Re: Controllerless distributed automation

["Paul Gordon" <paul_gordon@xxxxxxx>]

>
>That sounds pretty ambitious! I suppose that technically it shouldn't
be
>too
>difficult to have individual processors/programs for each device -
though I
>would guess that would involve a lot of design work and it'll make each
>unit
>fairly pricey (though relative to UK X10 modules.....) I would guess
your
>main problem would be communication - what kind of information does
each
>device need to have about others, and what doesn't it need. When a
device
>makes an announcement - "hello, I've just been switched on"
etc etc - how
>does it know that this device is related to it? You'd practically have
to
>have custom programs for each device. It would be impressive if you can
get
>it to work, though.

This is exactly the problem with all such distributed systems: if
individual
components of it are to co-ordinate their behavior in order to operate as a
coherent "system" (rather than a collection of individual control
systems),
then what is required is actually quite a bit more than mere communication
between those components. - Each and every component needs to be aware not
only of the presence of other components, but also of the capabilities, and
the state of them. This would mean that each component would require local
storage not just of it's own operating code, but also of the above
information. (otherwise, the actions of a device could not be determined in
relation to other devices without first polling them) - potentially this
means a single device might have to poll EVERY other connected device
before
deciding what action to take, and then having taken some action would have
to broadcast that back out to every other connected device, so that they
can
update their state tables. - this is all very messy, massively redundant
(excessively so, as every device has a local replica of the entire
"database"), and a communications nightmare. (as more devices are
added the
overhead increases)

THAT is why you haven't seen such systems around, - it just plain makes
sense to have a central controller as it negates all the above problems, it
keeps the intelligence in one place, keeps the management and programming
in
one place, and thereby reduces the overall cost (the "dumber"
modules are
obviously cheaper), eases setup, programming, management of the system and
future expansion.

>
>Of course, if you were *really* ambitious you could probably find some
way
>of linking all the devices to operate in parallel as one large computer
>with
>massive redundancy - one device is elected co-ordinator and this role
can
>be
>taken over by any other if it drops out. Hmmm.... Anyone got 5 years
spare
>to try that one? ;-).
>
>

THIS is exactly what I proposed a month or two ago in the thread I started
called "flight of fantasy" - I even went as far as to describe
(in loose
terms) how such elections could be coordinated and controlled, with
heartbeats, assigned priorities and so on.

This in my opinion, is actually the better "ideal" system - any
objections
(which are valid ones, don't get me wrong!) to having a "single point
of
failure" are negated by providing redundancy of the central controller
role.
- I wouldn't go as far as to suggest that >every< module needs to be
a
controller candidate though, as that is therefore almost the fully
distributed intelligence model that was originally postulated, and is not
ideal for the reasons I have outlined above. - How much redundabcy do you
realistically think you will need in any system? - 100%? - 200%? (Hint: the
space shuttle only has 3 main computers...)

How about a quick poll:

Q: How reliable is your HA controller? - how often (on average) does it
fail? What is the impact of such a failure?

In may case, my current HA system: Homeseer running on Windows ME - (Note
it
was on Win2K, but "experimenting" with Homeseer phone has forced
a TEMPORARY
downgrade to ME), has not failed since I built it (about 6 months ago on
the
current platform), and has only been down for Homeseer software
upgrades....

Just my £0.02

Paul G.


_________________________________________________________________________
Get Your Private, Free E-mail from MSN Hotmail at http://www.hotmail.com.