Design an ML job scheduler

Design a system that accepts user-submitted machine learning jobs and schedules them across multiple compute clusters containing heterogeneous CPU and GPU resources. The system should: - accept job submissions from users - choose an appropriate cluster for each job - allocate CPU and GPU resources efficiently - execute jobs and track their status - make good overall use of shared compute capacity Discuss the architecture, scheduling strategy, resource model, fairness, reliability, and how you would handle heterogeneous clusters and different job types.