Research On The End-to-End Model And Its Key Technologies In Reconfigurable Network

The current Internet architecture is ossified, thus it cannot support increasingly diverse end-systems, various heterogeneous networks and novel communication paradigms. To address these issues, the future Internet research community has been focusing on enhancing the inherent capabilities of the Internet to support network adaptability and evolution. Service centric network architecture like reconfigurable network, network service architecture and network function virtualization seeks to decompose and abstract the network functionality into fine-grained function units. Different compositions of function units can be used to customize end-to-end services to satisfy diverse user communication requirements. However, there are still serval unsolved problems in this new end-to-end service provisioning model.(1) The end-toend design philosophy of the Internet is changed due to the introduction of high-level network functions, hence a new end-to-end model needs to be designed.(2) Due to the coexisting heterogeneous networks, it is difficult to map the user-level service requirements onto the substrate networks.(3) Degradation in performance of constructing service paths is caused by the existing bottlenecks in the substrate network.(4) Previous service path constructing approaches primarily focusing on solving this optimization problem with single objective cannot allocate resources efficiently and improve the overall performance.To solve the above problems, this dissertation is supported by the 973 program – “Research on the Architecture of the Reconfigurable Fundamental Information Communication Network”, and focuses on the end-to-end model and its key technologies in reconfigurable network. First, the end-to-end autonomic management model in reconfigurable network is developed. Based on this model, a novel service provisioning schema differentiating service carrying from service adaptation is then proposed. Finally, a business-aware service carrying network constructing approach is proposed in the service carrying stage, and two approaches focusing on resource supplementing into the bottlenecks and multi-objective service path constructing respectively are designed for improving the performance of constructing service paths in the service adaptation stage. Specifically, the contributions of this dissertation are:1. Design of a two dimensional end-to-end autonomic management model in reconfigurable network with the horizontal dimension corresponding to the data plane and the vertical dimension corresponding to the management plane. It achieves intradomain and inter-domain autonomic management of resources, services, service carrying networks and service paths based on the agents' capabilities of perceiving environment, self-decision making and interacting with each other. To incorporate end systems in this autonomic management framework, a novel end system architecture is designed and an end system access scheme is proposed, which eventually enables automatic connections to service carrying networks. Finally, a real scenario based on the reconfigurable network test platform is designed to validate the proposed end-toend model.2. Design of a two-stage service provisioning schema differentiating service carrying from service adaptation based on the proposed end-to-end model. In service carrying stage, the service carrying network is constructed self-adaptively to serve diverse business types. In service adaptation stage, the service path is constructed to provide customized end-to-end service via decomposition and embedding of the service chain. Then a service carrying network constructing algorithm based on link selection strategy with hop constraint(SCNC-LSH) is proposed to minimize the constructing cost with assurance of the service carrying capability. Finally, simulation experiments show the effectiveness and efficiency of the designed service provisioning schema and proposed algorithms, and also show that the SCNC-LSH algorithm can determine the optimal service carrying network in terms of performance and cost for the specific reconfigurable network.3. Design of a sensitivity-based efficient resource supplementing algorithm aiming at the performance degradation caused by the bottleneck. First, the concept of sensitivity is introduced to evaluate the impact a node has on the performance. Second, two sensitivity computing approaches are proposed based on the poweroff-based performance evaluation method PSC and the workload-based statistical method USC respectively. The resource supplementing problem is formalized and the heuristic algorithm UFRS is proposed to reach a proper tradeoff between maximal revenue and minimal cost by iteratively supplementing resources on-demand into the most sensitive node and its adjacent links. Finally, the experimental results show the feasibility of the proposed sensitivity computing approaches and resource supplementing algorithms, and demonstrates the advantage of PSC in accuracy of computing sensitivity and the advantage of UFRS in efficiency of supplementing resources.4. Design of a multi-objective service path constructing algorithm based on particle swarm optimization. First, the service path constructing problem is modeled as an integer linear programming(ILP) with multiple constraints and the objective of maximizing the overall quality of the service path. Second, the discrete particle swarm optimization model is stated by redefining the parameters and operations of the particles, and a particle swarm optimization based multi-objective service path constructing algorithm is proposed. To promote the convergence rate and improve the performance, a particle position initialization and update strategy PIFC is designed. Simulation results show that the proposed algorithms can improve the overall performance comparing with the existing single objective approach, and the proposed PIFC strategy can achieve better performance comparing with the random strategy.