Deployment of Coffea-casa Analysis Facility at your Tier 2/Tier 3 grid site or Cluster
This page describes the infrastructure requirements and deployment details for the Coffea-Casa Analysis Facility.
Current Infrastructure @ T2 Nebraska
The production deployment at the University of Nebraska-Lincoln Tier 2 site runs on the following stack:
Component |
Details |
|---|---|
Operating System |
AlmaLinux 9.7 |
Kubernetes Version |
1.34.5 |
GPU Resources |
2× NVIDIA L40S |
HTCondor Pool |
Nebraska Tier 2 (autoscales worker nodes on demand) |
Storage |
XCache + EOS public (via XRootD redirector) |
GPU Support
The facility is equipped with two NVIDIA L40S GPUs for accelerated computing workloads. These are available to users who require GPU resources for their analyses. To request access to GPU nodes, please open a GitHub Discussion or contact the facility administrators directly.
The L40S cards provide 48 GB of GDDR6 VRAM each and are well suited for machine learning inference, histogram filling acceleration, and other GPU-aware HEP analysis workflows.
Deploying Your Own Instance
To deploy Coffea-Casa at your own Tier 2 or Tier 3 grid site you will need:
A Kubernetes cluster (tested on 1.34.5+)
HTCondor for worker node scheduling
Helm 3 for chart deployment
An XCache or XRootD redirector for data access
The Helm chart and full configuration reference are maintained in the
coffea-casa repository.
Refer to the values.yaml and values-prod.yaml files for a complete list of
configurable options.
Note
A detailed step-by-step deployment guide is in progress. For now, please open a GitHub issue or reach out via the e-group for deployment assistance.