On Wednesday, June 09, 1999 9:15 PM, Robert M. Hyatt
[SMTP:[EMAIL PROTECTED]] wrote:
> I don't follow. Linux _only_ has kernel-mode threads. Each thread is a
> unique process with its own PID and context. Linux is just real slick
> on doing the fork() by using the copy-on-write page-sharing trick...
Yes, yes, I know how Linux threads work. It *is* real slick!
My point (obviously not sufficiently well articulated) was that reasonable
people can make a good argument for also providing preemptively scheduled
threads that are _not_ processes, the way NT does.
> > So it's really a deep architectural debate, with pressure on Linux to
get
> > special handling for processes that are really threads (e.g. keep all
> > "process threads" on the same processor, etc.)
[...]
> linux _never_ keeps all threads on the same cpu. Unless you only have
one
> cpu...
I know, that was the point. You obviously view this as a plus. If one is
doing heavy-duty processing or data-crunching, this is often a good thing.
However, it is not *always* a good thing. If I want to spin a little logo
in the corner of an app's window, a thread is the simplest way of doing so.
Should this thread be running on the 2nd CPU, dragging the app's
instructions and data into its cache, too? It's no longer so
black-n-white, it starts getting complicated.
(Some may even go as far as to say that there are times when the evil of
cooperatively scheduled user threads could be justified, but I stop short.
Netscape is a good example of how horrible user threads can be).
> I hope NT comes toward Linux... :) Otherwise Linux will start
> to suck. :)
NT has its good and bad points. I think it's important to remember that
sometimes there are good reasons why it has some of the features it does,
religious and political conflicts aside.
-- Eugene
-
Linux SMP list: FIRST see FAQ at http://www.irisa.fr/prive/mentre/smp-faq/
To Unsubscribe: send "unsubscribe linux-smp" to [EMAIL PROTECTED]
