Dynamic Configuration of Reverse Proxy Cache based on Multi-Dimensional Time Series Prediction of Visit Traffic

This paper does research on the Internet visit traffic prediction by using time series and apply it to server configuration. After analyzing the periodicity and regularity of typical periods of time series, we propose the multi-dimensional model establishment method, which is both applicable for long-term and short-term prediction. In addition, for immediate prediction we propose dynamic change rate prediction and adjustment algorithm based on exponential smoothing and moving average. The experiment result show high precision of prediction. By applying the model and algorithm above to the cluster server reverse proxy for dynamic cache configuration, the average response time from server to the user's request is greatly reduced, accurately predict the Internet traffic warning line, and has a guiding significance for cluster resources optimization.