Comment

01 June 2002

Ethernet switching and network management

By Karl Glas and Bill King

The objective of factory automation is vertical integration

If the '80s was the decade of the PC, and the '90s was the decade of the LAN, the new decade will be the decade of the Internet, as industry figures out how to leverage its immense possibilities.

Information is now available from anywhere at any time. The question of many manufacturers is how to best utilize the Internet technology in automation.

The objective of factory automation is vertical integration-that is to say, the transparent connection of the different automation network levels. This would include maintaining continuity within the network and allowing easy access to data. Upon acquiring the data, it all immediately becomes available to Internet browsers.

The basic protocol for Internet technology is TCP/IP, and the network of choice is Ethernet. The wide distribution of Ethernet, which has a market share of more than 90% in data communication, and the continuing development of Ethernet technologies with scalable data rates of 10/100/1,000 megabits per second (Mbps) ensures Ethernet will excel in the future.

The potential for a dramatic increase in network performances, offered by switching and full duplex technologies, not only provides adequate resources for future applications but also accounts for Ethernet's real-time capabilities.

For many users of classic fieldbuses, the question is to how best utilize these technologies. Here are important concepts and functionalities to consider.

Establishing a connection

Switching is when data packets transmit directly from the input port, where the data originated, to the output port, which contains the destination address of the data packet.

Switches use the direct switching methodology in a way that's similar to the method telephone exchanges use. The switch thus establishes a connection between the input port and the output port for the duration of the data transmission.

Switches perform the following functions:

Connect collision domains/subnetworks:
Because repeaters and hubs function at only the physical level-they do not distinguish addresses-their use is limited to collision domains. These network domains, in which all nodes share the same signaling rate (usually 10 Mbps), are shared LANs. Switches connect collision domains. Their use is not limited to the maximum distance covered in a repeater network, as switches can create networks that extend for several hundred kilometers.

Load decoupling
Filtering the data traffic on the basis of the media access control (MAC) addresses ensures that local data traffic remains local. In contrast to repeaters/hubs, which send out broadcast messages to every port on the network, switches use the direct switching method.

Only data destined for nodes in another subnetwork transmits from the input port to the switch's relevant output port. To make this possible, the switch itself creates a table of MAC addresses for each port in a self-learn mode.

Limit errors from spreading to the relevant subnetwork
By checking the validity of a data package on the basis of the check sum each package contains, switches ensure that data packets containing errors don't go further. They also prevent collisions in one network segment from passing to another segment.

Parallel communication
Switches can transport multiple data packets between different network segments simultaneously. Depending on the number of ports, the switch establishes multiple temporary and dynamic connections between pairs of network segments.

This makes it possible to attain a tremendous increase in the network's data throughput and efficiency. Only when several simultaneously received data packages are heading to the same destination port do switches buffer, prioritize, and transport the data packets in succession.

Full duplex (FDX) and half duplex (HDX) are network operating modes. While nodes alternately send and receive data in HDX mode, the FDX mode allows them to send and receive simultaneously. When using FDX, the participating nodes' collision detection automatically deactivates, as collisions do not occur.

Even in Ethernet/Fast Ethernet (100 Mbps), full duplex is not a network topology but rather a method of data interchange agreed on between the two nodes.

A prerequisite for FDX is the use of transmission media with separate send and receive channels because different paths are required in order to transmit and receive at the same time. Fiber-optic and twisted-pair cables have this capability. The participating nodes must support full duplex as well.

In the case of HDX connections, the transmitter and receiver share the same physical medium (cable). At any given time, only one peer can send data while the other peer receives data. The peers in the communication link alternate use of the medium to send data.

The classic coaxial cable is a typical example of an HDX medium. In addition, fiber-optic and twisted-pair cables operate in the HDX mode, especially when the peers do not support FDX.

Nondeterministic, but . . .

Ethernet networks use a random-access method: carrier sense multiple access/collision detection (CSMA/CD). This makes it impossible to ascertain precisely when a node can carry out its communication request if a collision occurs.

Network collisions can interrupt communication, necessitating a retry. For this reason, Ethernet is nondeterministic. It is not possible to ascertain exactly how long it will take to transmit data from one node to another. There are no fixed cycle times.

However, when the network nodes connect individually to a switch port and operate in the 100-Mbps FDX mode, the peak-to-valley variation in the required transmission time is considerably shorter because there are no collisions in the network.

For all practical purposes, the time the switch needs to process the data packets is constant. The variable amount of time it takes to deliver the data packets, however, is important. This is influenced by the network load and thus the number of data packets in the output port's buffer, as well as by the length of the relevant data packets.

Although switching, 100-Mbps, and FDX still do not make Ethernet deterministic, suitable network design makes response times so short that a 100-Mbps, FDX, switched network is, in effect, real-time capable. And this response time is what makes Ethernet perfectly suitable for the factory floor.

Redundancy requires more

The IEEE 802.1-(d) standard describes the spanning tree algorithm, which serves the organization of meshed Ethernet structures made up of multiple switches. In order to prevent data packets from circling around aimlessly in a network configured in a loop, various connections switch and segments close, turning the ring structure into that of a tree structure.

The switches communicate with one another via the spanning tree protocol.

When a break occurs in one of the segments, the spanning tree protocol recognizes that this break occurs and switches over to a new segment. This allows data communication to continue.

The problem with the spanning tree algorithm is that it is very complex, sluggish, and time insensitive. Depending on the topology and the number of switches involved, the network usually requires more than 30 seconds to switch over when a break occurs.

The interruption can grow to more than 1 minute when considering a large, complicated network. The spanning tree protocol simply cannot meet the demands of industrial communication.

Broadcast routing protocols

In order to maintain the extremely short response times that automation requires, plants leverage specially designed algorithms for redundancy control. Using these, the reestablishment of a network after a break is less than 300 milliseconds.

The data terminal equipment's logical communication links are not severed even during reconfiguration of the network because of a fault or error, and the plant remains under operator control at all times.

This is simply not possible using commercial equipment.

Monitor network health

Network management describes the administration of a network and includes these areas:

• Configuration management entails collecting, representing, checking, and updating configuration files. This includes the configuration data for individual ports of a network component or the communication relationships among the network components themselves.
• The purpose of fault management is to increase the availability of the network. Detecting faults as soon as possible is a must in order to avoid failures in the network. Another purpose of fault management is to expedite fault diagnosis and recovery.
• Performance management helps monitor network performance by collecting network load data over specific time periods. This data in graphical form makes it possible to visualize the chronological development of the network load situation, which in turn allows for continuous evaluation.
• Security management makes it possible to give passwords and access authorizations to a data network and its valuable resources. Security management also includes the detection of unauthorized access attempts.

Running a simple network

Simple network management protocol (SNMP), a TCP/IP-based communication protocol, has established itself as the de facto standard for network management. The management station-usually a PC with special network management software-communicates with the so-called agents in the various network nodes such as network components, switches, hubs, PCs, network printers, and the like.

Network throughput

The agent, normally implemented in the form of a software module, makes the station-specific data available to the management system. The station data provided by the agents are stored in a structured form in the management information base (MIB).

Furthermore, the communication between the management station and the agents take place in the background and places very little load on network traffic. The management station reads and writes cyclically. However, the agent can also send an event-controlled message (trap) to the management station.

Web-based management enables access to individual network nodes (switch) via standard Internet browsers. An applet resides on each Web server, and the browser loads the applet. The applet accesses data from the MIBs of the addressed modules and represents this data on the browser as a hyptertext markup language page.

The familiar browser affords quick entry into network management with little additional training. However, Web-based management allows access only to a network node.

A complete overview of all nodes in the network and their topological arrangement is possible only with a comprehensive SNMP-based network management tool.

This vertical integration places new demands on time-tested Ethernet technology and will push development further ahead. The industrial aspects in particular will thus assume an ever-increasing importance. Switching and Fast Ethernet have brought Ethernet a long way toward real-time capabilities.

Office components cannot, however, adequately meet all the demands of an industrial automation network. Cost-efficient and powerful network management tools are available that provide a basis for simple mastery of even large networks. The intelligent inclusion of these options in the automation solution will be the automation engineer's primary job. WBJ

Behind the byline

Karl Glas is system manager for network components at Siemens AG SIMATIC NET Industrial Communications division in Germany. Bill King has a degree in industrial engineering and is product manager for wireless technology products at Siemens in Norcross, Ga. He is an ISA member. Write him at bill.king@sea.siemens.com.