(Warning: borderline OS/2-related material that adds little to the
information already presented on real-time capabilities of OS/2.)

In a message on 09-19-94, Carl Neal said to Peter Hansen:

CN> Any important control system with critical timing needs a dedicated
CN> processor to avoid disruption from other processes.  This would be
CN> true if the operating system was custom, OS/2, Unix, or DOS. 

While I don't want to stray too far from relevance to OS/2, I have to
point out that this statement is false provided your requirements are
not beyond the capabilities of the system.  If "critical timing" means
response no worse than 400 microseconds, for example, then an OS/2
device driver will certainly suffice.  With my custom real-time kernel
(using DOS as a disk device driver) I can achieve far higher response
and performance with no possibility of disruption such.

My timing is more "critical" than the 400 microseconds OS/2 guarantees,
but it's a fuzzy constraint and I don't know just how far OS/2 will get
me on decent hardware.  Anyway, I know what I want to do, it's just a
question of when I can do it and how complicated it might be to get
there.  Indications are that the future of OS/2 suits me well but that
to achieve acceptable performance _today_ may well be impossible.

CN> If a system is so critical that microsecond timing is important, the
CN> two PC approach should be able to be cost justified. 

This does not logically follow, I'm afraid.  I can think of many
potential applications which would require "microsecond timing" (once we
agree on the meaning of the term :) but which would not support the cost
of a two-PC approach.  Mine is bordering on being one of them.

I've already considered your "ideal" and "cost
effective" solutions from
many angles and have determined that they are not, in my particular
case, "cost effective" and therefore neither are they
"ideal."  I'm
trying to build a system that must have a nice user interface and
provide good performance at low cost.  The first requirement combined
with the last suggests a PC (commodity that it is), provided the middle
requirement can be met.  With the present custom kernel approach we have
achieved a satisfactory solution but at the expense of a really
attractive and versatile interface (DOS being what it is).  If OS/2 can
(now or in a Mach-based version) meet the performance aspect, we have
suddenly met all three requirements: i.e. the "ideal" solution.

  [regarding suggestions for dedicated controller, multi-PC, and
   networked solutions:]
These are very reasonable goals for the long run and OS/2 is exactly the
right tool around which to build such a system.  Unfortunately, I have
neither the budget nor the sales volume of National Instruments, so for
the time being I must concentrate my efforts on more humble goals.  Any
of these solutions adds at least $2k to the overall cost and it's mostly
money in somebody else's pocket. :(

In this case we can justify (barely) some fast PCs with basic analog I/O
boards and my custom software.  That software is where the money is made
so if I generously specify a whole slew of expensive PCs and equipment
from other companies, I don't do myself a lot of good.  

(Thanks for the comments, Carl.)

Peter Hansen  ***  Engenuity Corporation  ***  Guelph, Ontario, Canada
Internet: peter.hansen{at}canrem.com    RelayNet:->CRS    FIDO:(1:229/15)
___
 * MR/2 * ***  Learn Esperanto, the international language.  ***

--- QScan v1.12b / 01-0348