OPS.8.2:5 - Archetypal Grounding

OPS.8.2:5.1 - Practice during an operator’s reserve interval

An already-qualified operator has a 45-minute interval during which a call can require a return to protected service within five minutes. Once back, the operator and required equipment can sustain that service. A proposed practice exercise uses an independent training setup and can be stopped at short points.

The constructed timing assumptions are maximum durations under this arrangement:

This exercise fits the stated response requirement. The calculation does not establish its learning value; its practice and feedback still need to serve the selected development question.

A live lesson with a 25-minute segment that cannot be left can exceed the entire five-minute allowance before transfer or restoration begins. Its fit within the 45-minute interval does not make it suitable for this anytime-call protection.

Now change the workstation’s restoration to five minutes. The former exercise gives 1 + 1 + 1 + 5 = 8 minutes and no longer preserves the response. Preparing a usable handover in advance might reduce restoration, but that is a changed arrangement whose time and burden must be included. Otherwise choose another activity or preserve readiness.

OPS.8.2:5.2 - A free machine but an unavailable setter

A machine must be ready for its protected job at 09:30. There is no earlier-call requirement in this case. An auxiliary trial needs five minutes of setup by a setter, twenty minutes of unattended machine time and five minutes of restoration by that setter. The setter is available from 09:00 to 09:20 and then from 09:40. The protected job’s operator is available at 09:30.

The duration sum is thirty minutes. But setup at 09:00–09:05 and running at 09:05–09:25 leave restoration waiting until 09:40–09:45. The trial does not fit the protected start.

If a separately useful ten-minute trial is available, setup at 09:00–09:05, running at 09:05–09:15 and restoration at 09:15–09:20 fit both calendars. The machine is ready before 09:30. Alternatively, prepare a different unit or obtain another qualified setter; merely labeling the trial low priority changes none of these times.

This arrangement protects the named start. It has not established immediate recall during the trial’s indivisible machine cycle.

OPS.8.2:5.3 - Suspending a background AI job leaves its accelerator occupied

A protected computation needs an accelerator and its memory. A background job uses that allocation while the protected queue is empty. The scheduler can suspend the job’s execution, but under this configuration suspension retains the accelerator allocation. Starting the protected job requires that allocation to be released.

A “suspend on arrival” rule therefore fails to provide the needed resource even if the suspension command is immediate. For example, Slurm’s generic-resource documentation states that resources allocated to a suspended job remain unavailable to other jobs. Use a release mechanism supported by the actual application and scheduler, such as checkpointing followed by termination and later restart.

For a constructed alternative, allow one minute to receive the call, checkpoint and terminate; one minute to release and clean up the allocation; and one minute to load the protected computation. With those serial bounds and resources available, the three-minute return fits a three-minute start requirement. The protected computation’s subsequent capacity needs are also assumed to fit.

An extra cleanup wait breaks that bound. Keeping the allocation free may then be the usable choice. Separately retain the cost of restarting the auxiliary computation and any external actions that its earlier execution already caused.