UW-Madison CS Dept. Condor Pool Policies

The policies in effect for our Condor pool are described below. Contact condor-admin@cs.wisc.edu to request a customized policy for your workstation or to comment on these policies.

Start

Condor will start running a job on a workstation when:

the one minute CPU load average falls below 0.3 and
the console keyboard, mouse, and ttys have all been idle for over 15 minutes and
the workstation has more physical memory than the memory footprint of the job plus 15 MB and
the workstation has more physical memory than the job's MemoryRequirements plus 15 MB or the workstation has more than 128 MB of physical memory if MemoryRequirements are unspecified.

Preempt

Condor will preempt (i.e., stop running) a job when:

console keyboard, mouse, or tty activity is detected or
the CPU load average not attributed to the Condor job goes above 0.6

Condor may use three different mechanisms to preempt the job. Condor will suspend the job by sending a SIGSTOP to all job processes if:

the memory footprint of the job is less than 15 MB or
preemption was triggered by CPU load average, not console keyboard, mouse, or tty activity

Condor will unsuspend a suspended job by sending a SIGCONT to all job processes if:

the console keyboard, mouse, and ttys have all been idle for over 5 minutes and
the CPU load average not attributed to the Condor job falls below 0.3

Condor will request that the job perform a checkpoint and exit if:

the job has been suspended for 10 minutes or
the memory footprint of the job is 15 MB or more and preemption was triggered by console keyboard, mouse, or tty activity

Condor will kill the job by sending a SIGKILL to all job processes if:

the job has been checkpointing for 10 minutes or
the job has been running for less than 10 minutes (i.e., a checkpoint is not worthwhile)

Periodic Checkpointing

Condor jobs perform periodic checkpoints every 3 hours for fault tolerance.

Priorities

There are three tiers of user priorities in our pool. Local users have the highest priority, followed by users from outside the university, followed by "nice users" (i.e., with "nice-user" in their name). The Condor Manual describes the Condor priority mechanisms in more detail.

Job Submission

Job submission is currently disabled on the sol and nova instructional workstations due to limited disk space. We encourage users to submit Condor jobs from their desktop workstations. If that is not possible, contact condor-admin@cs.wisc.edu and we will make alternate arrangements for you.

condor-admin@cs.wisc.edu