Test result (capacity instance)
The following observations apply to all four test scenarios below.
When a shard has adequate resources, latency stays flat as the load increases — and QPS for random write operations continues to increase.
When a shard is under pressure, latency for random write operations grows as the load increases, causing a request queue on the server hosting that shard. In this state, QPS does not scale linearly.
Cache hit rate significantly affects random read and random get range performance. To isolate performance from cache effects, the tests used an extremely low hit rate. Actual production performance quite likely exceeds these results.
These tests measure baseline performance, not the service ceiling. No throttling was triggered on the Tablestore server during testing. Tablestore's automatic load balancing guarantees horizontal scaling of service capabilities for a single table.
Random write performance
Random read performance
Random batch write performance
Random get range performance
