(RP12) A Container-Based Approach to HPC Cloud
Clouds and Distributed Computing
Performance Analysis and Optimization
TimeTuesday, June 18th8:30am - 10am
DescriptionRecently, the VM-based HPC service has been provided in the cloud environment to satisfy portability, flexibility, scalability, and reduction of deployment costs in the HPC field. However, performance issues and workload management issues due to the limitations of VM are reducing the resource utilization of HPC users. Therefore, we aim to provide a lightweight container-based cloud environment to HPC users. This container-based approach consists of two main components: the image management system and the workload management system. We have designed and implemented the system workflow and architecture considering ease of use and efficiency of management. The results have been obtained by comparing network performance, MPI performance and a simple machine learning code – MNIST between bare-metal and container-based (both in Docker and Singularity) environments.