Question 770
An Azure Machine Learning compute cluster scales to zero after inactivity. A scheduled low-latency training job repeatedly violates its short startup deadline because nodes take several minutes to provision. What capacity policy should be evaluated?
Explanation
Correct Answer
A zero-minimum cluster saves idle cost but incurs cold-start allocation. Targeted prewarming before the known job window can reduce queue delay.
Incorrect Answers
Scoring input type is separate from the cluster boot time and allocation schedule.
Experiment metadata does not alter the time needed to allocate training compute.
Accuracy cannot measure scheduling and VM provisioning delay when queue telemetry is discarded.