Research Efficiency: Gamifying Resource Management on Duke's Compute Cluster

Description

Duke's shared research computing cluster is a finite resource, but when jobs fail with "out of memory" errors, frustrated researchers often respond by requesting the maximum possible resources—hogging capacity that others could use and creating a tragedy of the commons. The problem isn't malicious; it's a lack of feedback and incentives to learn better resource estimation. You'll build a system that changes the culture by making efficiency visible, measurable & rewarding. Your dashboard will track how well users estimate their resource needs by comparing requested vs. actual memory and CPU/GPU usage, flagging wasteful patterns (like always requesting 10× what you need) while celebrating users who thoughtfully tune their allocations. 

You'll design metrics, penalty systems and positive reinforcement mechanisms, experimenting with different modes from strict enforcement to gentle "learning mode" that awards badges for improvement. Think of it as creating a fitness tracker for computational citizenship: users see their efficiency scores, compete on leaderboards, and learn to be better cluster neighbors. You'll also gain experience with SLURM, an open-source workload manager & job scheduler for Linux HPC clusters that is used widely around the world - all while helping Duke's research computing infrastructure serve more scientists more effectively.