I was just reading this last night and thought you might find it interesting: https://lkml.org/lkml/2010/12/5/21
Con Kolivas, author of a CPU scheduler for the Linux kernel, had a very similar idea.
Con Kolivas, author of a CPU scheduler for the Linux kernel, had a very similar idea.
I believe priorities are a sucky way to express the developers desire to perform some tasks in a low-latency manner, while others are ok to delay for a larger time. Maybe latency would be a better thread parameter than priority. It has the advantage of being a hard metric, one you can easily measure and adjust your algorithm against.
But latency is slippery. Do you mean time-to-first-execution, or time-to-completion? Completion of what? The OS would have to be aware of some units of processing (message digestion time?). This is not a bad idea. I'm all in favor of the OS having a more complete model of what the developer is trying to accomplish.
Anyway my 2C worth.