Case Study

Provisioning an Auto-Scaling, On-Premises AI & Data Platform for NASA Scientists

Client: NASA LaRC & SAGE III project

NASA Langley Research Center (LaRC) supports the Stratospheric Aerosol and Gas Experiment III (SAGE III) mission on the International Space Station (ISS). SAGE III provides critical data on Earth’s atmospheric ozone, aerosols, and trace gases. To process and analyze complex experimental data, NASA data scientists required a collaborative, high-performance computing platform.

OpenTeams deployed Nebari, an open source AI and data science platform, across an on-premises, federally compliant server environment. Setup to analyze experimental atmospheric data for the International Space Station, this deployment delivered GPU-accelerated computing, standardized Python environments, and dynamic load balancing.

The Challenge

Prior to this project, researchers executed jobs directly on physical datacenter servers. This setup presented major operational and security hurdles:

No mechanism existed to govern CPU/GPU resource allocation or manage parallel job execution.

Lack of standardized software environments, leading to lack of reproducibility and potential for supply chain security issues.

The facility lacked an infrastructure layer, pre-existing container virtualization platform or Kubernetes cluster on internal hardware.

The new platform also needed to work in an air-gapped environment within security constraints. External access to NASA hardware and networking is strictly prohibited, which requires all deployment work to be pre-scripted or performed via paired sysadmin sessions

Solution
And Key Results

The Nebari platform provided the key features like reproducible and governable software environments, role-based-access-control to all resources, and observability.

Automated K3s Deployment

Developed Ansible playbooks to automate the setup and configuration of K3s lightweight Kubernetes on Red Hat Enterprise Linux 9 (RHEL 9) servers.

Nebari On-Premises Enhancements

Extended the open-source Nebari codebase to support K3s and non-cloud deployments, contributing core improvements back to the upstream codebase.

Virtualization & GPU Pass-Through

Guided NASA system administrators through setting up open-source virtualization (QEMU/LibVirt) with dedicated GPU pass-through (NVIDIA V100 & A100) for high-performance workloads.

Compliant Handover

Conducted interactive pairing sessions and training for system administrators and 15+ end-users covering JupyterLab, Dask parallel computing, and environment management.

Multi-Server Load Balancing

Achieved dynamic load balancing across a 3-server cluster with zero clustering overhead for end-users.

Centralized Governance

Implemented multi-tenant environment isolation, standardized Python dependencies, and resource controls.

Compliance

Run on federally compliant server operating systems (RHEL 8/9) and federal compliance standards for cybersecurity (FIPS).