Quick reminder about Auton cluster usage practices

[Piotr Bartosiewicz]

Dear All,

  We have recently observed an increase in practices that negatively impact
the overall performance, availability, and fair-share allocation of the
Auton cluster. Specifically, I'd like to address two primary concerns: *node
camping* and the *abuse of the debug partition*.
1. Debug Partition

  The debug partition is strictly intended for short-term testing, code
compilation, and rapid troubleshooting. It is configured with high priority
to provide fast turnaround times, but it has limited resources. Resources
are shared between debug and preempt partitions. If debug is not used you
are free to use those resources in the preempt partition.
2. Node Camping / Idle Resource Reservation

  Node camping occurs when users reserve resources and leave them idle,
underutilized, or fail to release them promptly upon completion. This also
includes submitting *dummy jobs*—such as sleep scripts, infinite no-op
loops, or scripts that artificially generate fake GPU load or hold idle
VRAM—solely to retain a node or GPU allocation for later use. With limited
and shared resources, this practice is unacceptable.

Users repeatedly violating these guidelines will have their jobs terminated
and may face temporary suspension of their cluster privileges.

Thank you for your cooperation in keeping our cluster fair and efficient
for everyone.

Piotr.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.srv.cs.cmu.edu/pipermail/autonlab-users/attachments/20261113/105d9014/attachment.html>