Em 05/12/2009, às 14:06, "Axel Dörfler" <axeld@xxxxxxxxxxxxxxxx> escreveu:
You can't know that. You can't even know if a fair scheduler is a good idea at all for Haiku. Because you just cannot compare it to other Unixes which have a very different use/understanding of process priorities.
Precisely. Which is why I'm using a different fairness criteria than the usual.
Unix schedulers tends to prefer a fair distribution of CPU time (which, on Unix, where most threads are created with the same priority level, has the effect of making a fair scheduler biased towards threads that block early and often, i.e., interactive threads), while I proposed the criteria of fair chance of being selected to run according to priority levels (i.e., after a sufficient number of scheduling rounds that establish a statistically relevant sampling interval, a thread with priority 2 will have been selected twice as many times as one with priority 1).
Cheers, A.
