Database latency impact on Jira Data Center performance

Summary

Database latency is the time it takes to send, execute and receive a response for a minimal database query.

The latency usually consists of the following components:

• database server latency itself

• network latency for a network path between the application and the database server (including network routers and switches)

• latency of proxies, load balancers or other components in the path (e.g. highly-available database solutions)

Naturally, the database latency does affect Jira – high latency would drop the performance.

The actual latency effect depends on dataset characteristics, usage patterns, hardware, etc. Still, we executed a number of performance tests against Jira 8.13.8 (with MySQL 5.7.32) to measure the latency impact at our test environments. The following graphs present the results.

To introduce the network delays, we used Toxiproxy and we used JDBC Tester to validate the delays are accurate. To measure the actions, we used Jira Performance Tests (JPT).

ℹ️ We have observed in support cases that latency might have a significantly higher impact in reindexing time than it does in normal operations seen below. A 2ms latency average could have the reindexing job take as much as 4 times longer than sub 0.5ms.

Solution

Per dataset (mixed actions)

Absolute response time

Relative slowdown

(latency effect comparing to a reference data point at 0.5 ms latency)

Per action (mixed datasets)

Absolute response time

Relative slowdown

(latency effect comparing to a reference data point at 0.5 ms latency)

Measurement data