How to compare the congestion control process of active congestion detection protocol and TCP protocol

Using simulation experiments, the congestion control process of CCA's active congestion detection protocol and TCP protocol (using Reno as an example) is compared. If the CCA's active congestion detection protocol has similar network congestion response behavior as the TCP protocol, RDP / CCA, UDP / CCA connections, and TCP connections should obtain approximate throughput under the same network conditions. Some end nodes implement the RDP / CA protocol and UDP / CA protocol, and the other end nodes use the TCP protocol and UDP protocol. Each connection packet is the same as 1000 bytes. Link L is a congested link, and the router uses a tail-drop buffer management algorithm. Change the network attributes and load conditions, and simulate separately to compare the bandwidth fairness of the RDP / CCA connection, UDP / CCA flow, and TCP connection in the steady state in each case. For comparison, the fairness factor (F) is defined as the average throughput ratio of all RDP / CCA flows or all UDP / CCA flows and all TCP connections.

Assuming that each connection has unlimited data packet transmission, the number of various network connections is 5, 10; the congested link L delay D is 20, 40, 60, 80, 100, and 120ms; and the routers R1 and R2 are used to buffer The size is twice the product of network bandwidth and round-trip delay to ensure that TCP has the appropriate window size and can perform slow start and congestion avoidance processes. The simulation results are shown in Figure 5. It can be seen from the figure that when the link delay is small, the RDP / CCA connection and UDP / CCA flow obtain slightly more network throughput than the TCP connection, and when the link delay increases, the RDP / CCA connection and UDP / CCA flow The fairness of using bandwidth with TCP connection is getting closer and closer 1. The reason is that when the link delay is small, the TCP connection congestion control window is relatively small, which cannot make the TCP connection recover quickly from data loss; when the network is increased at the same time The delay and router buffering increase the TCP connection congestion window, correspondingly reduce the number of times the TCP connection enters the timeout period, and increase the TCP connection throughput. This also shows. RDP / CCA connections and UDP / CCA streams make it easier to use network bandwidth.

Aiming at the problem that the UDP stream does not have the end-to-end congestion control mechanism, it is proposed to separate the congestion control mechanism and the TCP reliable transmission mechanism in the end node protocol stack and exist as an independent CCA. CCA actively detects the available bandwidth of the network, and when network congestion occurs, it controls each network application to adaptively reduce the data transmission rate, thereby ensuring that each network connection implements a TCP-friendly end-to-end congestion control mechanism.

At present, only through some simple simulation experiments to verify the effectiveness of the active congestion detection algorithm, further work is to modify the actual operating system, such as the TCP / IP protocol stack of Linux, to verify its effectiveness on the actual network; further work is to extend the control Agents include integrated services and corresponding mechanisms for differentiated services, such as shaping, marking, etc., to achieve high-quality services.

Vibratory Motor

Xinxiang Mina Import & Export Co., Ltd. , https://www.mina-motor.cn

Posted on