Slurm-V: Extending Slurm for Building Efficient HPC Cloud with SR-IOV and IVShmem

摘要

To alleviate the cost burden, efficiently sharing HPC cluster resources to end users through virtualization is becoming more and more attractive. In this context, some critical HPC resources among Virtual Machines, such as Single Root I/O Virtualization (SR-IOV) enabled Virtual Functions (VFs) and Inter-VM Shared memory (IVShmem) devices, need to be enabled and isolated to support efficiently running multiple concurrent MPI jobs on HPC clouds. However, original Slurm is not able to supervise VMs and associated critical resources, such as VFs and IVShmem. This paper proposes a novel framework, Slurm-V, which extends Slurm with virtualization-oriented capabilities such as job submission to dynamically created VMs with isolated SR-IOV and IVShmem resources. We propose several alternative designs for Slurm-V: Task-based design, SPANK plugin-based design, and SPANK plugin over OpenStack-based design, to manage and isolate IVShmem and SR-IOV resources for running MPI jobs. We evaluate these designs from aspects of startup performance, scalability, and application performance in different scenarios. The evaluation results show that VM startup time can be reduced by up to 2.64X through snapshot scheme in Slurm SPANK plugin. Our proposed Slurm-V framework shows good scalability and the ability of efficiently running concurrent MPI jobs on SR-IOV enabled InfiniBand clusters. To the best of our knowledge, Slurm-V is the first attempt to extend Slurm for the support of running concurrent MPI jobs with isolated SR-IOV and IVShmem resources. The capabilities of Slurm-V can be used to build efficient HPC clouds.