Re: [gentoo-embedded] Hard Real Time Systems with Gentoo?

[wireless]

momentics wrote:

> \\
> So, we're considering some system that (we think so here) can
> guarantee (by some way that is not important at this step) meeting its
> deadlines with a given (by the operator) probability. It uses some
> techniques that adapts to the given (read - required) probability and
> actually proves that it achieves the requirement with the probability.
> Does it mean a system is a HRT?    why/thoughts (?)

Um I alluded to this in the latest response to Bob, but I try to
state it as clearly as possible. You cannot prove determism with
statistics. Determinism can only be proven be defining, discovering,
characterizing and testing each and ever possible state AND
transition. You want to use another term, but just do not use
Deterministic.

So a statistical method may flesh out that a system is fully
deterministic, but it not a mathematical proof. If you do not
prove it, it's not fully deterministic, when in fact it just
might actually be fully deterministic. Fully deterministic
means you have determined or proved each and ever state and transition.

> We imply here that true HRT is impossible in reality and any single
> system that exists in the word meets hard real-time requirements with
> some probability. If it does not – HRT speaks of system failure.

Ah, how you are 'getting it'. YES, YES, YES, I absolutely agree
with this statement!

> Please note, we speak of free programmable generic system here,
> something like a programmable logic microcontroller for custom
> application development. Any systems like DSP/matrix/etc are out of
> scope.

As you probable guess, I'm old school on these matters, and I am
correct, just not very palpable. We build losts of systems that
are probably fully deterministic, but not one client wants to pay
us to verify the systems. They prefer to sell them and us when the
firmware need updating, new features and such. Economics is the
trump science we all live under.

We even develop DSP code here, using state machines, assembler
and ansi C. Rarely do we use things like 'TI bios' because
when we send a product out, we do not like bugs. it not a dump
on TI, it's more how you control your firmware products. Each
addtional layer exponentially increase the number of states and
transition, not to mention the usually unavailability of sourcecode.
Granted these are usually smaller systems (compared to some system)
that are very focused.  Never have we used OO based tools, language or
such, because they are not needed.  If somebody has a large system,
the first thing we do is move 99% of it to another processor
where timing and reliability are less stringent.  The current trend
to bloat embedded things and call it RT or HA or such,
just makes my skin crawl.......

That said, I love gentoo linux and keep a keen eye on embedded gentoo.
One day, these kids will put this engineer out to pasture,
but, that day has not arrive and it ain't even on the horizon.

ymmv,

James


-- 
gentoo-embedded@g.o mailing list