AWS Machine Learning Specialty - MLS-C01

1. 1.A team must store 40 TB of training images that SageMaker training jobs read repeatedly, with durable, cost-effective storage and no server management. Which storage should they use?

   • AAmazon S3Correct answer
   • BAmazon EBS volumes attached to the training instances
   • CAmazon Redshift
   • DAmazon EFS with a Max I/O performance mode

   Why

   Amazon S3 is the standard training data repository for SageMaker: it is durable, scales without provisioning, costs far less per gigabyte than block or file storage, and SageMaker training jobs read from it natively in File or Pipe mode. EBS volumes attach to a single instance and must be sized in advance, EFS is more expensive per gigabyte and is used when POSIX semantics or low-latency repeated reads are required, and Redshift is a data warehouse for structured analytics rather than image storage.

2. 2.Multiple SageMaker training jobs and notebook instances must read the same dataset concurrently through a POSIX file system with low first-byte latency, and the dataset is reused across many experiments. Which storage best fits?

   • AAmazon EFSCorrect answer
   • BAmazon EBS in Multi-Attach mode
   • CAmazon DynamoDB
   • DAmazon S3 Glacier Instant Retrieval

   Why

   Amazon EFS provides a shared POSIX file system that many training instances and notebooks can mount concurrently, which suits an actively reused dataset where per-epoch read latency matters. S3 is the cheaper default but is object storage without POSIX semantics, EBS attaches to one instance at a time in read-write mode, and DynamoDB is a key-value store rather than a file system.

3. 3.Clickstream events arrive continuously and must land in Amazon S3 in Parquet format with minimal code, buffered by size or time, with no consumer application to manage. Which service should be used?

   • AAmazon Data FirehoseCorrect answer
   • BAmazon Kinesis Data Streams with a custom consumer
   • CAmazon EMR with a Spark streaming job
   • DAmazon Athena

   Why

   Amazon Data Firehose is a fully managed delivery stream that buffers records by size or interval, converts records to Parquet or ORC using a Glue table schema, and writes them to S3 without any consumer code. Kinesis Data Streams stores records but requires a consumer application to write them out, EMR would mean managing a cluster for a delivery task, and Athena queries data already in S3.

4. 4.A streaming pipeline must retain records for 24 hours so that two independent consumer applications can each read the full stream at their own pace, and one consumer can replay from an earlier point. Which service provides this?

   • AAmazon Kinesis Data StreamsCorrect answer
   • BAmazon Data Firehose
   • CAmazon SQS standard queues
   • DAmazon S3 event notifications

   Why

   Kinesis Data Streams stores records in shards for a configurable retention period and allows multiple independent consumers to read the same records at their own offsets, including replay from an earlier position. Data Firehose delivers to a destination without offering consumer-managed replay, SQS deletes messages after acknowledgement by one consumer, and S3 is a destination rather than a stream.

5. 5.A team must run serverless ETL that discovers the schema of raw CSV files in S3, catalogs them, and transforms them into partitioned Parquet, with no cluster to manage. Which combination should they use?

   • AAn AWS Glue crawler plus an AWS Glue ETL jobCorrect answer
   • BAn Amazon EMR cluster running Spark
   • CAmazon Athena CREATE TABLE AS SELECT run manually
   • DAmazon Data Firehose with record format conversion

   Why

   An AWS Glue crawler populates the Data Catalog with the discovered schema, and a Glue ETL job written in PySpark reads the catalog table and writes partitioned Parquet, all serverless. EMR provides the same Spark capability but requires cluster sizing and management, Athena queries data without transforming and writing it in a managed job, and Firehose delivers streaming data rather than performing batch ETL.

Practice all 65 questions

Frequently asked questions