MLA-C01 Question 295
Single answerYou are building a machine learning model using Amazon SageMaker. During the training phase, you notice that your training jobs require varying amounts of compute resources depending on the dataset size and algorithm complexity. You also want to minimize costs and avoid managing unused capacity. Which resource option is most appropriate in this scenario?
- A
Use on-demand resources to dynamically allocate compute capacity as needed during the training process.
- B
Use provisioned resources to pre-allocate a fixed amount of compute capacity for all training jobs.
- C
Use on-demand resources to pre-allocate a fixed amount of compute capacity for predictable workloads.
- D
Use provisioned resources to automatically scale compute capacity based on training job requirements.
Show answer and explanation
Correct answer: A
Explanation
In this scenario, on-demand resources are the best choice because they dynamically allocate compute capacity as needed, which helps ensure cost efficiency and avoids the complexity of managing unused capacity. Provisioned resources, on the other hand, are better suited for predictable workloads with consistent resource requirements.
- A. Correct.
This is correct. On-demand resources allow you to allocate compute capacity dynamically based on the current workload, which is ideal for scenarios where resource demands fluctuate and cost efficiency is important.
- B. Incorrect.
This is incorrect. Provisioned resources involve pre-allocating a fixed amount of capacity, which can lead to underutilization or overprovisioning, especially when resource needs vary significantly.
- C. Incorrect.
This is incorrect. On-demand resources do not involve pre-allocating fixed capacity; rather, they are allocated dynamically based on real-time needs.
- D. Incorrect.
This is incorrect. Provisioned resources do not automatically scale. They are pre-allocated and fixed, which makes them unsuitable for fluctuating workloads.