The first Beowulf cluster was built in 1994 at NASA’s Goddard Space Flight Center. Thirty years later, most HPC shops still follow the same basic cluster paradigm (including the one at Goddard today). The intervening decades have witnessed many evolutions in computing: virtualization/cloud, software-defined networking, kubernetes, AI/ML; GitOps and Agile -- just to name some big ones. The traditional cluster paradigm, with its static network zones and security controls, make it all but impossible to “bolt” these new technologies onto an existing operational system. A new approach is needed essentially from the ground up – one that will increase interoperability between the on-premises solution and commercial clouds by enabling the reuse of code used to deploy infrastructure and services. This translates into faster time to science/discoveries for our end users. This is a User Story about one HPC organization’s journey to rapidly modernize in one giant leap. We will discuss the “Next Generation System” being built at Goddard now, which is designed from the start to leverage modern cloud-native methods and best practices to flexibly deploy baremetal, VMs, or containers from a single pool of resources. We will touch on network design, OpenStack & Ironic, Security considerations like DevSecOps, and cultural barriers to adoption.
