MLA-C01 Question 296
Single answerYou are building a machine learning model using Amazon SageMaker. Your workload is predictable, and you know you will require specific compute resources for training jobs at regular intervals. What is the MOST cost-effective approach to allocate compute resources in this scenario?
- A
Use on-demand instances for all your training jobs.
- B
Use provisioned resources with reserved capacity for your training jobs.
- C
Use spot instances for all your training jobs.
- D
Use a combination of on-demand and spot instances based on workload requirements.
Show answer and explanation
Correct answer: B
Explanation
Provisioned resources with reserved capacity are ideal for predictable workloads where you can anticipate resource requirements in advance. By reserving capacity, you benefit from lower costs compared to on-demand pricing, making it the most cost-effective solution for the given scenario. On-demand is more suitable for unpredictable workloads, while spot instances are better for cost savings on interruptible tasks.
- A. Incorrect.
This option is not cost-effective for predictable workloads, as on-demand instances are charged at a higher rate compared to provisioned resources.
- B. Correct.
This is the correct answer. For predictable workloads, provisioned resources with reserved capacity allow you to plan and reserve compute resources in advance, reducing costs.
- C. Incorrect.
Spot instances are cost-effective but are not reliable for predictable workloads since they can be interrupted at any time.
- D. Incorrect.
This option could work in some scenarios but does not fully align with the predictability of the workload described. Provisioned resources are a better option here.