Before the performance tests, need to ask with the development of clear purpose of performance testing is to find the maximum load the system can support or to check a certain load operation of the system. If the latter is the case, you also need to ask the maximum number of concurrent threads on the server side setting, to set out the appropriate load when this test. Pressure test I've done several times before the http request, is to find a system can support a maximum load. The pressure test, I first local test, to 50 threads and 100 cycles, the results will appear in the test run after a period of time to wait for a long time, the beginning thought it was a poor local network conditions, will this situation will still run on the script on the server, then the server to communicate with developers know sets the maximum concurrency limitations, and only need to test the system performance under the circumstances of a particular thread can be. With this experience later, before the performance test after the know to try to ask the test requirements.
Another point to note is that the cache for the performance, the impact on performance whether the cache is usually large, with the need to confirm the good development focus is still cache when no cache situation. The pressure test, a test article is the query class subject (topic) property, the topic of property is calculated in real time, if the incoming docid is the first time the method is called, will return to the subject property takes longer, after docid query topic again when the same property is taken directly from the cache data is returned, greatly reducing time-consuming. Performance testing of the method to get the topic of property, in 40 threads, repeated 100 times under pressure, there is no cache and cache TPS when there were 70 / s and 2200 / s, the large gap found in both cases.
General performance tests need to focus on when it comes to some other information, such as cpu and memory usage, in this case because of the small number of threads, little effect on the server cpu and memory usage, so there is no detailed records.
After the pressure test, you need to compile a more detailed feedback to the developers. The pressure measurement results are as follows, in the absence of buffer, the TPS is low, and did not achieve the desired development, the need for tuning.