Scavenger Queue
The new Scavenger Queue provides access to idle CPU/GPU nodes from participating Campus Cluster investors. Unlike the secondary queue where jobs have a 4-hour maximum wall time and no preemption, the scavenger queue allows jobs to run up to 24 hours but with 2-minute preemption when the node owner needs their resources back. You can view the full details of the new scavenger queue on our website.

How it differs from the secondary queue: The scavenger queue works similarly to the current secondary queue—when investor nodes are idle, anyone with a valid Campus Cluster login can submit jobs to use those resources. The key difference is the preemption model: scavenger jobs receive a 2-minute warning signal before being terminated when the primary investor/owner of that resource submits a job, allowing codes with checkpointing to save their progress. Jobs are then automatically requeued and will resume once another node becomes available.

How to use it:

  1. Create your job script with these key parameters:

bash

#!/bin/bash

#SBATCH --partition=scavenger        

#SBATCH --time=24:00:00              # Up to 24 hours

#SBATCH --requeue                    # Auto-restart if preempted

#SBATCH --signal=B:SIGTERM@120       # 2-minute warning before preemption

./your_program
  1. Submit and monitor:
    • Launch: sbatch job_script.slurm
    • Check status: squeue -u $USER
    • View available nodes: sinfo -p scavenger -t idle

Best practice: Use applications that support checkpointing to save progress regularly, minimizing lost work if your job gets preempted.