MLA-C01 Question 297
Single answerYou are building a machine learning pipeline on AWS SageMaker to train a large dataset using a custom algorithm. The training process is expected to run for several hours, but the frequency of training jobs is irregular and depends on new data availability. You want to minimize costs while ensuring sufficient resources are available for training. Which resource provisioning strategy should you choose for the training instance?
- A
On-demand instances
- B
Provisioned capacity
- C
Spot instances
- D
Reserved instances
Show answer and explanation
Correct answer: A
Explanation
On-demand instances are the best choice for irregular workloads with unpredictable resource needs. They allow you to launch resources when needed without long-term commitments or risks of interruptions, which aligns with the requirements of this machine learning training scenario.
- A. Correct.
On-demand instances are suitable for irregular workloads where you do not have a consistent or predictable usage pattern. They provide flexibility and eliminate the need for upfront commitments, making them ideal for your scenario.
- B. Incorrect.
Provisioned capacity is typically used for consistent and predictable workloads where resources need to be available at all times. This does not align with the irregular nature of your training jobs.
- C. Incorrect.
Spot instances could reduce costs, but their availability is not guaranteed, and they may be interrupted. This makes them less suitable for long-running training jobs that need to complete without interruption.
- D. Incorrect.
Reserved instances are cost-effective for long-term, predictable workloads but require upfront commitment. This is not appropriate for your irregular and unpredictable workload.