HPC Scheduler Delay

Scheduler admission delay is interpreted as a first-order systems variable for cross-facility federated learning. Production queue conditions varied substantially across the facilities used in the deployment. Queue delay does not alter the federated objective but changes when clients compute and return updates. Scheduler admission delay may range from minutes to hours and changes with system load, priority policy, and scheduler dynamics. Federated learning across HPC facilities is affected by batch scheduler admission delays from systems such as PBS or Slurm.