Upgrading an HPC Cluster and Solving Performance Issues: A Case Study

    

Keeping pace with a university’s need for computing resources to support its research is not an easy task. Our latest case study tells how the University of Colorado at Boulder upgraded and added to its high-performance compute cluster used for mechanical engineering research. After they replaced their original cluster management platform with Bright Computing’s Bright Cluster Manager they boosted cluster performance by more than 10 percent for some users, and eliminated a laundry list of annoying performance issues that had a negative effect on the important research being conducted.

The Department of Mechanical Engineering, University of Colorado at Boulder, uses “Prospero,” its high-performance compute (HPC) cluster for research projects, like improving decision support tools through inverse modeling and sensitivity analysis; constraining ammonia distributions through remote sensing, modeling and surface observations; accounting for costs of air quality and climate impacts on the future US electricity mix; and looking at sources of nitrate in Antarctica.

For the same cost as upgrading their old, problematic platform, they achieved a new expanded cluster that runs smoothly, got a handle on addressing issues as they arise, and improved cluster performance.

From this study, you can learn:

  • How Bright Cluster Manager helped the University eliminate issues with their management platform, like losing IPMI access, skipped nodes, trouble creating and moderating user accounts, and queue manager problems.
  • How Bright gave them an easy and effective way to monitor node health.
  • How the intuitive, user-friendly graphic user interface (GUI) made configuration and adding new nodes easy.
  • How the high level and quality of customer support resources eased the set up process.

Read the entire case study here.