Re: Applications "buying" resources



Hi Hans-Bernhard,

On 12/30/2010 6:47 AM, Hans-Bernhard Bröker wrote:
On 30.12.2010 09:53, D Yuniskis wrote:

First of all, note that a task can ignore ANY or ALL
of these "tactics"/requests (except, of course, the
last one :> )

That's exactly the problem. If they can, you have to assume they will.
Which renders the entire scheme moot.

By the same token, "voluntary multitasking" (each task
holding the CPU for as long as it wants before relinquishing
control to the next eligible task) "won't work". :>

Of course, it *does* work because the developer(s) "behave
responsibly".

This is fine in a closed-end, fixed functionality product. But,
if you open the product up to other applications, you run the
risk of those other applications misbehaving -- either intentionally
(to "look better" or to make their development effort easier)
or unintentionally (by not understanding the costs of each operation
they perform).

If you can develop a policy that lets tasks (applications)
express their relative resource needs "within a budget",
then you can have an external agency (the kernel) *enforce*
that sharing/cooperation.
.