MLS-C01 Question 331
Select 3You are building a deep learning model on AWS and want to reduce training costs by using Spot Instances. You decide to use AWS Batch to manage the training jobs. Which of the following configurations and practices should you implement to ensure cost efficiency and job reliability?
- A
Configure a Spot Fleet with a mix of instance types that meet your model's requirements.
- B
Set AWS Batch job retries to 0 to avoid unnecessary re-run of interrupted jobs.
- C
Use Spot Instances as the compute environment for AWS Batch, with a fallback to On-Demand Instances.
- D
Enable checkpointing in your deep learning training script to save progress during interruptions.
- E
Use a single instance type in your compute environment to simplify resource allocation.
Show answer and explanation
Correct answers: A, C, D
Explanation
To use Spot Instances effectively with AWS Batch for training deep learning models, you should configure a Spot Fleet with diverse instance types to maximize availability and minimize costs. You can set Spot Instances as the primary compute environment and configure a fallback to On-Demand Instances for reliability. Additionally, enabling checkpointing in your training script helps safeguard the training process against interruptions, ensuring that progress is not lost.
- A. Correct.
Configuring a Spot Fleet with a mix of instance types increases the chances of finding available capacity at the lowest price, improving cost efficiency and reliability.
- B. Incorrect.
Setting AWS Batch job retries to 0 is not recommended because it prevents jobs from being automatically retried if interrupted, which could result in incomplete training.
- C. Correct.
Using Spot Instances as the primary compute environment and enabling fallback to On-Demand Instances ensures both cost efficiency and job reliability if Spot capacity is unavailable.
- D. Correct.
Enabling checkpointing ensures that the training process can resume from the last saved state if a Spot Instance is interrupted, avoiding loss of progress.
- E. Incorrect.
Using a single instance type reduces flexibility and increases the risk of capacity shortages, which could lead to job delays or higher costs.