Re: The Scheduler Marc Feeley 15 Apr 2000 02:26 UTC

> Marc: What about weakening the definition of priority to allow other
> comparable values such as real numbers or (more specifically) time
> objects? With this, I am sure a RT-SRFI could be written without
> having to undo parts of SRFI-18.

I could see using real numbers for priorities, which would allow
deadlines to be expressed (i.e. you set the priority to
(- (time->seconds deadline)) for threads with deadlines, otherwise you
use a positive value for threads without a deadline).

My only worry is that this prevents certain efficient implementations.
For example, you can't represent the set of runnable threads as a
table of lists of threads indexed by priority (a common way to
implement priorities in other thread systems).  Also, with integer
priorities, a system that does not implement priorities can define
(thread-min-priority) and (thread-max-priority) to return the same
integer.  However, if priorities are real, then I see no reason to
keep (thread-min-priority) and (thread-max-priority).

Anybody have a strong opinion about this?