In the first part of this 3 part series on why intelligent activity metering is the only viable application performance management approach, a number of Java code profiling solutions were tested using a focused and somewhat contrived benchmark to determine the best case (lowest) unit cost in the measurement (profiling) of an instrumented method call. In closing part 1 it was put forward that there was only so much efficiency that could be squeezed out of measurement code and that it was never going to be enough to handle real-world usage, especially in low latency environments with high throughput requirements and large code coverage, unless measurement was not performed and instrumentation magically disabled (even removed) based on runtime intelligence (@see I.AM) gathered during the actual running of the application.

Read the full article here:

If you're not metering...  Part II