I was reading the article about HPROF:
As I couldn't leave a comment there (probably my proxy?), I came to this forum to ask my simple question:
Is the HPROF "cpu=old" option a sort of deprecated format and how does it work ?
After a try, I observed the same huge CPU overhead as with "cpu=times" option, which is ok, and as an output, a tabbed file. It looks like a list of methods sorted by execution time, with the following header "count callee caller time". This is clear, no extra explanation needed.
This option is very interesting so why isn't it described in this article ? Is it the old format of JVMPI /JVMDI ? However the history section shows that the output format has not changed.
This could seem a bit redundant with the cpu=times options, but after a quick comparison on a "real" application on a 2h internal standard perf test, it shows very different results, in terms of figures and top-consuming methods ranking.
Could you give me the beginnings of an explanation for this difference, by detailing a bit how it works ?
Which option would you advise to use ?