1. Analysis principles
From the outside to the inside, there are surfaces to the inside, layer by layer. The most direct manifestation of the performance of an application system beginning to decline is that the response time of the system becomes longer. Therefore, the system response time becomes the starting point for analyzing performance. Any complex system can be divided into two parts: the network and the server. · We can analyze other systems like WebTours, as shown in Figure 4-41.
Performance analysis is not done overnight, and needs to run through the beginning and end of the performance testing process. Performance analysis tuning is a step-by-step reasoning and verification process, just like the cartoon "Conan": bold conjecture, careful verification and troubleshooting.
2. Analysis process
Although performance analysis is a very complex process, there are rules to follow. It's like making a car is also a complicated process, but as long as it is carried out carefully in accordance with the predetermined steps, it is also the reason that a car is born.
The general analysis process is as follows.
1) Start with the summary of summary affairs. Judge whether the users are all running, whether the transaction response time is reasonable, and the transaction pass rate.
2) View the error statistics graph and the number of errors per second graph. Error statistics can intuitively see the errors that occur during operation. If the experience is enough, sometimes it can be directly located. The error number graph per second can check whether the error occurs regularly during the test. If there is, it may be that the program has problems in concurrent processing.
3) View system resources. For example, important indicators such as CPU, memory, 10, and queue change.
4) Detailed execution of virtual users and transactions. If a large number of users cannot pass, you need to check whether it is a script cause or a scene cause. If only one or a few virtual users are operating normally, there may be a problem with the script. Under normal circumstances, with the stability of virtual users, the transaction response time will not change much.
5) View the web resource map. You can stand on the server side for analysis and inference.
6) View the webpage breakdown. You can start with First Buffer Time to determine whether it is a network problem or a server problem, and then subdivide it for analysis. After mastering the above thoughts, only by thinking more, practicing more, and summing up more, Xiaobai secretly told himself.