Increasing Java Startup Performance

[Stefano Mazzocchi <st...@apache.org>]

I went to a this talk this morning

http://www.csail.mit.edu/events/eventcalendar/calendar.php?show=event&id=978

that presented the paper found at

http://www.research.ibm.com/people/m/marnold/arnold-welc-rajan.final.pdf

which implements a very interesting trick to speed up java startup 
performance: save the hotspot information in a repository (at JVM shut 
down) so that the JIT doesn't have to wait when it starts until it knows 
what is a hotspot to start compiling it. the performance improvements 
are not likely to change the perception that java is awefully slow to 
start up on the desktop, but it's a clever idea.

In the meanwhile, Sun's attempt to solve the same problem in JSE6 is this:

http://java.sun.com/developer/JDCTechTips/2005/tt1115.html#1

No comment.

-- 
Stefano.

Re: Increasing Java Startup Performance

[Paulex Yang <pa...@gmail.com>]

Does the Java 5 feature CDS(class data sharing) work on this issue?

pls. refer to
<Java SE 5 document root>/docs/guide/vm/class-data-sharing.html

Wes Felter wrote:
> Stefano Mazzocchi wrote:
>
>> http://www.research.ibm.com/people/m/marnold/arnold-welc-rajan.final.pdf
>>
>> which implements a very interesting trick to speed up java startup 
>> performance: save the hotspot information in a repository (at JVM 
>> shut down) so that the JIT doesn't have to wait when it starts until 
>> it knows what is a hotspot to start compiling it. the performance 
>> improvements are not likely to change the perception that java is 
>> awefully slow to start up on the desktop, but it's a clever idea.
>
> I'm surprised there's been so little discussion of the Java Executable 
> File Format (JEFF) (unfortunately, the Web site for it seems to have 
> disappeared), which should reduce memory usage and cold-start time by 
> allowing core classes to be mmap'ed. This would be mostly orthogonal 
> to the above paper.
>
> Wes Felter - wesley@felter.org
>
>


-- 
Paulex Yang
China Software Development Lab
IBM

Re: Increasing Java Startup Performance

[Wes Felter <we...@felter.org>]

Stefano Mazzocchi wrote:

> http://www.research.ibm.com/people/m/marnold/arnold-welc-rajan.final.pdf
> 
> which implements a very interesting trick to speed up java startup 
> performance: save the hotspot information in a repository (at JVM shut 
> down) so that the JIT doesn't have to wait when it starts until it knows 
> what is a hotspot to start compiling it. the performance improvements 
> are not likely to change the perception that java is awefully slow to 
> start up on the desktop, but it's a clever idea.

I'm surprised there's been so little discussion of the Java Executable 
File Format (JEFF) (unfortunately, the Web site for it seems to have 
disappeared), which should reduce memory usage and cold-start time by 
allowing core classes to be mmap'ed. This would be mostly orthogonal to 
the above paper.

Wes Felter - wesley@felter.org

Re: Increasing Java Startup Performance

[Geir Magnusson Jr <ge...@pobox.com>]

The wiki certainly works well for this...

Arzhan Kinzhalin wrote:
> On 2/16/06, Stefano Mazzocchi <st...@apache.org> wrote:
>> http://www.research.ibm.com/people/m/marnold/arnold-welc-rajan.final.pdf
>>
>> which implements a very interesting trick to speed up java startup
>> performance: save the hotspot information in a repository (at JVM shut
>> down) so that the JIT doesn't have to wait when it starts until it knows
>> what is a hotspot to start compiling it. the performance improvements
>> are not likely to change the perception that java is awefully slow to
>> start up on the desktop, but it's a clever idea.
> 
> Stefano,
> 
> this is very interesting. A little bit too advanced as of right now though. :)
> 
> I guess it makes sense to collect interesting papers and attend them
> later on when the time comes. Perhaps, Wiki is the right place? I've
> seen papers there, but they mostly high-level. "Idea stash" for that
> purpose could be worth creating.
> 
> --
> Arzhan
> Intel Middleware Platform Division
> 
> 

Re: Increasing Java Startup Performance

[Arzhan Kinzhalin <li...@gmail.com>]

On 2/16/06, Stefano Mazzocchi <st...@apache.org> wrote:
> http://www.research.ibm.com/people/m/marnold/arnold-welc-rajan.final.pdf
>
> which implements a very interesting trick to speed up java startup
> performance: save the hotspot information in a repository (at JVM shut
> down) so that the JIT doesn't have to wait when it starts until it knows
> what is a hotspot to start compiling it. the performance improvements
> are not likely to change the perception that java is awefully slow to
> start up on the desktop, but it's a clever idea.

Stefano,

this is very interesting. A little bit too advanced as of right now though. :)

I guess it makes sense to collect interesting papers and attend them
later on when the time comes. Perhaps, Wiki is the right place? I've
seen papers there, but they mostly high-level. "Idea stash" for that
purpose could be worth creating.

--
Arzhan
Intel Middleware Platform Division