Benchmark
Aerospike vs. ScyllaDB: Performance and cost at multi-terabyte scale
What you'll learn
As applications grow more complex with deeper dependency chains and unpredictable scale, databases must maintain predictable behavior under changing conditions, not just perform well in ideal scenarios.
To understand how these performance cliffs occur in practice, McKnight Consulting Group conducted an independent benchmarking study, which is available for download. The benchmark compared Aerospike and ScyllaDB's throughput, latency, stability, and scalability at 3TB and 6TB datasets. The results showcase why maintaining predictable performance is especially difficult at larger scales.
Key highlights
2.5× to 3× higher sustained throughput across all test scenarios
Sub-millisecond P99 latency for reads and writes, independent of cache effectiveness
Predictable tail latency, with peak latency under 7ms (vs. 200ms+ for alternatives)
Near-linear scaling when doubling datasets from 3 TB to 6 TB
40% lower infrastructure costs using one-third fewer resources