Research On The Key Techniques Of Transmission Control Optimization For Network Acceleration

With the popularity of the high-speed, wireless and data center networks, most existing transmission mechnisms are faced with the inefficient and unfair problems. In addition, the opening nature of Internet makes it more vulnerable againt most cyber attacks, espically the low-rate denial-of-service (LDoS) attacks against transmission control protocol (TCP). To solve these issues, we conduct four innovative studies. For the routers’ queue management, we first propose an enhanced active queue management (AQM) algorithm against TCP-targeted DoS attacks and design an approximately fair AQM algorithm. For the unfair and inefficient issues of TCPs in high bandwidth-delay product (BDP) and lossy networks, we propose an adaptive TCP protocol based on the network congestion level and enhance the less-than-best-effort protocol for the background transmission. In addition, we also propose an adaptive and asynchronous feedback-based protocol combining the load factor and rate factor in high BDP network. To improve the quality-of-service and solve TCP incast issue in data center network, we present a dynamic priority and scheduling framework. In summary, the main contributions of this dissertation are as following.Firstly, we design and implement the anti-attack and approximately fair AQM algorithms. In openning network, the cyber attacks, especially LDoS attacks, make a great influence on the performance of TCPs. In this dissertation, we present a flow trust model and design an anti-attack random early detection (RED) algorithm, named the RED with flow trust (RED-FT). In addition, we also design a low-complexity and approximately fair AQM algorithm, named LRURC, based on the sample-match and the improved least recently used (LRU) mechanisms. The experimental results show that these proposed schemes can enhance the performance of the TCP protocol.Secondly, we design and implement an asynchronous feedback-based transport protocol combining the load factor and rate factor. Due to the better performace and easy to deploy, the load factor based protocols attract a wide spread attation. However, some previous studies show that the convergence speed of the load factor based protocols is inefficient. To solve this issue, we propose an adaptive and asynchronous feedback-based protocol, which determine the coded feedback according to the flow rate and the load region. This can accelerate the convergence speed of the link utilization and fairness among different flows. The experimental results show that the proposed protocol has a better performance than other load factor based protocols in the convergence speed of the efficiency and fairness.Thirdly, we design an adaptive transport protocol based on the virtual parallelism to improve the efficiency of standard TCP in high-BDP and lossy networks. Other studies have shown that the deficiency of the additive increase multiplicative decreases of standard TCP. Based on the congestion level estimation using explicit congestion notification (ECN) bit stream, we propose an adaptive transport protocol based on the virtual parallelism. It can adjust the parallel degree dynamicly, which uses the estimated congestion level, to improve the efficiency of standard TCP and keep the TCP friendly. The experimental results show that the proposed protocol has better throughput in high BDP and lossy networks and can maintain the TCP friendly.Fourthly, we design and implement an adaptive low-priority transport protocol to improve the efficiency of the background transmission in high-BDP and lossy networks. For some non real-time applications, such as automatic backup and software updates, they can tolerate the longer flow completion time than the interactive web browsing. Hence, the low-priority transport protocols attaract more attention. However, the conservative congestion window adjustment (e.g. the linear increasing of LEDBAT) causes the inefficiency of existing low-priority transport protocols in high BDP and lossy networks. To solve this issue, we propose an adaptive low-priority transport protocol, which uses the one-way queueing delay for measuring the network congestion level and then updates the aggressiveness of the congestion window adjustment. It can improve the efficiency and maintain the low-priority feature of the low-priority transport protocol. In addition, using the non-congestion loss detection, the proposed protocol can obtain good performance in the lossy netwrok. The experimental results show that the proposed protocol can obtain better link utilization and maintain the low-priority feature.Finally, we design a fast and efficient transport framework, named FDCT, to improve the efficiency of the data center TCP. The many-to-one structure of data center network causes the TCP Incast that result in the poor TCP transmission performance. Although the modified end-host TCPs and switchs’ scheduling algorithms having mitigated this issue, the deficiency of the data center transmission still exists. Considering the constraint of the flow size and deadline, we propose a combined transmission framework that synthesizes the end-host TCP and the switchs’ scheduler. In this framework, the end-host TCP adopts the flow size and deadline to compute the flow priority and then adjust the congestion window. In order to ensure the transmission of high priority flows and avoid the death of low priority flows, the switch implements a priority-based weighted round robin scheduler. The experimental results show that the proposed FDCT can reduce the percentage of deadline-missed flows and mitigate the throughput collapse caused by TCP incast.