The search for an industrial network can take you in many directions. But lately the road most traveled leads to twenty year-old Ethernet.

Ethernet has many of the features installers have looked for, but so far not found, in other industrial buses. It's inexpensive. It's an off-the-shelf solution. It's hardware and platform neutral, connecting to a wide range of devices. It's easy to install. It interoperates with the Internet. And it's compatible with the dominant industrial control, the PC.

Despite these advantages, however, there are factors that can turn Ethernet's apparent strengths into weaknesses. Take a few minutes to become familiar with these factors, and hopefully, you'll be able to avoid many common interoperability problems.

Ethernet vs. Ethernet

After twenty years, Ethernet has been rid of most of its bugs. It's been around so long, though, that many parties have developed their own variations of the original specification.

These variations all go by the name of Ethernet. But they are not necessarily interoperable. Thus, it's important to determine which Ethernet version your clients are specifying for their motion applications.

Managing changes to the original Ethernet was the responsibility of the Xerox Palo Alto Research Center after the network was completed by engineers at Xerox, DEC, and Intel. They last made changes to the specification in 1982, after which the company gave up the trademark. This last version is known as Ethernet II or DIX V2.0.

A few years later, IEEE approved specification 802.3, which is commonly referred to as Ethernet and shares some features with DIX V2.0. One difference is in the methodology for controlling network traffic. IEEE 802.3 uses a method known as CSMA/CD, or carrier sense multiple access with collision detection.

Another crucial difference can be found in the messaging format, or frames. Frames in both versions have the same number of byte fields with the same total number of bits. But several byte fields between the versions hold different numbers of bits, making interoperability uncertain.

There are ways to reconcile these differences, but they don't go far enough to enable communication. That's because DIX and 802.3 describe just the physical media needed to transmit signals and how the signals access that media. The other tasks needed for communication have been left - by design - to other specifications, or protocols.

Those protocols conform to requirements for the network and transport layers, the third and fourth of seven possible layers in any network. (The various versions of Ethernet usually handle the tasks of the first and second layers.) It's their responsibility to move data from point A to point B. Probably the most wellknown protocols for these layers are the Internet Protocol (IP) and Transmission Control Protocol (TCP).

Ethernet combined with TCP/IP is often simply called Ethernet. But the Ethernet sublayer could be either DIX or 802.3. Even though most Ethernet- TCP/IP stacks in operation use the DIX frame format, it's important to verify the version to eliminate a possible source of interoperability problems.

DIX and 802.3 are not limited to working with TCP and IP. There are other protocols for the network and transport layers that can be grouped into a protocol stack that are just as valid and effective at moving messages. And, as you may have guessed by now, this stack group is also often called Ethernet.

One more protocol layer is required to establish communication among devices. It's known as the application layer or layer seven.

A network can have a total of seven layers. Most industrial buses, though, use a minimum number of protocols to fulfill the functions of layers one through four and layer seven. Adding more than necessary tends to decrease a network's data transmission speed. Each layer has a known delay, usually of nanoseconds, before passing data to the next layer. Depending on the type of data to be transmitted, this cumulative delay can be crucial.

The application layer is the latest focus of interoperability problems, and the next battlefield in the notorious "bus wars." While large control suppliers have not been keen on promoting any version of Ethernet for industrial control, they've adjusted to user demands and have taken the top layers of their protocol stacks and placed them over Ethernet protocols. The results are known as Profibus over Ethernet, DeviceNet over Ethernet, CIP protocol on Ethernet, Modbus/TCP, and so on.

As system integrators are discovering, a network with Ethernet in its name is no guarantee of interoperability with other "Ethernet" networks. Formats among protocols must be compatible all the way to the application layer. And none of the industrial bus application layers are. Each handles such tasks as error checking, statistics gathering, and configuring differently.

Continue to Page 2

When will data arrive?

The next factor that affects Ethernet's "plug-n-playability" is an application's need for deterministic performance. Neither DIX or 802.3 are deterministic. That is, you cannot determine exactly when a message will arrive at its destination. In addition, CSMA/CD prevents fair and predictable access to the network. For many discrete applications, this is a problem.

The methodology uses backoff algorithms and collision timers to negotiate access after detecting a collision between two devices. When one device gains access, its timer resets to zero. The other device's timer, however, continues to increment, increasing the wait time for the next transmission attempt, until it finally achieves a successful transmission. Thus, the access timing could be off just enough for the first device to transmit several times before the other device gets its chance.

Ethernet can be made deterministic. But usually there are tradeoffs.

One way is to add raw horsepower in the form of bandwidth. Ethernet comes in several transmission rates. Ten Mbps systems can be converted to 100 Mbps, also known as Fast Ethernet. At this speed, statistical chances of message collisions drop considerably. All attached devices must be able to send messages at this speed, however, to benefit.

Another way is to convert to full-duplex cabling to eliminate simultaneous sendand- receive message collisions. Ethernet was originally half-duplex, but newer versions are full-duplex.

Still another way is to give every critical device a dedicated network port, or switch. This technique improves the throughput of 10 Mbps Ethernet systems enough that an upgrade to Fast Ethernet may not be needed. The trade-offs, though, are cost and flexibility. Switch arrangements are usually limited to the star topology.

For even higher throughput, though, you can increase bandwidth as well as provide private ports.

Alternatively, you can use an industrial network like Modbus/TCP. On this network, devices can only send data when queried, providing guaranteed data transmission because this configuration turns the network into a master-slave arrangement. However, Ethernet's inherent multimaster features are lost.

The latest techniques for improving deterministic performance are known as Quality of Service (QoS) features. They allocate and prioritize a network's resources to ensure that data move to their destination consistently and reliably. They also provide a way for Ethernet switch developers to differentiate themselves.

These features are usually not available for older Ethernet systems. And some require the use of Windows 98 or NT.

Bits and bytes

Industrial applications may have problems with the size of Ethernet messages. Long messages are usually not a concern. The maximum size limit is 1,500 bytes without certain overhead formatting. Third and fourth layer protocols may restrict message sizes, though. IP, for example, allows no more than 576 bytes in a frame.

The problem comes in sending just a few bits of data. The shortest Ethernet frame is 64 bytes. The data field must be a minimum of 46 bytes, otherwise the frame is viewed as a message fragment, often called a runt, and the result of a transmission error. Therefore it's "dropped;" it is not forwarded to its destination. To avoid this problem, short data fields must either be padded with other small sized data messages or filler bits.

If sending bits of data is the norm in a control or real-time application, then Fast Ethernet may be the only Ethernet choice. Other versions may not provide the network speed and reliability needed to handle lots of minimum length packets.

Of course, an alternative is to use a network more geared to real-time motion and control, even though set up may not be plug-n-play easy. Such networks include IEEE 1394, Macro, and Sercos.

The bottom line is that no available industrial network offers the low cost, easy connection, and universality everyone wants. Not yet.