From: Andrew Lentvorski <[EMAIL PROTECTED]>
Gabriel Sechan wrote:
I fail to see how moving to a microkernel would have any effect on heavy
disk usage effecting overall system performance. Maybe I'm overlooking
something, but I see no inherent causality here. Care to explain your
logic?
Microkernels can do hard real-time. Thus even *drivers* get throttled
correctly so as not to chew up the system.
Well, yes. So can monolithics. A microkernel isn't a requirement to get
hard real-time. The only requirement for hard realtime is an OS that can do
the proper accounting and a lot of CPU/memory power (since hard real time
generally has poor overall performance in order to garuntee certain apps
having a minimum performance).
The problem is not "overall" system performance--that's throughput. The
problem is "interactive" system performance--that's latency. Something
which is rattling my disk should get throttled down when I am watching a
movie. That movie should get what it requests up until it would hang the
system. At which point, the system should pause it and throw an
error/warning/etc. Any time I interact with the system I expect the
computer to handle that stuff *NOW, DAMMIT*.
Ah, I get what you want now. This has nothing to do with realtime, it has
everything to do with prioritizing certain apps. The 2.6 kernel already did
a lot of tweaking on this. Basicly Linux has a huge number of internal
priorities in addition to the nice level, and tries to prioritize apps based
on CPU and IO usage. It seems to do pretty well. But its not, and never
can be, psychic. If you want to make sure your movie has priority, nice
down the other app. Otherwise, hope that the kernel correctly figures out
that your app is interactive (interactive apps get the most priority).
I'd give more details on how it works exactly, but I don't remember and the
book is packed away.
Gabe
--
[email protected]
http://www.kernel-panic.org/cgi-bin/mailman/listinfo/kplug-list
