摘要
该文详细描述了组播环境中基于层式结构的前向纠错和基于接收端驱动的 TCP友好拥塞控制策略 .采用层式结构发送数据可以满足尽量多的异构接收用户 ,不同的用户可以根据网络状态和计算的 TCP友好发送速率选择接收不同的数据层和纠错层的数据流 .采用接收端驱动和接收端计算包往返时间的方法可以有效地防止大量的反馈信息和计算阻塞发送端 .
Multicasting is an efficient way to distribute massive data to multiple users simultaneously, especially for the multipoint multimedia system. But in a multicasting system, different users may have different bandwidth, different jitter & delay tolerance and may suffer different loss rate during data transmission. The traditional way is that the multicast sender only sends its data stream to the multicast address and does not care about the heterogeneity of the users. In the meantime, the no congestion controlled multimedia stream can suck up all the bandwidth and can block the TCP connections which share the same link. This paper introduces a joint Layered Forward Error Correction (FEC) and Receiver Driven TCP friendly congestion control scheme for multicast to solve the above problems. Layered data and FEC in the sender side can feed as many as the heterogeneous receivers. TCP friendly congestion control scheme can make the UDP multimedia data stream share the bandwidth friendly with congestion controlled TCP connections.The sender uses a layered FEC scheme to separate the data stream into four classes, including two classes of data: high speed for high bandwidth users and low speed for low bandwidth users, and two classes of FEC: high protection and low protection for different bandwidth and loss rate users, and forwards each class into different multicasting addresses. The sender only updates its sending scheme according to the feedback statistics after a period of time. During transmission, every receiver calculates the available bandwidth using the TCP equations, and then it selectively receives the data stream of any combination of FEC layers and data layers by subscribing or un subscribing the different layer according to the network conditions and the calculated TCP friendly rate at receiver side. For example, the receiver can chooses to receive the high speed data and two FEC layers if it has large bandwidth and high loss rate. Almost all the calculations are done at the receiver side and the calculation results are fed back to the sender. The receivers use a special feedback scheme to reduce the amount of feedback times. The receiver driven method can effectively prevent the large amount of feedbacks overloading the sender. Both theory analysis and experiments done at INRIA France show that the scheme works well.
出处
《计算机学报》
EI
CSCD
北大核心
2002年第5期508-513,共6页
Chinese Journal of Computers
