Re: [Geotools-devel] Interesting findings about JAI tile cache

[Andrea A. <aa...@li...>]

On Sunday 25 May 2003 14:03, Martin Desruisseaux wrote:
> Andrea Aime a =E9crit:
> > Uhm... I've tried with 32 MB too, but it takes forever to load (once
> > loaded it works fine thougth). Maybe the pyramid settings should be
> > changed according to available memory?
>
> Maybe. It worth a try.
>
> The Linux version of J2SE is know to consume more memory than the
> Windows version (at least in 1.3 I think. Don't know if it is fixed
> now). Maybe is explain the difference.
>

Sorry, I've found the reason... it's not java, it's not my processor, it's =
my stupidity!
I forgot in the GTopo30Example code the following two lines:

PlanarImage pi =3D (PlanarImage) gc.geophysics(false).getRenderedImage();
pi.getAsBufferedImage();

no wonder it didn't worked with 32 MB. Now it works fine. If I really want =
something
to complain about, well, there still are some performance offenders:
* the legend bar in the bottom of the map really takes a lot of time to be =
computed...
* why at the first display the image is rendered three times? is it just a =
window manager
  problem or there is some redraw call in the code that gets triggered with=
 no good
  reason? When zoomin in or out there is no real problem (just on redisplay)
* the MapPane should really be initialized always with  mapPane.setPainting=
WhileAdjusting(false);
  it really needs a good processor o a light map to be of any use with the =
"true" setting.
* Is the pyramid composed of four or five levels?. I keep getting messages =
like:
  Select an image of "topo" decimated to level 0 of 4.
  ...
  Select an image of "topo" decimated to level 3 of 4.

  but I've never seen a level 4 of 4... an off by one error somewhere in th=
e logging code?

Uh, by the way, I've tried various settings of the MIN_SIZE value in the Re=
nderedGridCoverage
while keeping the heap to 32MB and it seems that a value of 1024 performs b=
etter both
than 512 and 2048, especially when the image is zoomed out to show the whol=
e map.
Can you reproduce that? I wonder if this is just the case of this image, th=
at is, if this
value should be set somehow according to the image size, the tile cache dim=
ension and
the available memory... experience with more complex maps will tell us, I g=
uess.

Best regards
Andrea