Towards tenant demand-aware bandwidth allocation strategy in cloud datacenter

Jiuxin Cao, Zhuo Ma, Jue Xie, Xiangying Zhu, Fang Dong, Bo Liu

Research output: Contribution to journalArticleResearchpeer-review

6 Citations (Scopus)

Abstract

As a critical resource for tenants in cloud datacenter, network bandwidth is shared and competed by tenants at the same time. Previous static bandwidth allocation strategies have a good performance in the sharing case. However, for the competing case where bandwidth oversubscription causes conflicts in network resources, existing bandwidth allocation strategies cannot offer a satisfactory solution. In this article, we propose an auto pre-allocation strategy to solve the bandwidth oversubscription issue in cloud datacenter. Our proposal aims to design and implement a bandwidth allocation system embedded in cloud platform using the technology of software-defined networking (SDN). We employ two sampling methods in bandwidth collection and adopt the ARIMA model to make the prediction. Firstly, the virtual machines (VMs) are divided into predictable and unpredictable groups based on ARIMA model, and each predictable VM has three states in terms of its loading status. After that, corresponding bandwidth allocation strategy is produced to limit the bandwidth utilization in a proper range by adjusting the bandwidth for next period. The experimental results show that the auto pre-allocation strategy improves network performance of cloud datacenter, in both bandwidth utilization ratio and network capacity.

Original languageEnglish
Pages (from-to)904-915
Number of pages12
JournalFuture Generation Computer Systems
Volume105
DOIs
Publication statusPublished - Apr 2020

Keywords

  • Bandwidth prediction
  • Resource allocation
  • sDN
  • Time series

Cite this