Re: [CAD_CAM_EDM_DRO] black box specs

<unlurk>

Jeff, it's not as bad as you think.

Jon, with the exception of ATM (and Allen-Bradley's ControlNet), few
networks are deterministic. All are described in probabilistic.
I've worked with _many_ fieldbus type I/O system. Ethernet is
becoming very popular because of it price, ease of use, distances,
reliability, etc.. Probability is based on loading, network speed, etc..
Do a quick lookup of industrial Ethernet, you may be surprised at what
is being done with it in real-time control environments.

Jeff, you are correct that you would not have to attach a 'name brand'
protocol stack of any type. Likewise, most management traffic comes
from 'higher' layers such as Netware & Microsoft network clients which
you would not bind to those cards. Even if you did use a full blown TCP/IP
stack, you'll see ARP traffic but if it is an isolated subnet that's all.

Multiple devices are attached to a single subnet via a switch. With 100Mb
PCI Ethernet cards costing $20 and 10/100Mb 5 port Ethernet switches
costing $80, this is very economical.

The only real problem is the complexity of creating you own protocol
stack so that you can use the NIC manufacturers low-level drivers.
Still eminently doable.

Perhaps a better suggestion is CAN. Inexpensive CAN interfaces are
available. While not deterministic, collisions do not result in a loss of
data.
CAN is the basis for DeviceNET which is a _widely_ used standard for
real-time I/O. No real need for the DeviceNET overhead (it's not that much
tho)
and it is _very_ reliable and has a top speed of 1Mb. Microchip now sells
CAN interfaces standalone (for a few bucks) and finally integrated with a
PIC.
Making a serial to CAN bridge would be walk in the park. They are available
commercially for a couple hundred bucks from people like AB. There are also
developer kits even cheaper. The CAN spec is free from BOSCH. As far
as the 'black box' side, more powerful processors with CAN integrated are
available (Siemens C167, C505).

<lurk>

> Jon,
>
> Congratulations on being the first flaming arrow archer.
>
> I've been designing PC interfaces for years and I'm not convinced any of
> them are all that deterministic.
>
> I am thinking in terms of only the control PC and the motion box on the
> "network". Also there would be no real reason to run a full TCP/IP stack
> on this so a lot of the other noise could go away.
>
> Yes, this is a very half baked idea.
>
> Jeff
>
> >
> >Ugh! Ethernet is specifically designed to almost NEVER be deterministic.
> >For the most part, it is fast enough that if you coordinated the message

flow,

> >and only had the control PC and the motion box there, it probably would
> >work. Even still, there would be network management software that
> >would be sending out multicasts saying "I'm so-and-so, and I'm here!"
> >every second or so, and that would cause an occasional collision.
> >
> >Jon
> >

__________________________________________________
Do You Yahoo!?
Talk to your friends online with Yahoo! Messenger.
http://im.yahoo.com