Re: [java-gnome-hackers] Having benefited from YourKit...

[Vreixo F. L. <met...@ya...>]

--- Andrew Cowie <an...@op...>
escreveu:

> YourKit LLC very kindly donates licences for its
> Java profiling tool to
> individual hackers within Open Source projects,
> which is, as ever, a
> very legitimate way of contributing to the FOSS
> movement.
> 
> YourKit was recommended to me by one of my clients,
> and so about a year
> ago I requested such a licence, and was kindly
> granted one. I have
> hardly used YourKit (early on it told me that trying
> to fix java-gnome
> 2.x was going to be impossible, but I already knew
> that) ... but a
> casual "hey, that could be interesting" run *very*
> recently turned out
> to be a great help.
> 
> And then the other day I received a courtesy ping
> from Vladimir
> Kondratyev. You may be interested in my reply, which
> I enclose below.
> 
> AfC
> Sydney
> 
> 
> P.S. If any of the java-gnome 4.0 hackers are
> interested in such a
> licence, I would certainly be willing to vouch for
> your contribution.
> Character references are a confidential matter, so
> contact me privately
> if you would like such a referral.
> 
> -------- Forwarded Message --------
> From: Andrew Cowie <an...@op...>
> To: Vladimir Kondratyev <sa...@yo...>
> Subject: Re: Request for an open-source project
> licence for use with
> java-gnome
> Date: Thu, 15 Feb 2007 16:57:34 +1100
> 
> On Mon, 2007-02-12 at 18:53 +0300, sa...@yo...
> wrote:
> > Congratulations on releasing new Java bindings for
> GNOME
> >
>
http://research.operationaldynamics.com/blogs/andrew/software/java-gnome/java-gnome-4.0.2-release.html
> 
> > Did YourKit Java Profiler help to find performance
> problems or memory
> > leaks?
> 
> I was about to say "no", but then I thought about
> it, and actually, the
> answer was "yes"! 
> 
> So thank you! I owe you a blog post. :)
> 
> Also, one day fairly soon I will be putting a
> "supporters" page up on
> the website, and you will most certainly have the
> mention there you
> originally requested.
> 
> We don't have any large applications in java-gnome
> 4.0 yet, so
> performance impact analysis is still a bit
> contrived. But the much to my
> surprise, it was the memory allocation side of yjp
> that helped us.
> 
> About two weeks before 4.0.2 actually came out, we
> were doing clean-ups
> and and thinking we were ready to release. One of my
> staff suddenly
> realized that they were not seeing any debug
> statements about the GValue
> proxies freeing the memory that they had allocated.
> 
> Ooops.
> 
> So I ran yjp to see if I could figure out why.
> YourKit ended up
> reminding me that I had not gotten around to
> implementing the memory
> management algorithms. I knew that, of course, but
> had (being honest)
> forgotten about it with a "TODO later" comment. By
> showing me a vast
> quantity of these Value objects piling up and not
> being garbage
> collected, yjp showed where the strong references
> were piling up, and
> that made me realize it was high time to do the
> TODO. So I did, and with
> the enormous help of Vreixo Formoso Lopes, we have
> implemented fully
> compliant cross boundary memory and reference
> handling in java-gnome 4.0
> 
> Some feedback:
> 
> As is not unusual with profilers, the problem is
> ever to swim past the
> allocations of objects internal to the Java VM so
> you can see things
> that are "interesting". To be honest, it was only an
> accident that I was
> able to realize from yourkit that something was
> wrong - most of the way
> I was either battling signal to noise ratio or not
> recognizing that the
> data I was being shown actually *was* significant.
> 
> I'm sure you know this already, but some way to
> treat Java Collections
> as a single entity would be really cool. And even
> better, a default
> configuration which automatically filtered stuff
> like ClassLoader
> loading classes and sun.* internal classes would be
> nice - there is
> nothing I can do about any of that, and so I'm not
> really interested in
> the time it takes or the objects it allocates.
> 
> Finally, for completeness, I would observer that the
> CPU side of YourKit
> (the fun part!) hasn't been as useful, primarily
> because it's hard to
> see past the fact that 99% of time is spent in GTK's
> main loop - we make
> one call to a native method to start gtk_main(), and
> that blocks, and as
> far as YourKit is concerned, that's CPU usage.
> 
> yjp is very good at showing top down analysis, but
> in the case of our
> event driven programming, where the code paths that
> you want to look
> into start with a JNI invocation of an arbitrary
> method somewhere deep.
> and then go back up from there, it can be difficult
> to get to that
> method to then start doing merged callees and
> backtraces. I'm not sure
> what to suggest in this case, but I imagine anyone
> debugging an event
> driven program that mixes Java and native C code and
> where the events
> originate from the native side has run into this.
> 
> I realize feedback from a non-paying-customer isn't
> top of the priority
> list, but that's been my experience so far. YourKit
> _did_ help out, and
> so I thank you very much.
> 
> AfC
> Sydney
> 
> 
> -- 
> Andrew Frederick Cowie
> 
> Technology strategy, managing change, establishing
> procedures,
> and executing successful upgrades to mission
> critical business
> infrastructure.
> 
> http://www.operationaldynamics.com/
> 
> Sydney   New York   Toronto   London
> >
-------------------------------------------------------------------------
> Take Surveys. Earn Cash. Influence the Future of IT
> Join SourceForge.net's Techsay panel and you'll get
> the chance to share your
> opinions on IT & business topics through brief
> surveys-and earn cash
>
http://www.techsay.com/default.php?page=join.php&p=sourceforge&CID=DEVDEV>
_______________________________________________
> java-gnome-hackers mailing list
> jav...@li...
>
https://lists.sourceforge.net/lists/listinfo/java-gnome-hackers
> 


__________________________________________________
Fale com seus amigos  de graça com o novo Yahoo! Messenger 
http://br.messenger.yahoo.com/