NTRON技术白皮书:网络自动协商机制

2024年2月8日发(作者：)

820 S. University Blvd. Suite 4E, Mobile, AL 36609

PH: 251-342-2164 • FAX: 251-342-6353

N-TRON

Ethernet Auto-negotiation

In today’s world of computer networks auto-negotiation is an important plug-and-play

technology. Auto-negotiation as an algorithm was defined by Section 28 of the IEEE 802.3

standard and first introduced in 1997 as part of the IEEE 802.3u standard on Fast Ethernet.

Auto-negotiation was designed to be backward compatible with original Ethernet networking

standards as well. Auto-negotiation was further enhanced in 1999 by the IEEE standard

802.3ab with the introduction of Gigabit Ethernet. Auto-negotiation is best defined as the mutual

agreement by two network devices sharing a wire on the speed, duplex, and controls to govern

the use of that wire. As a protocol auto-negotiation exists strictly at the PHY (physical) layer of

the OSI (Open System Interconnection Reference Model) and is implemented by software,

hardware, or a mixture of both. Specifically this white paper will detail how the protocol

negotiates speed, duplex, Auto-MDIX (cable termination), and flow control.

As you will see in the technical discussions that follow, auto-negotiation is an extremely

important setting on today’s wired Ethernet networks. For a link to function properly the devices

on either side of the wire must be configured in the same manner; either both set to auto-negotiation or both set to the same hard-coded speed and duplex settings. In an environment

where one device is set to auto-negotiate and the other device is set to a hard-coded speed and

duplex the auto-negotiate algorithm can detect speed and set that appropriately. The duplex

setting of the remote device is indeterminable by the auto-negotiating device. Following the

IEEE standard, the auto-negotiating device falls back to half-duplex. This presents an issue if

the remote device is set to full-duplex. Typically in such a scenario, users complain of slow

network connectivity and application timeouts. These symptoms will be explained in detail in

the section discussing Duplex.

Lastly, it should be noted that, according to the IEEE specification the use of gigabit Ethernet

requires the use of auto-negotiation therefore 1000Mb/s is not a valid hard-coded option in a

true IEEE compliant networking device.

Speed

IEEE 802.3u introduced 100Mb/s to what was previously only a 10 Mb/s Ethernet networking

world. Now that computers had a choice of what speed to communicate a procedure needed to

be introduced to govern this decision. With the introduction of a third speed, 1000 Mb/s or

Gigabit Ethernet, this procedure became even more important. Thus the auto-negotiation

protocol was created and the NWay algorithm adapted to provide a plug and play solution to this

decision making process while still maintaining complete backwards compatibility with the 10

Mb/s protocol.

The 10 Mb/s standard detects an active link with another network device through the

transmission and reception of Link Integrity Test (LIT) pulses whenever the device is not actively

sending or receiving data. These LIT pulses or Normal Link Pulses (NLP), as the name was

later changed to, consist of a single uni-polar positive-only pulse for the duration of 100ns at an

interval of 16ms with a +/-8ms window.

The auto-negotiation protocol introduced with the 100 Mb/s standard transmits a Fast Link Pulse

(FLP) instead of an NLP. A single FLP burst consists of a series of 33 pulses. Each burst of 33

pulses is 2ms long in total and fall into the same transmission interval of 16ms +/- 8ms. The

individual pulses are 125 µs with 62.5µs +/- 7µs between pulses. Diagram 1 illustrates this

Page 1 of 6

820 S. University Blvd. Suite 4E, Mobile, AL 36609

PH: 251-342-2164 • FAX: 251-342-6353

N-TRON

timing. The individual pulses alternate between clock pulses and data pulses with the first and

all successive odd numbered pulses being clock pulses. Each of the 16 data pulses (with each

pulse or lack of pulse representing a 1 or a 0, respectively) consist of a single bit of data and

collectively add up to 16 bits of 2 bytes of data. These 2 bytes make up the link code word

(LCW) which contains the information needed for auto-negotiation.

16 ms802.3u StandardFast Link Pulse (FLP) Burst16 Data Pulses (two Bytes)

alternated with 17 Clock Pulses33 Pulses per BurstData PulseClock Pulse16 ms802.3 StandardNormal Link Pulses (NLP) orLink Integrity Test Pulses (LIT) 2 msDiagram 1. NLP and FLP Timing.

There are multiple LCW formats but the most important LCW is the base page. This base page

is the transmission stating the capabilities of that device. The first 5 bits only have two valid

values. They state either to use IEEE 802.3 (Ethernet) or IEEE 802.9 (IsoEthernet over Cat3

twisted pair). The next 5 bits state what speed and duplex combinations that a device can

communicate. Bits A5 and A6 are used for Flow Control and D14 is used to acknowledge a

negotiation. The last bit, D15 is used to denote the need to use Next Page, a more advanced

LCW used to negotiate Gigabit speeds and controls. Diagram 2 illustrates the Base Page.

D0S0D1S1D2S2D3S3D4S4D5A0D6A1D7A2D8A3D9A4D10A5D11A6D12A7D13RFD14AckD15NPSelector FieldS0 thru S4802.3 = 00001802.9 = 00010Technology Ability FieldA0 thru A7 A0

A1 A2 A3 A4 A5 A6 10BaseT 10BaseT-FD 100BaseTX 100BaseTX-FD 100BaseT4 PAUSE Asymmetric Pause

operation for full

duplex links A7 ReservedOther Fields RF Remote Fault Ack Acknowledge NP Next PageDiagram 2. LCW Base Page.

Page 2 of 6

820 S. University Blvd. Suite 4E, Mobile, AL 36609

PH: 251-342-2164 • FAX: 251-342-6353

N-TRON

In order for two devices to agree on a speed for transmission over the wire six identical LCW’s

must be transmitted and received over the wire, three from each side. Once a device has

received three identical LCW’s contained within an FLP from the remote end the local device

will transmit an FLP with an Acknowledgement bit (ACK) set. It should be noted that each

device is only stating what its own capabilities are therefore the two devices need to use the

same order of priority to agree upon a speed. This order of priority is part of the IEEE standard

and can be found in diagram 3. Once both sides have received a response ACK the speed is

agreed upon.

Priority123456789LowestHighestTechnology Type1000BaseT – Full Duplex1000BaseT – Half Duplex100BaseT2 – Full Duplex100BaseTX– Full Duplex100BaseT2 – Half Duplex100BaseT4100BaseTX – Half Duplex10BaseT – Full Duplex10BaseT – Half DuplexDiagram 3. Priority Resolution table

FLP’s were designed to line up with the NLP so that a 10Mb/s device will detect signal on the

line at the usual interval and be able to communicate. An auto-negotiation capable device will

detect the existence of NLP’s and, due to the backward compatibility standards, be able to fall

back and communicate via NLP’s to work at 10Mb/s.

Duplex

With the introduction of the IEEE 802.3u Fast Ethernet standard the possibility of simultaneous

bi-directional communication became available. Again a protocol and method needed to be

introduced to govern this decision. As discussed in the previous section, duplex negotiations

are handled in the Base Code Word for 100Mb/s networks and are a part of the Next Page and

Message Page LCW’s in a 1000Mb/s network.

Duplex mismatch is the most common cause for network link problems outside of physical

cabling or hardware failure. Duplex mismatches are caused by the inability of an auto-negotiation device to predict the settings of a hard-coded device. This is because the

transmission of FLP’s is disabled when a device is hard-coded in accordance with the IEEE

specification. Also in accordance with the IEEE specification the auto-negotiation device will

connect with the Half-Duplex setting when the duplex setting of the other device cannot be

determined.

Page 3 of 6

820 S. University Blvd. Suite 4E, Mobile, AL 36609

PH: 251-342-2164 • FAX: 251-342-6353

N-TRON

Duplex mismatches can be difficult to identify because they do not cause a complete link drop.

Often the link will perform sufficiently to prevent any alarms initially. This is especially the case

when the link is used sparsely. Problems will crop up once link activity increases however. The

specific problem is that a half-duplex device believes that only one device can talk at a time so it

will not talk as long as the other (full-duplex) device is talking. The full-duplex device is not

under these restrictions and believes that both devices can transmit simultaneously. If, while

the half-duplex device is transmitting, it detects a transmission from the other device it will

immediately stop transmitting, throw away all incoming transmissions as invalid, and initiate a

stand-off timer to quiesce the medium. In the meantime the full-duplex device completes

transmission and assumes reception. The full-duplex device will also receive the truncated

packet from the half-duplex device, determine that it is bad, and flag the Cyclical Redundancy

Check (CRC) error counter. The half-duplex device will attempt a retransmission of its packet

once the stand-off timer is done but the full-duplex device feels no need to retransmit

(remember it doesn’t know that the other device threw away its packet) and so the half-duplex

device will never receive that packet unless a higher level of the OSI layer requires an

acknowledgement and causes a retransmission.

The symptoms of this situation will most often show up as a sluggish network link or an

application or applications with excessive timeouts. On a correctly configured connection CRC

errors should be negligible so an excessive CRC count is often considered symptomatic of a

duplex mismatch.

Duplex mismatches can be a particularly difficult problem with unmanaged switches. By

definition, an unmanaged switch does not have the ability to hard-code a port to a particular

speed or duplex setting and is always in auto-negotiate mode. If a device is hard-coded to a

particular speed or duplex the unmanaged switch will not be able to create a fully functioning

link with this device and problems will eventually occur. Please refer to diagram 4 for an

illustrated example.

100 MbpsFull DuplexThis mismatch duplex connection

will cause communication collisions

and may cause data loss.100 MbpsHalf Duplex100 MbpsFull Duplex100 MbpsFull DuplexPLC Hard Coded for

100 Mbps Full Duplex(will not Advertise)100 MbpsFull Duplex100 MbpsFull DuplexHMI ConfiguredTo Auto negotiateUnmanaged Switches will always Auto

Negotiate resulting in a missmatch with any

Hard Coded Network Interface Card (NIC)

coded to 100Mbps Full DuplexDrive ConfiguredTo Auto negotiateDiagram 4. Duplex mismatch scenario

Page 4 of 6

820 S. University Blvd. Suite 4E, Mobile, AL 36609

PH: 251-342-2164 • FAX: 251-342-6353

N-TRON

Auto-MDIX (Media Dependent Interface with Crossover support)

The introduction of twisted pair cabling also allowed for the possibility of multiple ways to plug

the cable in. A category-5e twisted pair cable contains 8 distinct copper wires but there are two

industry standard ways to wire the RJ-45 connectors. The goal of these standards is to provide

the same wire on each of the pins 1 to 8 on either side of the cable. In order to wire a cross-over cable the cable maker installs one end upside down thereby presenting pins 1 to 8 on one

side and pins 8 to 1 on the other side.

In order for one device to connect to another device the transmit (TX) on one must be

connected to the receive (RX) on the other device and vice versa. This system is required in

order for the two devices to communicate. Since most cables are usually wired to be a straight-through cable it was decided to fix this problem at the device level. Traditionally, network

devices and computer network cards are wired opposite of each other. Media Device Interface

(MDI) is the orientation that a computer card is traditionally cabled to and Media Device

Interface-Crossover (MDIX) is the orientation used for a switch or other network device. This

was sufficient in the old days but required the use of a special cross-over cable for computer-to-computer and switch-to-switch communications.

Auto-MDIX is a procedure developed and patented by two engineers at HP and is included in

the Gigabit Ethernet standard from the IEEE, IEEE 802.3ab. The Auto-MDIX protocol removes

the need of a specific cross-over or straight-through cable by attaching both the receiver and the

transmitter to both wires in the pair. Per the Gigabit Ethernet standard the receiver knows what

the transmitter is sending.

It electrically subtracts that signal from what it senses on the wire and

using echo cancellation the receiver is able to compute what is being transmitted from the

remote end.

Flow Control

With the ever increasing speed at which devices can transmit data it is important that back-planes, buffers, and switch to switch ports keep up with this speed escalation. If a switch’s

backplane speed is greater than the sum of the cumulative speed of all of that switch’s ports we

often call the switch a wire speed switch. This is often unattainable with higher density

switches. A link will become saturated when the connection between two devices has more

data to transmit than bandwidth to transfer that data. This is an easy scenario to create on a

link between two switches if the uplink port is the same speed as the user ports. This

introduces the need for flow control, a process that allows one device to ask the other to pause

in order to let it catch up. This pause procedure could include a timer until a restart, require a

unpause notification, or simply be a stall tactic with dummy data to delay communications.

In 10 Mb/s networks devices that require a pause in the network simply fill the media with a

dummy packet after each packet is received to prevent data from coming through. This

technique is called backpressure. Backpressure is also the pause process in 100 Mb/s half-duplex networks. In 100 Mb/s full-duplex and 1000 Mb/s full-duplex networks the auto-negotiation protocol’s pause control is implemented. A device that needs a pause sends an

FLP with the appropriate pause bit set (either A5 or A6).

Page 5 of 6

820 S. University Blvd. Suite 4E, Mobile, AL 36609

PH: 251-342-2164 • FAX: 251-342-6353

N-TRON

Summary

In conclusion, the auto-negotiation standard allows for a plug and play environment to exist in a

networking world with multiple speeds, duplexes, cabling standards, and flow controls. N-TRON

recommends leaving all devices in a network set to auto-negotiate to allow for ease of

deployment and to minimize the possibility of introducing configurations into your network now

or at a later date. If hard-coding is necessary then N-TRON recommends hard-coding both

sides and documenting these hard-coded settings to ensure against future problems if network

changes are made.

Globally recognized as a market leader in the Industrial Ethernet marketplace, N-TRON’s

products are used throughout the world in a wide variety of applications including maritime,

process control, wind farms, wastewater treatment plants, nuclear power plants, solar energy,

and security and surveillance where reliability is an absolute necessity.

N-TRON is headquartered in Mobile, Alabama, with operations located throughout the United

States, Canada, EMEA, India and the Pacific Rim. N-TRON products are distributed in over 75

countries worldwide.

Page 6 of 6