This post received a lot more attention than I thought it would; if I had known, I might have explained a few things more carefully. If you didn't immediately dismiss it as stupid and found it at least a little thought-provoking, I highly recommend checking out the technical report http://research.microsoft.com/pubs/173709/submitted-nsdi13.p... (PDF), especially Figure 2 (which describes the entirety of the syscall interface), Section 3 (which describes it) and Section 4 (which talks about how to reimplement traditional browser user interface paradigms in this new world.) It's all quite interesting, and at the very least, might say something about how we should take mobile native apps forward.
It's very interesting, thank you for posting. One question I have is inspired by some of the discussion about WebGL. You know as well as anyone that one of the most obnoxious and pernicious side-effects is non-termination. This is bad enough in the CPU, but when DoS-like attacks are perpetrated on the GPU shaders or other I/O ports, we can create unfortunate lockups and non-responsiveness. Have you heard of any academic work on this issue?
As I understand, there is work on the hardware end going on to allow things like running untrusted GL on only part of the graphics card's processing units, as well as possible preemption of code on GPUs, precisely to deal with these problems...
