Cloud-native applications require optimal NoSQL database selection to balance performance, scalability, and operational costs. This study presents a comprehensive empirical evaluation framework comparing five NoSQL databases (Redis, MongoDB, Cassandra, DynamoDB Local, and Couchbase) using YCSB benchmarks across datasets from 100K to 10M records. Our containerized testing environment on modern hardware reveals distinct performance profiles: Redis achieves exceptional throughput (18,450 ops/s) with ultra-low latencies (95 µs) for small datasets but shows 34% degradation at scale. MongoDB provides balanced performance (8,740-11,200 ops/s) with predictable scaling and consistent latencies (320-441 µs). Cassandra demonstrates inverse scaling, improving 21% with larger datasets while maintaining write-optimized characteristics. DynamoDB Local shows stable performance (6,200-6,800 ops/s) across scales, while Couchbase exhibits moderate scaling (5,400-8,900 ops/s). Through comprehensive analysis including concurrent workload testing, resource profiling, and tail latency evaluation, we establish quantitative selection criteria providing evidence-based guidance for NoSQL database selection in production environments.
Abbasi et al. (Thu,) studied this question.