eekee wrote:
Would it help to introduce a bit of randomness into the scheduler timing, or does it just result in a "less accurate clock"?
Help with what? Also, yes, a more random clock is just a little less accurate. And modern CPUs are so fast that a thread incrementing a global variable can get many, many cycles in before being preempted. Again, it's not perfect, but it is good enough. Overall, the timer thread is always runnable and will therefore consume most of one CPU core. If not much else is going on, there may not even be a need to ever preempt this thread.
eekee wrote:
I'm also almost sure there was a Linux kernel option to randomize the TSC, last time I configured it, (2.6.??) but now I'm not sure that makes any sense, let alone whether it would help.
Well, it no longer exists as far as I could find. To my knowledge there is no way to change the TSC. Even if there was, there would be no point, since most of the code using the TSC is only interested in its differences over time. And if the Linux team somehow found a way to make the TSC completely useless, there are other methods of timing things.