Performance Testing Process, Plans, and Best Practices for High‑Traffic Events

The author, a test engineer at 新达达, introduces performance testing (压测) as a method of simulating real‑world traffic by gradually increasing request load to optimize system behavior and meet target goals.

Four testing schemes are compared: full‑link testing in production, proportional deployment (cloning the production environment), proportional scaling down, and test‑environment testing. Full‑link offers the most realistic results but requires system redesign; proportional deployment was chosen for the Double‑11 event due to its balance of cost and accuracy.

Preparation steps include defining the testing scope (identifying critical services and interfaces), scheduling the testing timeline (ensuring completion before the event), and setting up a testing environment that mirrors production hardware, architecture, and monitoring platforms.

Three types of test scripts are required: data‑generation scripts, traffic‑generation scripts for single‑interface and mixed‑scenario phases, and analysis scripts to collect and process results efficiently.

Key performance metrics to analyze are TPS, response times (average, 95th, 99th percentiles), CPU/memory usage across servers, databases, and middleware, as well as connection counts and slow queries.

Practical tips emphasize understanding business logic, involving all stakeholders, disabling protective mechanisms during testing, and meticulously recording each test iteration.

In conclusion, the Double‑11 load test enabled 新达达 to handle up to 8 million daily orders smoothly, and the article invites readers to follow the series for more technical insights.