Review request for 6857194

[Mandy Chung]

Keith,

Thanks for reviewing the first version.
    http://cr.openjdk.java.net/~mchung/6857194/webrev.01/

I made a fix in jvm.cpp line 759 that missed to name the variable in 
webrev.00 and thus caused the small defineClass time. I also changed 
parseClassFile to use the same solution as oops/instanceKlass.cpp using 
the recursion_counter.

Here is the result of extracting from the perfdata of a few apps runs:
    http://cr.openjdk.java.net/~mchung/startup_measurement/perfdata.summary

Thanks
Mandy

Mandy Chung wrote:
> Karen, Keith,
>
> I add several new perf counters to collect time spent in various 
> aspects about class loading.
>
> Webrev:
>  http://cr.openjdk.java.net/~mchung/6857194/webrev.00/
>
> Changes:
> 1. Counters to collect the time spent in defineClass and the byte size 
> read.
> 2. Counters to collect the number of bytes read when parsing a class 
> file stream.
> The existing sun.cls.loadedBytes and sharedLoadedBytes counter (in 
> classLoadingService.cpp - that was carried from 1.4.2) represent the 
> footprint of a instanceKlass.  The new counters measure the numbers of 
> bytes in the classfile read.
>
> 3. Existing class init, class link, and class verify counters include 
> the class loading and resolution.  I add the counters to measure the 
> exclusive time (i.e. self time) as well.
>
> I ran startup3 on Windows XP.  No significant overhead.
> ============================================================================== 
>
> mchung.baseline:
> Benchmark           Samples        Mean     Stdev             Geomean 
> Weight
> startup3                 15        2.36      0.04
> ============================================================================== 
>
> mchung.hsfix:
> Benchmark           Samples        Mean     Stdev   %Diff     P  
> Significant
> startup3                 15        2.35      0.05    0.24 
> 0.714            *
> ============================================================================== 
>
>
> Thanks
> Mandy