AWS Certified Solutions Architect – Associate (SAA-C03)Design Resilient ArchitecturesMedium
An analytics company processes large datasets daily using Amazon EMR clusters. The processing is time-sensitive but can tolerate occasional interruptions to save costs. The company also needs to ensure that the data processed is consistent and durable. Which combination of AWS services should be used for the EMR cluster's compute instances and its input/output data storage?
- AOn-Demand Instances for compute, Amazon EBS for data storage
- BDedicated Hosts for compute, Amazon Glacier for data storage
- CSpot Instances for compute, Amazon S3 for data storage
- DReserved Instances for compute, Amazon EFS for data storage
Show answer & explanationAnswer & explanation
Correct answer: C. Spot Instances for compute, Amazon S3 for data storage
Spot Instances are ideal for fault-tolerant, flexible workloads like EMR, where interruptions are tolerable, offering significant cost savings. Amazon S3 provides highly durable and scalable object storage, suitable for EMR input and output data, ensuring data consistency and durability even if compute instances are interrupted.
Why the other options are wrong
- A. On-Demand is more expensive; EBS is instance-attached and not ideal for distributed, transient EMR data.
- B. Dedicated Hosts are for specific licensing requirements and are very expensive; Glacier is for archival, not for active EMR processing.
- D. Reserved Instances are for predictable, long-term workloads; EFS is file storage, generally not the primary choice for large-scale EMR data I/O.
EMR with Spot & S3
A common and cost-effective architecture for Amazon EMR clusters, using EC2 Spot Instances for compute to reduce costs and Amazon S3 for durable, scalable storage of input and output data.
- Spot Instances offer up to 90% savings for interruptible workloads.
- S3 provides high durability and availability for data, decoupling storage from compute.
- EMR is designed to be resilient to instance failures (including Spot interruptions).
- Ideal for batch processing, analytics, and big data workloads.
Memory trick: Spot's the Deal, S3's the Data Seal.