On Saturday, 28 November 1998,
     JEFF DUNLOP wrote to IVAN TODOROSKI about PM drawing priorities...

 IT>>   So you now have a higher priority thread which is not releasing any
 IT>>   CPU cycles! And if you've read Denis' explanation, you now what this
 IT>>   means. It means that any lower priority threads will NOT get any CPU
 IT>>   cycles (except by means of the starvation boost) till the dragging
 IT>>   stops.

 JD> ...on a single machine that represents the degenerate worst case.
 JD> Sorry, but in my scores of OS/2 installs, I've never run into
 JD> anything quite that pathetic, and if I did I wouldn't hesitate to
 JD> replace the unsupported video card or at least turn off full
 JD> window dragging. In a previous message I asserted that any program

         Yes... it is kinda pathetic, isn't it? :(

 JD> that runs at an elevated priority _must_ ensure regular blocking,
 JD> and that I viewed this as the most glaring flaw of OS/2's priority

    Yes, this is was exactly my point. It has three glaring flaws:

1)  That somewhat too trusting priority scheme.

2)  Those somewhat illogical and unfair boosts an application gets when
    in foreground or having its window dragged.

3)  The fabled SIQ.

 JD> scheme. It turns out that you've simply confirmed this with a
 JD> program that fails in precisely the way I described.

  Except that it wasn't the program's or the video driver's fault at
  all!

  The program runs at normal priority and the only thing that is wrong
  with the driver is its slowness.

  OS/2 itself is to blame for giving that driver higher priority than it
  deserves. It boosts PM drawing beyond Foreground Server, delta 31
  priority, higher than the program actually requesting it... :(

  This boost went away when I patched the kernel priority table. So the
  scheduler works ok, but is somewhat poorly misguided by some of the
  exagerated boosts in the table. I eliminated all of them, and all of
  the problems went away (and I haven't changed the driver or the card).

  I'll try to bring back all of the boosts except the one(s) responsible
  for that window dragging problem.

                                                            - Ivan -
.!. Die Hard: With Windows 3.1  
--- Terminate 5.00/Pro [OS/2]
 þ TerMail/QWK þ