I read a lot of articles about performance testing. The more I see the disorder, the more I cannot figure out how to implement real concurrency testing.
1. Whether to re-establish an HTTP connection every time
After reading the log documentation of LR, I found that if I had 10 concurrent users, it would have been running for half an hour. The running mechanism of LR may be like this. At first 10 threads were established to establish an HTTP connection. After the first request of the first user returned the result, the connection was not closed, continue to use this connection to send the second request. This is not the same as the actual situation. Maybe the next second users are new users. They need to re-establish HTTP connections and send requests and other operations. The establishment of an HTTP connection is omitted in LR. In fact, we have been running the first 10 connections in the last 30 hours, but they are continuously sending requests. I don't know if this will affect the results of our concurrent tests.
2. How to determine the number of concurrent jobs
The development team puts forward testing requirements and concurrency of 20, depending on the response time. Is the so-called concurrency 20 running 20 requests per second or 20 threads? If LR is used for testing, I usually adopt a target-oriented test, with the target being 20/sec trancations. I don't know how everyone works?
I saw an article a few days ago. If you join the thinking time between each request, your test results will be greatly affected. The article is in my blog, "Adjust stress testing tools". My understanding is that if the server is in the load status, that is, the speed of entering the queue is greater than the departure speed, it indicates that your service is converged and the response time includes the queuing time, this time is actually a strong addition of improper use of testing tools, rather than a real response time. Now you should join the thinking time so that the speed of entering the queue is less than the departure speed. This test is performed without server load and should be a benchmark test. This is a relatively endless loop test method. For example, if there are two concurrent users in 1 s, if there is no thinking time and an endless loop is executed, the actual request in 1 s is 100, and I want to test the number of requests per second = 50. In this case, we should add the thinking time, make requests per second = 50
3. indicators used for Capacity Testing
I think it is throughput. No matter how many users you add, your throughput can no longer be increased. This point should be the maximum server capacity. Then, based on this point, debug and obtain the number of concurrency. What do you think?
