All of lore.kernel.org
 help / color / mirror / Atom feed
* JVM Profiling: Does hotspot emit frame pointer, and other Q's
@ 2015-01-10  9:03 Dennis
  2015-01-10  9:37 ` Dennis
  0 siblings, 1 reply; 2+ messages in thread
From: Dennis @ 2015-01-10  9:03 UTC (permalink / raw)
  To: linux-perf-users

Hi,

http://www.brendangregg.com/perf.html mentions the following** in
regards to profiling JVMs.

1) Is this still an issue, if not when was it fixed?
2) Are there workarounds? (Other JVMs?)
3) Does it even matter? What is the status quo? Are symbols just not
resolved or are parts of the stack omitted completely?

3 The lightweight-java-profiler seems to be able to resolve most
symbols, but I don't understand why it can and perf cannot.

** Note that Java may not show full stacks to begin with, due to
hotspot omitting the frame pointer (just like gcc). There is currently
no option to disable this optimization (either in OpenJDK or Oracle
JDK), meaning that profiling is basically broken. I've heard that the
performance benefit for omitting the frame pointer in hotspot is
negligible, meaning, we lose the incredibly useful ability to profile
stack traces for no real gain. Hotspot, at least, need to provide us
with an option to not omit the frame pointer.

P.S. I love the perf tool!


Regards,

Dennis Ordanov

^ permalink raw reply	[flat|nested] 2+ messages in thread

* Re: JVM Profiling: Does hotspot emit frame pointer, and other Q's
  2015-01-10  9:03 JVM Profiling: Does hotspot emit frame pointer, and other Q's Dennis
@ 2015-01-10  9:37 ` Dennis
  0 siblings, 0 replies; 2+ messages in thread
From: Dennis @ 2015-01-10  9:37 UTC (permalink / raw)
  To: linux-perf-users

Apologies for tempting fate and not reading the archives :-).


"Perf support for interpreted and Just-In-Time translated languages"
http://thread.gmane.org/gmane.linux.kernel.perf.user/1817
and
"A hotspot patch for stack profiling (frame pointer)"
http://mail.openjdk.java.net/pipermail/hotspot-compiler-dev/2014-December/016546.html

Would it please the mailing list if I merge my questions into the
perf-users thread above or continue with this one?  Most of my
questions still stand though and I don't want to derail that thread.

Between either or both the light-weight-java profiler and perf I can
find a way to come as close to possible to always on / continuous
profiling of a JVM.


Regards,

Dennis O.

On Sat, Jan 10, 2015 at 1:03 AM, Dennis <daodennis@gmail.com> wrote:
> Hi,
>
> http://www.brendangregg.com/perf.html mentions the following** in
> regards to profiling JVMs.
>
> 1) Is this still an issue, if not when was it fixed?
> 2) Are there workarounds? (Other JVMs?)
> 3) Does it even matter? What is the status quo? Are symbols just not
> resolved or are parts of the stack omitted completely?
>
> 3 The lightweight-java-profiler seems to be able to resolve most
> symbols, but I don't understand why it can and perf cannot.
>
> ** Note that Java may not show full stacks to begin with, due to
> hotspot omitting the frame pointer (just like gcc). There is currently
> no option to disable this optimization (either in OpenJDK or Oracle
> JDK), meaning that profiling is basically broken. I've heard that the
> performance benefit for omitting the frame pointer in hotspot is
> negligible, meaning, we lose the incredibly useful ability to profile
> stack traces for no real gain. Hotspot, at least, need to provide us
> with an option to not omit the frame pointer.
>
> P.S. I love the perf tool!
>
>
> Regards,
>
> Dennis Ordanov

^ permalink raw reply	[flat|nested] 2+ messages in thread

end of thread, other threads:[~2015-01-10  9:37 UTC | newest]

Thread overview: 2+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2015-01-10  9:03 JVM Profiling: Does hotspot emit frame pointer, and other Q's Dennis
2015-01-10  9:37 ` Dennis

This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.