Research On Key Technology Of Routing In Logically Centralized Control Network

The emergence of new network services and applications put forward increasing demand of network capability. More functionality must be added to routing system to accomodate these requirements. The network architecture nowadays only allow patch-like increment of new mechanisms and functionalities to be incorporated into the routing system, which imposes challenges such as incapable of achieving fast convergence and routing optimization, rising problem of network control and management, lack of customization. We claim that the fundamental reason for these problems resides in the distributed implementation of the routing system, and the control plane and data plane are tightly coupled with each other considering the present network architecture. This has led to a sophisticated routing system with increasing difficulties and complexities of control and management. Centralized control netwok is a brand new network architecture, which separates control plane from data plane and implements the functionality of control plane in a centralized fashion. According to this architecture, centralized routing control system can solve current problems concerning on routing, management, optimization, policy, and so on. It presents an available technique to promote the ability of routing system. Centralized control netwok can directly control the network, simplify the operation of routing control and decrease the complexity of network management. New control functionality can be easily integrated into centralized control network, through which the network development and evolvement can be significantly promoted.This dissertation focuses on investigations about four issues of centralized routing control system: reliability and scalability of control plane, intra-domain routing/traffic delivery optimization, and cross-plane communication. We propose a routing optimization framework based on logically centralized control, which guarantees reliability and scalablity by logical centralization. We investigate the problems of link state routing protocol, and propose an intra-domain routing protocol based on the routing optimization framework. We research on the traffic engineering mechanisms based on centralized control, and present proposals to optimize the traffic delivery considering both intra-domain and inter-domain situation. We propose a method for constructing dissemination paths, which can establish robust dissemination paths in centralized control network with logically centralized-physically distributed control plane. The major contributions and innovations are summarized as follows.(1)We investigate the demand of large-scale network routing optimization. We emphasize the logically centralization to ensure reliability and scalability, and take the ability of flexible centrliaztion policy support as one of the design principles. According to the principles and objectives, we propose a routing optimization framework based on logically centralized control. This framework uses separate control network and forwarding network to implement control plane and data plane respectively, and introduce the concept of virtual area. The framework has several notable featues such as logically centralized-physically distributed control plane, load balancing of control workload, information dissemination based on unicast and deployable ability of routing control.(2)In order to solve the poblems of link state protoocl, we propose LSC2, an extended link state protocol with centralized control. LSC2 takes advantage of routing optimization framework to achieve controllable link state dissemination, which eliminates flooding and the limitation on timers introduced by distributed control. LSC2 implements a routing control algorithm to suppress redundant routing computation. We implement a LSC2 prototype based on Quagga, and perform evaluations using emulation platform and real topologies. The results show that LSC2 can achieve faster convergence than OSPF, and reduce the packet overhead by over 60%. The routing control algorithm can avoid routing computations on specific routers, which further decreases the computation and packet overhead. LSC2 is also scalable by dynamically adds more CEs to balance the routing control workload across all CEs.(3)Traffic engineering based on link weight configuration has several problems such as error-prone configuration, transient convergence status and unexpected traffic forwarding behavior. Based on routing optimization framework, we propose an optimization scheme for Traffic engineering based on link weight configuration. We deploy a Routing Management System(RMS) on CE to integrate the link weight optimization algorithm and configuration process. RMS can acquire global topological and traffic information, which are used to compute optimized weight for each link. Then RMS configures all the link weights through single convergence, thus it can avoid transient network state and protocol convergence overhead, as well as largely decreaces the human involvement during configuration.(4)We propose an algorithm called TMP-TIE to solve the problem of large traffic migration during egress selection. TMP-TIE predicts the volume of traffic migration casued by egress change. TMP-TIE aims to avoid prominent traffic migration caused by egress change, which can decrease the impact on inter-domain traffic delivery optimization. We use simulation experiments to compare the performance of hot-potato, TIE and TMP-TIE. The results show that TMP-TIE has the minimum routing sensitivity and traffic sensitivity. TMP-TIE can reduce network cost and the probability of congestion in presence of failures, thus it leverages the traffic engineering goal.(5)Based on routing optimaztion framework, we propose a method of dissemination paths construction(DPCM) supporting centralized routing control system with logically centralized control plane. DPCM uses a sophisticated Probe packet which carries the path information it has traveled to discover the network topology and available path. DPCM can automatically divide the network into several control domains during the construction procedure. DPCM also provides mechanism for path aliveness verification. The robustness is guaranteed by instant path reconstruction after primary dissemination path has failed. We build a real network to evaluate DPCM. The results verify DPCM functionality, the dissemination paths robustness and the feature of lightweight overhead.