Performance / Measured results
Performance at scale
Compare AtalantaDB CE and EE with Aerospike CE on the same hardware and workloads. Explore measured throughput and read the full 21-test report.
Aerospike CE vs AtalantaDB
Three-node cluster load test
- 3 nodes
- Replication factor 2
- 10 fields
- 50–150 bytes per field
| Operation | Aerospike CE | AtalantaDB CE | AtalantaDB EE | Details |
|---|---|---|---|---|
| INSERT | 271.3Baseline | 295.31.09× baseline | 291.81.08× baseline | |
| UPDATE | 204.7Baseline | 216.21.06× baseline | 238.91.17× baseline | |
| GET | 528.9Baseline | 530.21.00× baseline | 502.60.95× baseline | |
| Operation | Aerospike CE | AtalantaDB CE | AtalantaDB EE | Details |
|---|---|---|---|---|
| INSERT | 461.0Baseline | 779.61.69× baseline | 779.11.69× baseline | |
| UPDATE | 414.4Baseline | 430.51.04× baseline | 579.11.40× baseline | |
| GET | 681.1Baseline | 973.61.43× baseline | 1,720.62.53× baseline | |
Means include startup and completion samples. Each run is averaged over its own available samples; blank cells after completion are excluded. Ratios use unrounded means.
Batch workloads show the largest gains. Single-record results are closer: CE is nearly level with Aerospike CE on reads, while EE is slightly slower in that test.
Read the full technical article online
Architecture, benchmarks, deployment and memory sizing.
Methodology & test environment
YCSB runs INSERT, UPDATE and GET separately. Ten load-generator hosts use 50 worker threads each. UPDATE results use the second pass to include the effect of compaction.
- Database servers
- 80 hardware threads at 3.1 GHz · 750 GB RAM · 4 × 6 TB NVMe · XFS
- Load generators
- 10 hosts · 80 hardware threads at 2.3 GHz per host · 50 workers per host
- Dataset
- 150 million records per generator · 1.5 billion records in total · 10 fields of 50–150 bytes each
- Deployment
- 3 database instances · replication factor 2 · Aerospike CE 8.2.0
How the table is calculated
Each value is the mean of that database’s available 10-second samples, including startup and completion samples. Blank cells after a run finishes are excluded, not counted as zero. Throughput is per node; batch throughput counts individual record operations, not batch requests. Ratios compare unrounded means with Aerospike CE in the same workload.
Scope of the comparison
This table covers six measured three-node workloads. The full report contains 21 test cases, Cassandra comparisons and latency results. The modelled AS_EE_AF series is separate from measurements and is not included here. Results describe these test conditions; small differences do not establish a statistically significant advantage.
The log archive contains YCSB logs and operations-per-second measurements from an AtalantaDB instance. Individual source CSV files are available under each chart.
View the YCSB implementation ↗