[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
|
[news.eclipse.tools.hyades] Re: Extreme performance degradation while profiling
|
Dave,
Thanks for the feedback. Currently it is not possible to set 'triggers' in
the UI to target method slicing the way you have suggested. What you could
do as an alternative is the following:
1. Set an include filter for just the method you want and exclude
everything else.
2. Turn the boundry depth of to something like 10.
This will cause a slice to occur as you suggested.
The only other mechanism for controlling how much data is collected is by
setting the profiling limits page to limit collection to something like x
method invocations (I'de recommend around 100,000 to start with) and/or
telling it only to collect for the first y seconds .
"Dave Kelsey" <dkelsey@xxxxxxxxxxxxxxxxx> wrote in message
news:cfa8pd$gg2$1@xxxxxxxxxxxxxx
> Thanks for all the responses.
> I did get it writing to a file (not using the destination tab, and so not
> using RAServer) but the performance
> was still pretty bad and it generated a file over 6Gb in size before I
> stopped it, I did try to import what
> it had generated so far but it took a long time.
>
> I tried to use the destination tab, but my eclipse setup must have got
> corrupted somewhere as no matter
> what I did, launching it through the hyades profiler it would not load the
> specified plugins. Launching it
> normally from run or debug did load the correct plugins. So I don't know
> what caused the problem
> because originally I did have it at least launching via hyades and
> collecting ok.
>
> One thought. It would be nice to have the profiler on start profiling when
a
> particular method is called and to
> stop profiling when the method is exited, I couldn't find this facility in
> hyades
>
> Dave Kelsey
>
> "Richard Duggan" <rduggan@xxxxxxxxxx> wrote in message
> news:cetj0d$3g2$1@xxxxxxxxxxxxxx
> > Dave,
> >
> > Once you have the data going to a file there are a few other knobs you
can
> > turn for performance as well. One of these is in the configuration file
> for
> > the data collection engine you can increase the channel size if your
> system
> > has enough physical memory. The default channel size is 64Mb, I often
> will
> > crank this to 128 or higher when I want to get a intensive slice of
data.
> > The immediate beinifit of increasing its size is the response time of
the
> > application under test increases. This coupled with redirecting data to
a
> > file and incrementally loading it should get you up and running quickly.
> >
> >
> > <curtispd@xxxxxxxxxx> wrote in message news:cetet5$q92$1@xxxxxxxxxxxxxx
> > > Actually there is an easier way to profile to a file other than
> > > standalone. When launching normally from the UI, under the Profiling
tab
> > > there is a Destination tab; you can specify the file here. Here you
> won't
> > > see any data in the UI because it is all being rerouted to the file.
> > >
> > > As for standalone, try running "java -XrunpiAgent:help" in a console
and
> > > it prints out the usage with the options. There might be some more
info
> in
> > > the getting_started.html in the data collector install dir. And there
> are
> > > some sample input files in the /Resources directory for filters, etc.
> > >
> > > Thanks,
> > > Curt
> >
> >
>
>