Red Hat, US DOE Team For HPC and Cloud Environments

June 1, 2022

Red Hat is collaborating with multiple U.S. Department of Energy (DOE) laboratories to bolster cloud-native standards and practices in high-performance computing (HPC), including Lawrence Berkeley National Laboratory, Lawrence Livermore National Laboratory, and Sandia National Laboratories.

Adoption of HPC is expanding beyond traditional use cases. Advancements in artificial intelligence, machine learning and deep learning, as well as compute and data-driven analytics, is driving greater interest and need for organizations to be able to run scalable containerized workloads on traditional HPC infrastructure. According to industry analyst firm Hyperion Research, roughly one-third of all HPC system revenue will be dedicated to AI-centric systems by 2025, showing nearly 23% CAGR over the five year period1, driven by the influx of AI workloads. Additionally, nearly 20% of HPC users' HPC-enabled AI workloads are currently being run in the cloud.

Red Hat is a leader in cloud-native innovation across hybrid and multicloud environments, while laboratories understand the needs and unique demands of massive-scale HPC deployments. By establishing a common foundation of technology best practices, this collaboration seeks to use standardized container platforms to link HPC and cloud computing footprints, helping to fill potential gaps in building cloud-friendly HPC applications while creating common usage patterns for industry, enterprise and HPC deployments.

Together with the laboratories, Red Hat will focus on advancing four specific areas that address current gaps and help lay the groundwork for exascale computing, including standardization, scale, cloud-native application development, and container storage. Examples of collaborative projects between Red Hat and DOE laboratories includes:

Bringing standard container technologies to HPC

Red Hat and the National Energy Research Scientific Computing Center (NERSC) at Berkeley Lab recognize the importance of standard-based solutions in enabling computing innovation, especially when technologies must span from the edge to the cloud to HPC environments. From container security to scaling containerized workloads, common, accepted practices help HPC sites to get the most from container technologies. To better meet the unique requirements for large scale HPC systems and pave the way for organizations to be able to take advantage of containers in exascale computing, Red Hat and NERSC are collaborating on enhancements to Podman, a daemonless container engine for developing, managing and running container images on a Linux system, to enable it to replace NERSC’s custom development runtime, Shifter.

Running Kubernetes at massive scale

Red Hat has been collaborating with Sandia National Laboratories on the SuperContainers project for several years, working to make Linux containers and other building blocks of cloud-native computing more readily accessible to supercomputing operations. In this expanded collaboration, Red Hat and Sandia National Laboratories intend to explore the deployment scenarios of Kubernetes-based infrastructure at extreme scale, providing easier, well-defined mechanisms for delivering containerized workloads to users.

Bridging traditional HPC jobs with cloud-native workloads

Red Hat and Lawrence Livermore National Laboratory are collaborating to bring HPC job schedulers, such as Flux, to Kubernetes through a standardized programmatic interface helping IT teams supporting supercomputing operations to better manage traditional parallel workflows alongside containerized jobs, including how this mix of technologies operates with low-level hardware devices, like accelerators or high-speed networks.

Reimagining storage for containers

For containers to be used effectively across both HPC and commercial cloud resources, a set of standard interfaces is needed in order to manage various container image formats and for providing access to distributed file systems. Red Hat and the three DOE National Laboratories aim to define the mechanisms by which container images can be migrated from and deployed with other container engines, allowing users to freely move their applications across popular container runtime platforms, as well as create mechanisms that allow containers to use distributed file systems as persistent storage.

Through the collaboration and Red Hat's experience supporting some of the most powerful supercomputers in the world, HPC sites will be able to abstract the immense complexities their environments can present, benefiting the range of United States exascale machines being deployed by DOE.

Chris Wright, senior vice president and chief technology officer, Red Hat

“The HPC community has served as the proving ground for compute-intensive applications, embracing containers early on to help deal with a new set of scientific challenges and problems. That led to the lack of standardization across various HPC sites creating barriers to building and deploying containerized applications that can effectively span large-scale HPC, commercial and cloud environments, while also taking advantage of emerging hardware accelerators. Through our collaboration with leading laboratories, we are working to remove these barriers, opening the door to liberating next-generation HPC workloads.”

Earl Joseph, Ph.D., chief executive officer, Hyperion Research

“High performance computing infrastructure must adapt to the requirements of today's heterogeneous workloads, including workloads that use containers. Red Hat’s partnership with the DOE labs is designed to allow the new generation of HPC applications to run in containers at exascale while utilizing distributed file system storage, providing a strong example of collaboration between industry and research leaders."

Shane Canon, senior engineer, Lawrence Berkeley National Laboratory

“The collaboration with the Podman community and Red Hat engineers is helping us to explore and co-develop enhancements that will allow Podman to scale and perform for the largest HPC workloads. We have already demonstrated this across 512 GPU nodes on Perlmutter. NERSC sees a convergence of HPC and cloud-native workloads, and Podman can be an important tool in helping to bridge between these two worlds.”

Bronis R. de Supinski, chief technology officer, Lawrence Livermore National Laboratory

“High performance computing infrastructure is becoming more diverse and is increasingly being used to run non-traditional HPC workflows. We need to provide mechanisms for scheduling various types of workflows and expect container orchestration frameworks like Kubernetes and Red Hat OpenShift to be a significant part of the software ecosystem effectively contributing to the convergence of the HPC and cloud realms.”

Andrew J. Younge, Ph.D., R&D manager and computer scientist, Sandia National Laboratories

“Sandia and the DOE are seeing an increased need to support more diverse HPC workloads, beyond traditional batch-based modeling and simulation codes. This requires us to find new and innovative ways to enabling services, tasks, and data persistence models together within tight coordination with current simulation capabilities. Furthermore, workload portability remains an important consideration where containers are now a key component to our code deployment strategy. Sandia’s collaboration with Red Hat on Podman and Kubernetes-based OpenShift enables us to investigate approaches for delivering modeling and simulation capabilities as a service to Sandia’s designer and analyst communities.”