Swarm cluster scheduling algorithm based on combined prediction model

Abstract

The invention provides a Swarm cluster scheduling algorithm based on a combined prediction model. For the defect that the prediction accuracy of a traditional unary linear regression model and a greymodel is relatively low, the problems existing in a Swarm cluster built-in scheduling strategy are comprehensively considered, the two prediction models are combined, and a prediction model for performing weighted comprehensive calculation on two weights to form combined optimization is established. The combined prediction model is superior to a traditional unary linear regression model and a greymodel, and has high prediction accuracy.

Description

technical field [0001] The invention belongs to the field of cluster scheduling, in particular to a Swarm cluster scheduling algorithm based on a combination prediction model. Background technique [0002] Docker occupies an important position in the cloud ecology, and Swarm also occupies an important position in the PaaS (platform and service) / IaaS (infrastructure and service) layer of the cloud architecture. The main task of Swarm is to run the container on the appropriate node according to the scheduling policy. Due to the difference of the container running on the node, the resource utilization rate is also different. The resource utilization of each node determines the load of the entire cluster. Therefore, the optimization of the cluster scheduling strategy is particularly important. , [0003] There are three built-in scheduling strategies in Swarm, namely Random (random), Spread (diffusion) and Binpack (boxing). The advantage of the Random strategy is that it is s...

AI Technical Summary

Problems solved by technology

[0006] The resource scheduling strategy provided by the Swarm cluster is relatively simple, and there will be unbalanced resource utilization on the nodes. A Swarm cluster scheduling algorithm based on a combined prediction model is provided.

Images