Reply by Kenneth Porter●February 5, 20042004-02-05
Richard <rh86@azglobal.com> wrote in news:401C08D7.32DC7227@azglobal.com:
> [numerous design questions]
All excellent questions. I'll take those back to my team and try to
establish more precisely what we need.
> Ethernet is a decent fit for a lot of your requirements, and it works in
> fairly hostile environments over good distances. Media converters are
> readily available / cheap for fiber use in high EMI environments.
It does look attractive as a dedicated private bus.
Reply by Richard●January 31, 20042004-01-31
Kenneth Porter wrote:
> Ethernet and USB2 have the needed bandwidth but the
> latency is too long
How much bandwidth do you need per node?
How much latency is too long?
Do you require a bus or star cabling configuration?
What's the rate and size of each data block?
Polled or asynchronous?
> and Ethernet would be non-deterministic
How much jitter can you tolerate?
Ethernet is a decent fit for a lot of your requirements, and it works in
fairly hostile environments over good distances. Media converters are
readily available / cheap for fiber use in high EMI environments.
Ethernet switches are the cure to a lot of Ethernet's negatives - they
let you run in full-duplex mode to eliminate contention. If a single
100Mbps at the master is enough, it connects to the switch like any node
and the switch buffers the egress port. If that won't do, then make the
master the center of the cabling star; use several 1:5 ethernet switch
chips or a separate MAC/PHY for each port.
There are a couple 100Mbps chipsets with ISA bus; most are PCI, and
you'll probably find that all Gig-E are PCI.
Reply by Paul Keinanen●January 29, 20042004-01-29
On Thu, 29 Jan 2004 20:30:03 GMT, Kenneth Porter
<ken.blacklist@sewingwitch.com> wrote:
>Multi-drop, single-master, all nodes in one room. I'm trying to build a
>multi-input multi-output system where several output nodes depend on
>several input nodes and system state, with a high update rate. The
>environment might be electrically hostile. (Stupid customers, bad building
>wiring, electrostatic issues, etc.)
If you have only a single master, you do not have to worry about
determinism in an Ethernet network. With about 50 bytes minimum frame
size (5 us @100 Mbit/s) you should be able to transfer about 200 000
frames/s or 10 frames at 20 kHz.
If that is not enough, an you have a large number of nodes, look at
systems similar to Interbus-S and EtherCAT, in which data can be
collected from a large number of nodes into a single message frame.
Paul
Reply by Kenneth Porter●January 29, 20042004-01-29
thierry.moreau@connotech.com (Thierry Moreau) wrote in
news:3e2ff5a.0401280716.51ac7f5c@posting.google.com:
> These unique requirements appear to call for a proprietary protocol
> (including isolation solution based on the parts that are readily
> available for high speed transitions) based on a serial link
> connection (clock, data, and framing signal) found in processors with
> built-in serial-to-DMA integrated peripherals.
Yep. I'm looking at Analog Devices' SPORT in TDM mode at 30 Mbps for a few
nodes, but will need something else for more.
> Is your remote device able to provide the clock for the serial
> transmission and 20 KHz sampling? Is this a multi-drop configutation,
> in which case you would require timing information broadcast by the
> master station?
Multi-drop, single-master, all nodes in one room. I'm trying to build a
multi-input multi-output system where several output nodes depend on
several input nodes and system state, with a high update rate. The
environment might be electrically hostile. (Stupid customers, bad building
wiring, electrostatic issues, etc.)
> If you are interested, I can provide the details based on the Motorola
> MPC8xx communications capabilities.
Thanks, I've got an 8xx manual around here somewhere. I'm hoping to find
something less tied to a specific CPU but that's not imperative.
> Alternatively, or as a special-case, the Ethernet solution using the
> *full-duplex* twisted pair connection (no multi-drop) should be
> feasible, and the Motorola MPC8xx communications capabilities should
> meet the latency requirements.
Are you suggesting gigabit Ethernet, wired point-to-point to eliminate non-
determinism? That might work.
Reply by Thierry Moreau●January 28, 20042004-01-28
Kenneth Porter <ken.blacklist@sewingwitch.com> wrote in message news:<Xns947DA2D645AB3shivawellcom@64.164.98.51>...
> I'm shopping for a serial bus that will give me lots of bandwidth (10's of
> Mbps) and low latency (20 kHz update cycle).
>
> I'd also like electrical isolation. An optical fiber transport would be
> ideal but some kind of high-speed opto-coupled or magneto-coupled system
> would also be workable.
>
> What's available? I've heard FireWire mentioned and see that it can do 400-
> 1200 Mbps, but haven't found a latency spec and suspect its frame time is
> too long for my application. Ethernet and USB2 have the needed bandwidth
> but the latency is too long and Ethernet would be non-deterministic. Are
> there other alternatives I should look at?
These unique requirements appear to call for a proprietary protocol
(including isolation solution based on the parts that are readily
available for high speed transitions) based on a serial link
connection (clock, data, and framing signal) found in processors with
built-in serial-to-DMA integrated peripherals.
Is your remote device able to provide the clock for the serial
transmission and 20 KHz sampling? Is this a multi-drop configutation,
in which case you would require timing information broadcast by the
master station?
If you are interested, I can provide the details based on the Motorola
MPC8xx communications capabilities.
Alternatively, or as a special-case, the Ethernet solution using the
*full-duplex* twisted pair connection (no multi-drop) should be
feasible, and the Motorola MPC8xx communications capabilities should
meet the latency requirements.
Regards,
- Thierry Moreau
CONNOTECH Experts-conseils inc.
9130 Place de Montgolfier
Montreal, Qc
H2M 2A1
Tel.: (514)385-5691
Fax: (514)385-5900
e-mail: thierry.moreau@connotech.com
Reply by Peter C. Wallace●January 27, 20042004-01-27
On Tue, 27 Jan 2004 16:00:27 -0800, Kenneth Porter wrote:
> I'm shopping for a serial bus that will give me lots of bandwidth (10's
> of Mbps) and low latency (20 kHz update cycle).
>
> I'd also like electrical isolation. An optical fiber transport would be
> ideal but some kind of high-speed opto-coupled or magneto-coupled system
> would also be workable.
>
> What's available? I've heard FireWire mentioned and see that it can do
> 400- 1200 Mbps, but haven't found a latency spec and suspect its frame
> time is too long for my application. Ethernet and USB2 have the needed
> bandwidth but the latency is too long and Ethernet would be
> non-deterministic. Are there other alternatives I should look at?
If its point-to point (no possibility of collision), Ethernet will give you
full duplex, electrical isolation and determinism - plus low cost.
Latency is more complicated as this depends on where the data needs to go,
host CPU/OS interrupt issues etc etc
Peter Wallace
Reply by Pete Fenelon●January 27, 20042004-01-27
In comp.arch.embedded Kenneth Porter <ken.blacklist@sewingwitch.com> wrote:
> I'm shopping for a serial bus that will give me lots of bandwidth (10's of
> Mbps) and low latency (20 kHz update cycle).
>
> I'd also like electrical isolation. An optical fiber transport would be
> ideal but some kind of high-speed opto-coupled or magneto-coupled system
> would also be workable.
>
> What's available? I've heard FireWire mentioned and see that it can do 400-
> 1200 Mbps, but haven't found a latency spec and suspect its frame time is
> too long for my application. Ethernet and USB2 have the needed bandwidth
> but the latency is too long and Ethernet would be non-deterministic. Are
> there other alternatives I should look at?
Might be worth looking at MOST.
Fibre-optic, pretty high-speed, and 25mbit.
http://www.mostnet.de/
pete
--
pete@fenelon.com "there's no room for enigmas in built-up areas"
Reply by Kenneth Porter●January 27, 20042004-01-27
I'm shopping for a serial bus that will give me lots of bandwidth (10's of
Mbps) and low latency (20 kHz update cycle).
I'd also like electrical isolation. An optical fiber transport would be
ideal but some kind of high-speed opto-coupled or magneto-coupled system
would also be workable.
What's available? I've heard FireWire mentioned and see that it can do 400-
1200 Mbps, but haven't found a latency spec and suspect its frame time is
too long for my application. Ethernet and USB2 have the needed bandwidth
but the latency is too long and Ethernet would be non-deterministic. Are
there other alternatives I should look at?