🚀 go-pugleaf

In article <434ei3$ncu@oddball.sjc.MENTORG.COM>,
 ewill@hpewill.sje.mentorg.com (Eric Williams @ PCB x5577) writes:
> In article <42qclc$3bu@sashimi.wwa.com>, blackbob@sashimi.wwa.com (G) writes:

[...]

> |> Your analogy is worthless.  A supercomputer costs several million
> |> dollars.  Preemptive multitasking is no more expensive to  imple-
> |> ment  than  cooperative.  It might be cheaper, especially from an
> |> application developer's perspective.
> 
> PMT is (slightly) more expensive, because those systems that
> use it must also implement and require a method of
> synchronization between processes; CMT has built-in
> synchronization.  The Amiga Exec, for example, which had PMT
> from the beginning, had to implement signals and messages
> (and in v1.2, semaphores) in order to keep its processes
> from tripping over each other.  The programmer is required to
> learn this complexity (though actually it's pretty simple;
> Amiga Exec is very clean).  Unix also has signals
> and waiting in the kernel (as well as at the user
> level).

About the only significant thing an application programmer must
do, unless they are using shared resources, is make sure that
they don't do busy waits when waiting for events to occur. (Disk
access only really counts as a shared resource when processes
share a single file -- other accesses are looked after by the
OS.)

As another feather in the Amiga's cap, the vast majority of its
kernel is re-entrant, allowing processes to be pre-empted during
system calls, unlike Unix.

-- 
 -- Duncan Grisby                                    --
  -- dpg1001@cam.ac.uk                                --
   -- finger dpg1001@cus.cam.ac.uk for PGP public key. --