In 1886, Matt Kilroy of the Baltimore Orioles struck out 513 batters. Kilroy accomplished this feat toward the end of a seven-year period during which Albert Michelson and Edward Morley conducted a series of experiments to measure the speed of light very accurately. In spite of their precise interferometer and the care with which they made the measurements, they found little evidence that the speed of light was affected by the earth’s rotation.
This negative result effectively disproved the existence of the ether (or aether), a substance that had been postulated to fill space to support the propagation of light waves. At the time, the difference between longitudinal and transverse waves was not realized: the former required a medium in which to propagate but the latter didn’t. Einstein’s theory of relativity was influenced by this unexpected result, and Albert Michelson was awarded the 1907 Nobel Prize in physics for his work.
As an electronics engineering undergraduate many decades later at Case Institute of Technology, I regularly came close to the site of the famous experiments. Michelson’s lab was long gone from the basement of the Old Main building, replaced by rows of student mailboxes. But the old pictures hanging in the building and the original Case School of Applied Science name carved high on the ivy-covered stone façade reminded you that the ether had been a huge subject of debate at one time.
This is the same ether that forms part of the familiar Ethernet local area network (LAN) name. Keeping with the original medium-related meaning of the term ether, Ethernet can use many types of media such as copper cables, fiber- optic cables, and radio links. When David Boggs and Robert Metcalf of Xerox developed the original multidrop, carrier-sense-multiple-access-with-collision-detect (CSMA/CD) Ethernet in the 1970s, it had a transmission rate of 2.94 Mb/s and supported up to 256 devices over a 1-km cable.
Ethernet As It Has Been
Transmission speed, range, and accuracy are closely related to the topology of a network and the means devices must use to access it. For example, there are modern spread-spectrum systems (not Ethernet) that allow all users to transmit simultaneously, relying on the orthogonality of the channel encoding for unambiguous message recovery at the receivers. This approach makes maximum use of the bandwidth of the medium although cost and complexity are high.
In contrast, Ethernet supports only one talker at a time and uses a more structured bus or star configuration. Because it is a broadcast system, all other devices on the network can listen but cannot start to talk themselves until no one else is talking. In its half-duplex CSMA/CD form, Ethernet is only bandwidth efficient when lightly loaded. But it is relatively simple and low cost, and it is a well-proven technology. Today, Ethernet accounts for 85% of the world’s LAN market.
Figure 1
shows the details of typical serial data frames defined by IEEE 802.3 and Ethernet II standards. The two implementations are very similar. The most significant difference between them is that Ethernet defines services corresponding to both the physical or layer 1 and data link or layer 2 within the open systems interconnection (OSI) reference model. IEEE 802.3 LANs also specify layer 1 but only the media access control (MAC) portion of layer 2.
The logical link control (LLC) protocol is not specified in the 802.3 specification, hence, the Ethernet type field referring to the type of LLC. Regardless of the subtle differences between the two standards, in common usage today, the term Ethernet generally refers to the 802.3 system unless stated otherwise.
The name given to a particular combination of medium, speed, and range is defined in Figure 2. For example, 10Base-T is a 10-Mb/s system, running at baseband, having segments up to 100-m long, and using unshielded twisted-pair (UTP) cable. 10Base-5 is a 10-Mb/s system, running over thick 50-W coaxial cable with 500-m segments. 10Base-5 is the 802.3 variant closest to Ethernet.
The range of the system is inversely proportional to its speed; that is, 10Base systems have approximately 10 times the range of 100Base systems, based on the workings of the CSMA/CD protocol. Should two devices at opposite ends of the network start talking, it will take a finite amount of time for the collision to be detected.
The maximum length of a network must be short enough that device A at one end can detect that device B at the other end has started talking within the first frame A sends. Allowing for repeater delays as well as cable delays, the maximum network size is 2,460 m for 10Base-5. Cable loss and crosstalk also enter into the equation, so 10Base-T, for example, allows only 100-m segments of UTP compared to 500 m for 10Base-5 on thick 50-W coax.
If there is a collision, the first device to detect it will send a jamming signal. The intention is to make sure that all devices detect the collision and stop transmitting. Each node trying to transmit will back off for a time proportional to its priority or for a random time and then try again. Back-off algorithms are efficient enough that multiple back-offs are seldom required unless the network is heavily loaded or there has been a failure in a node.
100Base-T, or Fast Ethernet, is defined in the 802.3u standard. Both 10-Mb/s and 100-Mb/s rates are supported and the same 802.3 MAC runs at both speeds (Figure 3). Three types of physical media are specified: two pairs of category 5 UTP, two strands of 62.5/125-µm multimode optical fiber, or four pairs of category 3, 4, or 5 UTP if an alternate 4T+ signaling scheme is used.
Gigabit Ethernet, defined in the IEEE 802.3z standard, retains compatibility with 10Base and 100Base systems and offers higher bandwidth with switch-to-switch and switch-to-end full-duplex operating modes. The 1000Base-X specification covers the use of long-wavelength lasers with single and multimode optical fibers (1000Base-LX), short-wavelength lasers with multimode fiber (1000Base-SX), and shielded, balanced 150-W copper cable (1000Base-CX). There also is a 1000Base-T specification that runs over four pairs of Cat 5 UTP.
Increases in bandwidth have been accomplished by a combination of changes. Running the system faster certainly is part of the answer, but more sophisticated modulation schemes that pack more bits per symbol also are required.
Three-level Manchester encoding is used with 10Base systems, guaranteeing a synchronization transition in the center of each bit period. The copper medium carries 0.85 V for a 1, -0.85 V for a 0, and 0 V for idle.
The fiber distributed data interface (FDDI) 100-Mb/s protocol is the physical medium dependent (PMD) layer used to carry 100Base Ethernet. A new convergence sublayer adapts the half-duplex, start-stop signaling of 10Base Ethernet to FDDI’s full-duplex, continuous signaling. Also, FDDI encodes every 4 original bits into 5 bits, giving rise to the 4-b/5-b terminology associated with Fast Ethernet.1
The 1000Base-T variant of Ethernet uses four pairs of UTP rather than the two pairs called for by 10-Mb/s or 100-Mb/s systems. It achieves high speed by using five-level phase amplitude modulation (PAM-5).
The 150-W copper and optical fiber Gigabit systems are based on a Fibre Channel PMD layer that requires 8-b/10-b encoding. Because 10 is an even number, each group of transmitted bits can be balanced to have zero DC offset. This is not the case with the 4-b/5-b system and is important at high data rates.
Ethernet As It Is Changing
Groups of devices are within a collision domain if they are attached to the same assembly of cables and repeaters, also called hubs. Up to 1,024 nodes can be part of a collision domain. Of course, if very many of these devices try to communicate with each other at the same time, a large number of collisions occur and network efficiency will drop.
Although the same CSMA/CD protocol will operate for Gigabit Ethernet as for slower systems, the minimum frame length has been extended. At the high speeds of 1000Base systems, the original 64-byte minimum packet size restricted the end-to-end bus distance too much. Also, to improve efficiency, bursts of frames now can be sent without relinquishing control of the bus.
With regard to these necessary changes to accommodate 1-Gb/s operation, “It is important to point out that the issues surrounding half-duplex Gigabit Ethernet, such as frame size inefficiency, which in turn drives the need for carrier extension, as well as the signal round-trip time at Gigabit speeds, indicate that, in reality, half-duplex is not effective for Gigabit Ethernet.”
2
Bridges and switches separate collision domains in those networks that do operate in the half-duplex mode. Using these devices, large networks can be subdivided to ensure higher actual speeds within each of the smaller parts. Multiport switches even allow individual nodes to be isolated to achieve the highest possible performance with full-duplex operation. But, breaking up a network to this extreme creates other problems.
According to Neal Allen, a product manager with Fluke Networks Division, “A greater impact was felt when switching technology was introduced than when Gigabit Ethernet became available. Switches removed the capability to use traditional tools like protocol analyzers as general troubleshooting tools. For example, unless you are prepared to log into a switch and set up port mirroring or aliasing, then a protocol analyzer is of diminished value in a switched environment.”
Agreeing with Mr. Allen was Othmar Kyas, the NSTD product marketing section manager at Agilent Technologies. “The biggest problem from the perspective of network testing is the strong trend toward Ethernet switching. In these environments with today’s protocol analyzers, it is only possible to monitor one port at a time, which in many cases means just being able to monitor the traffic to and from a single network node. Reading out port statistics may not provide the necessary information so troubleshooting can become a major challenge,” he concluded.
There also is a trend toward using Ethernet in real-time control applications. Traditionally, Ethernet has only been used for process-related information, but not for time-critical control of the process itself. This is because it is impossible to determine the exact time a command will reach an actuator if a collision occurs requiring the contending nodes to back off before trying to transmit again.
Put another way, Ethernet is nondeterministic. One way to improve performance eliminates the likelihood of collisions by isolating nodes with switches. Another approach uses a very fast network to minimize the delays caused by collisions.
These reasons were cited for using Ethernet as an industrial control technology:
The ready availability of Ethernet-literate engineers and technicians.
Ethernet’s low cost.
Critics reply that it is misleading to compare very low-cost network interface cards (NICs) intended for an office PC environment to the ruggedized NICs needed on the shop floor. Also, there is no quality-of-service mechanism inherent in Ethernet to guarantee levels of packet latency. Should a node become overloaded because of a fault in the network, packets could be lost. And because all data has the same priority level, very important messages have to wait for less important ones.
However, the most convincing argument is that Ethernet networks require an additional layer of one or more communications protocols running on top of them to provide complete data transfer and network management functionality. These protocols are very often transport control protocol and Internet protocol (TCP/IP), but can also be DECnet™, MAP™, AppleTalk™, or Novell IPX™. Even if all industrial manufacturers used the Internet standard protocols, TCP/IP, a problem still would exist.
According to an Allen-Bradley white paper, “Each vendor of automation equipment that runs over Ethernet-TCP/IP has implemented its own application layer protocol. As a result, equipment from different automation vendors connected to the same plant-floor intranet can physically coexist on the LAN but cannot interoperate.”
3
Resolving this Tower of Babel problem is the goal of the various Fieldbus groups, and Fast Ethernet may be their lower-level technology of choice.When Ethernet Becomes MichelsonNet
Because of its cost and very high-speed capabilities, Gigabit Ethernet is more appropriate today for the network backbone than for desktops. Following closely behind 1000Base systems is the development of 10-Gb/s Ethernet, rivaling optical wide area network (WAN) rates. “Terabit Ethernet is looming on the horizon,” said Dave Ushler, LAN product manager at Digitech-LeCroy. “To take advantage of these higher-speed networks, protocols will become more complex to handle the various framing and packet difficulties inherent in the exchange of data at higher speeds.”
“Testing of higher-speed networks does not lend itself to real-time packet analysis,” he continued. “Statistical and post-capture analysis will be the only methods of determining the health and efficiency of a network. Analysis tools will require extremely elaborate filtering algorithms to minimize and zoom into the large amount of information.”
As an example of high-speed Ethernet-based networks, some metropolitan area network (MAN) cable systems are based on the EtherRing developed by Codenoll Technology. Using pairs of counter-rotating optical fiber rings and dense wavelength division multiplexing (DWDM), EtherRing networks transmit native-mode Ethernet data packets between switches up to 45 miles apart and at overall speeds as high as 16 Gb/s.
According to the company, this technology consolidates routing and provides distributed switches in place of hubs. It no longer is necessary to convert Ethernet packets to frame relay or asynchronous transfer mode (ATM) formats, for example. As a result, the cost of the conversion equipment is eliminated, and the transmission capacity is maximized.
This particular implementation of a MAN combines features of token-ring networks defined in IEEE 802.5 and the 802.3 version of Ethernet. It uses switches to eliminate the collisions inherent in a CSMA/CD system. The optical fibers carry DWDM signals to achieve the high data rate. So, is it still Ethernet, or is it something else—maybe MichelsonNet?
The answer is that data communications is changing at a very fast rate. Ethernet, as initially defined by Xerox, underwent changes before it was adopted by the IEEE as 802.3, when it was changed again. However, the basic frame definition has not been altered. This is the remaining core specification that seems to be fundamental to Ethernet being Ethernet.
References
1. “Ethernet/IEEE 802.3,” Internetworking Technology Overview, Chapter 7, Cisco Systems, June 1999, p. 5,
www.cisco.com/univercd/data/doc/cintrnet/ito/55771.htm.
2. Ibid, p. 19.
3. Ethernet for Industrial Control, An Ethernet White Paper, Allen-Bradley, April 1998, www.ab.com/networks/enetpaper.html.
LAN Test Products
Voice Over IP Simulator
The H.323 Simulation Software Package benchmarks, loads tests, and verifies proper protocol implementation in Voice over IP equipment. Features include simulated H.323 terminal call generation with or without a gatekeeper; control of codec formats for real-time transport protocol (RTP) voice packet transmission; and pinpoint testing of registration, administration, and status (RAS), Q.931, and H.245 protocols during the call setup process. Detailed control of simulation parameters and analysis of call information are provided. The software package runs under Windows NT on a PC with a standard 10/100 Ethernet adapter card. Starts at $8,000. RADCOM Equipment, (800) 723-2664.
Intelligent 10/100 Tester
The OneTouch™ Network Assistant automatically proves connectivity from desktop stations to local routers and Novell
®
, Windows NT, and NetBIOS servers. An icon-based interface and a touch-sensitive display screen improve use for front-line network support personnel. In addition to providing 10/100-Mb/s Ethernet criteria such as collision and utilization rates, errors, and broadcasts, the Network Advice™ feature assists in determining acceptable performance limits. Capabilities include IP Ping; detection of network connections; audio location of cable faults; and identification of opens, crossed pairs, and split pairs. 10-Mb/s: $3,695; 10/100-Mb/s: $4,995. Fluke Networks Division, (800) 443-5853.Modular LANs Tester
The Internet Advisor helps isolate and solve problems on LANs. The Model HP J3446D offers Ethernet and Fast Ethernet analysis in the mainframe and can be enhanced with undercradles for Gigabit Ethernet, FDDI, or token ring. Dual connectors support switched-environment testing. Features include expert analysis, statistical measurements, active stimulus/response tests, and traffic generation. The drill-down capability provides access to a specific protocol layer, to only the selected protocol stack, or to field-by-field protocol decode for every frame. $20,895. Agilent Technologies, (800) 452-4844.
On-Time Protocol Analysis
The SmartTime
®
LAN Protocol Analysis System collects and time-stamps traffic using a global positioning system (GPS) world clock. The system provides time-synchronized accuracy to <10 µs between separate analyzers. In addition, secure, web-based remote-control capabilities support access by staff at any location. They can simultaneously view the performance of each monitored segment and analyze packet delay. Measurements include bandwidth, utilization, accuracy, throughput, latency, delay variation, and response time. Contact company for price. Digitech-LeCroy, (800) 821-2265.
Copyright 2000 Nelson Publishing Inc.
March 2000
|
|