A company ingests a continuous stream of market data at a constant rate and stores it. A nightly aggregation job runs for about 4 hours to compute statistics; the analysis is noncritical and can be retried on the next run if it fails. The current system uses a pool of 1-year Reserved Instances running constantly to ingest and store data on attached EBS volumes. A scheduled script launches On-Demand EC2 instances nightly to run the aggregation, which access data over NFS from the ingestion servers, and terminates them when work finishes. The Reserved Instance reservations are expiring. The company wants the most cost-effective option. Which solution should the architect recommend?

This option is the most cost-effective. Kinesis Data Firehose is a managed service for streaming data ingestion, eliminating the need for EC2 instances and EBS volumes for ingestion and storage, reducing operational overhead and cost. Delivering data to S3 leverages S3's low-cost, highly durable storage. For the nightly aggregation, AWS Batch with Spot Instances is ideal because the analysis is noncritical and can be retried. Spot Instances offer significant cost savings (up to 90% off On-Demand prices) and are suitable for fault-tolerant, flexible workloads. Limiting Spot bids helps manage costs further. The first incorrect option keeps On-Demand instances for processing, which are more expensive than Spot Instances. The third incorrect option continues using Reserved Instances for ingestion, which is less cost-effective than Kinesis Data Firehose and S3 for this use case, and also continues to use EC2 instances for ingestion. The fourth incorrect option uses Amazon Redshift, which is an analytical data warehouse and can be significantly more expensive than S3 for raw data storage and batch processing, especially for noncritical, retryable workloads. Lambda for heavy aggregation might also hit performance or cost limits.