Performance Prediction and Fine-Grained Resource Provision of Virtual Machines via LightGBM

Autor: Jia Hao, ZhaoXiang OuYang, Jun Wang
Rok vydání: 2021
Předmět:
Zdroj: Data Mining and Big Data ISBN: 9789811674754
DMBD (1)
DOI: 10.1007/978-981-16-7476-1_24
Popis: It is significant to accurately predict the performance of virtual machines (VMs), and then provide the corresponding fine-grained resources according to users’ requirements for both users and cloud resource providers in IaaS cloud computing. In this paper, based on the idea of LightGBM, we first analyze the hardware/software, configuration and then runtime environmental features that may have impacts on the VM performance, and then propose a VM performance prediction model with Gradient-based One-side Sampling (GOSS) method, called VPGB. VPGB pays more attentions on the data instances that with the larger gradients so as to speed up the model training process and then predicts the VM performance accurately. In addition, based on the prediction results, we apply the genetic algorithm to find the optimal fine-grained resources configuration and then provide for users. Experimental results show that VPGB-based method can predict the VM performance accurately and provide the fine-grained VM resources for users effectively.
Databáze: OpenAIRE