Sampling
This mode periodically accesses the stack and evaluates the code that is running slowly. This mode has no method call count, only CPU time. This pattern is the best model for providing a bottleneck analysis, with minimal additional overhead for the server.
Tracing
Using tracing mode, each method of the monitored application is converted into bytecode to record its CPU consumption. The number of method calls and CPU time are recorded.
However, tracing mode has a large impact on the server.
The more methods of service being detected, the greater the impact on server performance
Call counting
The most lightweight CPU performance analysis mode with almost zero performance impact on the server
The pattern technology is not necessarily completely accurate. In order to ensure that the performance impact on server opportunities is zero,
This mode does not apply any of the locks and Test-and-set atomic operations. Therefore, if the same method is multi-threaded, and at the same time on different cores of different CPUs are called, a count error occurs.
The mode does not support the call count of auto activity mode, and does not collect hight-level-statistics
After you start monitoring, the real-time results are just basic information about the CPU, you want to see the details, after you save the snapshot, view the snapshot
Yourkit study notes-start CPU profiling. Sampling, tracing, call counting.