The Research Of Forecast Feedback Control Based Algorithm For Congestion Control

The computer network has had an explosive development in the past years, but at the same time, it bring a sever problem: congestion. Congestion control has arisen great attention since Jacobson brought forward the concept of congestion control in 1988. Since that, a lots of congestion control algorithms have been put forward. Currently, with the development of the science of flow control and management, it has become an important branch of the computer science: flow control engine. Its main content of research is the detection and avoidance of congestion and the real-time mechanism of the recovery from congestion. There are lots of difficulties in the design of congestion control. Such as the effective and fair allocation of resource. Another problem is the characteristic of distribution of networks information that brings us the difficulty for distributed design of algorithm.The algorithm of AIMD, namely additive increase and multiplicative decrease, which is thought a good algorithm for congestion avoidance, has been applied widely in TCP congestion control. Its principium is that the receiver feedback a binary bit that indicates the information of overload or under load to the sender, and the sender adjusts the load with AIMD algorithm according to the binary information. In this paper, the main points of AIMD algorithm: stability, convergence, fairness are discussed.The algorithm of AIMD is strict synchronous. It assumes that all senders receive the feedback simultaneity. When the receivers have different delay of feedback, additional constraint conditions are required. In this paper, it simulates the system action under the different initial conditions and different round trip time.The AIMD has got a widely apply for its briefness and ease. At the other hand, it contain limited information because it has only two status: overload or under load. It cannot represent the true demand ofthe receiver and at the same time the policies of window adjust destroy the demand of smoothness of load, so it cannot afford the request of real-time streaming media. AIMD algorithm exhibits a tradeoff between performance and complexity. In this paper, at the base of feedback, the theory of forecast is introduced: certain amount of data is stored, the neck resource forecast the load of current scope of time according the stored data, then it feedback the information to the sender. The senders adjust its load according the information. Forecast feedback control meets the requirement of fairness and efficiency and overcome the disadvantage as above.