System capacity prediction method based on service volume

Abstract

The invention discloses a system capacity prediction method based on service volume. The method is characterized by comprising the following steps; collecting original data of at least one month or more; first rejecting seasonal factor effects, selecting a model group based on a linear model, a moving average model and an ARIMA model; calculating a weight index of each model by utilizing a multiple regression model; and then introducing genetic algorithm machine learning to iteratively optimize the model group to obtain optimal model group parameters, calculating a daily dimension prediction result and a minute dimension prediction result, and calculating a system capacity demand based on the prediction results of the two dimensions by utilizing a fitting relationship between the service and historical data of the system capacity. By applying the prediction method provided by the invention, the influence of external factors on prediction is comprehensively considered, the parameters are automatically optimized through model group prediction and introduction of a genetic algorithm, not only can the change of a transaction structure and a service volume trend be flexibly coped with,but also the prediction result is improved, so that the average error is less than 5%, and the service quality of Internet application is favorably guaranteed.

Description

technical field [0001] The present invention relates to the technical field of software for computer and network products, and in particular to a method for exploring the relationship between business access and hardware resources occupied by application software systems, and for the required hardware resources (such as CPU, memory, network traffic, etc.) A predictive approach to capacity planning. Background technique [0002] With the rapid development of IT technology, Internet applications have gradually become an important driving force for business innovation and enterprise quality improvement. Therefore, it is very important to guarantee the service quality of Internet applications. With the rapid popularization of mobile and the Internet, various holidays and marketing days have put forward higher requirements for Internet application business access; Internet applications under high frequency and high concurrent access require more hardware resources to support, G...

AI Technical Summary

Problems solved by technology

Although the current machine learning and deep learning frameworks include Caffe (http: / / caffe.berkeleyvision.org / ), Trensorflow (https: / / tensorflow.org), etc., which provide general algorithms, models, and frameworks, due to the limited capacity Prediction has a strong correlation with Internet application business access conditions, and cannot directly apply algorithms to directly adapt, lacks direct application in actual business scenarios, and lacks a general method for system capacity prediction for business

Images