ArtsAutosBooksBusinessEducationEntertainmentFamilyFashionFoodGamesGenderHealthHolidaysHomeHubPagesPersonal FinancePetsPoliticsReligionSportsTechnologyTravel
  • »
  • Technology»
  • Computers & Software

The History of Ethernet

Updated on January 28, 2016
The RJ45 endpoint of a twisted-pair Ethernet cable
The RJ45 endpoint of a twisted-pair Ethernet cable | Source

The early days of data networking

The Internet has its roots in the development of the ARPANET. This fledgling packet-switched network was a revolutionary break from the circuit-switched network that dominated the telecommunications industry at that time.

Computers passed these packetized messages to each other via IMPs (predecessor to today's routers); computers and IMPs connected via point-to-point links. This strategy worked well enough in the early days of computer networks when nodes were separated by long distances.

Once data networks reached a high enough density, point-to-point links were no longer the most efficient means of communication.

Intrigued by the problem of scaling up local area networks in the early 1970s, Bob Metcalfe, a researcher at Xerox's Palo Alto Research Center (PARC), invented a method for connecting computers to a common networking medium. Named for the mythical essence by which light travels through space, Metcalfe's Ethernet has grown from humble beginnings to become the dominant method of interconnecting modern networks.

The influence of ALOHAnet

Metcalfe spent a lot of time on the road, training military personnel and other clientele on the intricacies of how ARPANET worked. On one road trip, he came across a paper describing a packet-switched radio-based network from Hawaii, called ALOHAnet. Sparked by the idea of a common carrier, Metcalfe soon put together a proposal for a CSMA/CD (carrier sense, multiple access, collision detection) technology that became Ethernet.

Within a few months, the first prototype connected 100 computers over a common 1 km cable. The initial demonstration reached a data transmission bandwidth of an impressive 2.94 megabits per second, more than 50 times faster than the best serial ARPANET interconnect.

A single collision domain can only scale so far

The CSMA/CD scheme allowed many-to-many communication over a common medium, or "collision domain." Each participant on the LAN segment received every transmission. To distinguish which packet belonged to whom, the preamble of the Ethernet packet contained a unique identifier for both the sender and the receiver. By convention, some addresses were set aside to be "broadcast" or "multicast" to signify more than one recipient.

As illustrated in the flowchart below, network participants would "sense" the carrier before attempting to transmit. Due to signal propagation, however, collisions still occurred, and were "detected" when the signal amplitude rose above the threshold of a single transmission.

To prevent collision-related data corruption, participants would stop sending their original packets, send a jamming signal over the carrier for the duration of a minimum packet size, then wait a random time before attempting to send the packet again. Upon receiving a jamming signal, all participants drop the current packet being processed and wait for a new transmission.

Shared media Ethernet (10Base5, 10Base2) was a leap forward over point-to-point links for local area networking. However, as LAN sizes continued to scale higher, the CSMA/CD signaling protocol could not keep up with too many participants in a single collision domain.

Because all participants received all transmissions sent over the shared medium, each transmission had to be separately handled by each participant, regardless of who the addressee was. Although packets had destination addresses in the Ethernet header, participants never knew which packet was addressed to whom until after the packet was read and processed. The efficiency of a single collision domain scaled up poorly as network population grew.

Flowchart of CSMA/CD algorithm


Duplex: half or full?

The signaling capacity of a communication system can be classified by the direction of signaling between parties.

Broadcast media (radio-based communications, such as television, FM, AM, satellite radio) are considered simplex: one sender, many receivers, communication only flows in one direction.

Point to point communications between HAM operators, CB radios, or other similar systems are considered duplex: many senders, many receivers, communication flows both directions.

If you are familiar with radio communications, you see a caveat coming up: radio communication can only flow in one direction at a time, due to the common carrier. Similarly, shared Ethernet technologies can only send or receive in a single direction at a time. This limitation is designated as half-duplex.

Full-duplex communication means that there is no common collision domain between sending channel and receiving channel. Both ends can send simultaneously. While rare in radio technology, this could be accomplished by designating separate channels (frequency bands) for transmission direction.

For Ethernet, full-duplex is possible in twisted pair and fiber. In twisted pair, one or two of the four pair send in one direction, while the other pair(s) send the opposite direction. Fiber has two strands, each sending in the opposite direction.

Break up the collision domain

The next major development in Ethernet came in 1990 in the form of network switches. These active electronic components breathed new life into LANs by separating the collision domain from the broadcast domain.

LAN participants no longer connected directly together, but interconnected via network switches. Switches streamlined communications by reducing the number of Ethernet packets sent to each participant. Instead of sending all packets to all members, switches sent packets only out the segment of the designated recipient.

The added complexity of multiple collision domains brought up an entire new set of problems: loops, spanning trees, VLANs, bonded links, and more. The IEEE introduced several new standards to address these problem spaces, highlighted in the table below.

Ethernet standards by IEEE code

Ethernet problem space
IEEE Standard
Initial Release
MAC bridges, spanning tree, loop detection and avoidance
VLANs: more than one broadcast domain per switch, how to signal these between switches
Link aggregation: treat multiple physical links between switches as one logical link (a. k. a., LACP, LAG)
Highlights of IEEE standards developed in the rapid expansion of Ethernet technologies

How a broadcast domain is different from a collision domain

Each LAN segment was said to be a "broadcast domain." The collision domain refers to the physical medium (typically copper) that is shared between network participants. The broadcast domain refers to the logical separation between LANs. In IP networks, a LAN typically has the same logical boundaries as a subnet.

A broadcast packet is meant to be repeated out from the sender to every other participant within the LAN. Many networked services use broadcast to discover what resources are available (such as ARP and DHCP).

To prevent scaling issues with large networks, LANs are typically kept to communities of less than 1,000 participants. There is no set rule for how many participants can be in a single LAN, but performance will decrease once enough participants begin broadcasting to discover each other.

The steady progression of Ethernet technologies

Let's rewind back to 1979. Flush with the success of Ethernet, Metcalfe left Xerox that year to found his own company, 3Com. Partnered with Digital, Intel, and Xerox (the DIX consortium), his pioneering work led the industry to form the IEEE committee for standardizing local and metropolitan area networks. By 1983 the now-famous IEEE 802.3 standard was adopted.

The shared medium coax-based Ethernet technologies (the various 10Base-X standards) reached a bandwidth of 10 megabits per second (Mbps) or 1.25 megabytes per second (MBps) but were limited in distance to 100m. By 1987, fiber optics extended LANs to 2000m (10BaseFL).

Improved electronics and signaling led to 100Mbps transfer rates for Fast Ethernet (100Base-X) by 1995. In 1997, the introduction of full-duplex communication meant that data could pass in both directions simultaneously, effectively doubling throughput.

Gigabit (1000Mbps) transfers over fiber arrived in 1998, followed by twisted-pair in 1999. Four years later, 10Gbps fiber hit the scene in 2003, and appeared in twisted-pair form by 2006. As of this writing, 40G and 100G are already in production, with 400G specifications still in draft.

Ethernet over Fiber


Milestones in the life of Ethernet

Medium Type
Max Throughput
Initial Ethernet prototype
Thick Coax
2.94 Mbps
Ethernet II standard (DIX 2.0)
Thick Coax
10 Mbps
Official IEEE 802.3 standard (includes some changes to header fields)
Thick Coax
10 Mbps
thin net (IEEE 802.3a)
Thin Coax
10 Mbps
IEEE 802.3d introduces fiber for repeaters
Fiber Optics
10 Mbps
IEEE 802.3i introduces twisted pair
Twisted Pair
10 Mbps
IEEE 802.3j brings fiber to the end station
Fiber Optics
10 Mbps
IEEE 802.3u introduces Fast Ethernet and auto-negotiation
Twisted Pair, Fiber Optics
100 Mbps
IEEE 802.3z describes Gigabit over fiber
Fiber Optics
1 Gbps
IEEE 802.3ab brings Gbps to twisted pair
Twisted Pair
1 Gbps
IEEE 802.3ae is 10G over fiber
Fiber Optics
10 Gbps
IEEE is 40G and 100G over fiber
Fiber Optics
40 Gbps, 100 Gbps
Synopsis of IEEE 802.3 timeline from Wikipedia

Other forms of Ethernet

First released in 1997, the IEEE 802.11 standard describes Ethernet over radio, or Wi-Fi. Similar to the wired technology, speeds and throughput have steadily increased over time. The initial release had effective throughput of less than 1 Mbps, while current/proposed bandwidth exceeds 800Mbps. Because transmission occurs over a shared medium, communication is only half-duplex (reminiscent of the earlier Ethernet standards).

Ethernet has grown to form the backbone of much of the Internet. Research networks (such as Internet2) have utilized long haul fiber networks, combined with WDM, to span the continental US. Commercial Internet providers interconnect with peers using Ethernet at various exchanges points around the world.


    0 of 8192 characters used
    Post Comment

    No comments yet.