X

An Oracle blog about Education and Research

  • November 2, 2020

1.5 PetaFLOPS with the next generation of hybrid cloud HPC clusters

Mari Brands
Principal Technical Program Manager
This is a syndicated post, view the original post here

Oracle has partnered with Osaka University, one of the top public research universities in Japan, and NEC Corporation, on Osaka University’s OCTOPUS supercomputer. Osaka University Cybermedia Center Over-Petascale Universal Supercomputer (OCTOPUS) is a hybrid-cluster system with 1.463 PetaFLOPS peak performance, designed by the university’s Cybermedia Center (CMC). OCTOPUS users can now extend their performance and scalability with Oracle Cloud Infrastructure’s (OCI) high-performance computing (HPC) platform. A research team led by Osaka U Associate Professor Susumu Date is using OCTOPUS to combat the spread of new COVID-19 infections.

Highlights

  • Cloud bursting without compromise: The Osaka U, Oracle Corporation Japan, and NEC partnership enables cloud bursting for OCTOPUS users with OCI’s bare metal computing resources. Oracle Cloud’s unique HPC bare metal compute instances are especially suited for performance-intensive workloads that require extra low-latency.

  • Operational efficiency provided by NEC: The job scheduler NEC NQSII is used to centrally manage a multicloud bursting mechanism, streamlining various cloud compute resource deployment processes to provide manageability and operational efficiency for OCTOPUS.

  • Osaka University and Oracle for Research partnership: Oracle for Research, a global program that helps researchers, scientists, and university-associated innovators tackle highly complex problems by using Oracle Cloud technology, is providing Osaka University the HPC through cloud bursting to help accelerate their research.

With greater scalability and performance, cloud bursting with Oracle’s HPC platform saves users money with the ability to stop HPC nodes when resources aren’t needed and restart when demand returns. With Oracle Cloud’s FastConnect, a dedicated, private network connection service through Science Information Network (SINET) at ultra-high speed, users will benefit from unlimited and free data transfer between the CMC’s OCTOPUS and OCI.

 

The same computational and network performance as on-premises HPC

With eight nodes of OCI HPC instances on OCI’s RDMA cluster network, Osaka University was able to confirm bandwidth between 91.16 Gbps to 92.72 Gbps. The communication latency was in the range of 1.44–1.62 microseconds at RoCEv2 level, and 1.63–1.65 microseconds using OpenMPI. This exceptional performance and scaling are especially important for OCTOPUS users because most of their workloads require parallel computing using multiple compute nodes.

Research results have a significant impact on society

With cloud bursting, OCTOPUS users can now cost-effectively scale out their workloads on-demand, using Oracle Cloud, without compromising performance. Research on COVID-19 infections requires highly complex simulation workloads that need massive computational resources. Osaka U’s flexible, scale-out, low-cost solution enabled by their technology partnership with Oracle and NEC make this type of research possible. Future collaborations help accelerate industrial and commercial use of HPC in working on important issues that our society faces.

References

Be the first to comment

Comments ( 0 )
Please enter your name.Please provide a valid email address.Please enter a comment.CAPTCHA challenge response provided was incorrect. Please try again.