On 01/02/2019 11:19, Lindenmaier, Goetz wrote:
:
To kick in here, too:
The existing solution tells you which threads spend a lot of time
in IO operations. This is a good hint to problems in the infrastructure.
You might just write a few bytes, but still get long delays because the
disc is too slow or saturated etc.
The numbers Gunter collects give information about what your
code intends to do: how much bytes does the code write? Should
you optimize your application to communicate less? ... to write
less to file?
If you have both numbers, you can even better recognize threads
that wait too long on the network: they write few bytes but still
spend a lot of time in the calls.
So I see enough arguments to have both metrics.
Erik Gahlin put in support in JDK 12 to allow for code in java.base to
generate JFR events. I think the only user so far is in the security
area where there are events to log usages of weak crypto. I think this
is one of the approaches that will need to be looked at as part of the
exploration on the right way to add serviceability support to the libraries.
-Alan
