from Wikipedia, the free encyclopedia
Ethernet in the TCP / IP protocol stack :
application HTTP IMAP SMTP DNS ...
transport TCP UDP
Internet IP ( IPv4 , IPv6 )
Network access Ethernet
Ethernet in the AppleTalk protocol stack (EtherTalk)
application AFP ADSP
transport ATP
Internet DDP
Network access ELAP AARP

Ethernet ([ ˈeːtɐˌnɛt ] or English [ ˈiːθərˌnɛt ]) is a technology that specifies software (protocols, etc.) and hardware (cables, distributors, network cards, etc.) for wired data networks, which was originally intended for local data networks ( LANs ) and therefore also known as LAN technology. It enables data to be exchanged in the form of data frames between the devices (computers, printers and the like) connected to a local area network (LAN). Currently, transmission rates of 1, 10, 100 megabit / s (Fast Ethernet), 1000 Mb / s (Gigabit Ethernet), 2.5, 5, 10, 40, 50, 100, 200 and 400 Gigabit / s specified. In its original form, the LAN only extends over one building; Ethernet variants via fiber optics have a range of up to 70 km.

The Ethernet protocols include specifications for cable types and connectors as well as for forms of transmission (signals on the physical layer, packet formats). In the OSI model , Ethernet defines both the physical layer (OSI Layer 1) and the data link layer (OSI Layer 2). Ethernet largely corresponds to the IEEE standard 802.3 . It became the most widely used LAN technology from the 1990s and has replaced other LAN standards such as token ring or, as in the case of ARCNET in industrial and manufacturing networks or FDDI in highly available networks, made it into niche products for specialty areas. Ethernet can be the basis for network protocols such as B. AppleTalk , DECnet , IPX / SPX and TCP / IP .

Real-time Ethernet is used for applications in which high demands are placed on the reliability of communication .


Ethernet was originally developed at the Xerox Palo Alto Research Center (PARC). Metcalfe says he first outlined Ethernet in 1973 in a memo to his superiors about the potential of Ethernet. He derived the protocol from the radio-based ALOHAnet developed at the University of Hawaii . Hence the name Ether -net (English for “ ether ”, which according to historical assumptions would be the medium for the propagation of (electromagnetic) waves). Metcalfe himself says that Ethernet was developed over several years and therefore no starting point can be determined.

So it was originally a company-specific and non-standardized product. This first version of the Ethernet still worked at 3 Mbit / s. In 1976, Metcalfe and his assistant David Boggs published an article entitled Ethernet: Distributed Packet Switching For Local Computer Networks.

Robert Metcalfe left Xerox in 1979 to promote the use of personal computers and LANs , and founded 3Com . He convinced DEC , Intel, and Xerox to work with him to make Ethernet the standard. Its first Ethernet version 1 was further developed in 1980 by the IEEE ( Institute of Electrical and Electronics Engineers ) in working group 802. Originally only a LAN standard for transmission rates between 1 and 20 Mbit / s was planned. Also in 1980 a so-called “token access method” was added. From 1981 the IEEE pursued three different technologies: CSMA / CD (802.3), Token Bus (802.4) and Token Ring (802.5), of which the last two were soon drowned in a veritable flood of Ethernet products. 3Com became a big company in the process.

Work on the Cheapernet standard (10BASE2) was published in June 1983. At the same time, work began on the specifications for Ethernet-on-Broadband ( 10BROAD36 ) and for the StarLAN (1BASE5). When the Ethernet standard was also published as the international ISO / DIS 8802/3 standard in 1985, it was quickly supported by over 100 manufacturers. In 1986, some smaller companies began to transmit data in Ethernet format on four-wire lines from the telephone sector (CAT-3). The IEEE then stepped up its activities in the areas of Ethernet-on- Twisted Pair , which became the standard for 10BASE-T in 1991, and Ethernet on fiber optic cables , which in 1992 led to the 10BASE-F standards (F for fiber optics). In the mid-1990s, there was a tug-of-war over the successor standard; On the one hand there were AT&T and HP , who were striving for a technically more elegant solution according to IEEE 802.12 (100BASE-VG), on the other hand there were the manufacturers of the Fast Ethernet Alliance , consisting of around 35 well-known companies such as Bay Networks , 3Com , Intel , SUN , Novell etc., which propagated 100 Mbit / s according to the tried and tested IEEE 802.3 standard.

Ultimately, the 100 Mbit / s standard for Ethernet was adopted in 1995 at the end of the Fast Ethernet Alliance in accordance with IEEE 802.3u, roughly at the same time as the standard for a wireless LAN called 802.11. In the meantime, work on the 10 Gbit / s Ethernet and the Ethernet in the First Mile (EFM) is already targeting university and city ​​networks instead of purely local operations .

Nowadays, in the form of Industrial Ethernet , the Ethernet cabling standard is also being used more and more in industrial production systems. Global networking and the resulting growing demands on data transmission - not only for professional, but also for private purposes - has led to high-performance networks being installed in private buildings and even cruise ships .

Robert Metcalfe was awarded the "National Medal of Technology" in 2003 for his services to the development of Ethernet.

On March 21, 2019, DE-CIX became the first Internet node worldwide to offer 400 Gbit / s Ethernet.

Physical layer

Ethernet is based on the idea that the participants in a LAN transmit messages by high frequency, but only within a common line network. Each network interface has a globally unique 48-bit key known as the MAC address . This ensures that all systems in an Ethernet have different addresses. Ethernet transmits the data on the transmission medium using the so-called baseband method and digital time division multiplex .

CSMA / CD algorithm

An algorithm called " Carrier Sense Multiple Access with Collision Detection " ( CSMA / CD ) regulates the systems' access to the shared medium. It is a further development of the ALOHAnet protocol that was used in Hawaii in the 1970s .

In practice, this algorithm works visually like a discussion round without a moderator, on which all guests use a common medium (the air) to talk to each other. Before starting to speak, politely wait for the other guest to stop talking. If two guests start speaking at the same time, they both stop and wait for a short, random period of time before trying again.

The point that wants to send data listens to the medium ( Carrier Sense ) to see whether it is already occupied and only sends when the line is free. Since two points can start sending at the same time, collisions can still occur, which are then detected ( collision detection ), whereupon both points briefly generate a "fault detected" signal pattern, then stop sending and wait a random time until you try to send again. To do this, a sender must listen to the medium while it is sending to see whether another sender collides with it. Media are therefore unsuitable for CSMA / CD if a high transmission power is required and a very weak reception signal is faced, which then "goes down".

In order for the collision to be detected and a retransmission to be initiated, the data frames must have a certain minimum length, depending on the line length - the interference signal from the second transmitter must reach the first before it has finished its data packet (and is considered to be "sent without collision"). This minimum length results from the signal speed and the transmission rate. With a transmission rate of 10 Mbit / s and a maximum distance of 2.5 km between two stations, a minimum length of 64 bytes (14 byte header, 46 byte user data, 4 byte CRC) is required. Smaller data frames must be filled accordingly. A maximum segment length of 100 m and four repeaters are allowed for a transmission rate of 10 Mbit / s (standard Ethernet). This allows two stations to be connected directly up to a distance of 500 m. With higher transmission rates and maximum segment length, the number of repeaters is reduced due to the physical dependencies. For example, with Fast Ethernet (100 Mbit / s) only two repeaters and with Gigabit Ethernet (1000 Mbit / s) one repeater are allowed. With 1 Gbit / s Ethernet (1000 Mbit / s) in (but rather hypothetical) half-duplex mode, small frames in the Ethernet packet are extended to 520 bytes in order to still allow reliable collision detection with a reasonable physical network size.

Even if the IEEE 802.3 standard has the name “CSMA / CD” in its title, collision resolution is only of minor importance today. Most networks today are operated in full duplex mode, in which participants (routers, switches, end devices, etc.) can use the send and receive directions independently of one another by means of a point-to-point connection and thus no more collisions occur. Nevertheless, the frame format, in particular the frame header and the minimum frame length prescribed for collision detection, up to 400 Gbit / s Ethernet, remained unchanged.

Broadcast and Security

In the first Ethernet implementations, all communication was handled via a common bus, which was implemented in the form of a coaxial cable . Depending on the type of cable, all workstations were connected to these either with a T-piece or "invasive plug" (also called vampire clamp, vampire branch or vampire tap ). Any information sent from one computer was received by everyone. The devices connected via Ethernet must constantly filter out information that is not intended for them.

This fact can be used to send broadcast (German: Rundruf) messages to all connected systems. In the case of TCP / IP, for example, the ARP uses such a mechanism for the resolution of the layer 2 addresses. This fact is also a security problem for Ethernet, since a participant with bad intentions can log all data traffic on the line. One possible remedy is the use of cryptography (encryption) at higher protocol levels. The confidentiality of traffic relationships (who exchanges data with whom to what extent and when?) Cannot be protected in this way.

The use of (repeater) hubs to create multi-segment Ethernet networks does not change anything here because all data packets are replicated in all segments.

In modern Ethernet networks , bridges , nowadays switches, were initially used to split the collision domains . This divides an Ethernet into segments, in each of which only a subset of end devices can be found. If only switches are used, then network-wide communication can take place in full-duplex mode , which enables simultaneous sending and receiving for each end device. Data packets are usually sent directly from the sender to the recipient via switches - the packet is not delivered to uninvolved participants. Broadcast (German: Rundruf-) and multicast messages, however, are sent to all connected systems.

This makes spying on and eavesdropping more difficult, but the lack of security is only reduced and not remedied by setting up a “switched” environment. In addition to the broadcast messages, the first packets are also sent to all connected systems after a transmission pause - when the switch does not (yet) know the destination MAC address. This condition can also be brought about maliciously by MAC flooding . Packets can also be maliciously redirected through MAC spoofing .

The security of operations in terms of the trouble-free availability of data and services is based on the good behavior of all connected systems. Intentional or accidental misuse must be detected in an Ethernet environment by analyzing the data traffic ( LAN analysis ). Switches often provide statistical information and messages that allow malfunctions to be identified at an early stage or give rise to a more detailed analysis.



Ethernet in its early forms (e.g. 10BASE5, 10BASE2), with a cable used jointly by several devices as a transmission medium ( collision domain / shared medium - in contrast to the later switched Ethernet), works well as long as the traffic volume is relative to the nominal Bandwidth is low. Since the chance for collisions is proportional to the number of transmitters (English " transmitter ") and increases to be transmitted amount of data occurs above a 50% duty cycle increases a as Congestion known (jam) phenomenon, whereby capacity overloads occur and thus a good efficiency the transmission performance within the network is prevented.

In order to solve this problem and to maximize the available transmission capacity, switches were developed (sometimes also called switching hubs , bridging hubs or MAC bridges ), one also speaks of switched Ethernet . Switches store packets / frames temporarily and thus limit the range of the collisions (the collision domain ) to the devices connected to the corresponding switch port. With twisted pair or fiber optic cabling, connections between two devices ( Link ) can also be operated in full duplex mode ( FDX ) if both devices support this (this is the rule).

If (all) hubs / repeaters are removed from a network and replaced by full-duplex-capable components, one speaks of a (pure) switched Ethernet , in which there are no more half-duplex links and therefore no more collisions. The use of switches enables collision-free communication in FDX mode, i.e. That is, data can be sent and received at the same time without collisions. Despite collision-free processing, however, packet losses can occur, for example if two senders each use the bandwidth to send data packets to a common receiver. The switch can buffer packets for a short time, but if the recipient does not have double the bandwidth or if the data flow cannot be slowed down, it has to discard data if the buffer overflows so that it cannot be delivered.

Ethernet flow control

Ethernet flow control is a mechanism that temporarily stops data transmission on Ethernet. In CSMA / CD networks, this special signaling could be dispensed with, because here the signaling of a collision is practically the same as a stop or pause signal ( back pressure ).

Since Fast Ethernet and the introduction of switches , data transmission has practically only taken place collision-free in full duplex mode. Since this means that CSMA / CD is not used, an additional flow control is required, which enables a station, for example in the event of overload, to give a signal that it does not want to have any further packets sent at the moment - unlike with CSMA / CD, there is no way to to indicate a loss and thus the need to re-broadcast. Flow Control was introduced for this purpose. This allows a station to request the remote stations to take a pause in transmission, thus avoiding packets (at least in part) having to be discarded. To do this, the station sends a PAUSE packet with a desired waiting time to another station (a MAC address) or to all stations (broadcast). The pause is 0 to 65535 units; one unit corresponds to the time required for the transmission of 512 bits.

Ethernet Flow Control improves the reliability of the delivery - since the requested breaks act directly on the sending node, however, there may be a loss of performance. If, for example, a destination node can only accept the data to be received more slowly than the transmission rate and therefore sends pause frames, it brakes the sending node as a whole, and it also supplies other destination nodes with data more slowly than would actually be possible ( head-of-line blocking ).

Flow control is optional and is often not used to avoid head-of-line blocking. Most networks use protocols for important data in the higher network layers that can compensate for slight transmission losses, in particular the Transmission Control Protocol . If this is not possible, the network architecture or other mechanisms must ensure that important packets cannot be lost, for example with Quality of Service or with Fiber Channel over Ethernet .

Formats of the Ethernet frames and the type field

Historical formats

There are four types of Ethernet data blocks (English ethernet frames ):

  • Ethernet version I (no longer used, definition 1980 by consortium DEC , Intel and Xerox )
  • The Ethernet version 2 or Ethernet II data block (English ethernet II frame ), the so-called DIX frame (definition 1982 by the consortium DEC , Intel and Xerox ).

The IEEE 802.3 standard has been in existence since 1983. Ethernet is practically a synonym for this standard. IEEE 802.3 defines two frame formats:

  • IEEE 802.3 3.1.a Basic MAC frame
  • IEEE 802.3 3.1.b Tagged MAC frame

The original Xerox Version 1 Ethernet data block had a 16-bit field in which the length of the data block was stored. Since this length is not important for the transmission of the frames, it was used as an Ethertype field by the later Ethernet II standard. The format of Ethernet I with the length field is now part of the 802.3 standard.

The Ethernet II format uses bytes 13 and 14 in the frame as the Ethertype. There is no length field as in the Ethernet I-Frame. The length of a frame is not transmitted by a numerical value, but by the bit-precise signaling of the end of transmission. As with Ethernet I, the length of the data field is limited to 1500 bytes. The Ethernet II format is now also part of the 802.3 standard, only the Ethernet types with numerical values ​​less than 1500 have been omitted, because the numerical values ​​less than or equal to 1500 in this field are now interpreted as length and are checked against the actual length.

IEEE 802.3 defines the 16-bit field after the MAC addresses as a type / length field. With the convention that values ​​between 0 and 1500 indicated the original Ethernet format and higher values indicate the EtherType , the coexistence of the standards on the same physical medium was made possible. The permitted values ​​for Ethertype are administered by IEEE. This administration is limited to the assignment of new Ethertype values. When reassigning, IEEE takes into account Ethertype values ​​already assigned for Ethernet II, but does not document them. So it happens that, for example, the value 0x0800 for IP data is missing in the IEEE documentation of the Ethertype values. Ethertype describes the format or the protocol for interpreting the data block. The LLC field and a possible SNAP field are already part of the MAC frame data field. In the tagged MAC frame, four bytes with the QTAG prefix are inserted after the source MAC address. This field is defined by the 802.1Q standard and enables up to 4096 virtual local area networks (VLANs) on a physical medium. The total permitted length of the Mac frame is extended to 1522 bytes, the length of the data field remains limited to 1500 bytes.

IEEE 802.3 Tagged MAC Frame

Data frame

The Ethernet data block format Ethernet-II according to IEEE 802.3 (with 802.1Q VLAN tag), which is almost exclusively used today


Ethernet transmits the data serially, starting with the lowest, least significant bit (the "ones digit") of a byte. This means, for example, that byte 0xD5 is traveling as a bit sequence (left to right) “10101011”. The bytes of the wider fields are transmitted as BigEndians, i.e. H. with the byte with the higher significance first. For example, the MAC address in the picture 0x0040F6112233 is transmitted in this order as "00 40 F6 11 22 33". Since the first bit of a frame is the multicast bit, multicast addresses have a first byte with an odd number, e.g. B. 01-1B-19-00-00-00 for IEEE 1588.

One difference concerns the FCS ( Frame Check Sequence , CRC): Since all transmitted bits are shifted from the LSB to the MSB by the CRC generator , the most significant bit of the most significant byte of the CRC must be transmitted first. A calculated CRC value of 0x8242C222 is thus appended as "41 42 43 44" to the transmitted data bytes as an FCS checksum for transmission.

In contrast to the Ethernet frame, with some other types of LAN (for example Token Ring or FDDI ) the most significant bit of a byte is in the first position in a frame. This means that when bridging between an Ethernet LAN and another type of LAN, the order of the bits of each byte of the MAC addresses must be reversed.

The preamble and SFD

The preamble consists of a seven-byte long, alternating bit sequence “101010… 1010”, followed by the Start Frame Delimiter (SFD) with the bit sequence “10101011”. This sequence was once used for bit synchronization of network devices. It was necessary for all those device connections that could not maintain bit synchronization by transmitting a continuous carrier wave even during periods of rest, but had to re-establish it with each frame sent. The alternating bit pattern allowed each receiver to correctly synchronize with the bit spacing. Since a certain part of the preamble is lost when forwarding via repeaters (hubs), it was chosen large enough in the specification that a minimal settling phase remains for the recipient when the network is maximally expanded.

The bus network architectures that rely on such transient processes are rarely used today, which means that the preamble, as well as the access pattern CSMA / CD, the minimum and maximum frame length and the minimum packet spacing ( IFG , also IPG) only make up Compatibility reasons are in the specification. Strictly speaking, the preamble and SFD are packet elements that should be defined on a level below the frame and thus also the MAC so that their use would depend on the specific physical medium. Modern wired network architectures are star-shaped or ring-shaped and use permanently oscillating (synchronous) point-to-point connections between end users and network distributors ( bridges or switches), which signal packet boundaries in a different form and therefore actually make preamble and SFD unnecessary. On the other hand, IFGs and minimum frame lengths for network distributors also result in certain maximum packet rates to be processed, which simplifies their design.

Destination and source MAC address

The destination address identifies the network station that is to receive the data. This address can also be a multicast or broadcast address. The source address identifies the sender. Each MAC address in the two fields is six bytes or 48 bits long.

Two bits of the MAC address are used to classify them. The first transmitted bit and thus bit 0 of the first byte decide whether it is a unicast (0) or broadcast / multicast address (1). The second transmitted bit and thus bit 1 of the first byte decides whether the remaining 46 bits of the MAC address are administered globally (0) or locally (1). Purchased network cards have a globally unique MAC address that is managed globally by a consortium and the manufacturer. However, you can choose individual MAC addresses at any time and assign them to most network cards via the driver configuration, in which you select the value (1) for bit 1 and manage the remaining 46 bits locally according to the specifications and keep them clearly in the broadcast domain.

MAC addresses are traditionally represented as a sequence of six two-digit hex numbers separated by colons, e.g. B. as "08: 00: 01: EA: DE: 21", which corresponds to the transmission sequence on the medium. The individual bytes are sent starting with the LSB.

VLAN tag

In the tagged MAC frame according to IEEE 802.1Q, there are also four bytes as a VLAN tag . The first two bytes contain the constant 0x8100 (= 802.1qTagType), which identify a tagged MAC frame as such. In terms of position , the Ethertype field would appear here in the Basic MAC frame . The value 0x8100 can thus also be viewed as an Ethertype for VLAN data, but the actual Ethertype follows after the tag (see below). The next two bytes ( TCI Tag Control Information) contain three bits for the priority ( Class of Service , 0 lowest, 7 highest priority), one bit Canonical Format Indicator (CFI), which ensures the compatibility between Ethernet and Token Ring (This 1-bit data field indicates whether the MAC address is in a recognized or unrecognized format. If the bit set is a 0, then it is not legal, if it is 1, it is legal. For Ethernet switches it is always 0. If an Ethernet port receives a 1 as CFI information, the Ethernet switch does not connect the tagging frame to a non-tagged port.), and 12 bits for the VLAN ID . This VLAN tag is followed by the type field (EtherType) of the actual frame, originally at the position of the VLAN tag, with a value other than 0x8100 (e.g. 0x0800 for an IPv4 packet in the picture).

The VLAN tag is transmitted as a sequence of two bytes "81 00". The 16 bits of the TCI are sent in the same way Big-Endian with the higher byte first.

The type field (EtherType)

The type field provides information about the protocol used for the next higher layer within the user data. The values ​​are greater than 0x0600 (otherwise this is an Ethernet I-frame with a length field in this position). The special value 0x8100 for identifying a VLAN tag is reserved in the value set of Type . If a VLAN tag is available, the subsequent type field must not be 0x8100.

Values ​​in the type field ( EtherType ) for some important protocols:

Type field protocol
0x0800 IP Internet Protocol, Version 4 ( IPv4 )
0x0806 Address Resolution Protocol ( ARP )
0x0842 Wake on LAN ( WoL )
0x8035 Reverse Address Resolution Protocol ( RARP )
0x809B AppleTalk (EtherTalk)
0x80F3 Appletalk Address Resolution Protocol ( AARP )
0x8100 VLAN Tag ( VLAN )
0x8137 Novell IPX (old)
0x8138 Novell
0x86DD IP Internet Protocol, Version 6 ( IPv6 )
0x8847 MPLS unicast
0x8848 MPLS multicast
0x8863 PPPoE Discovery
0x8864 PPPoE session
0x8870 Jumbo Frames (obsolete)
0x888E 802.1X Port Access Entity
0x8892 Real-time Ethernet PROFINET
0x88A2 ATA over Ethernet Coraid AoE
0x88A4 Real-time Ethernet EtherCAT
0x88A8 Provider bridging
0x88AB Real-time Ethernet Ethernet POWERLINK
0x88B8 IEC61850 GOOSE
0x88CC Link Layer Discovery Protocol LLDP
0x88CD Real-time Ethernet Sercos III
0x88E1 HomePlug AV
0x88E5 MACsec
0x8906 Fiber Channel over Ethernet
0x8914 FCoE Initialization Protocol (FIP)
0x8947 GeoNetworking protocol

In Ethernet 802.3 frames, the length of the data content can be specified in the DATA part (length field) instead of the type field for compatibility with Ethernet I. Since the data field must not be longer than 1500 bytes in any Ethernet frame , the values ​​1536 (0x0600) and above can be used as protocol types ( Ethertype ). The use of the values ​​1501 to 1535 is not specified. The use as length has practically completely disappeared - to signal the end of a frame, all Ethernet variants either use a special control symbol (100 Mbit / s upwards) or end the carrier cycle (10 Mbit / s).

The type field is interpreted as a big-endian byte sequence and sent with the most significant byte in front.


A maximum of 1500 bytes of user data can be transferred per data block. The user data are interpreted by the protocol specified under Type. So-called jumbo frames , super jumbo frames and jumbograms also allow larger data blocks, but these special modes are officially separate from Ethernet or IEEE 802.3.

The data bytes are sent in ascending byte order.

PAD field

The PAD field is used to bring the Ethernet frame to the required minimum size of 64 bytes. This is important with old transmission methods in order to reliably detect collisions in the so-called collision domain. The preamble and SFD (8 bytes) are not included in the required minimum frame length, but a VLAN tag is. A PAD field is required if less than 46 or 42 bytes (with or without an 802.1Q VLAN tag) are to be transmitted as user data. The protocol specified in Type must ensure that these bytes added as pads (also called "padding bytes") are not interpreted, for which it usually has its own user data length specification.

FCS (Frame Check Sequence)

The FCS field represents a 32-bit CRC checksum . The FCS is calculated over the actual frame, ie starting with the destination MAC address and ending with the PAD field. The preamble, the SFD and the FCS themselves are not included in the FCS. When a packet is created by the sender, a CRC calculation is carried out using the bit sequence and the checksum is appended to the data block. The receiver performs the same calculation after receiving it. If the received checksum does not match the self-calculated checksum, the recipient assumes that the transmission was faulty and the data block is discarded. To calculate the CRC-32 checksum, the first 32 bits of the MAC address are inverted and the result of the checksum calculation is also inverted (avoidance of the zero problem ).

In common CRC implementations as feedback shift registers, data bits are sent in the transferred order, i.e. from the LSB to the MSB, through a shift register, which is, however, loaded from the LSB itself. In the shift direction, the MSB of the CRC is therefore available first and, in contrast to all other data, is first on the line. If the data stream is now written to the shift register at the receiver including the received CRC value, the CRC will contain the value zero if there are no errors. A value other than zero indicates a transmission error.

By inverting the first 32 bits and the CRC sum, the result is no longer zero. If no transmission error has occurred, the shift register always contains the same number, also known as the magic number. With Ethernet it is 0xC704DD7B.

Order of bits and bytes

With Ethernet, bytes ( octets ) are always transmitted with the least significant bit first (with the exception of the frame check sequence). However, many faster variants do not transmit individual bits, but rather multi-bit symbols or entire octets in one step. Fields that consist of several bytes are always transmitted with the most significant octet first.

Conversion into a data stream

After the data stream has been provided as a sequence of bytes, one or more bits are encoded in a line code, depending on the physical medium and the transmission rate , on the one hand to take into account the physical properties of the medium and on the other hand to enable the receiver to recover the clock . Depending on the code, the permitted frequency bandwidth is limited downwards (freedom from DC voltage) and upwards.

In non-transmission times, that is between two frames, there are, by definition, pauses (“inter-frame spacing”) with a certain minimum length. In the physical half-duplex mode, the transmitter switches itself off during this time to allow other stations to access the shared medium. With more modern media types with physical full duplex mode, a carrier oscillation is maintained, which enables the receiver to synchronize faster with the data stream. In addition, out-of-band information can be exchanged between the stations when there is no transmission.

With some physical full duplex media types such as 10BASE-T, the sending station deactivates itself between frames despite exclusive access to the medium. The transmission-free time is used here for out-of-band signaling (link pulses, fast link pulses) of the link parameters.

Ethernet media types

The different Ethernet variants ( PHY s) differ in transmission rate, the cable types used and the line coding . The protocol stack works identically for most of the following types.

A successful connection between two connections ( ports ) is called a link . Some variants divide the data stream into several channels ( lanes ) in order to adapt the data rate and frequencies to the medium. The respective range is the maximum possible length of a link within the specification. With a higher quality of the medium - especially with fiber optics - significantly longer links can also function stably.

The variants get their names from the specifications used:

  • 10, 100, 1000, 10G, ... - the nominal, usable speed on the bit level (no suffix = megabit / s, G = gigabit / s); the line- coded sublayers usually have a higher data rate
  • BASE, BROAD, PASS - baseband , broadband , or passband signaling
  • -T, -S, -L, -C, -K, ... - Medium: T = twisted pair cable , S = ( short ) short wavelength approx. 850  nm over multimode fiber , L = ( long ) long wavelength approx. 1300 nm, mainly singlemode fiber , E / Z = extra long wavelength approx . 1500 nm (singlemode), B = bidirectional fiber with WDM (mostly singlemode), P = passive optical network , C = ( copper ) twinaxial cable , K = backplane , 2/5 = coaxial cable with 185/500 m range
  • X, R - PCS coding (depending on the generation), for example X for 8b / 10b block coding ( 4B5B for Fast Ethernet), R for large blocks ( 64b / 66b )
  • 1, 2, 4, 10 - number of lanes per link or range for 100/1000 Mbit / s WAN PHYs

With 10 Mbit / s Ethernet, all variants use Manchester code throughout , no coding is specified. Most twisted pair variants use special coding, only -T is specified.

The following sections provide a brief overview of all of the official Ethernet media types. In addition to these official standards, many manufacturers have developed proprietary media types, often to achieve greater ranges with fiber optic cables.

Some early variants of Ethernet

  • Xerox Ethernet (Alto Aloha System) - The name arose from testing the concept on Alto computers . Xerox Ethernet is the original Ethernet implementation that had two versions during its development. The version 2 data block format is currently mainly used.
  • 10Broad36 (IEEE 802.3 Clause 11) - Obsolete. An early standard that supported Ethernet over longer distances. It used broadband modulation techniques similar to cable modems and worked with coaxial cables .
  • StarLAN , standardized as 1BASE5 (IEEE 802.3 Clause 12) - the first Ethernet implementation using twisted pair cables , developed by AT&T . 1 Mbit / s via the already widespread (mostly) Cat 3 cabling with a link range of 250 to 500 m. A commercial failure, but one that provided the technical basis for 10BASE-T.

10 Mbit / s Ethernet

With 10 Mbit / s Ethernet, a simple Manchester coding is used, which transmits two line bits per data bit (thus 20 Mbaud). With this doubling of the signaling rate and thereby alternately transmitted data bits, the DC voltage is effectively suppressed and at the same time the clock recovery is tracked in the receiver, the spectrum extends up to 10 MHz. The line is only seized when an Ethernet packet is actually sent.

10 Mbit / s Ethernet with coaxial cable

T-pieces and terminating resistors for 10BASE2
EAD cable for 10BASE2
10BASE2 , IEEE 802.3 Clause 10 (formerly IEEE 802.3a)
(also known as Thin Wire Ethernet , Thinnet or Cheapernet ) - A coaxial cable (RG58) with a wave impedance of 50 ohms connects the participants with each other, each participant uses a BNC T-piece to connect their network card. Terminating resistors attached to both ends of the line ensure reflection-free signal transmission. A segment (that is, all coaxial cable pieces connected to one another by the BNC T-pieces) may be a maximum of 185 meters long and supply a maximum of 30 participants. Any two participants on the bus must maintain a distance of at least 0.5 meters from one another. In contrast to 10BASE5, which also uses coaxial cable, the transceivers are integrated in the NIC (Network Interface Card) and must be connected directly to the T-piece (without any additional coaxial cable). Additional network segments can be connected via repeaters so that the maximum expansion of the network includes 5 network segments in a chain. The number of segments can be further increased with structured cabling. A maximum total spread of 925 m in diameter can thus be achieved. There were also Ethernet jacks (EAD) is used. With 10BASE2, the entire network segment fails if the cable or a plug connection, especially the terminating resistor, is defective at one point. Manually assembled coaxial cables are particularly susceptible if the BNC connector is not correctly attached.
Thick Ethernet transceiver
10BASE5 , IEEE 802.3 Clause 8
(also Thicknet or Yellow Cable ) - an early IEEE standard that uses a 10 mm thick coaxial cable (RG8) with a wave impedance of 50 ohms. To connect devices, a hole must be drilled in the cable at a marked point using a drilling template, through which a contact of a special terminal ( vampire terminal ) of the transceiver is inserted and clamped. The network card of the computer is connected to this transceiver by means of the AUI interface and a connection cable. This standard offers 10 Mbit / s data rate for transmission in the baseband and supports a maximum of 500 m cable length and 100 participants on each segment. Like 10BASE2, the line has no branches, and there are 50-ohm terminating resistors at the ends. As with 10BASE2, the network can be extended to a max. Length of 2.5 km. This type is actually obsolete, but is still in use in some systems due to its widespread use in the early days.

10 Mbit / s Ethernet with twisted pair cable

8P8C modular plug and socket (socket is on the right)
  • StarLAN 10 - developed from 1BASE5, 10 Mbit / s, almost identical to 10BASE-T
  • 10BASE-T , IEEE 802.3i Clause 14 - uses four wires (two twisted pairs) of a CAT-3 or CAT-5 cable (cabling according to TIA-568A / B ). A hub or switch is located in the middle and each participant is connected via a dedicated port . The transmission rate is 10 Mbit / s and the maximum length of a segment is 100 m. Physically, the plug connections are designed as 8P8C modular plugs and sockets, which are usually referred to as "RJ-45" ​​or "RJ45" plugs / sockets. Since 1: 1 cables are normally used, the connectors of the computer ( MDI ) and uplink (hub, switch, MDI-X ) are assigned the same number. The following assignment applies to the computer: Pin1 - Transmit +; Pin2 - Transmit−; Pin3 - Receive +; Pin6 - Receive−.
  • 10BASE-T1L , IEEE P802.3cg - is transmitted via the two-wire cables according to IEC 61158-2 cable type A, which are common in process automation. The transmission takes place with a data transmission rate of 10 Mbit / s, is coded 4B3T and modulated as PAM-3 and transmitted with 7.5 M baud full duplex . The participants can be supplied with up to 60 W power via the same cable. The structure consists of a " trunk " cable with a maximum of 1000 m between the field switches and the "track" cable with a maximum of 200 m between a field switch and a field device.

10 Mbit / s Ethernet with fiber optic cable

  • FOIRL - Fiber-optic inter-repeater link. The original standard for Ethernet over fiber optic cable.
  • 10BASE-F , IEEE 802.3j (IEEE 802.3 Clause 15) - General term for the new family of 10 Mbit / s Ethernet standards: 10BASE-FL, 10BASE-FB and 10BASE-FP. The only one more widely used is 10BASE-FL.
  • 10BASE-FL (IEEE 802.3 Clause 18) - A revised version of the FOIRL standard.
  • 10BASE-FB (IEEE 802.3 Clause 17) - Intended for backbones that connect multiple hubs or switches. Is now technically obsolete.
  • 10BASE-FP (IEEE 802.3 Clause 16) - A passive star-shaped network that does not need a repeater. There are no implementations.

100 Mbit / s Ethernet

With the transition from 10 to 100 Mbit / s Ethernet ( Fast Ethernet ), the signaling layer was further subdivided in order to come to a clearer definition of what separates the PHY (the physical layer, OSI layer 1) from the MAC. Was there PLS (physical layer signaling, Manchester coding, identical for all 10 Mbit / s standards) and PMA (physical medium attachment, coaxial, twisted pair and optical connections) for 10 Mbit / s Ethernet , Fast Ethernet now uses PCS (Physical Coding Sublayer) with PMA and PMD (Physical Medium Dependent). PCS, PMA and PMD together form the physical layer. Three different PCS-PMA combinations were designed, of which those for 100BASE-T4 and 100BASE-T2 (IEEE 802.3 Clauses 23 and 32) could not acquire any economic significance.

Only 100BASE-TX (IEEE 802.3 Clause 24) for twisted pair cables has prevailed for copper cables, which, like the fiber optic variants, uses the more efficient 4B5B code instead of Manchester coding . This is not free of DC voltage, but enables clock recovery from the signal and the symbol rate of 125 Mbaud is only slightly higher than the data rate itself. The line code words used guarantee a minimum frequency of line status changes that is sufficient for bit synchronization at the receiver. The DC voltage component is removed by additional coding with MLT-3 and a scrambling process, which also ensures a (statistically) even frequency spectrum regardless of the line load. Since there are no physical buses here, only point-to-point connections, a continuous transmission was favored, which limits the time-consuming transient processes of the receiver to the start-up phase of the segment.


General term for the three 100 Mbit / s Ethernet standards over twisted pair cables : 100BASE-TX, 100BASE-T4 and 100BASE-T2 (cabling according to TIA-568A / B ). As with 10BASE-T, the maximum length of a segment is 100 meters. The plug connections are designed as 8P8C modular plugs and sockets and are usually referred to as "RJ-45".
100BASE-T4, IEEE 802.3 Clause 23
100 Mbit / s Ethernet over Category 3 cable (as used in 10BASE-T installations). Uses all four wire pairs in the cable. It is now obsolete as Category 5 cabling is the norm today. It is also limited to half-duplex transmission.
100BASE-T2, IEEE 802.3 Clause 32
There are no products, but the basic technology lives on in 1000BASE-T and is very successful there. 100BASE-T2 offers 100 Mbit / s data rate over Cat-3 cable. It supports full duplex mode and only uses two wire pairs. It is therefore functionally equivalent to 100BASE-TX, but supports older cable installations.
100BASE-TX, IEEE 802.3 Clause 25 (formerly IEEE 802.3u)
Like 10BASE-T, it uses a twisted pair of wires per direction, but requires at least unshielded Cat 5 cables .
The use of conventional telephone cables is possible if the range is limited. The decisive factor here is the correct assignment of the two Ethernet pairs to a twisted pair of the telephone cable. If the telephone cable is stranded as a star quad, the opposing cores each form a pair.
In the 100 Mbit / s market, 100BASE-TX is the standard Ethernet implementation today. 100BASE-TX uses 4B5B as the line code and the coding MLT-3 to halve bandwidth at PMD level . Not only are two states (positive or negative differential voltage) differentiated on the wire pair, there is also a third state (no differential voltage) (ternary code). This means that the data stream is transmitted at a symbol rate of 125 Mbaud within a bandwidth of 31.25 MHz.

While the 4B5B code guarantees a sufficient number of signal changes for bit synchronization at the receiver, the MLT-3 cannot contribute anything to the required freedom from DC voltage. Transmission patterns known as “ killer packets ” can compensate for scrambling and superimpose a significant DC voltage on the transmission pattern ( baseline wander ), which makes scanning more difficult and leads to the end device breaking the connection. In order to be immune to such attacks, the PHY components of the network cards therefore implement DC voltage compensation.

glass fiber

100BASE-FX, IEEE 802.3 Clause 26
100 Mbit / s Ethernet over multimode glass fiber . Maximum segment lengths over multi-mode cables: 400 meters in half-duplex / repeater operation, 2000 meters in full-duplex / switch operation. The scrambled 4B5B data stream is sent directly via an optical light modulator and received in the same way using a pair of fibers. A wavelength of 1300 nm is used, so it is not compatible with 10BASE-FL (10 Mbit / s over fiber), which uses a wavelength of 850 nm.
100BASE-SX, TIA-785
Cheaper alternative to 100BASE-FX, as a wavelength of 850 nm is used; the components for this are cheaper. Maximum segment length: 550 meters over multimode fiber. Due to the wavelength used, it is optionally backwards compatible with 10BASE-FL. A pair of fibers is required.
100BASE-BX10, IEEE 802.3 Clause 58
In contrast to 100BASE-FX, 100BASE-SX and 100BASE-LX10, the send and receive directions are transmitted via a single single-mode fiber. A splitter is required for this, which divides the data to be sent / received between the wavelengths 1310 and 1550 nm. This splitter can be in the transmission component, e.g. B. an SFP module. This standard achieves ranges of 10 km, extended versions 20 or 40 km.
100BASE-LX10, IEEE 802.3 Clause 58
Fast Ethernet over a single-mode fiber pair. Wavelength: 1310 nm, segment length: 10 km.

Gigabit Ethernet

With 1000 Mbit / s Ethernet (Gigabit Ethernet; GbE or GigE for short), two different coding variants are essentially used. With 1000BASE-X (IEEE 802.3 Clause 36), the data stream is broken down into 8-bit units and brought to a symbol rate of 1250 Mbaud using the 8b10b code . This creates a continuous, DC-free data stream that flows to the receiver via a transformer on a twisted wire pair with 1000BASE-CX or modulates the optical carrier wave with 1000BASE-SX / LX / ZX. With 1000BASE-T, on the other hand, the data stream is divided into four sub-streams, each of which is shaped with PAM-5 and trellis coding in its bandwidth and sent and received via the four wire pairs at the same time.

The repeater hubs, which were still widespread in the early Fast Ethernet, were initially defined in the standard for Gigabit Ethernet, but no hubs were produced, so the standard was frozen in 2007 and GbE actually only exists via switches in full duplex mode.


  • 1000BASE-T , IEEE 802.3 Clause 40 (formerly IEEE 802.3ab) - 1 Gbit / s via copper cable from Cat-5 UTP cable or better Cat-5e or Cat-6 (cabling according to TIA-568A / B ). As with 10BASE-T and 100BASE-TX, the maximum length of a segment is 100 meters. Important features of the procedure are:
Basically, 1000BASE-T is an "upscaled" variant of the then unsuccessful 100BASE-T2, only that it uses twice as many wire pairs (namely all four pairs of a typical Cat 5 installation) and the available bandwidth of a Cat -5 cable is used.
  • 1000BASE-TX , 1000BASE-T2 / 4 (not standardized in IEEE 802.3) - Unsuccessful attempts by various interest groups to compensate for the complex modulating / demodulating and echo cancellation circuits of 1000BASE-T with a higher signaling rate. Instead of class D cabling with 1000BASE-T, these transmission methods require class E and class F installations in return. The main argument for the emergence of this transmission method, the high costs for network connections with 1000BASE-T support, has long been invalid.
1000BASE-SX transceiver in SFP -version
  • 1000BASE-CX , IEEE 802.3 Clause 39 - Two wire pairs of a shielded twisted pair cable (STP) with a maximum cable length of 25 m and an impedance of 150 ohms are used as the transmission medium . The connection is made via 8P8C modular plugs / sockets (usually referred to as "RJ45" / "RJ-45") or DE-9 in a star topology . Compared to 1000BASE-T, the 1000BASE-CX places significantly higher demands on the cable. For example, the bandwidth used is 10 times higher (625 MHz compared to 62.5 MHz). The components are also incompatible with one another.
  • 1000Base-T1 , IEEE 802.3bp 1Gbit / s via an unshielded wire pair. The maximum length of a segment is 15 m and is intended for use in vehicles .

glass fiber

  • 1000BASE-SX , 1000BASE-LX , IEEE 802.3 Clause 38 (formerly IEEE 802.3z) - 1 Gbit / s over fiber. The two standards differ fundamentally in the wavelength of the optical infrared laser used and the type of fibers: 1000BASE-SX uses short-wave light with a wavelength of 850 nm and multimode glass fibers, with 1000BASE-LX the lasers emit long-wave light with a wavelength of 1310 nm . The length of a fiber optic cable must be at least 2 meters, the maximum spread depends on the characteristics of the fiber optic used. Multimode fiber optic cables can reach between 200 and 550 meters, depending on the fiber cross-section and modal attenuation, while 1000BASE-LX are specified for singlemode fiber optic cables up to 5 km.
  • 1000BASE-LX10 , sometimes also 1000BASE-LH (LH stands for Long Haul ) - Singlemode fiber optic cables with a maximum length of 10 km are used here. The remaining properties are the same as those of 1000BASE-LX.
  • 1000BASE-BX10 uses a single singlemode fiber with a range of up to 10 km with different wavelengths in each direction: downstream 1490 nm, upstream 1310 nm.
  • 1000BASE-EX and -ZX are not IEEE standards - single-mode fiber optic cables with a maximum length of 40 km (-EX) or 70 km (-ZX) are used. The light used has a wavelength of 1550 nm.

2.5 and 5 Gbps Ethernet

2.5GBASE-T and 5GBASE-T , also abbreviated to 2.5GbE and 5GbE and sometimes referred to together as NBASE-T or MGBASE-T , are transmitted over copper cables like 1000BASE-T or 10GBASE-T.

2.5GBASE-T and 5GBASE-T are downscaled versions of 10GBASE-T with 25% and 50% of the signal rate. Due to the lower frequencies, it is possible to use lower quality cables than the Cat6A required for 10GBASE-T.

For 2.5G, cabling at least according to Cat5e and for 5G according to at least Cat6 is used. When IEEE 802.3bz was officially adopted, products from some manufacturers, including Broadcom , Intel and Marvell , were already available before that .

10 Gbit / s Ethernet

The 10 Gbit / s Ethernet standard (short: 10GbE or 10GE) brings ten different transmission technologies , eight for fiber optic cables and two for copper cables . 10 Gbit / s Ethernet is used for LAN , MAN and WAN . The standard for fiber optic transmission is called IEEE 802.3ae , the standards for copper are IEEE 802.3ak and IEEE 802.3an .

glass fiber

  • 10GBASE-SR bridges short distances via multimode fibers, using long-wave light with a wavelength of 850 nm. The range depends on the cable type, so 62.5 µm "FDDI-grade" fibers go up to 26 m, 62.5 µm / OM1 fibers up to 33 m, 50 µm / OM2 up to 82 m and 50 m µm / OM3 up to 300 m.
  • 10GBASE-LRM uses a wavelength of 1310 nm in order to cover a distance of up to 50 µm / OM3 over all classic multimode fibers (62.5 µm fiber “FDDI-grade”, 62.5 µm / OM1, 50 µm / OM2, 50 µm / OM3) to bridge to 220 m.
  • 10GBASE-LX4 (Clause 53) uses wavelength division multiplexing to enable ranges between 240 m and 300 m over the multimode fibers OM1, OM2 and OM3 or 10 km over single mode fibers. In this case, the wavelengths 1275, 1300, 1325 and 1350 nm are used simultaneously .
  • 10GBASE-LW4 uses singlemode fibers to transmit light at a wavelength of 1310 nm over distances of up to 10 km.
  • 10GBASE-LR uses a wavelength of 1310 nm to bridge a distance of up to 10 km via single-mode fibers.
  • Like 10GBASE-LR, 10GBASE-ER uses singlemode fibers for transmission, but at a wavelength of 1550 nm, which increases the range to up to 40 km. Since 10GBASE-ER with this wavelength has the rare property of being compatible with CWDM infrastructures, it avoids replacing the existing technology with DWDM optics.
OC-192 - STM-64
  • The standards 10GBASE-SW , 10GBASE-LW and 10GBASE-EW use an additional WAN-Phy in order to be able to work together with OC-192 ( SONET ) or STM-64 equipment ( SDH ). The physical layer corresponds to 10GBASE-SR or 10GBASE-LR or 10GBASE-ER, i.e. they also use the same fiber types and achieve the same ranges. There is no corresponding variant for 10GBASE-LX4 with an additional WAN-Phy.

In the LAN reach the standards 10GBASE-SR and 10GBASE-LR caused an increasing spread by the availability of the products.


The advantage of copper cabling compared to fiber optic systems lies in the faster assembly and the different usability of the cabling (many applications via one cable). In addition, the longevity of copper systems is still higher than that of fiber optic systems (burn-out and wear of the LEDs / laser) and the costs of additional (expensive) electronics.


10GBASE-CX4 uses double- twinaxial copper cables (such as InfiniBand ), which can have a maximum length of 15 m. For a long time, this standard was the only one for copper cabling with 10 Gbit / s, but is becoming increasingly less important due to 10GBASE-T, which is downward compatible with the slower standards and can use existing cabling.


Like 1000BASE-T, 10GBASE-T uses four pairs of twisted pairs. The used for structured cabling is the global standard ISO / IEC 11801 as well as in TIA / EIA-568 described. The permissible link length depends on the type of cabling used: In order to achieve the desired link length of 100 m, the requirements of CAT-6a / 7 must be met. With the CAT-5 cables (Cat-5e) used for 1000BASE-T, only half the link length can be achieved. The standard is described in 802.3an and was adopted in mid-2006.

During the transmission, the data stream is divided between the four wire pairs, so that 2.5 Gbit / s are transmitted on each wire pair in the sending direction and in the receiving direction. As with 1000BASE-T, each wire pair is used in full duplex mode. The modulation method 128-DSQ (a kind of double 64QAM ) and finally PAM16 are used for coding , which reduces the Nyquist frequency to 417 MHz.

Due to the high signal rate, various precautions had to be taken to ensure transmission security. Disturbances within the cable are passively reduced by a cross web in the cable, which ensures distance between the wire pairs. In addition, digital signal processors are used in the active components to subtract the interference.

So-called alien crosstalk (Alien crosstalk), so the crosstalk of adjacent, tightly bundled over longer distances, unshielded cable may, however, are not prevented in this way. That is why the standards provide for cables of category Cat  6 A (class E A ). These are either shielded or otherwise adequately suppress the alien crosstalk (e.g. by using a thicker or specially shaped jacket). Unshielded Cat 6 cables (class E) do not reach the usual 100 m cable length when bundled tightly (and only then). On the other hand, a minimum distance between the plug-in connections must be maintained.

10GBASE-T is also possible to a limited extent via Cat 5e cables, see table with cable lengths .

Converged 10 GbE

Converged 10 GbE is a standard for networks where 10 GbE and 10 Gb FC are merged. The converged approach also includes the new Fiber Channel over Ethernet (FCoE). These are FC packets that are encapsulated in Ethernet and for which the Converged Ethernet topology can then also be used; z. B. then appropriately updated switches (due to packet sizes) can be used transparently for FC and iSCSI storage as well as for the LAN.

25 Gbit / s and 50 Gbit / s Ethernet

25 Gigabit (25GbE) and 50 Gigabit Ethernet (50GbE) were proposed for standardization by an industrial consortium and examined by IEEE 802.3 in the form of a study group.

25 / 50GbE are intended to provide higher performance than 10GbE in data centers at significantly lower costs than 40GbE by using technology that has already been defined for those 100GbE variants that are based on 25 Gbps lanes (IEEE 802.3bj). In addition, 25/50 Gbit / s connections can be scaled directly to 100 Gbit / s. In addition, the higher production volume of 25 Gbit / s components could lead to a faster price drop in the 100 Gbit / s range.

Transmission bandwidth for 25GBASE-T is 1250 MHz, which means that Cat 8 cables are required.

40 Gbps and 100 Gbps Ethernet

The fastest generation to date supports 40 and 100 Gbit / s both via copper cables ( Twinax ) and fiber optic cables (single and multimode). The information comes from the specification 802.3ba-2010 of the IEEE and defines the following ranges (lines per direction):

  • 40GBASE-KR4 40 Gbit / s (40GBASE-R with 4 lines of a backplane) at least 1 m
  • 40GBASE-CR4 40 Gbit / s (40GBASE-R with 4 lines of a shielded twinax copper cable) at least 7 m
  • 40GBASE-T 40 Gbit / s (Category-8 twisted pair) at least 30 m, bandwidth 2000 MHz
  • 40GBASE-SR4 40 Gbit / s (40GBASE-R with 4 OM3 glass fibers, multimode) at least 100 m
  • 40GBASE-LR4 40 Gbit / s (40GBASE-R with 1 OS2 fiber and four colors / wavelengths, singlemode, CWDM ) at least 10 km
  • 100GBASE-CR10 100 Gbit / s (100GBASE-R with 10 lines of a shielded twinax copper cable) at least 7 m
  • 100GBASE-SR10 100 Gbit / s (100GBASE-R with 10 OM3 optical fibers, multimode) at least 100 m
  • 100GBASE-SR4 100 Gbit / s (100GBASE-R with 4 OM4 glass fibers, multimode) at least 100 m (IEEE 802.3bm)
  • 100GBASE-LR4 100 Gbit / s (100GBASE-R with 1 OS2 fiber and four colors, singlemode) at least 10 km
  • 100GBASE-ER4 100 Gbit / s (100GBASE-R with 1 OS2 fiber and four colors, singlemode) at least 40 km

200 Gbit / s and 400 Gbit / s Ethernet

Speeds and expected standards faster than 100 Gbit / s are sometimes referred to as Terabit Ethernet .

In March 2013 the IEEE 802.3 400 Gb / s Ethernet Study Group started work on the next generation with 400 Gb / s , in March 2014 the IEEE 802.3bs 400 Gb / s Ethernet Task Force was formed. In January 2016, 200 Gbit / s was added as an additional development goal. The new standards were published in December 2017:

200 Gbit / s
  • 200GBASE-DR4 (Clause 121): 500 m over four single-mode fibers each
  • 200GBASE-FR4 (Clause 122): 2 km via single mode fiber, four wavelengths / colors each (CWDM)
  • 200GBASE-LR4 (Clause 122): 10 km via single mode fiber, four wavelengths / colors each (CWDM)
400 Gbit / s
  • 400GBASE-FR8 (Clause 122): 2 km via single-mode fiber, eight wavelengths / colors each (CWDM)
  • 400GBASE-LR8 (Clause 122): 10 km via single mode fiber, eight wavelengths / colors each (CWDM)
  • 400GBASE-SR16 (Clause 123): 70 m (OM3) or 100 m (OM4) over 16 multimode fibers each
  • 400GBASE-DR4 (Clause 124): 500 m over four single-mode fibers each

electric wire

Cable lengths

Lengths for copper pair
Cable category Transmission
(according to ISO / EN)
default Link length Transmission
Cable standardized up to
(according to TIA / EIA 568
and EN 50288)
Cat-3 Class C 10 BASE-T, 100 BASE-VG 100 m 2 x 10 MHz 16 MHz
Cat-5 - 100 BASE-TX 2 x 31.25 MHz 100 MHz
Cat-5 - 1000 BASE-T 4 x 62.5 MHz 100 MHz
Cat-5e Class D. 100 MHz
Cat-5e, unshielded 10G BASE-T *) 45 ...? m 4 × 417 MHz 100 MHz
Cat-5e, shielded over 45 m 100 MHz
Cat-6, unshielded Class E. *) 55… 100 m 250 MHz
Cat-6, shielded 100 m 250 MHz
Cat-6A Class E A 500 MHz
Cat-7 Class F 600 MHz
Cat-8.1 class 1 40G BASE-T 30 m 1600 MHz
Cat 8.2 2nd grade 40G BASE-T 1600 MHz

The total permissible length of the transmission path is usually 100 m. This includes:

  • 90 m installation cable
  • 10 m patch cable (2 × 5 m)
  • 2 plug connections (e.g. socket and patch panel)

Patch cables have poorer transmission properties. If the patch cables are longer than 10 m, the permissible length of the installation cable is reduced by 1.5 m for each meter that is exceeded.
If the route consists only of patch cables, the permissible standard length is approx. 70 m.

Unless otherwise specified, the lengths for shielded and unshielded cables apply equally.
The values ​​for 10 Gbit / s Ethernet correspond to IEE 802.3-2008, Table 55-13 .
*) Reduced lengths at 10 Gbit / s result from external crosstalk between several cables and only apply unshielded if they are tightly bundled over a length of many meters.

The value for 10 Gbit / s over Cat 5e was proposed in a draft, but not included in the final IEEE 802.3 standard. However, numerous hardware manufacturers confirm the function over 45 m Cat 5e UTP.
Shielded CAT 5e is uncommon outside of Europe and has not been tested by the US-dominated body. This results in considerably longer lengths because the length-limiting parameter is the alien crosstalk. However, shielded cables are practically unaffected.

Lengths for multimode fiber optic cables
speed cabling Distance (max)
10 Mbit / s 10BASE-FL / -FB OM1 FO multimode 62.5 / 125 µm 2000 m
100 Mbit / s 100BASE-FX OM1 / OM2 FO multimode 62.5 / 125 µm / 50/125 µm HDX 412 m
OM1 / OM2 FO multimode 62.5 µm / 50 µm FDX 2000 m
1 Gbit / s 1000BASE-SX OM1 FO multimode 62.5 / 125 µm 220 m
OM2 FO multimode 50/125 µm 550 m
OM3 fiber optic multimode 50/125 µm > 550 m
10 Gbit / s 10GBASE-SR OM1 FO multimode 62.5 / 125 µm 26 m
OM2 FO multimode 50/125 µm 82 m
OM3 fiber optic multimode 50/125 µm 300 m
10 Gbit / s 10GBASE-LRM OM1 / 2/3 fiber optic multimode 62.5 / 125 µm / 50/125 µm 220 m
Single-mode fiber optic cable lengths
speed cabling Distance (max)
10 Gbit / s 10GBASE-LR FO single mode 8-10 µm OS2: 10 km
10 Gbit / s 10GBASE-ER FO single mode 8-10 µm OS2: 30-40 km
10 Gbit / s 10GBASE-ZR (not IEEE 802.3) FO single mode 8-10 µm 80 km

8P8C connector for Ethernet

  • Token Ring wire pair 1 and 3
  • 10BASE-T wire pair 2 and 3
  • 100BASE-TX wire pair 2 and 3
  • 100BASE-T4 wire pair 1, 2, 3 and 4
  • 100BASE-VG (VG-AnyLAN) wire pair 1, 2, 3 and 4

Assignment and cable color code for 8P8C connector

Wire pair Pins EIA / TIA 568A IEC REA DIN 47.100
1 4/5 blue White White blue White blue White Brown
2 3/6 white / orange Red orange turquoise / purple green yellow
3 1/2 white / green black / gray white / orange gray / pink
4th 7/8 White Brown yellow / brown turquoise / purple blue red

Assignment of 8P8C connector according to EIA / TIA 568B

10BASE-T and 100BASE-TX
signal Pin code colour
TX + 1 white / orange
TX- 2 orange
RX + 3 white / green
4th blue
5 White blue
RX- 6th green
7th White Brown
8th brown

Metro ethernet

Metro Ethernet networks (MEN) are Ethernet-based Metropolitan Area Network (MAN) networks that are based on Carrier Grade Ethernet. Now that the length restrictions for Ethernet networks have practically been lifted with the introduction of sophisticated fiber optic technology, Ethernet is also gaining importance in wide area networks such as MAN. On the customer side, MANs are based on cheaper, well-known technology and guarantee comparatively high efficiency with low complexity.

Power over Ethernet

IEEE 802.3af (IEEE 802.3 Clause 33) also belongs to the family of Ethernet standards . The procedure describes how Ethernet-capable devices can be supplied with energy via the twisted pair cable. Either the unused wires of the line are used or a direct current component is transmitted over the four wires used in addition to the data signal. A logic ensures that only PoE-capable devices are supplied with energy. According to 802.3af, appropriately designed devices are supplied with 48 V and up to 15.4 watts. Up to 30 W at 54 V at the end of 2009 reached ratified 802.3at standard or PoE + . In 2018, the third generation 4PPoE was adopted as 802.3bt, which can supply devices with up to 100 W via all four wire pairs.

Related standards

The following network standards do not belong to the IEEE 802.3 Ethernet standard, but support the Ethernet data block format and can work together with Ethernet:

  • WLAN ( IEEE 802.11 ) - A technology for wireless networking via radio technology over short distances (distances depend on the local conditions and comparable to LAN ), initially with transmission rates from 1 Mbit / s, currently (2010) with up to 600 Mbit / s .
  • VG-AnyLan ( IEEE 802.12 ) or 100BASE-VG - An early competitor to 100 Mbit / s Ethernet and 100 Mbit / s TokenRing. A method according to the Multimedia Extensions owns and example as FDDI knows guaranteed bandwidth, it is based on a demand priority access procedure referred ( Demand Priority Access Method , collision-free, all requests are prioritized by the Hub / Repeater centrally controlled), with which the disadvantages of CSMA are eliminated . 100BASE-VG also runs over category 3 cables, but uses four wire pairs. Hewlett-Packard and AT&T played a leading role in the development ; commercially, VG-AnyLan was a failure.
  • TIA 100BASE-SX - Standard promoted by the Telecommunications Industry Association . 100BASE-SX is an alternative implementation of 100 Mbit / s Ethernet over fiber optics and is incompatible with the official 100BASE-FX standard. A prominent feature is the possible interoperability with 10BASE-FL, as it masters autonegotiation between 10 Mbit / s and 100 Mbit / s. The official standards can not do this due to the different wavelengths of the LEDs used . The target group are organizations with an already installed 10 Mbit / s fiber optic base.
  • TIA 1000BASE-TX also comes from the Telecommunications Industry Association. The standard was a commercial failure and there are no products to implement it. 1000BASE-TX uses a simpler protocol than the official 1000BASE-T standard, but requires Cat 6 cables (opponents claim that this standard, primarily promoted by the cable industry, was not intended for product development at all, but only for initial application for this cable class, which until then had no advantages over Cat-5 , was able to demonstrate).
  • InfiniBand is a fast, high-performance method that has been specified since 1999 for bridging short distances (over copper cables up to 15 m). It uses bidirectional point-to-point connections for cost-effective and low-latency data transmission (under 2 µs) and achieves theoretical data transmission rates of up to 2.5 Gbit / s in both directions per channel and 5 Gbit / s in the newer DDR variant. With InfiniBand, several channels can be bundled transparently, whereby a common cable is then used. Four channels (4 ×) are usually 10 Gbit / s or 20 Gbit / s. The main areas of application are supercomputers (HPC clusters) as can also be found in the TOP500 list.

See also


Web links

Commons : Ethernet  - collection of pictures, videos and audio files

Individual evidence

  1. Ethernet. In: Duden online. Dudenverlag , accessed on March 24, 2015 .
  2. J. Jasperneite : Real-time Ethernet at a Glance , atp 3/2005, pp. 29–34, ISSN  0178-2320 .
  3. The History of Ethernet on YouTube
  4. The first graphic about the function of the Ethernet
  5. ^ List of the winners of the "National Medal of Technology" in 2003
  6. DE-CIX first Internet Exchange worldwide to offer 400-Gigabit Ethernet access technology. Retrieved April 3, 2019 .
  7. IEEE 802.3 Clause 3.2.10 Extension Field, IEEE 802.3 Clause Carrier extension (half duplex mode only)
  8. ^ Objectives . IEEE 802.3bs Task Force. March 20, 2014. Retrieved August 27, 2015.
  9. Extended Ethernet Frame Size Support . May 26, 2000.
  10. Coraid AoE Protocol Specification ( Memento from December 27, 2012 in the Internet Archive )
  11. IEEE Std 802.3-2005, 3.2.6
  12. IEEE 802.3 Table 24-1 4B / 5B code groups
  13. Embeddet Contrl advanced module WS 2005/06 ( Memento from January 31, 2012 in the Internet Archive ) (PDF; 1 MB)
  14. IEEE 802.3 1.2.3 Physical layer and media notation
  15. Advanced Physical Layer APL. In: PI White Paper. Profinet International, 2018, accessed October 9, 2019 .
  16. Suitability of telephone cables as Ethernet network cables , accessed April 15, 2012.
  17. IEEE 802.3 2008 Section 3: 41
  19. ( Memento from December 4, 2014 in the Internet Archive )
  21. Broadcom offers PHYs for 2.5G and 5G Ethernet speeds . November 18, 2015. Archived from the original on November 18, 2015.
  23. (PDF)
  24. a b c Enabling 10 GB deployment in the Enterprise
  25. a b John George, BICSI (en) : 10 Gigabit Ethernet over Multimode Fiber  ( page no longer available , search in web archives ) (PDF)@1@ 2Template: Dead Link /
  26. Datwyler White Paper: 10 Gigabit Ethernet over shielded copper cable systems ( Memento of May 16, 2013 in the Internet Archive ) (January 2009; PDF; 109 kB), accessed April 15, 2012
  27. BICSI FAQ: Can Category 6 Run 10G in Distances , accessed on April 15, 2012.
  28. Overview 25G & 50G Ethernet Specification, Draft 1.4 (PDF)
  29. IEEE 802.3 25 Gb / s Ethernet Study Group Public Area
  30. a b IEEE 802.3 adopted 25 / 40GBase-T ▪︎ LANline. Retrieved September 9, 2018 (German).
  31. IEEE Standard 802.3ba-2010 Part3 No. 80 ff.
  33. 400 Gb / s Ethernet Study Group , IEEE 802.3, May 23, 2013
  35. 10GBASE-T Objective Proposal: July 2003 10GBASE-T Study Group Objectives (PDF; 665 kB), accessed April 15, 2012
  36. SMC: TigerCard 10G User Guide (PDF; 1 MB), accessed April 15, 2012
  37. Standards and Industry Groups - Standards & Initiatives . In: Helping Define 802.11n and other Wireless LAN Standards . Early 2010. Accessed December 27, 2010.
This version was added to the list of articles worth reading on August 8, 2006 .