4 ms·
Concurrencys Shysters
- losethos 15y agoSMP is good if you have multiple applications running. If you have just one application, give it complete control of scheduling, master/slave. This is obviously optimal. A video game can divide-up the screen update between cores. The scheduler has no idea what size jobs are and jobs are created based on loading--if you have more CPU, you make it dynamically render the screen better. I park one task on each core and my update routine syncronizes and divides-out what part of the screen and how much quality each core is responsible for. (It's funny how I'm talking like an arrogant shit-for-brains Indian.) You're never gonna get compilers to make things parallel, except matrix math.
- nwmcsween 15y agoMost concurrency is now achieved by a somewhat M:N threading model where n kernel threads map to m 'slots' (actors.. etc). Why does this method of M:N threading work yet kernel to userspace M:N threading doesn't?