By providing a load change file, all threads are ready before running, the threads that are not needed are in the idle state, and the threads after use return to the sleep or idle state. The runtime report allows you to view parameters such as response time at runtime, rather than waiting until all benchmarks have been run.
The format of the load change file is: <runtime in secs>,<thread count> . The file contains many pairs of runtime and thread count. Configure <fa:variableLoad> true </fa:variableLoad> in the run.xml file
<fa:variableLoadFile>myvariationfile</fa:variableLoadFile> ，
where myvariationfile is the location of the file on this machine. Note: The total time in the load change file must be equal to the steadytime in the configuration file. And the maximum thread count in the load change file must be the same as the running thread, not the same as the number of scales, because a scale may have multiple agnets specified in the Driver class.
If the agent is distributed, then a load change file must be placed manually on each agent machine. Moreover, the thread count specified in the load change file is for each agent. If there are multiple agents, it is equivalent to n*thread count.
But now a problem is that I experimented with two benchmarks, one of which is 20 and the other is 200, but the requests per second are the same. What is this asking? And that load change time is not implemented as expected. Also take a look at how this function is implemented in the faban code.
Looking at the source code, there is roughly a switching time, agent.loadSwitchTime += load.runTime * 1000000000l; then loadswitchtime is required for a time, is it because of this that the time for a small number of users is too short?