The concept of software test was earliered from the teacher from the teacher (all of which lost light): Development is to pass as much as possible; and test, it is as much as possible to make the program. The difference between the two is to select a test instance to be different in design guiding ideas. Although this sentence is simple, but it is easy to remember, it is true that it is really a profit. There was still no story in JavaWindows, so-called software, in fact, C language and compilation, but this idea I think it can be used to develop and test almost aspects of the test.
Generally speaking, usability testing and black box testing This kind of job can't afford my interest and don't be my job responsibility. In addition to the receiving views, normal conditions will be more eye-catching. Moreover, in order to make the usability test less trouble, it has long been developed, as long as it is not the project in the Requirement Specifics, don't do more, do more wrong. (However, if the project is not standardized, it is another matter when there is no specifications, and I will adhere to the demand since the demand, it is also reasonable requirement, which is a reasonable requirement. . It truly needs to worry about the cell test, and pressure test in the operation; both of which are mutually conflict.
Holding the unit testing from the "Ekka Autobiography": The product is discovered in front of the factory, the cost of spending is to discover the fault after the factory. The unit component is my product, the unit test is the test before the factory. It is called back to the factory after the factory: "Your program is wrong", it is better to test it with himself. So my habits I don't know if it is good, but the test is in the heart, basically every spending Manday is developed, it will spend a manday test, including writing test routines, organizing test data; (both together That is the test example); the time of the document must be counted. JUnit, as a tool for unit test, I only used in EJB environments; this is not easy to test the EJB test, the buried breakpoint of usually used, writing test classes in the EJB container, EJB is actually There is no feature for the class setting output log for the EJB itself. It is a mistake that is full of system logs. It may be just an error log in dozens of EJBs. In contrast, JUnit and its extended like Junitee provide a function of impanding other parts to test EJB components. This is actually the basic idea of JUnit, and the components of the hierarchical development implementation of the fixed interface, with its own write test environment and then MAIN, it is better to use a unified test framework. This is also an indispensable part of efficient development, but for other classes, I can't see any benefits for using junit: The class itself can output breakpoints, and can run a few lines of code in main, what is better than this Is the unit test more effective? Why do you run to JUNIT?
However, the larger the intensity of the unit test, the more strong, but the more cases, the more IF-Else, the more IF, but it is actually consumed system resources; more components One point, in fact, the load capacity of the entire system has dropped very visible. Therefore, the unit test intensity is conflict with the last pressure test. It's just in a short time, I just want to manage the unit, I can't bite me, and I don't think it is necessary to consider it. I don't think it is. Until the system performance decreases at runtime, I can't ask for the overall performance when I decrease in the run. At this time, the interest in the pressure test litered. Load test is a luxury in a considerable period of time. Loadrunner is not only a big guy, it is not easy to pirate, one of the reasons is not my main business, so I will not specifically play. This; anyway, I never get a version that can be used for a long time, nor does it really learn this. In addition, I am worried that I am worried, I really don't have much time when my load is too heavy, and there is always a way to respond. If I don't have a new project, it is estimated that I will not spend too much in this. The tool originally written in a pressure test ended in failure: the thread was carried out as required, but also issued an HTTP request, but the system is normal as if there is no matter; but it is loaded on the working environment. Obviously, this is not the purpose of the test program, not the system is really so strong. Reason, in fact, I am not very much today. Not only that, these two days of trial of WebCT and Webrunner two domestic software, the result is exactly the same as I wrote, no matter how I set up pressurization, the system is also the wind does not blow. Don't think it is systematic. The later way is to use Jakarta HttpClient, call proxy, a big amount of these proxy online, as long as the collection is always a considerable amount; this subtitative local area network does not measure, but the online website is ok; it also replicates large visits In the case, the system hangs half-hook, or you can also check how many proxy when calling, how to reach the hang-up effect, how to optimize it will not hang up, and slow down and recover. Bry hard to use, in fact, a tool that has been done, check the DDOS response; but collect proxy is really troublesome and lacks log statistics (need to write a procedure); no session, COKKIE, etc. Web download, then verify the keyword, make sure the download is complete. Moreover, since it is necessary to start repeatedly, the agent server response has some time, and the system limit recovery is also a period of time; so use this approach to 700 more than 700, and the latter has ended.
In these days, in order to test the load limit of the website, according to a friend's recommendation, I used a few pressure tools. The above WebCT / Webrunner is one of them. I didn't have to ran in a few years. Microsoft's WAS and JMeter became the main way of trial, which is also a relatively perfect load tool that can be found in easily load tools. WAS settings are simple, and the results are also repeatable. However, the report is not intuitive, but it is not possible to run and report continuously; worse, it does not verify the function of the request (or I will not use, no found), which just issued a request, as for the request? Complete execution is still returned to an empty file, I really know. In some cases, it is also a lot of request to turn the webpage. So I understand that the understanding of WAS is not to see its own output log, but focus on whether there is abnormality from the test log from the tester itself in its own record log, and can be restored in the limit state The result is judged. In contrast, Jmeter is better, this thing is not difficult, and the related articles will basically use it. It has the function of verifying whether the request is properly executed, and can also be easily discovered by continuous requesting to observe the response and departure conditions, and it is easier to find the fine difference between performance to performance; to optimize the server, it is better than WAS. However, Jmeter seems to have bugs. Anyway, I have tried several times to start startup, and I can't turn off the closure.
Whether using WAS or JMeter, it is easy to put a server session count for thousands. But blame, if you use my own long-selling proxy server to simulate DDoS attack, press the request frequency, soon, the target target can enter the limit state, even applying Crash, must be restarted. But use WAS / JMETER, no matter how much load is transferred, it is only a slow response, and the target machine can't die; until my own test machine is dead, the target machine will live and dance. There is no spectrum there; it is estimated that it is related to the working method of the proxy server; I found that the target of the target machine is twice as high as the thread request I actually issued.
Anyway, if you plan to use a unit test for a long time to test the web application load, Jmeter should be a good choice; another WAS is not a waste. Loadrunner and other people who have no money to do with money, they will be exempt from the time being.
Corel
