Research On The Key Protocols For Congesion Control In Next-Generation Networks

With the rapid development of Internet, the diversity of multimedia services, and the access of several heterogeneous networks, many problems affect the network system. If we want to provide effective and fair resources allocations for high-speed network and to improve the comprehensive QOS performance, we must solve the congestion control problems in next-generation network. Therefore, network congestion control topics under the new environment are still hot research to be focused. Internet congestion control concerns three main aspects:(1). the issues on congestion control based on source systems.(2). the issues on AQM based on links.(2). the issues on congest control based on explicit congestion control. Our thesis focuses on the last two aspects (2) and (3).It is noteworthy that active queue management (AQM) has received many research attentions as a key technology to effective control the network congestions for the last two decades. AQM treated the packet loss as the signal of congestion. Based on the signal, the senders take proper packet strategies to alleviate the link congestion. However, the performance of most AQM algorithms such as Random early drop (RED) recommended by IEIF is sensitive to parameter configurations. Moreover, as the deployment of AQM in wireless networks, AQM is still a hot issue in recent years.Explicit control protocols, as a promising protocol that takes place of TCP in next-generation network, are proposed to solve the problems that TCP will become instability in high bandwidth-delay product network. However, it still faces challenges in deployment and standardization process for next-generation network.Our dissertation focuses on two issues on congestion control mentioned above. We made some contribution work as follows,(1) The research on the famous AQM algorithmsConsidering RED, we establish the relationship between the TCP congestion window and the router queue by utilizing the fluid model, and obtain the uncertain linear time-delay system described by differential equations. We take the2-D Laplace-z transform to the time-delay system, and perform stability analysis. Then we obtain a novel sufficient condition for TCP/RED system’s stability. The condition provides the upper and lower bounds of the parameter pmax for the system stability. We validity our stability conditions by performing extensive NS2simulations. The simulation results show that the system can attain the stable performance in term of router queue size if the network parameter configuration satisfies our condition. Second, we propose a simple scheme based on Particle Swarm Optimization (PSO) algorithm to test the time-delay system. Our experiments show that it is simple, but effective.(2) The proposed novel AQM algorithm iDroptailTo address the problem that most AQM algorithms’performance is sensitive to parameter configuration, we propose a novel and simple active queue management algorithm called iDroptail (improved Droptail algorithm). We directly introduce the active random drop packet strategies to Droptail, and cancel the AQM feedback mechanism. The active random drop packet ratio p will be taken as input signal to TCP system in terms of router size. The p can be configured according to the users’QOS demands. We also model the system, and obtain the linear time-delay system. Based on2-D Routh-Schur time-delay stability analysis, we prove that the system can be stabilized by the proposed algorithm. Via extensive NS2simulations, our algorithm shows its effectiveness. Our simulations show that the performance of iDroptail can match the one of the most AQM algorithm such as RED, PI.(3) The proposed novel explicit control protocol iMLCPWe first point out that load factor based congestion control protocol MLCP has the limitation that in multi-bottleneck networks, MLCP will penalize the long flows much than short flows, and cause link under-utilization. Second, we propose an improved version of MLCP (iMLCP) to address this problem. We introduce an additional field in packet header to record the minimum capacity link, and modify the MLCP router’s mechanism. Our simulation results show that our algorithm not only guarantees high link utilization but also preserves the good properties of MLCP. We also present stability condition for the iMLCP by utilizing frequency-sweeping theorem, which shows that the iMLCP can still stabilize the system effectively.(4) The proposed novel explicit control protocol eACPThe original design of ACP (Adaptive Congestion Protocol) does not support dynamic non-ACP traffic, and will cause unstable performance in a network with ACP and non-ACP traffic. To address the problem, we propose an improved version of Adaptive Congestion Protocol (eACP) that operates efficiently for dynamic non-ACP traffic. We modify the number of queues maintained in the router, and extend the architecture of ACP. And we propose stability conditions for the eACP system. The NS2simulation results verify our scheme is effective. Second, we achieve DiffServ based on eACP.