John Tinker wrote:
> I would think that a properly layered system would look after its own
> integrity. To say that an app took down a system is by definition an
> indictment of the system, in my, perhaps idealistic, opinion.
Generally, but not always. I certainly agree with the assertion when applied
to _most_ things that people think of when using the term "operating system",
but there are certainly exceptions for some classes of real-time or
embedded systems, where the "applications" are really tasks and protection
hardware may be unemployed or simply lacking.
[snip]
> I've always wanted to hear a os programmer tell me whether allowing multiple
> preemption is a fundamental flaw in certain operating systems.
Again, in _most_ things that fit into the OS bucket the preemption model
is both defined by and enforced by the operating system itself. At best a
process or task within a process can elect to give up the CPU by forcing
a reschedule. Again, it's real-time and some embedded systems where
one generally finds the notion of a task either seizing the processor or
a task relinquishing the processor to a specified task. A common place where
this occurs is in telephone systems, and in such cases the distinction
between "application" and "system" is blurry at best.
> I'm thinking
> that multiple preemption might be a contradiction in terms. Intuitively I
> would expect the operating system itself to have the ultimate preemption,
> expressed as its servicing of hardware interrupts.
Since _most_ operating systems don't allow code at the application level
to manpiulate honest-to-god interrupts (but may do so in a virtual machine
context) what you describe is generally the case.
> I would expect that it
> would only be able to delegate its authority to be preemptive to *any*
> application if *every* other application running agrees to the proposition.
The notion of the OS delegating scheduling authority is, in most contexts,
bizarre. It usually suggests that the OS isn't an OS but rather an overgrown
disk monitor. Oh wait, this thread _was_ talking about Windows, wasn't it...
--
Chris Kennedy
chris_at_mainecoon.com
http://www.mainecoon.com
PGP fingerprint: 4E99 10B6 7253 B048 6685 6CBC 55E1 20A3 108D AB97
Received on Sat Jan 13 2001 - 23:11:02 GMT