figure out the best way to ring buffer

[haysmike]

shouldn't be shifting 15 4K buffers during every draw callback

further performance tweaks:
mach memory mapping stuff?
use lower-level semaphores