More memory tricks

[Patrick Nichols <pnichols@graphics.csail.mit.edu>]

Hi Vitaly,

I redesigned some data structures and managed to load up to 68,000 
triangles into main memory; turning on some exotic VM options (maxing out 
the heap size of the machine, turning on incremental garbage collection) 
seems to get past 70,000 but perf is horrendous. 

It does turn out from a memory perspective that my implementation would 
greatly benefit from collecting triangles into spaces. The reason is that 
each space maintains lists of its portals and neighboring spaces, which I 
suppose is pretty big. Sure, it also maintains a copy of its own 
triangulation, but this is a pretty small double[]. So, collapsing 
triangles is looking prety good right now.

What are you thoughts on feasibility and such?

-Patrick

-- 
* * *

Patrick James Nichols II
Graduate Student, MIT Computer Graphics Group
http://graphics.csail.mit.edu/~pnichols