AWS Certified Solutions Architect – Professional — Question 708

A company is using an Amazon EMR cluster to run its big data jobs. The cluster's jobs are invoked by AWS Step Functions Express Workflows that consume various Amazon Simple Queue Service (Amazon SQS) queues. The workload of this solution is variable and unpredictable. Amazon CloudWatch metrics show that the cluster's peak utilization is only 25% at times and that the cluster sits idle the rest of the time.
A solutions architect must optimize the costs of the cluster without negatively impacting the time it takes to run the various jobs.
What is the MOST cost-effective solution that meets these requirements?

Answer options

• A. Modify the EMR cluster by turning on automatic scaling of the core nodes and task nodes with a custom policy that is based on cluster utilization. Purchase Reserved Instance capacity to cover the master node.
• B. Modify the EMR cluster to use an instance fleet of Dedicated On-Demand Instances for the master node and core nodes, and to use Spot Instances for the task nodes. Define target capacity for each node type to cover the load.
• C. Purchase Reserved Instances for the master node and core nodes. Terminate all existing task nodes in the EMR cluster.
• D. Modify the EMR cluster to use capacity-optimized Spot Instances and a diversified task fleet. Define target capacity for each node type with a mix of On- Demand Instances and Spot Instances.

Correct answer: D

Explanation

Option D is correct because using a diversified task fleet with capacity-optimized Spot Instances allows Amazon EMR to provision capacity from the most available pools, minimizing Spot interruptions while drastically reducing costs for unpredictable workloads. Combining On-Demand and Spot Instances ensures that baseline processing is guaranteed while scaling out cheaply. Options A, B, and C are incorrect because they either introduce unnecessary costs (such as Dedicated On-Demand Instances or Reserved Instances for highly variable workloads) or degrade performance by removing task nodes entirely.