multi-routing generationalgorithm very well, and can effectively balance network link resource utilization.3. From the perspective of network users, network congestion control problem in multi-pathnetwork is considered. Researching on multi-path network flow control algorithm under thetheoretical framework of network utility maximization, this paper extends the single-path flow control algorithm to multi-path network. Considering the problem of multi-path network utilitymaximization which is not a strictly convex function, it revises the objective function of theutility maximization, therefore it ensures the differential coefficient of dual problem.It also offersa source end and link end flow control algorithm and verifies the convergence of the algorithmby theoretical analysis and simulation. As traffic engineering and traffic control are separated, anoptimization model and algorithm to prevent network congestion from the perspective ofcross-layer is proposed. Experiments show that the algorithm can improve the throughput andthe load balancing performance of the network.4. According to the multi-services carrying problems in the evolution of network, optimizedresource allocation algorithm in multi-path network is analyzed. By analyzing the characteristicof each service’s utility function, we divide the services in network into elastic service, softreal-time inelastic service and hard real-time inelastic service, and bring forward a resourceallocation algorithm of elastic service.By introducing the concept of equivalent bandwidth inmulti-path aggregation, the sufficient condition of optimized resource allocation of inelasticservice and multi-service co-existed can be gained. Based on the non-convex optimizationspeciality of resource allocation when multi-service co-existed, a genetic algorithm based onpenalty function is proposed. The results of simulations show that the algorithm converges fastand can get the optimization solution of resource allocation in multi-service co-existingconditions.5. To satisfy the engineering requirement of the863project “Fast Self-Recovery RoutingProtocol and Experiment System”, a forwarding engine that supports multi-next hop routing isdesigned and implemented. The forwarding engine mainly consists of searching structure withthree levels forwarding, multi-next hop routing result processing module based on keepingin-order forwarding, and a128bits parallel CRC32hash algorithm based on FPGA. After adetailed circuit design and performance analysis for each module, the forwarding engine PCBboard based on ATCA architecture is implemented. Testing with the commercial SpirentTestCenter instrument the performance of the engine shows that the packet loss rate, theprocessing delay and balancing performance satisfy the requirements of10Gbps processingcapacity.