Text 167, 52 rader
Skriven 2005-05-09 00:44:42 av Tony Ingenoso (1:379/45)
    Kommentar till text 165 av Geo (1:379/45)
Ärende: Re: This is not an Anti-OSS Flame
=========================================
From: "Tony Ingenoso" <admin@spamcop.net>

This is the nature of all (non-MT aware) DOS apps.  When there is nothing else
to do they scan for keystrokes.  The problem is not that the CPU meter goes to
100%.  The problem would be if it impacts the rest of system performance when
something really is going on elsewhere -- i.e. if its keyboard polling loop is
doing OTHER things that trick the OS's idle detection logic into thinking the
app is really doing something.  When this happens, the app will be a genuine
CPU burner impacting other stuff in the system.

If it just pegs the meter with nothing else running on the system, then its
actually a normal condition - the scheduler would choose to run the DOS app
rather than its own idle loop.  The DOS app will burn the CPU polling, and the
meter(running at an idle time priority) will naturally be starved and think the
machine is at 100% - which it is, but the CPU is really available for something
else even though it appears to be 100%. If the idle detection logic works OK,
the first time something else shows up in the system, the DOS app would be
knocked back.

There are a few DOS apps out there that are hard to do proper idle detection on
because of the mix of other API calls and activity they mixed in with their KBD
polling.  Not a lot of them, but there are a few.  One of the Word Perfect
releases was a PITA this way as I recall.  DOS app idle detection is really a
bunch of heuristic black magic<g>

"Geo" <georger@nls.net> wrote in message news:427ec4ef$1@w3.nls.net...
> "Tony Ingenoso" <admin@spamcop.net> wrote in message
> news:427eb908$1@w3.nls.net...
>
> > Don't laugh, I've seen an editor (one of Borland's first OS/2 apps)
under
> > the OS/2 kernel debugger that allocated and freed a 1 byte segment on
> every
> > keystroke.  Every keystroke was two expensive trips through kernel
memory
> > management.  Programmers do some pretty dumb things...
>
> this I can believe, I remember some borland dos app that used to run NT to
> 100% cpu usage because of it's key poling routine and as a programmer I've
> done some pretty dumb things so I have no problem believing that either.
>
> But there are other issues. Most of the apps that I use don't have a
problem
> with cpu speed being the bottleneck, the times I spend waiting for the app
> are usually when it's talking to the disk or network.
>
> Geo.
>
>

--- BBBS/NT v4.01 Flag-5
 * Origin: Barktopia BBS Site http://HarborWebs.com:8081 (1:379/45)