Re: [pcgen_developers] Re: Benchmarking Chris' PCGen Memory Improvements
- Hi Paul,
I halted my project. The performance was too poor to be viable without a
massive overhaul, which was beyond what I wanted to do in my hobby time.
Before I gave up, I got the app to load and render a sample PC on my
JellyBean tablet and my ICS phone, but it took 7 and 35 minutes
respectively. Most of that time was spent in the Dalvik garbage collector,
I believe, while parsing the many .lst files. For comparison, a nearly
identical load-and-render Swing app ran in about 10 seconds under the
Oracle JDK on my desktop.
All of my code (my PCGen mods and my Android test app) is in github. I've
posted the URLs a few times to this list.
On Wed, April 10, 2013 10:53 am, Paul wrote:
> Out of curiosity how far along are you on the android app Chris? I see the
> last check-if was 2 months ago. Also is this aimed for tablets only, or
> would it also work with smartphones?
> From reading the code tracker this will significantly impact subsequent
> performance, but initial loading will take a similar time?
> --- In firstname.lastname@example.org, James Dempsey <jdempsey@...>
>> I've just finished checking in Chris Dolan's work on Lst loading memory
>> optimisation. The work is tracked in
>> As there was an expectation that this would have an impact on
>> performance I ran some simple benchmarks for each change - ant inttest
>> run 3 times and the run times averaged.
>> Change Avg Time Improvement
>> Base code 487
>> With 1st attempt 495 -2%
>> CDOM lazy map init 265 46%
>> Intern strings 301 38%
>> If that isn't legible, a nicely formatted version is in CODE-2126.
>> Basically we get a huge improvement in performance from these changes.
>> You can try out the changes in trunk now.
>> So a big thank-you to Chris for this work!
> Yahoo! Groups Links