Server cluster dynamic scaling method based on RNN time sequence prediction

Abstract

The invention relates to a method for carrying out dynamic scaling scheduling on a server cluster based on RNN neural network time sequence prediction. The method comprises a technical scheme for predicting the resource load of a server cluster based on an RNN neural network, a method for solving an optimal solution of an RNN model based on a back propagation theory, and a method for dynamically scheduling the server cluster according to a prediction result. The historical load of the server cluster is detected through the RNN neural network, the possible load condition of the server cluster at the next time point is predicted, and the number of server cluster nodes is increased or decreased according to the prediction result. The server cluster has elastic, telescopic and stable computingcapability is ensured.

Description

technical field [0001] The invention belongs to the field of cloud computing. Background technique [0002] Cloud Computing (Cloud Computing) is a new type of computing driven by the rapid development of the Internet. Through high-speed Internet transmission, shared hardware infrastructure, software resources and information can be provided to various mobile phones, computers and other terminals on demand. . The cloud can be said to be an abstraction of network links, and it can also be used to represent the abstraction of the underlying infrastructure. Through the Internet, a large number of computing resources are linked together. Users no longer need to care about the details of the infrastructure in the "cloud" and do not need to know Hardware operation and maintenance knowledge, you only need to request computing resources from the cloud according to your own needs. It can be said that elastic scaling is an important feature of cloud computing, but the traditional ser...

AI Technical Summary

Problems solved by technology

It can be said that elastic scaling is an important feature of cloud computing, but the traditional server expansion scheme, one is vertical expansion, directly modifying the configuration of the extended virtual machine, which generally requires restarting the server, but most companies' services do not allow intermediate stops

The other is to organize server resources through the cluster method, and increase server nodes appropriately when needed. However, this has a certain lag, and the server load peak is often sudden. If the server is expanded when the peak arrives, the server may no longer be able to Handling load and even downtime

[0003] Traditional machine learning prediction generally requires a large number of features to build a suitable model, and the server load status generally only has two features of CPU utilization and memory utilization, machine learning cannot model and predict this data set well

Images