One of the questions anyone faces when deploying OpenStack is how to dimension the hardware resources for the cloud platform. The amount of resources required for 'compute' nodes is rather straightforward: it depends on the number and flavor of the VMs that are to be deployed on the cloud as well as the overcommit ratio allowed for the various resources on the compute nodes. However, dimensioning of hardware resources for 'controller' nodes is not as straightforward. This talk presents a simple measurement-based model that can be used to predict the resource demand of the various OpenStack services under different usage conditions. The talk also shows the applicability of this model in (1) dimensioning hardware resources when designing an OpenStack cloud and (2) predicting scalability limits given the available hardware resources for the cloud platform.
