Real-world measurements on a single server with 64 GB RAM (recommended specification).
All figures below are the median of three runs in a reproducible environment.
Initial cube build aggregates data from the source database.
All subsequent queries use cube serving — millisecond response, independent of source DB load.
| Data Scale | Cube Build (one-time) | Cube Serving | Peak Memory |
|---|---|---|---|
| 100K rows | 0.3 s | < 1 s | < 200 MB |
| 1M rows | 0.4 s | < 1 s | < 200 MB |
| 10M rows | 1.7 s | < 1 s | < 200 MB |
| 100M rows | 14.4 s | < 1 s | < 200 MB |
About these numbers: Cube build time includes source-database aggregation and varies with source DB performance.
Peak memory is measured for a single query (100M rows → 200-row result).
The cube serving path maintains consistent response times as concurrent user count increases.
Measured on 10M rows with 30-column OLAP aggregation.
| Concurrent Users | Median Response (p50) | 95th-Percentile (p95) |
|---|---|---|
| 1 user | < 200 ms | < 300 ms |
| 10 users | ~ 300 ms | ~ 400 ms |
| 20 users | 417 ms | 499 ms |
Streaming processing writes results in chunks, keeping memory usage independent of data size.
All figures on this page were measured in the following environment, with full reproducibility.