A Sorting Hat For Clusters. Dynamic Provisioning of Compute Nodes for Colocated Large Scale Computational Research Infrastructures

Abstract

Current large scale computational research infrastructures are composed of multitudes of compute nodes fitted with similar or identical hardware. For practical purposes, the deployment of the software operating environment to each compute node is done in an automated fashion. If a data centre hosts more than one of these systems – for example cloud and HPC clusters – it is beneficial to use the same provisioning method for all of them. The uniform provisioning approach unifies administration of the various systems and allows flexible dedication and reconfiguration of computational resources. In particular, we will highlight the requirements on the underlying network infrastructure for unified remote boot but segregated service operations. Building upon this, we will present the Boot Selection Service, allowing for the addition, removal or rededication of a node to a given research infrastructure with a simple reconfiguration.