Project case study
Cloud Scheduler AI
Reduce queue wait time while placing jobs under CPU, RAM and GPU constraints.
The approach
Simulates synthetic jobs and a two-server environment. Compares First Fit, Best Fit, Best Fit RAM and two PPO variants on matched seeds, with action masks for invalid placements.
Results and scope
What the project demonstrates
Across 100 synthetic scenarios, mean wait was 14.5425 steps for First Fit and 13.2940 for PPO. PPO did not show a clear advantage over Best Fit methods.