AWS Certified Solutions Architect – Professional — Question 907

A company has a large on-premises Apache Hadoop cluster with a 20 PB HDFS database. The cluster is growing every quarter by roughly 200 instances and 1
PB. The company's goals are to enable resiliency for its Hadoop data, limit the impact of losing cluster nodes, and significantly reduce costs. The current cluster runs 24/7 and supports a variety of analysis workloads, including interactive queries and batch processing.
Which solution would meet these requirements with the LEAST expense and down time?

Answer options

• A. Use AWS Snowmobile to migrate the existing cluster data to Amazon S3. Create a persistent Amazon EMR cluster initially sized to handle the interactive workload based on historical data from the on-premises cluster. Store the data on EMRFS. Minimize costs using Reserved Instances for master and core nodes and Spot Instances for task nodes, and auto scale task nodes based on Amazon CloudWatch metrics. Create job-specific, optimized clusters for batch workloads that are similarly optimized.
• B. Use AWS Snowmobile to migrate the existing cluster data to Amazon S3. Create a persistent Amazon EMR cluster of a similar size and configuration to the current cluster. Store the data on EMRFS. Minimize costs by using Reserved Instances. As the workload grows each quarter, purchase additional Reserved Instances and add to the cluster.
• C. Use AWS Snowball to migrate the existing cluster data to Amazon S3. Create a persistent Amazon EMR cluster initially sized to handle the interactive workloads based on historical data from the on-premises cluster. Store the data on EMRFS. Minimize costs using Reserved Instances for master and core nodes and Spot Instances for task nodes, and auto scale task nodes based on Amazon CloudWatch metrics. Create job-specific, optimized clusters for batch workloads that are similarly optimized.
• D. Use AWS Direct Connect to migrate the existing cluster data to Amazon S3. Create a persistent Amazon EMR cluster initially sized to handle the interactive workload based on historical data from the on-premises cluster. Store the data on EMRFS. Minimize costs using Reserved Instances for master and core nodes and Spot Instances for task nodes, and auto scale task nodes based on Amazon CloudWatch metrics. Create job-specific, optimized clusters for batch workloads that are similarly optimized.

Correct answer: A

Explanation

AWS Snowmobile is the most appropriate service for transferring a 20 PB dataset, as using AWS Snowball would require a massive fleet of devices and Direct Connect would take too long and cost too much. For the architecture, using a combination of a persistent EMR cluster with Spot Instances for interactive queries and transient, job-specific EMR clusters for batch workloads optimizes costs compared to running a single, massive persistent cluster.