Research And Improvement Of The Congestion Control Algorithm Based On Queuing Delay And Explicit Feedback

With the expansion of Internet'scale, the number of users and applications using Internet increases very quickly. Congestion has become an important issue. When the requirements of traffic in a network are larger than the network capacity, congestion occurs. The role of congestion management is to avoid congestion collapse of networks. Congestion collapse can be approached by improving end-to-end protocols, and designing a new style of congestion signaling scheme.By giving an insight into a modern end-to-end congestion control protocol-FAST TCP, I find that, this protocol is affected by an unfairness problem in certain scenarios due to an inaccurate measurement of propagation delay. In this thesis, the unfairness problem is quantitatively assessed by mathematical analysis and NS2 simulations. The thesis also reports that FAST TCP suffers from frequent buffer overflows and unstable transmission in a situation of small buffer. A novel parameter adjusting algorithm grounded in a redesigned FAST TCP pacing scheme is proposed to solve this problem. In addition, a queue draining algorithm is introduced to minimize this unfairness. FAST TCP and other TCP variants may not work well partially because they reply on packet loss or delay to estimate the congestion level. This thesis investigates an Adaptive Deterministic Packet Marking (ADPM) algorithm which is used to signal information from routers to end systems. This algorithm is able to signal congestion by marking only a single bit per packet.By theoretical analysis and validated by NS2 simulations, the proposed parameter adjusting algorithm is proved to be effective in avoiding heavy packet losses and unstable throughput when buffer overflows occur. ADPM is compared with other congestion signaling schemes and demonstrates the capability to convey information more accurately with a given number of packets.