Is there a perfect scheduler that is non-optimal in the Big(O) sense but is optimal if you’re looking at maximizing hardware utilization? In other words, scheduler that takes a long time to determine CPU utilization for each process, but provides an optimal total CPU utilization? I realize that it would not be ideal since we’d essentially have these “sudden stops” as it recalculates the schedule. I’m just more interested in the theory.
Is there a perfect scheduler that is non-optimal in the Big(O) sense but is optimal if you’re looking at maximizing hardware utilization? In other words, scheduler that takes a long time to determine CPU utilization for each process, but provides an optimal total CPU utilization? I realize that it would not be ideal since we’d essentially have these “sudden stops” as it recalculates the schedule. I’m just more interested in the theory.
How would you deal with iowaits in a system like that? I can perfectly burn 100% of CPU time running a poll(), but that’s not useful work…