Re: Safety/Reliability

"Terry L. Ridder" wrote:

> <snip>
> jon>
> jon> I really don't understand the desire to break with what has been proven
> jon> to work to forge into VASTLY uncharted territory. And, trying to
> jon> bring up the RT patches on non-Intel architectures just to do a
> jon> machine tool conversion is going SO far afield as to be ludicrous!
> jon>
>
> it is not clear to me whether this is being addressed to me

Yes, it was.

>
> or the rtlinux.org/fsmlabs.com folks.
>
> assuming for a moment that you are addressing me.
> the rtlinux.org/fsmlabs.com folks have all ready
> stated they will not do the work of getting rtlinux
> to run on other alpha platforms. therefore, if someone
> wants to use rtlinux on the alpha platform and they
> do not have a ds10/tsunami they need to make it work
> either with the help of others or one their own.
>
> getting the rtlinux-3.0-pre8 release to run on other
> alpha hardware not only benefits me directly but others
> who would like to have the ability to run rtlinux on
> their alphas as well. therefore, it is not just to do
> a machine conversion. therefore, it is not ludicrous.
>
> this is about giving bakc and not just taking.

Well, OK, I see the point of giving back. But, it has been shown that
the Pentium machines CAN run RT-Linux reliably, and run machine
tools with them. If the point of the task is to spend MANY months
working on an essentially academic project, here it is! If the
task is to convert machine tools to more modern CNC controls,
the porting RT to Alpha architecture is a VERY big detour!
That's the point I was trying to make, originally.

>
> would i trust rtlinux to control a machine?
> probably not.

Well, now I'm really lost! You are going to spend months of work,
and then decide it can't be trusted, so it will never be used.

<Horror story snipped to spare the audience.>

> with rtlinux there is much more than a slight chance.

I really don't think a programmer's commenting style is a direct
indication of his competence, or the reliability of his software.
Linux as a whole is more reliable than most other systems out
there (about the only thing I've ever used that compares is VMS,
on either VAX or Alpha hardware.) It also was written by very
small groups with minimal documentation (and NO committees,
endless meetings, performance review boards, etc.) And, yes,
sometimes things are hard to figure out because the documentation
IS terse, to say the least.

Using known stable versions of EMC on ordinary commercial
Pentium motherboards, I have run EMC for hundreds of hours
without any real problems showing up. There are a few known
quirks in the software, but I haven't had any wild motions.
One of the most disturbing is when switching between manual
jogs and MDI commands, and having interrupted a jog with the
esc key, you can enter an MDI move in a single axis, and have
another axis also move, due to the 'remembered' incomplete
move. This has been fixed and brought back a few times.
I should also state I'm still running a 20-DEC-1999 version of
EMC because of other bugs that provided greater problems
(like system crashes).

I do some fairly substantial testing of these systems before I
start using them in production to find out about such quirks.

If you need absolute assurance that such a system will NEVER
make an unintended move, or hurt someone, I suggest you
get out of the business, and avoid even hand tools, as they
can pinch fingers, hand-powered drills can still break off with
flying fragments, etc. When computers are added to the
system, it gets to be impossible to be absolutely sure of
ANYTHING!

I'm not belittling your injury in any way, and I would strongly
recommend anyone who wants to hook EMC up to a large
machine (perhaps anything larger than a Bridgeport Series I)
should perform VERY intensive testing of all hardware and
software in the system. Software could be tested on a desktop
machine for any combinations of operations that could
cause unintended movements.

Jon