Can gigabit Ethernet improve industrial control network performance?
These days many machine builders, manufacturers, and integrators may be using one, or more, popular fieldbus systems based on standard Ethernet. Currently, most if not all of these are based on 100 Mbit/sec line speed, or “fast Ethernet.” Depending on the protocol, some users may find the delivered performance and speed to be somewhat less than anticipated. Has 100 Mbit Ethernet been eclipsed in performance, speed, precision, or throughput by the promises of gigabit Ethernet? Are there additional costs or real-world limitations imposed by moving to gigabit Ethernet? Where does it really make sense to use gigabit Ethernet and 100 Mbit Ethernet in the field today? If you are responsible for answering these difficult questions, you need practical answers.
While a popular buzzword in some device vendor circles, gigabit Ethernet has some cost and technological limitations when used in industrial applications in 2011. Efficient 100 Mbit Ethernet protocols are available today that avoid the current pitfalls of gigabit Ethernet while delivering industry-leading speed, performance, precision, and cost controls. This discussion will examine the benefits and challenges associated with gigabit Ethernet and 100 Mbit Ethernet when used in the realm of machine control and industrial automation. We’ll look at technical considerations that will help you decide which approach is more appropriate for your systems today.
There is no doubt that the use of Ethernet as a fieldbus in industrial controls systems has gained tremendous ground in the past 10 years and has become a well-accepted practice. These systems allow the control system user or designer to integrate many of the same benefits that those in the office network environment have enjoyed for years: the cost-effective ability to utilize standard cables and interconnect devices, standard infrastructure components, and above all, the promise of blazing fast speed. For some industrial Ethernet systems, this speed translates into sub-millisecond communication times to thousands of devices including controllers, I/O, drives, sensors, and more. Additionally, the same know-how of the IT department can be used to manage automation devices. However, the standard application of fast Ethernet, which we perceive as incredibly fast in an office environment, may not work very well when communicating to industrial devices, depending on the approach.
One must consider that Ethernet was originally designed to move large amounts of data around, such as hypertext Web pages or files in a LAN. This is precisely where Ethernet is most efficient, when each frame is moving a sizeable amount of data. Unfortunately, in a node-based industrial system, most automation devices handle only a small amount of data per device, such as drives, I/O devices, encoders, or transducers. These usually pass just a few bytes of data or less. According to the IEEE standard (IEEE802.3) each fast Ethernet frame must be at least 84 bytes because it typically requires two frames to complete the poll and response of each device (one frame to send output data from the controller to the device, and one to receive input data back to the controller and to tell the controller that it is powered and functioning). As a result, the majority of the system bandwidth is eaten up by the overhead of the mechanism that moves the data around—the frame itself. This method of applying Ethernet for industrial purposes is simply inefficient because of the architecture itself and the approach to data transfer.
Gigabit Ethernet to the rescue?
Some experts in the automation and controls industry have begun promoting gigabit Ethernet as the ideal solution in the near-term for these standard Ethernet challenges. The rationale commonly provided for moving from 100 Mbit Ethernet to gigabit Ethernet often falls into two major categories:
1. Many 100 Mbit-based Ethernet protocols are characterized as slow and inefficient. Unfortunately, several of the 100 Mbit protocols have been found to deliver very poor update rates (some on the order of 10’s of milliseconds or higher), very limited ability to perform even the most trivial of synchronization tasks, and are generally poor performers compared to certain promises that have been made and what common sense would dictate. Very simply speaking, what should you do if you want something to go faster? You just add a bigger engine, right? This is the main question being put to the various 100 Mbit industrial Ethernet technology providers: “When will you move your protocol to gigabit Ethernet?” Many think gigabit is the bigger engine that will solve problems for industrial Ethernet today.
2. Adding horsepower (line speed) is the solution for inefficient systems. Gigabit is offered as a panacea for inefficient systems, but this does not correct the fundamental issues that make them slow in the first place. Because the delivered performance has come up short of promises in many cases, in typical more-is-better thinking, the first reaction is to simply suggest a move to the next fastest Ethernet protocol, from 100 Mbit to gigabit, for example. However, going back to the automotive analogy, this may be akin to putting a larger engine in a car with square wheels. It may drive a little faster, but the underlying cause of the lack of speed and the foremost problems with efficiency were not addressed. The question that must be asked may not be, “Should I move to gigabit,” but, “Why is this implementation too slow if it’s supposedly based on a very fast 100 Mbit foundation?”
Gigabit Ethernet—ready today?
The technological advancements made by system suppliers related to gigabit Ethernet certainly have their place in the world, but there are compelling arguments as to why it may not be ready yet for industrial automation and controls implementation. Here are some considerations based on where gigabit technology is in 2011:
1. Most industrial Ethernet protocols suffer from problems that moving to gigabit cannot fix, such as stack delays and infrastructure delays. Several of the Ethernet-based fieldbuses are simply an implementation of an Internet Protocol (IP) stack with another underlying protocol, such as Modbus/TCP or EtherNet/IP as examples. These systems require a CPU running a protocol stack in each slave node, along with active infrastructure components (switches), which are a drag on performance and increase overall system cost.
Perhaps the most unfortunate thing about these types of systems is that they suffer more from stack delays and infrastructure delays (switches, routers, etc.) than can be accommodated with a faster line speed. In the past, information published for a proposed gigabit Ethernet fieldbus, which has subsequently been canceled before being finalized, led many of those reading the information to assume that the gigabit version of this protocol would be 10 times faster than the 100 Mbit version. Once the actual stack and infrastructure delays were calculated and accounted for, the end result was only a 30% improvement in scan time, and the network was still slower than many serial-based legacy fieldbuses. The main point here is that the limiting factor is not always the speed of the physical layer; more often the bottlenecks in the system are the weak links.
2. Actually moving industrial applications to gigabit Ethernet could require new versions of fieldbus protocols, which, unless managed properly, could cause issues with standardization, backwards compatibility, and version control. Just like a drop of water hitting a calm pond sends ripples in all directions, any change to a fieldbus protocol standard causes many waves of change that can be hard to manage throughout the community of device vendors, users, master providers, and everyone else involved.
As an example, imagine that you are a vendor that supplies drives based on a particular fieldbus protocol version 1.0. You go through the necessary training and develop your new drive based on the new protocol, perhaps even develop circuit boards based on specific properties needed by the protocol and the associated software. When a new version of the protocol comes out, which may not be compatible with the old protocol at all, they could call it 2.0, or even 1.1. Now you are required, within a certain amount of time, to have all your current products available on the new standard. Availability of the old version, and eventually multiple old versions, is typically only for spare parts for previously existing customer projects. Now new circuits, new software, and new firmware versions must be developed, tracked, and tested.
Granted, these efforts are no different than those required for developing new products for any new fieldbus protocol. However, there must be a compelling reason and a real resulting solution in order to make this effort practical—especially if the change is a gigabit “upgrade” to an existing industrial Ethernet system.
Solving practical issues
While there are some limitations to gigabit Ethernet technology today that still largely prevent its successful implementation in the field of automation and controls, it isn’t ruled out forever. Here are some practical considerations that must first be addressed:
1. Gigabit devices today typically consume up to six times the power of 100 Mbit devices. Power consumption creates heat which must be dissipated. That calls for heat management techniques and/or more expensive devices because more cooling (active or passive) is required.
This consequence of increasing the communication speed of industrial devices is often overlooked. When loading a small enclosure full of fast Ethernet I/O versus gigabit Ethernet I/O, the difference in heat management for those devices will become very important. Heat considerations will be significant for the device vendor, who must design the new device to work in the same temperature specifications as the cooler-running 100 Mbit device. This will most certainly require some major modifications to the design, which will impart cost, either directly in the form of thermal and physical properties of designs or higher space requirements—having to install components further apart to alleviate heat buildup in those devices.
2. Gigabit cabling requires four pairs of wires instead of two pairs for 100 Mbit, which adds wiring complexity. Cables become more difficult to terminate, new tools are required, and installation time increases. For standard Ethernet cables, which account for a much greater percentage of cabling solutions for interconnecting industrial Ethernet devices than fiber optics, the number of conductors required for Gigabit is doubled when compared with 100 Mbit. The 100 Mbit cables (100BASE-TX physical layer) require two pairs of conductors. Gigabit cables, on the other hand, require four pairs of conductors. Although not a showstopper, this makes field terminations slightly more complicated and another thing to consider when evaluating the move to a newer technology.
3. Gigabit communication methods could be more susceptible to noise (ESD, EMC) in industrial environments. Because gigabit Ethernet (1000BASE-T) uses more conductors and more signal levels than 100BASE-TX, extra care must be given when designing devices and for field cabling to ensure noise-free transmission. This requires extra diligence on the part of vendors to create devices which can effectively channel any EMC noise to ground. Also cable selection, terminations, and connectors need to specified and used with more caution and care to prevent electrical noise from causing performance issues with gigabit communications.
So, if gigabit industrial Ethernet has some challenges to overcome before it can offer real-world results to controls engineers, what should a user look for in a good, fast Ethernet solution today? As stated before, there are fast and efficient fieldbus platforms available today based on 100 Mbit Ethernet, which exceed the requirements and capabilities of scan rate, synchronization, and bandwidth efficiency for most users. So what should a vendor or user look for in a capable Fast Ethernet fieldbus?
1. Efficiency—The bandwidth of the 100 Mbit physical layer offers excellent speed capabilities if implemented efficiently. The use of standard IP protocol stacks, for instance, although seemingly inviting because of familiarity with the IT community, are actually a hindrance when implemented in a fieldbus because of poor utilization of the available bandwidth. Any system that requires a separate and complete Ethernet frame to communicate to each node in the network performs very inefficiently, typically at less than 5% of the possible data rate that the physical layer could be communicating.
2. Eliminating bottlenecks in the system—Once again, there can be many bottlenecks in an Ethernet-based system. Typical areas where Ethernet communication can be slowed include: software stacks in slave devices, switches, hubs, or routers. Very few of these devices are cut through; in other words, they store the complete Ethernet frame, then forward it.
3. Communication in hardware—Consider an alternative to handling the fieldbus network via a software stack running on an expensive CPU with a great deal of RAM. You can replace the data handling with an inexpensive ASIC or FPGA. This way, the network speed is not determined by one bad or slow slave implementation. Also, hardware can typically handle the protocol stack much faster than a software implementation.
Technology for today
Certainly, there will be a time when gigabit technology will develop sufficiently that it can be effectively applied for all Ethernet-based fieldbus protocols. But for now, most of the gigabit challenges have yet to be solved, including heat dissipation, wiring complexity, and lack of streamlined performance that will be required to make the most of these technologies in industrial applications. It makes more sense to choose one of the efficient, cost-effective systems based on today’s 100 Mbit technology. One such example is EtherCAT, a 100 Mbit-based industrial Ethernet system that provides the ability to update hundreds of servo axes in less than a millisecond—not because it breaks any rules of the IEEE802.3 standard, but because it efficiently uses the 100 Mbit bandwidth to communicate to many nodes with one frame, instead of using the inefficient node-based communication that comes from basing the protocol on the Internet protocol.
Such a system can communicate to 1,000 distributed I/O points in 30 microseconds (µs), implement a network of almost unlimited size, and ensure optimum vertical integration because it still uses Ethernet and Internet technologies. It can also replace the costly Ethernet star topology with a simple line or tree structure—no expensive infrastructure components are required. All types of Ethernet devices can be integrated via a switch or switch port. Hubs and switches can be eliminated and IP addresses are not required on any of the devices in an EtherCAT network. In addition, time-stamping and oversampling is possible, which permits temporal information with significantly higher precision and resolution. With this level of speed, performance, and precision available now from an inherently efficient 100 Mbit network, the case for utilizing gigabit technology today is less compelling.
Joey Stubbs, PE, PMP, is North American representative for the EtherCAT Technology Group.
For more information, visit:
Control Engineering webcasts include more on Ethernet optimization.