Because all the stress tests use synchronous requests, the request QPS is correlated with the number of runners and the average latency. This relationship is as follows:

Total QPS = 1000/Average latency * 5 (number of runners) * Number of threads per runner

  • Runner: quantity of runners
  • Thread: number of YCSB threads created by each runner
  • QPS (row): number of rows written or read per second
  • Avg Latency (ms): average latency calculated by runners, in milliseconds

For example, in a scenario with four partitions, five runners, five threads per runner, and an average latency of 4 ms, the theoretical QPS for a single thread would be 1000/4 = 250. The total QPS in this scenario would be 250 * 5 (number of runners) * 5 (number of threads per runner) = 6250.

Note This performance test is not a limit test of the service performance. The test does not trigger throttling measures on the Table Store server. The automatic load balancing feature of Table Store guarantees horizontal scale-up of the service capabilities provided by a single table.