Re: scheduler interactivity: timeslice calculation seem wrong

[Eric St-Laurent <ericstl34@sympatico.ca>]

> So, if you give low priority processes bigger timeslices than high prio
> ones, the low priority processes will be allowed to consume more CPU than
> high. Obviously not the intended result. I agree with your intended result,

Thanks for the crystal clear explanation.

that 'switched-arrays timeslice distribution' is good for fairness but
maybe it add unwanted scheduling latency to a high priority task that
sit with it's timeslice expired...

i know it's more a real-time os thing, but i always liked the concept of
pure priority scheduling with priority boost (calculated from aging) to
prevent starvation. in a multi-level feedback queue scheduler, a
processor share percentile could be assigned to each priority level.
anyway i'm sure there is some proven fair-share scheduling algos out
there that's better than this old stuff.

> I don't think you need that much grainularity. Might be a benefit though.

personally, i not a fan of the jiffies/tick concept; conversions, lost
ticks problems, drifts, sub-tick high-res-posix-timers etc. everything 
should use the highest resolution timer/counter in the system (TSC, ACPI
PM counter, ...) directly. it's a major cleanup and many old PCs don't
have the newer timers.

> >- lastly, it may be usefull to better encapsulate the scheduler to ease
> >adding alternative scheduler, much like the i/o schedulers work...

Well, i was looking at TimeSys scheduler, trying something like that in
2.6 requires modifications to many files and it's a PITA to maintain a
diff with frequents kernel releases. having a structure in place to
plug-in other schedulers sure helps.

Eric St-Laurent