Research On Congestion Control Algorithm In High-speed Networks

The development of network technology such as optical communication technology leads to the appearance of many high-speed networks with bandwidth larger than 1Gbps, or even 10Gbps. Through high-speed networks, applications like scientific collaboration, telemedicine, and real-time environment monitoring can transfer high-bandwidth real time data, images, and video captured from remote sensors such as satellite, radars, and echocardiography. What is more, data intensive grid application and SAN network can benefit from high-speed networks too. In the future, more applications will appear in high-speed networks.Widely adopted as a data transfer protocol in current Internet, TCP performs badly in high-speed networks because of its congestion control algorithm. Some new congestion control algorithms have been proposed.In this paper, we firstly gave an overview of the related work, and made some contribution work as follows:(1) Analyzed the RTT unfairness of HSTCP with model. HSTCP proposed by S. Floyd is a high-speed protocol, which is recommended by IETF. However, severe RTT (Round Trip Time) unfairness of HSTCP was found through simulation. We gave an analysis with a model to show that severe synchronized loss and scability of HSTCP lead to RTT unfairness.(2) CW-HSTCP was proposed to alleviate the RTT unfairness. CW-HSTCP adds fair factor to eliminate the difference caused by RTT. To evaluate the performance of proposed algorithm, a relative fair criterion was proposed. Theoretical analasis and simulation results show that CW-HSTCP can keep RTT unfairness within the range of RTT reverse ratio meanwhile keeping advantages of HSTCP, such as TCP friendliness.(3) ACWAP alogorithm was proposed to reduce the loss rate of high-speed TCP. Our experiments showed that the number of dropped packets in one congestion event is related to a(w) and that packets are not dropped successively to result in severe synchronized loss. On the basis of a deep analysis, we propose ACWAP (Adaptive Congestion Window Adjustment plus Pacing). The basic idea is that the sender adaptively change a(w) to 1 before congestion according to the variation of RTT, which is a reflection of network state. To avoid burstiness, we also adopt pacing algorithm. ACWAP-HSTCP is the combination of HSTCP and ACWAP algorithm.Simulation results showed that ACWAP can reduce the number of dropped packets in one congestion event, alleviate synchronized loss and RTT unfairness. Particularly, ACWAP-HSTCP can coexist with HSTCP fairly and can be deployed gradually in Internet. Briefly speaking, our algorithm is scalable. In addition to HSTCP, ACWAP algorithm can be combined with some other algorithms, such as STCP, LTCP.(4) HRED, an active queue management algorithm, was proposed to achive a better overall performance. HRED is realized in router. Based on well-known RED (Random Early Detection) algorithm, HRED distinguishs high bandwidth flows through the packet drop history in RED, and then adopts corresponding punishment before packets enter queue. HRED overcomes the shortage of RED in dealing with RTT unfairness and unresponsive flows, and can be adaptively used in the high-speed networks and ordinary networks. Simulation results show that HRED can effectively achieve a better RTT fariness and fairly allocate bandwidth between responsive flows and unresponsive flows.