Research On Triffic Prediction Algorithm In Sampling Environment

The computer network is one of the greatest inventions in20th century, it strongly promotes the development of economy and society, and however, it also brings issues of network security and management. Using network monitoring technology to monitor network status and combining with some management mechanism, has been proven to be an effective way to solve these problems. The biggest deficiency of traditional network monitoring technology is hysteresis, that is, you can find the problems only when they come out, the harm is often already begun, and is difficult to eliminate. For the reason above, the concept of network pre-management been proposed. Predicting future state of the network, and taking preventive measures before possible problems, we can eliminate their harm in a latent state. In this context, a traffic prediction method is presented in this paper.Research work in this paper is carried out around network traffic prediction. network measurement is Firstly declared in the paper. Network measurement is the basis of network monitoring technology. Understanding network measurement knowledge will help us have a general understanding of the technical methods used and the purpose want to achieve in this paper. Secondly, paper describes network traffic sampling measurement. Traffic measurement will provide necessary historical data for prediction algorithm, it’s a key content of this paper. Traditional traffic measurement method will be invalid in a high-speed network environment, sampling measurement provides a direction to resolve this problem. Thirdly, paper describes network traffic prediction. Traffic prediction algorithm is the core content of this paper, some related knowledge such as flow model, flow characteristics will contribute to a comprehensive understanding of principles and design of predication algorithm. The last research work of this paper is application of machine learning on this subject. In order to achieve a better adaptability, machine learning is introduced into prediction algorithm. In this paper, libSVM is used to build a machine learning model for smoothing parameter so that smoothing parameter value can be dynamically adjusted in prediction process.Some experiment results are shown in the end of this paper, the results prove that the algorithm works well, and comply with the requirements of system accuracy.