UNIX for Advanced & Expert Users

>If unix simply lacks the functionality that windows has in this area--fine. Idle tasks suddenly running in the middle of your benchmark is usually a rare event. Its certainly more important that you do obvious simple things like make sure that no other non-idle tasks are running. But having an api for committing idle tasks like windows does certainly is icing on the cake, as it can eliminate the rare possibility.

Well the point is: This is the true difference between a true preemptive multitask/multiprocessing OS (like ****x) and almost and not quite yet (still has some cooperative behaviour) OS like Win*...
E.G. when Windows has decided to scan your drives, you cannot work till it gives you back the hand (cooperative...), at this point it never happens on a normally tuned unix... the time sharing is efficient under unix, not all that under windows... So you cant talk of lack of functionality that windows has... windows has added something to cure some insane behaviour...

Again my 2 cents...

While, like any good person, I share your dislike of Microsoft, I am pretty sure that you are wrong for WinNT and successors: I was under the impression that they have always been fully preemptive; see
Preemption (computing) - Wikipedia, the free encyclopedia

People sometimes complain about the scheduler in NT, just as they do in, say, Linux, but I have never seen anyone claim that the NT kernel partly operates in cooperative multitasking mode. Find a good reference that proves otherwise, and I will be grateful for the enlightenment.

What usually happens in windows if an idle task kicks in during benchmarking is that both run concurrently (on a multi cpu box) and/or apparently concurrently (i.e. time sliced, especially if there is just a single cpu) and the presence of that now active idle task distorts your results to some extent.

I imagine that unix is just as vulnerable, at least in principle, tho perhaps not in practice (e.g. maybe typical unix boxes just don't have idle tasks that can suddenly become active; then again, they usually seem to have all sorts of background daemons...).

Ok. I don't want to speak for the entire Unix world , but I guess we are more concerned with performance under regular load, rather than matching numbers within platforms...

One method that you could use to achieve something vaguely similar would be to run your unix test on single user mode; only a portion of services run when the box is running on single user mode. Any service that your particular application may need can be started, and you could kill whatever service may pop up during your benchmark (mostly killing the cron daemon would do).