Is Ethernet Deterministic? Does it matter?

If you are not familiar with Ethernet, here is an abridged version of how it works:  Every device on the network listens.  If traffic is detected, a device will wait until the line is clear before transmitting on the network.  While a message is being transmitted, a device will listen to make sure that the message it is sending isn’t interfered with, or collide with a message that might have been sent at the same time from another device. If a collision is detected, the device will resend its message again after a time delay, and hopefully not have another collision. The potential of collisions is part of what makes Ethernet non-deterministic.To eliminate collisions,  the physical medium must be restricted, as is the case for telephone cabling – where one line is used to talk and another to listen. A similar design would allow an Ethernet network to operate without collisions.  But, like a telephone, you would be able to connect only two devices together with this “duplex” link.  Switches are commonly used today to create this “duplex” link – so that the only two devices connected to one another are the switch and one other device.  The switch internally manages all other aspects of the traffic between all the devices on the network.However, while a switch eliminates collisions, it does not ensure that time sensitive data will arrive the moment it is expected.  Switches do not provide a Real Time Deterministic network because all the messaging still needs to be scheduled. Switches can receive messages from many devices addressed to one device all at once, but must transmit them serially to the receiving device.  Since you can’t know how far back in line your message is going to be, you’ve lost the time-critical predictability for the time-sensitive data.There are two other options that allow you to avoid the use of switches and create a true, deterministic network: 1) modify either the physical and datalink layers of the Ethernet standard, or 2) modify the architecture of your overall network.   Both of these solutions, however, result in a trade-off since your network is now comprised of some non-standard, specialized hardware which may or may not be compatible with other standardized networking equipment.  Or, the network is confined to a specific architecture.

Does it matter?

Let’s examine why a deterministic network is important.  For most control applications, information transmission or receipt is not time-sensitive.    However, for faster applications such as motion control, the very nature of the control loop requires a predictable amount of time between the receipt of new data for it work properly.  This control scheme requires a deterministic network: one for which the exact time to expect the message is known and predictable.  For time-sensitive variables in any control loop, such as motion control, a message arriving late can be very bad news.  This can also be the case for messages arriving early.

EtherNet/IP is an Application Layer managed by ODVA that sits on top of the lower network layers. This allows the use of standard networking hardware.  The application layer it uses is called CIP (Control and Information Protocol).  Utilizing the correct physical layout and the correct switches as well as ensuring that non-EtherNet/IP devices remain off of the subnet, will yield a fast, real-time network.  However, this configuration will not provide the level of determinism necessary for motion control.  For those applications that need more determinism, EtherNet/IP adds CIPSync, which is an implementation of IEEE 1588, to synchronize clocks throughout the network.

EtherCAT uses a Master/Slave model and some specialized hardware to manage network traffic and is managed by ETG (EtherCAT Technology Group). This model uses specialized hardware.  Therefore, in order to maintain the nature of the network, the types of devices that can be used is limited.  Also, it is difficult to connect two segments of the network together due to the Master/Slave configuration.  However, by locating routers and switches between the Master and Slave, the configuration will yield a fast, real–time, Deterministic network. http://www.ethercat.org/

Ethernet Powerlink (EPL) uses a Master/Slave model without any specialized hardware to manage network traffic and is managed by EPSG (Ethernet POWERLINK Standardization Group).  Like EtherNet/IP, EPL does not use specialized hardware, therefore EPL is easier to implement, but must be well engineered. By ensuring that non-EPL devices are never connected to the network segment, this model will provide a real-time, Deterministic network.  EPL segments can be connected but must comply with the IEEE 1588 standard to synchronize data between segments.  http://www.ethernet-powerlink.org/

PROFInet uses a hardware based solution to manage network traffic and is managed by PROFIBUS & PROFINET International (PI).  Each node of a PROFInet network is highly synchronized at the hardware level to allow determinism.  Because the devices on the PROFInet network use specialized hardware once connected together they form a real-time, Deterministic network. http://www.profibus.com/

These implementations make deterministic Ethernet on the factory floor possible.  With the speeds that Ethernet can now achieve and the addition of the IEEE 1588 synchronization, Ethernet is by far a better choice for control networks than any of the slower proprietary deterministic networks of the past. The evidence of this is that these Ethernet networks are now being certified for safety applications.  The IEC 61508 standard applies to these safety systems.  Look for manufacturers that comply with this standard and provide a safety integrity level (SIL) rating of their network model.