Talk about stress testing outside the pedestrian's eye

Not a full-time stress-testing profession, you can only use your own experience. The stress test is not just a stress test process, but a fairly systematic and complex project, and I think the stress test is designed to allow the system to achieve the desired performance and withstand the desired pressure. This also requires that stress testing should help the performance Tuning team, to provide a degree of guidance, where I do not make the stress test and performance tuning so clear, in my view, the stress testing process includes: Clear the goal of stress testing, develop a stress test plan, stress testing, analysis of stress test results, Looking for bottlenecks and tuning to achieve the goal, in this blog to look at these processes and common methods.

Identify the target of the stress test

Usually (note is usually), the goal of the stress test is to have these points:

1, the evaluation system to meet the requirements of pressure support

To gauge whether the system meets the requirements of pressure support, it is also necessary to define exactly how much pressure the system needs to support, such as:

Machine configuration: CPU, memory, hard disk, etc.

Network condition: 100M

Operating system: Linux core:2.6

When the number of concurrent numbers is 10 users, the system should be able to respond in 20ms (this time the TPS is 500), the system's load must be under 2, when the concurrent number of 100 users, the system should be able to respond in 50ms (this time the TPS is 2000), the system load must be under 4 ; When the number of concurrent numbers is 200 users, the system should be able to respond in 80ms (this time the TPS is 2500), allowing 1 per thousand of the error rate, the system load needs to be under 6, in the process of stress testing, as long as any of these indicators did not meet, can be judged the system has not reached the target of pressure.

The actual stress test will be a lot more complicated than the example I have here, such as network traffic, memory consumption, IOPS, number of connections, and so on.

This stress test hidden goal is to provide some guidance for capacity planning, such as the current system in a certain configuration of a single machine can withstand the maximum concurrent number of 100 users, if the system peak pressure is 1000, If the system supports lossless horizontal scaling, it means that 10 machines of this type are required, and this step is also tested.

2. Estimate the status of the system running on line

After all, there's a lot of difference between the stress test environment and the online environment, pressure, data volume, hardware environment, and so on, basically can only be based on the environment under the circumstances of a certain proportion of the comparison after the calculation to estimate, here is very important to predict the system after the normal performance of the situation, The health and risk points after a certain growth rate (for example, problems can occur when the number of concurrent users grows, and how much the system is load).

I think it is very difficult to achieve this goal, but with experience, I believe it can be done, if it can do this effect will be a great help.

The above two goals are basically a stress test to achieve or want to achieve, and specific may be due to the specific circumstances of the system's business will set other different goals.