A server load prediction method and system based on a Markov chain and a time sequence model

Abstract

The invention discloses a server load prediction method and system based on a Markov chain and a time sequence model. The method comprises the following steps: step 1, periodically sampling and updating load information of a server by utilizing a sliding window method, and generating a load value time sequence; Step 2, establishing an ARIMA model according to the load value time sequence, and correcting the predicted load error of the ARIMA model by using a Markov chain to obtain a corrected load predicted value; And step 3, each host reports the corrected load prediction value to the dispatcher of the server cluster where the host is located, and a decision basis is provided for the dispatcher to perform task distribution. The server load prediction method based on the Markov chain and the time sequence model can provide a decision basis for subsequently realizing reasonable scheduling of computing resources.

Description

technical field [0001] The present invention relates to the technical field of servers, in particular to a server load prediction method and system based on a Markov chain and a time series model that can provide decision-making basis for subsequent reasonable scheduling of computing resources. Background technique [0002] With the popularization and development of cloud computing technology, more and more institutions and enterprises use remote server networks hosted on the Internet to build cloud platforms to store, manage and process business data. During the operation of the cloud platform, the utilization rate of resources such as server CPU and memory fluctuates up and down all the time. If the server is kept under high load, the response speed of access requests will be affected; on the contrary, if the server is kept under low load, computing resources will be wasted. In view of the above problems, the present invention proposes a server load prediction method base...

AI Technical Summary

Problems solved by technology

The key problem in realizing the above-mentioned on-demand dynamic scheduling of computing resources is that resource scheduling takes a certain amount of time. If the demand increases sharply, the resource allocation will be insufficient during this period; idle

Due to the nonlinear, non-stationary and dynamic random characteristics of host load changes in the cloud platform, the traditional method does not fit the load trend well, and the prediction accuracy is low

The other category is the prediction method based on the neural network model. Compared with the traditional prediction method, this type of method improves the prediction accuracy, but the neural network model is greatly affected by the complexity of the sample, and its training takes more time, and there is convergence. Slow speed, sensitive parameter selection and easy to fall into local optimal solution, etc.

Images