Throughput measures how much work a system completes over time. In ScyllaDB testing, it appears as operations, partitions, or rows processed per second. section 11.2.2
Use throughput to interpret database performance:
- Increase test traffic gradually and watch whether throughput continues to grow. section 11.2.2
- Read throughput can rise while latency worsens, so evaluate both measurements together. section 11.2.2
- During migration, raise source or destination throughput only until either database begins degrading. section 12.2
A load-test comparison
One test reaches 5,476 operations per second with 4.5-millisecond 99.9th-percentile latency. A heavier run reaches 24,619 operations per second, but latency rises to 44.9 milliseconds at the same percentile.
