Architecture of virtualized computational resource allocation on SDN-enhanced job management system framework

摘要

Nowadays, users' computation requests to a high-performance computing (HPC) environment have been increasing and diversifying for requiring large-scale simulations and analysis in the various science fields. In order to efficiently and flexibly handle such computation requests, resource allocation of the virtualized computational resources on an HPC cluster system such as Cloud Computing service is attracting attention. Currently, we aim to realize a novel resource management system (RMS) that enable to handle various resources of an HPC cluster system, and have been studying and developing the SDN-enhanced Job Management System (JMS) Framework, which can manage an interconnect as network resources by integrating Software Defined Networking (SDN) concept into a traditional JMS. However, the current SDN-enhanced JMS Framework cannot allocate virtualized computational resources to a job because the computational resource management is performed by the mechanism of a traditional JMS. In this paper, we propose a mechanism to handle virtualized computational resources on the SDN-enhanced JMS Framework. This mechanism enables to deploy virtual machines (VMs) requested by the user to the computing nodes allocated to a job and execute job's processes in the VMs.