[jdom-interest] Toward beta 9

Vadim.Strizhevsky at morganstanley.com Vadim.Strizhevsky at morganstanley.com
Tue Apr 15 06:56:47 PDT 2003


Philip,

Did you discount JIT by ignoring a first few runs? I find that a first few
(could be as high as 30 depending on the code) runs are "polluted" by JIT.
At first the code is not JITed. then it will get JITed gradually, but
JITing itself takes time  and adds to apparent time of the code. Also
depending on the code JIT itself may take different amount  of time,
which may not correlate to the actual  performance of the code. After I
let JIT be done howerver I find the runs be quite consistent over time.

-Vadim

On Mon, 14 Apr 2003, Philip Nelson wrote:

> Not a good night 8^(
>
> I tried a few different approaches of using subclasses to skip verification.
> As a benchmark I used Dennis Sosnoskis tests comparing to the b8, b9 rc1, b9
> rc1 + Rolfs Verifier, and b9 + my SAXHandler mods.
>
> The best I did with any of these was to very slightly improve with the
> SAXHandler mods on b9.  b9 slightly degraded compared to b8 (jdk1.3 on linux).
>
>
> Some other surprises, at least to me.  Use of inner classes for node elements
> was slower than classes in their own source files.  Using an anonymous inner
> class though to subclass DefaultJDOMFactory in SAXHandler was somewhat (~15%)
> faster than a direct class.
>
> I had high hopes for the Verifier and if there is some way to skip the
> "compile" step, perhaps it would be faster.  I had my suspicions though when
> the test suite for the Verifier ran slower with the Rolfs version than the
> original did.
>
> So, back to the drawing board and so much for theory once again....
> _______________________________________________
> To control your jdom-interest membership:
> http://lists.denveronline.net/mailman/options/jdom-interest/youraddr@yourhost.com
>
>




More information about the jdom-interest mailing list