MLA-C01 Question 28
Single answerYou are building a machine learning pipeline to process large amounts of image data for training a deep learning model. The image data is stored in Amazon S3. During the training phase, the model will be trained using multiple Amazon SageMaker instances that need shared access to the data. Which approach should you take to ensure efficient and scalable access to the data?
- A
Mount the Amazon S3 bucket directly as a file system in SageMaker using the AWS CLI.
- B
Use Amazon FSx for NetApp ONTAP to create a shared file system and copy the data from Amazon S3 into the file system.
- C
Use Amazon Elastic File System (Amazon EFS) to create a shared file system and copy the data from Amazon S3 into the file system.
- D
Use Amazon SageMaker's built-in support for directly accessing data from Amazon S3 during training.
Show answer and explanation
Correct answer: D
Explanation
Amazon SageMaker is designed to work seamlessly with Amazon S3, making it the best choice for accessing training data stored in S3. SageMaker's direct integration eliminates the need for additional services like FSx or EFS and avoids unnecessary data transfer or duplication. This ensures efficient and scalable training workflows.
- A. Incorrect.
Amazon S3 cannot be mounted directly as a file system. Accessing data from S3 requires using SDKs, APIs, or built-in integrations, and mounting it directly is not supported.
- B. Incorrect.
While Amazon FSx for NetApp ONTAP can provide shared file system access, it would require additional effort to copy the data from S3 and manage the file system. This is an unnecessary step when SageMaker can directly access S3.
- C. Incorrect.
Amazon EFS also provides shared file system access, but similar to FSx, it would require copying the data from S3, which adds complexity and data transfer overhead.
- D. Correct.
Amazon SageMaker has built-in support for accessing training data stored in Amazon S3. This approach is highly scalable, efficient, and eliminates the need for additional file systems or copying of data.