Re: [patch] "HT scheduler", sched-2.5.63-B3

[Martin J. Bligh]

>> Right, being able to control this interactivity knob programmatically
>> seems like a useful thing. That way, the window manager can boost the
>> interactivity of the foreground window for example. It does seem that
>> figuring out that something is interactive in the scheduler is tough,
>> there is just not enough information, whereas a higher layer may know
>> this for a fact. I guess this reduces my argument to just keeping the 
>> interactivity setting separate from priority.
> 
> No no no.  Martin's point shows exactly that nothing but the kernel can
> ever know whether a task is I/O or CPU bound.  What is bash?  Is it
> interactive (when you are typing into it) or CPU bound (when its running
> a script or doing other junk)?
> 
> Only the kernel knows exactly the sleep patterns of tasks, which is
> essentially whether or not a task is interactive.

Exactly ... all this tweaking, and setting up every app individually is bad.
It should "just frigging work" ;-) We seem to be pretty close to that
at the moment - 2.5 feels *so* much better than 2.4 already (2.4 degenerates
into a total slug overnight, presumably when things like man page reindexes
thrash the page cache).

The fact that the debian renice of the X server actually breaks things is
probably good news ... we're actually paying real attention to the nice
value ;-)

M.