You are designing storage for a new global application that ingests millions of small records per minute from devices. Each record is under 4 KB, must be stored durably with low-latency retrieval, and is retained only for 120 days. Estimated annual storage is 10–15 TB. Which storage approach is the most cost-effective while meeting requirements?

DynamoDB is ideal for this use case because it offers low-latency retrieval and can handle millions of small records per minute. Its fully managed nature simplifies scaling, and the Time to Live (TTL) feature automatically deletes records older than 120 days, making it cost-effective for data with a defined retention period. Storing each record as an individual S3 object is inefficient for small files due to S3's object overhead and the cost of millions of PUT requests. Using RDS for MySQL would struggle with the high ingestion rate and the nightly cron job for deletion would be inefficient and resource-intensive. Batching records into S3 objects and using metadata search is not a practical or performant solution for retrieving individual records, as S3 metadata search is limited and not designed for this type of query.