This presentation addresses the following topic(s):
- When is cloud more efficient than on-premises computing, and vice versa? How should these hybrid environments be managed? (Alternately: Discuss the efficiencies and inefficiencies of operating all on-prem, all cloud, and hybrid.)
HPC has been a late-moving workload to the cloud. Early design of network and virtualization of the OS resulted in poor performance of HPC applications in the cloud, making it impractical. Modern public clouds have addressed these issues with bare metal processors, elimination of hypervisor, and specialized networks targeted to tightly coupled parallel applications. As a result, an increasing number of on-premises HPC users are moving to the cloud to gain flexibility, run their workloads on the latest hardware, and scale clusters to their needs, all while paying only for the correct hardware at the correct time and scale. In this talk I will present the Oracle Cloud Infrastructure (OCI), which provides availability, resiliency, and performance at scale, allowing HPC users to easily choose the best option for their workloads. I will discuss the state of the art of cloud for high-performance computing, share users’ experiences, as well as some of the challenges and existing limitations that still need to be addressed.
