Ethernet Overview

Gigabit LAN Standards Gigabit Ethernet Leads the Way Over Fiber

Gigabit LAN Standards


Specifications

Bus-Standard Ethernet (Coax)

10BASE5

  1. The maximum length of a segment is 500 m (1640 feet).A maximum of 2 IRL (Inter-Repeater Links) is allowed between devices
  2. The maximum length of cable is 2.5 km (1.5 miles).
  3. Devices attach to the backbone via transceivers.
  4. The minimum distance between transceivers is 2.5 m (8.2 feet).
  5. The maximum length of a transceiver cable is 50 m (164 feet),
  6. Up to 100 transceiver connections can be attached to a single segment.
  7. Only transceivers without SOE ("heartbeat") test enabled should be used with repeaters.
  8. Both ends of each segment should be terminated with a 50-ohm resistor.
  9. One end of each segment should be grounded to earth.

Twisted-Pair Ethernet (Unshielded Twisted Pair):

10BASE=T, UTP

  1. There are two versions of Ethernet over unshielded twisted pair-. 10BASE-T (the standard) and its predecessor UTP.
  2. 10BASE-T and UTP segments can coexist on the same network when each hub is attached to a common segment, via a transceiver and transceiver cable, or converter.
  3. The cable used is 22 to 26 AWG unshielded twisted pair (standard telephone wire), at least Category 2 with two twists per foot. Category 3 or 4 is preferred. Category 5 supports 100 BASE-T (Fast Ethernet).
  4. Workstations are connected to a central concentrator ("hub") in a star configuration. Concentrators can be attached to a fiber optic or coax network, and can be daisy chain to form larger networks.
  5. A hub can have AUI port, BNC ThinNet, 10baseT, Fiberoptic for up steam standard Ethernet connections and 10baseT down steam.
  6. The maximum distance of a segment (from concentrator to node) is 100 m (328 feet).
  7. The maximum number of devices per segment is 2. One device is the hub port; the other is the 10BASE-T or UTP device.
  8. Ethernet network interface cards (NICS) are available with built-in 10BASE-T transceivers. Devices with standard AUI ports may be attached with a twisted-pair transceiver. Twisted pair is the most economical cable type, especially since it may already be installed, and it is the easiest to work with. But it is not recommended for installations with much EMI/RFI interference (for example, in industrial environments).

Bus-Thin Net Ethernet (Coax):
10BASE2

  1. The maximum length of a segment is 185 m (607 feet).
  2. A maximum of 2 IRL (Inter-Repeater Links) is allowed between devices, the maximum length of cable is 925 m (3035 feet).
  3. Typically, devices use Ethernet network interface cards (NICs') with built in BNC transceivers. This eliminates the need for separate transceivers, as connections can be made directly to the Thin Net cable.
  4. Devices are connected to the cable with T-connectors, which must be plugged directly into the card. Each dead end must be terminated with a BNC 50-ohm terminator (typical is at the server end and at the ends of the daisy chain) No cable is allowed between the T and the card. Workstations are daisy chained, with an "In-and-out" (unless it is terminated) cabling system.
  5. The minimum distance between T-connectors is 0.5 m (1.6 feet).
  6. If the interface card does not have its own built-in BNC transceiver, a BNC transceiver and transceiver cable are required. The maximum length of a transceiver cable is 50 m (164 feet).
  7. Up to 30 connections can be attached to a single segment.
  8. Both ends of each segment should be terminated with a 50-ohm resistor.
  9. One end of each segment should be grounded to earth.

Star-Fiberoptic Ethernet
FOIRL or 10BASEN-FL

There are two versions of Ethernet over fiberoptic cable, meeting the older FOIRL (Fiberoptic Inter-Repeater Link) and the more recent 10BASE-FL standards.

  1. FOIRL and 10BASE-FL fiberoptic Ethernet differ only in how far each will transmit (the maximum length of a segment). For FOIRL it is 1 km (0.6 miles); for 10BASE-FL it is 2 km (1.2 miles).
  2. The maximum number of devices per segment is 2. One device is the hub port, the other device is the 10BASE-FL device.
  3. Fiber optic cable provides the best signal quality as well as the greatest point-to-point distance typical cable is 62um core 125um cladding.
  4. Fiber optic cable is completely free of EMI/RFI interference.
  5. Fiber optic cable runs point to point only; it cannot be tapped or daisy chained which adds to the security of the network. A fiber optic hub or multiport repeater is required to carry the signal to multiple devices (for FOIRL, a FOIRL multiport repeater and transceivers).
  6. Since fiber optic cable does not carry electrical charges, all electrical cable problems disappear. When fiber optic cable (outdoor quality) is used to link buildings, grounding problems, ground loops, and voltage spikes are eliminated and fiberoptic cable is all so immune to electronic eavesdropping.

Ethernet is the most widely used network topology. You can choose between bus and star topologies, and coaxial, twisted-pair, or fiber optic cabling. But with the right connective equipment, multiple Ethernet-based LANs (local area networks) can be linked together no matter which topology and/or cabling system they use. In fact, with the right equipment and software, even Token Ring, Apple Talk, and wireless LANs can be connected to Ethernet.

The access method Ethernet uses is CSMA/CD (Carrier Sense Multiple Access with Collision Detection). In this method, multiple workstation access a transmission medium (Multiple Access) by listening until no signals are detected (Carrier Sense). Then they transmit and check to see if more than one signal is present (Collision Detection). Each station attempts to transmit when it "believes" the network is free. If there is a collision, each station attempts to retransmit after a preset delay, which is different for each workstation.

Collision detection is an essential part of the CSMA/CD access method. Each transmitting workstation needs to be able to detect that simultaneous (and therefore data-corrupting) transmission has taken place. If a collision is detected, a "jam" signal is propagated to all nodes. Each station that detects the Collision will wait some period of time and then try again.

The two Possible topologies for Ethernet are bus and star. The bus is the simplest (and the traditional) topology. Standard Ethernet (10BASE5) and Thin Ethernet (1OBASE2), both based on coaxial cable systems, use the bus.

In this one-cable LAN, all workstations are connected in succession (a "bus" arrangement) on a single cable. All transmissions go to all the connected workstations. Each workstation then selects those transmissions it should receive, based on the address information contained in the transmission.

In a star topology, all attached workstations are wired directly to a central hub, which establishes, maintains, and breaks connections between them (in the event of an error). The advantage of a star topology is that it is easy to isolate a problem node. The disadvantage is that if the hub fails, the entire system is compromised.

Twisted-Pair Ethernet (10BASE-T), based on unshielded twisted pair, and Fiberoptic Ethernet (FOIRL and 10BASE-FL), based on fiberoptic cable, use the star.


Ethernet Topologies

  1. Fast, reliable throughput speed of 10&100Mbps.
  2. Accurate transmission-CSMA/CD access method.
  3. More LAN components match Ethernet standards than any other.
  4. Maximum flexibility-two topologies (bus or star) and five kinds of cable (Standard
  5. or Thin coax; unshielded twisted pair; FOIRL or 10BASE-FL fiberoptic).

Switched Ethernet

Switched Ethernet relies on centralized multiport Switches to provide physical link between multiple LAN segments. Inside each intelligent S" high-speed circuitry supports wire-speed virtual connections between all the segments, for maximum bandwidth allocation on demand. Adding new segments to a switch increases the aggregate network speed while reducing overall congestion, so Switched Ethernet provides superior configuration flexibility. It also gives you an excellent migration path from 10- to 1 00-Mbps Ethernet, since both segments can often operate via the same Switch.


Benefits of Switched Ethernet

It is a cost-effective technique for increasing the overall network throughput and reducing congestion on a 10-Mbps network. Other than the addition of the switching hub, the Ethernet network remains the same the same network interface cards, the same client software, the same LAN cabling.


100BASE-T (IEEE 802.3u)

100BASE-T retains the familiar CSMA/CD media access technique used in 1 0-Mbps Ethernet networks. It also supports a broad range of cabling options: two standards for twisted pair, one for fiber. 100BASE-TX supports 2-pair Category 5 UTP or Type 1 STP cable. 100BASE-T4 uses 4-pair Category 3 or 4 cable. And 100BASE-FX allows fiber optic links via duplex multimode fiber cable.


Benefits of 100BASE-T

It retains CSMA/CD so existing network management systems don't need to be rewritten. It can easily be integrated into existing 10-Mbps Ethernet LANs so your previous investment is saved.


100VG (IEEE 802.12)

100VG uses an encoding scheme called Quartet Signaling to transmit data simultaneously over all four pairs in the network cable, so it achieves a full tenfold increase in transmission speeds over 1 OBASE-T. It also replaces the CSMA/CD media access control protocol with Demand Priority to optimize network operation and eliminate the overhead of packet collisions and recovery. Demand Priority works like this: The hub directs all transmissions, acknowledging higher-priority packet requests before normal-priority requests. This effectively guarantees bandwidth to time-sensitive applications like voice, video, and multimedia applications.


Benefits of 100VG

It uses a transmission frequency very similar to traditional Ethernet, and works on any conventional cabling system (Category 3, 4, or 5 UTP, Type 1 STP, and fiber optics) and uses the same connectors. In addition, 100VG may soon support Token-Ring networks-a potential advantage over its rival standard 100BASE-T.


ATM

Asynchronous Transfer Mode (ATM) is a cell-based fast-packet communication technique that supports data-transfer rates ranging from sub-Tl speeds (less than 1.544 Mbps) up to 10 Gbps. Like other packet-switching services (Frame Relay, SMDS), ATM achieves its high speeds in part by transmitting data in fixed-size cells, and dispensing with error-correction protocols. Instead, it relies on the inherent integrity of digital lines to ensure data integrity.


Benefits of ATM

Networks are extremely versatile. An ATM network can be treated as a single network, whether it connects points in a building or across the country. Its fixed-length cell-relay operation, the signaling technology of the future, offers more predictable performance than variable-length frames. And it can be integrated into an existing network as needed, without having to upgrade the entire LAN.


The Advantages of 100BASE-T


 

100Base-T (IEEE 802.3u)

Variations of This Standard

100BASE-TX
100BASE-T4
100BASE-FX

Supported Cable Type 100BASE-TX: 5 (2-Pair)

IBM Category Type 1 (2-Pair)

100BASE-T4: Category 3 or 4 (4-Pair)

100BASE-FX: Duplex Multimode or Single-Mode Fiber

Maximum Cable Segments

(HUB-TO-NODE)

100BASE-TX or T4: Category 3, 4, or 5-1 00 m

IBM Type 1-100 m

100BASE-FX: MultimodeFiber-2km,Single-Mode-l0km

(HUB-TO-NODE)

100BASE-TX or T4: Category 3, 4, or 5-100 m

IBM Type 1-100 m

100BASE-FX: MultimodeFiber-2km,Single-Mode-l0km

Best Applications Backbone utilizing Ethernet switches to provide increased throughput

Small to medium workgroups using applications (i.e.-. CAD, CAM)

which output huge data files.


By Erica Roberts

Gigabit LANs at a Glance


Gigabit LAN Standards, It Takes Two to Tangle

Do 1-Gbit/s versions of 100Base-T and 100VG mean double trouble for net managers?

Chalk it up to dj vu all over again. In 1993, when the IEEE couldn't make the call on a single 100-Mbit/s LAN transport, it "decided" on two: 100Base-T and 100VG-AnyLAN. Now the standards body seems to think that playing doubles is a smart bet: It's getting ready to define 1-Gbit/s versions of both high-speed LAN specs.

At first glance, two new gigabit standards look like a good deal. Both standards have the horsepower needed to turbo-charge overtired LANs. Both run over copper and fiber (naturally enough, since they're both based on parts of the ANSI Fibre Channel spec). Both make it possible to build so-called scalable Ethernets. And both are relatively cheap. Vendors are already hinting at prices of $1,500 per node, even though products aren't slated to ship until next year.

Too bad it all could add up to double trouble for net managers. Two standards are bound to lead to compatibility questions. What's more, neither spec guarantees delivery of time-sensitive voice or video. And for all the big talk about big bandwidth, it may not be possible to run either standard over copper without severely cutting speed or limiting the distance between nodes.


Given these shortcomings, "it's unlikely that gigabit Ethernet will ever become a dominant desktop technology," says Paul Sherer, vice president of technology development at 3Com Corp. (Santa Clara, Calif.). Nevertheless, it could make its mark on the backbone--as a fat fire hose for async data.

But does the world really need two 1-Gbit/s Ethernet standards? (Did it really need two 100-Mbit/s LAN specs?) At this point, 100Base-T and 100VG vendors have too much invested in their technologies to let an opportunity like this slip by.

They better move fast. ATM is an obvious--and formidable--rival for the backbone. Prices for ATM (asynchronous transfer mode) hardware are coming down even as we speak, and gigabit ATM gear is on the way. So even if the IEEE manages to push through both very-high-speed specs in record time, ATM could still leave them in the dust.


DIVIDE AND STANDARDIZE

The 802.12 working group has already received an IEEE Project Authorization Request (PAR) to define a gigabit version of 100VG. Not surprisingly, the push is headed up by Hewlett-Packard Co. (HP, Palo Alto, Calif.), the driving force behind the original AnyLAN. Joining HP are Compaq Computer Corp. (Houston), Texas Instruments Inc. (Dallas), and the Semiconductor Division of Motorola Inc. (Austin, Texas). The spec should receive "full and final approval" by the summer of 1997, according to Patricia Thaler, chair of the 802.12 working group and principal engineer for LAN architecture and standards with Hewlett-Packard Co. Roseville Networks Division (Roseville, Calif.).

Like its 100-Mbit/s predecessor, the gigabit version of 100VG will handle both Ethernet and token ring frames. It also will boast several new features, including burst mode and redundant links between repeaters. The plan now is to define transmissions of 500 Mbit/s and 1 Gbit/s, with 4 Gbit/s possibly coming in the future. The group also is working on a full-duplex version of the spec.

The 802.3 working group is still waiting to receive its PAR for the gigabit version of 100Base-T. A task force that includes Bay Networks Inc. (Santa Clara, Calif.), Cisco Systems Inc. (San Jose, Calif.), Packet Engines Inc. (Union City, Calif.), and 3Com Corp. (Santa Clara, Calif.) is working on a proposal that will be presented at the plenary meeting of the 802 LAN/MAN Standards Committee this month.


"Assuming we get through the politics of the plenary meeting, we could have a standard out as early as 1998," says Brian MacLeod, director of business development for Packet Engines.

The gigabit 100Base-T group is looking to keep things simple. It will retain the minimum and maximum frame sizes defined by the 802.3 spec and accommodate full-duplex communications for point-to-point switched connections. It also may incorporate new developments like flow control as they are added to 100Base-T.


SHARE TACTICS

Both specs will retain the MAC protocols employed by their 100-Mbit/s forerunners. Thus, gigabit 100Base-T will use CSMA/CD (carrier-sense multiple access with collision detection), while gigabit 100VG will go with the more recently developed demand-priority mechanism.

CSMA/CD and demand priority are both shared-media schemes, which means sending stations contend for bandwidth. CSMA/CD uses a back-off algorithm to prevent more than one device from sending information at a time. Demand priority relies on a round-robin polling sequence to give each network node the opportunity to transmit.

Shared-media schemes suffer from a simple shortcoming: As the number of nodes on the network rises, the bandwidth available to each drops. To make sure that the new gigabit networks don't wind up suffering from bandwidth starvation, vendors say they'll deliver switched versions of their technologies.

RECYCLED FIBRE

For all the apparent differences between the two gigabit specs, there's one underlying similarity: Both rely on Fibre Channel's physical (PHY) layer as their transmission technology. For gigabit 100Base-T, that means mapping the 802.3 MAC layer to the PHY, so Ethernet packets can be carried via Fibre Channel encoding. For gigabit 100VG, both the 802.3 and the 802.5 (token ring) MAC layers must be mapped.

"The Fibre Channel PHY is a good piece of engineering," comments Thaler. "Why reinvent it?" Recycling Fibre Channel in this way reduces time to market. What's more, it should help bring down prices. Gigabit gear will use Fibre Channel transceivers; as silicon shipments climb, chip prices should plummet. It's still very early in the game, but gigabit over copper could cost between $1,500 and $2,000 per switched connection (which includes the price of the adapter and the switch port). Fiber is expected to cost $3,000 to $4,000.

Fibre Channel aficionados view the gigabit vendors' recycling plans as a tacit acknowledgment of the merits of their technology. They also argue that if a little bit of Fibre Channel is good, all of it would be even better. Scot Ruple, director of product marketing for Emulex Corp. (Costa Mesa, Calif.), expects to see Fibre Channel take off for storage and backbone applications: "Gigabit Ethernet is our chance to slip through the back door and onto the LAN." Thaler disagrees, saying Fibre Channel's MAC layer would add too much overhead to LAN applications.

THE CABLING CATCH

Fiber Channel's PHY layer defines transmissions over fiber and copper. Fiber is more expensive, but it shouldn't have any trouble handling gigabit data. Multimode fiber should be able to carry gigabit transmissions to 500 meters; with single-mode fiber, that distance should reach 2 kilometers.

But net managers who expect to see those sorts of speeds and distances over copper are going to be disappointed. It's a question of physics. Electromagnetic radiation increases in proportion to the speed at which data is carried and the length of the cable. In order to stay within the EMI limits set by the FCC for UTP (unshielded twisted pair), net managers must reduce data rates or shorten cabling runs (or both).

That leads to two immediate questions: How slow? How short?


Proponents of gigabit 100Base-T say they can hit 1 Gbit/s over UTP by locating the switch within 50 meters of the end-station. And that's an optimistic estimate. Some vendors think things could get even tighter. Remember, gigabit 100Base-T is a CSMA/CD scheme. As the speed of a CSMA/CD network increases, so does the likelihood of collisions. In effect, this means that the length of the cabling runs are inversely proportional to the bit rate. That's why some gigabit 100Base-T backers are saying link distances will likely be held to 25 meters on Category 5 UTP.

Even with these reduced runs, getting 1 Gbit/s over UTP is going to be a good trick. The Fibre Channel spec is silent when it comes to unshielded cabling. It does, however, call for a top speed of 100 Mbit/s over shielded twisted pair.


Packet Engine's MacLeod has an answer: "Technology has moved ahead since the Fibre Channel specs were ratified." He's hopeful that new connectors and transceivers will make it possible to push Ethernet frames at higher speeds over longer distances on Category 5.

Where does all this leave gigabit 100VG? Its backers say they can hit 500 Mbit/s over 100-meter UTP runs. But to do so, they're developing a new physical layer and transmission scheme expressly for four-pair UTP. They'll use the Fibre Channel PHY strictly for fiber. HP recently submitted a new physical layer to the 802.12 working group.

BACKBONE BONUS

Questions about copper are one of the reasons that vendors are talking up gigabit 100Base-T and 100VG as high-speed fiber links for point-to-point applications. On the backbone, the gigabit standards are being pitched as an alternative to FDDI or ATM. "As fast Ethernet picks up momentum over the next few years, we're going to need bigger pipes to servers and for interswitch connections," says Peter Tarrant, vice president of product management at Bay Networks.

What's more, since both gigabit specs field conventional Ethernet frames, they give net managers a way to build "scalable" networks. These configurations run at different speeds in different spots but use the same frame format end to end. For instance, a company could run 10-Mbit/s Ethernet to the desktop, 100-Mbit/s 100Base-T between departments, and gigabit 100Base-T on the backbone--all without having to install pricey, performance-impairing equipment that converts between one format and another. This one-frame-fits-all approach isn't possible with FDDI or Fibre Channel.

THE ATM ARGUMENT

Backers of the new gigabit LANs also argue that their standards are out ahead of ATM when it comes to price and performance. They point out that a 1-Gbit/s fiber connection costs $3,000 to $4,000 per switch port (including adapter). A 622-Mbit/s ATM connection, in contrast, today costs $15,000 per port (not counting the adapter).

But this is a good case of the figures not telling the entire story. Only one LAN vendor offers 622-Mbit/s ATM. Prices are sure to fall as competition works its market magic.

"By the time gigabit Ethernet arrives many of the ATM issues will be resolved," says Esmeralda Silva, LAN analyst with International Data Corp. (Framingham, Mass.). "ATM prices are already down. I see them dropping even more over the next few quarters, with bigger savings in 1997."

And ATM has a very strong argument in its favor--advanced multimedia capabilities. Gigabit 100Base-T does nothing to guarantee the delivery of time-sensitive traffic. Gigabit 100VG lets net managers assign two priority levels to traffic. But if the network gets busy there's still a good chance that voice and video won't arrive on time.

HOLD THE LINE

Given all of the outstanding issues, it's not surprising that some net managers are taking a wait-and-see attitude. "With gigabit 100Base-T, it's going to be very interesting to see what the actual performance is," comments John Scoggin, supervisor of network operation with Delmarva Power and Light (Newark, Del.). He's been testing gigabit Fiber Channel but notes, "We're not getting anywhere close to a gigabit over it."

Scoggin knows that it's easy to play a waiting game when he doesn't need the throughput. But net managers won't be able to put off their decisions forever. "Right now I can't imagine what I'd need gigabit Ethernet for. But 10 years ago I couldn't imagine networks running faster than T1."


TOP OF PAGE


ARC Electronics

301-924-7400 EXT 25 / 301-924-7400 EXT 25 / Fax 281-392-3657

E-Mail us at - arc@arcelect.com ... Home Page