Workload regression-based resource provisioning for small cloud providers
Bruno Yuji Lino Kimura, Roberto Sadao Yokoyama, Thiago Oliveira Miranda · 2016
Unlike giant clouds, small cloud providers (SCPs) focus attention on offering personalized services to better fit business needs of regional customers. However, SCPs exploit such a business niche with no massive computing infrastructure. To ensure well provisioned resources, capacity planning and resource management are pivotal tasks. To this end, this paper describes a regression-based resource estimation method for selecting and provisioning virtual machines (vms) with minimal configurations to meet tenants' web application workloads in SCP. From hundreds of http workload tests performed with dozens of vm configurations on the SCP's virtualization platform, a non-linear regression model was obtained to estimate not only CPU demands, but also amount of RAM required to handle workloads in different scales. A resource utilization improvement was verified from a case study of a tenant's web application monitored for a month in an SCP. According to the observed workload, estimated configurations have indicated that resources were overprovisioned for the tenant's vm, so that the underutilized residual resources could provision, at least, a new vm for all the monitoring time.