HP 180 Degree Turn User Manual

The Ethernet Evolution

The 180 Degree Turn

2005/03/11(C) Herbert Haas

1

“Use common sense in routing cable.
Avoid wrapping coax around sources

of strong electric or magnetic fields.

Do not wrap the cable around

flourescent light ballasts or

cyclotrons, for example.”

E

thernet Headstart Product, Information and Installation Guide,

Bell Technologies, pg. 11

2

History: Initial Idea

 Shared media  CSMA/CD as access algorithm
 COAX Cables
 Half duplex communication
 Low latency  No networking nodes

(except repeaters)

 One collision domain and also one broadcast domain

10 Mbit/s shared

by 5 hosts  2

Mbit/s each !!!

(C) Herbert Haas

2005/03/11

The initial idea of Ethernet was completely different than what is used today
under the term "Ethernet". The original new concept of Ethernet was the use of
a shared media and an Aloha based access algorithm, called Carrier Sense
Multiple Access with Collision Detection (CSMA/CD). Coaxial cables were
used as shared medium, allowing a simple coupling of station to bus-like
topology.

Coax-cables were used in baseband mode, thus allowing only unicast
transmissions. Therefore, CSMA/CD was used to let Ethernet operate under the
events of frequent collisions.

Another important point: No intermediate network devices should be used in
order to keep latency as small as possible. Soon repeaters were invented to be the
only exception for a while.

An Ethernet segment is a coax cable, probably extended by repeaters. The
segment constitutes one collision domain (only one station may send at the same
time) and one broadcast domain (any station receives the current frame sent).
Therefore, the total bandwidth is shared by the number of devices attached to the
segment. For example 10 devices attached means that each device can send 1
Mbit/s of data on average.

3

Ethernet technologies at that time (1975-80s): 10Base2 and 10Base5

3

History: Multiport Repeaters

 Demand for structured cabling (voice-grade

twisted-pair)

 10BaseT (Cat3, Cat4, ...)

 Multiport repeater ("Hub") created
 Still one collision domain

("CSMA/CD in a box")

(C) Herbert Haas

Later, Ethernet devices supporting structured cabling were created in order to
reuse the voice-grade twisted-pair cables already installed in buildings. 10BaseT
had been specified to support Cat3 cables (voice grade) or better, for example
Cat4 (and today Cat5, Cat6, and Cat7).

Hub devices were necessary to interconnect several stations. These hub devices
were basically multi-port repeaters, simulating the half-duplex coax-cable, which
is known as "CSMA/CD in a box". Logically, nothing has changed, we have still
one single collision and broadcast domain.

Note that the Ethernet topology became star-shaped.

2005/03/11

4

4

History: Bridges

 Store and forwarding according destination MAC

address

 Separated collision domains
 Improved network performance
 Still one broadcast domain

Three collision

domains in this

example !

(C) Herbert Haas

2005/03/11

5

Bridges were invented for performance reasons. It seemed to be impractical that
each additional station reduces the average per-station bandwidth by 1/n. On the
other hand the benefit of sharing a medium for communication should be still
maintained (which was expressed by Metcalfe's law).

Bridges are store and forwarding devices (introducing significant delay) that can
filter traffic based on the destination MAC addresses to avoid unnecessary
flooding of frames to certain segments. Thus, bridges segment the LAN into
several collision domains. Broadcasts are still forwarded to allow layer 3
connectivity (ARP etc), so the bridged network is still a single broadcast domain.

5

History: Switches

 Switch = Multiport Bridges with HW acceleration
 Full duplex  Collision-free Ethernet  No CSMA/CD

necessary anymore

 Different data rates at the same time supported

 Autonegotiation

 VLAN splits LAN into several broadcast domains

Collision-free

plug & play

scalable Ethernet !

6

1

0 Mbit/s

(C) Herbert Haas

2005/03/11

100 Mbit/s

1000 Mbit/s

1

00 Mbit/s

Several vendors built advanced bridges, which are partly or fully implemented in
hardware. The introduced latency could be dramatically lowered and
furthermore other features were introduced, for example full duplex
communication on twisted pair cables, different frame rates on different ports,
special forwarding techniques (e,g, cut through or fragment free), Content
Addressable Memory (CAM) tables, and much more. Of course marketing rules
demand for another designation for this machine: the switch was born.

Suddenly, a collision free plug and play Ethernet was available. Simply use
twisted pair cabling only and enable autonegotiation to automatically determine
the line speed on each port (of course manual configurations would also do). This
way, switched Ethernet become very scalable.

Furthermore, Virtual LANs (VLANs) were invented to split the LAN into several
broadcast domains. VLANs improve security, utilization, and allows for logical
borders between workgroups.

6

Today

 No collisions  no distance limitations !
 Gigabit Ethernet becomes WAN

technology !

 Over 100 km link span already

 Combine several links to "Etherchannels"

 Acts as single link from the spanning-tree view

• Cisco: Port Aggregation Protocol (PAgP)

• IEEE 802.1ad: Link Aggregation Control Protocol

(LACP)

1 Gbit/s or even 10 Gbit/s long reach connection !!!

(C) Herbert Haas

Today, Gigabit and even 10 Gigabit Ethernet is available. Only twisted pair and
more and more fiber cables are used between switches, allowing full duplex
collision-free connections. Since collisions cannot occur anymore, there is no
need for a collision window anymore! From this it follows, that there is virtually
no distance limit between each two Ethernet devices.

Recent experiments demonstrated the interconnection of two Ethernet Switches
over a span of more than 100 km! Thus Ethernet became a WAN technology!
Today, many carriers use Ethernet instead of ATM/SONET/SDH or other rather
expensive technologies. GE and 10GE is relatively cheap and much simpler to
deploy. Furthermore it easily integrates into existing low-rate Ethernet
environments, allowing a homogeneous interconnection between multiple
Ethernet LAN sites. Basically, the deployment is plug and play.

If the link speed is still too slow, so-called "Etherchannels" can be configured
between each two switches by combining several ports to one logical connection.
Note that it is not possible to deploy parallel connections between two switches
without an Etherchannel configuration because the Spanning Tree Protocol (STP)
would cut off all redundant links.

2005/03/11

7

Depending on the vendor, up to eight ports can be combined to constitute one
"Etherchannel".

7

What About Gigabit Hubs?

 Would limit network diameter to 20-

25 meters (Gigabit Ethernet)

 Solutions

 Frame Bursting
 Carrier Extension

 No GE-Hubs available on the market

today  forget it!

 No CSMA/CD defined for 10GE (!)

(C) Herbert Haas

Remember: Hubs simulate a half-duplex coaxial cable inside, hence limiting the
total network diameter. For Gigabit Ethernet this limitation would be about 25
meters, which is rather impracticable for professional usage. Although some
countermeasures had been specified in the standard, such as frame bursting and
carrier extension, no vendor developed an GE hub as for today. Thus: Forget GE
Hubs!

The 10 GE specification does neither consider copper connections nor hubs. 10
GE can only run over fiber.

At this point please remember the initial idea in the mid 1970s: Bus, CSMA/CD,
short distances, no network nodes.

Today: Structured cabling (point-to-point or star), never CSMA/CD, WAN
capabilities, sophisticated switching devices in between.

2005/03/11

8

8

MAC Control Frames

 Additional functionality easily integrated
 Currently only Pause-Frame supported

Always 64 bytes

8 bytes 6 6 2 2 44 4

preamble FCSMAC-ctrl parametersMAC-ctrl opcode8808hSADA

MAC-ctrl opcode ........... Defines function of control frame

MAC-ctrl parameters .... control parameter data (always filled up to 44 bytes)

(C) Herbert Haas

2005/03/11

Different data rates between switches (and different performance levels) often
lead to congestion conditions, full buffers, and frame drops. Traditional Ethernet
flow control was only supported on half-duplex links by enforcing collisions to
occur and hereby triggering the truncated exponential backoff algorithm. Just let
a collision occur and the aggressive sender will be silent for a while.

A much finer method is to send some dummy frames just before the backoff
timer allows sending. This way the other station never comes to send again.

Both methods are considered as ugly and only work on half duplex lines.
Therefore the MAC Control frames were specified, allowing for active flow
control. Now the receiver sends this special frame, notifying the sender to be
silent for N slot times.

The MAC Control frame originates in a new Ethernet layer—the MAC Control
Layer—and will support also other functionalities, but currently only the "Pause"
frame has been specified.

9

9

Auto Negotiation

 Enables each two Ethernet devices to

exchange information about their
capabilities

 Signal rate, CSMA/CD, half- or full-duplex

 Using Link-Integrity-Test-Pulse-Sequence

 Normal-Link-Pulse (NLP) technique is used

in 10BaseT to check the link state (green LED)

 10 Mbit/s LAN devices send every 16.8 ms a

100ns lasting NLP, no signal on the wire
means disconnected

(C) Herbert Haas

Several Ethernet operating modes had been defined, which are incompatible to
each other, including different data rates (10, 100, 1000 Mbit/s), half or full
duplex operation, MAC control frames capabilities, etc.

Original Ethernet utilized so-called Normal Link Pulses (NLPs) to verify layer 2
connectivity. NLPs are single pulses which must be received periodically
between regular frames. If NLPs are received, the green LED on the NIC is
turned on.

Newer Ethernet cards realize auto negotiation by sending a sequence of NLPs,
which is called a Fast Link Pulse (FLP) sequence.

2005/03/11

10

10

Fast Link Pulses

 Modern Ethernet NICs send bursts of

Fast-Link-Pulses (FLP) consisting of
17-33 NLPs for Autonegotiation
signalling

 Each representing a 16 bit word

 GE sends several "pages"

(C) Herbert Haas

A series of FLPs constitute an autonegotiation frame. The whole frame consists
of 33 timeslots, where each odd numbered timeslot consists of a real NLP and
each even timeslot is either a NLP or empty, representing 1 or 0. Thus, each FLP
sequence consists of a 16 bit word.

Note that GE Ethernet sends several such "pages".

2005/03/11

11

11

100 Mbit Ethernet Overview

(C) Herbert Haas

Signaling Schemes

Fast Ethernet

100BaseX

Signaling

"100BaseT"

2005/03/11

IEEE 802.3u

100BaseTX100BaseFX

Fast Ethernet

100Base4T+

Signaling

100BaseT4

(half duplex)

IEEE 802.12

Demand Priority

100VG-AnyLAN

HP and AT&T

invention for real time

applications

12

The diagram above gives an overview of 100 Mbit/s Ethernet technologies,
which are differentiated into IEEE 802.3u and IEEE 802.12 standards. The IEEE

802.3u defines the widely used Fast Ethernet variants, most importantly those
utilizing the 100BaseX signaling scheme. The 100BaseX signaling consists of
several details, but basically it utilizes 4B5B block coding over only two pairs of
regular Cat 5 twisted pair cables or two strand 50/125 or 62.5/125-Fm multimode
fiber-optic cables.

100Base4T+ signaling has been specified to support 100 Mbit/s over Cat3 cables.
This mode allows half duplex operation only and uses a 8B6T code over 4 pairs
of wires; one pair for collision detection, three pairs for data transmission. One
unidirectional pair is used for sending only and two bi-directional pairs for both
sending and receiving.

The 100VG-AnyLAN technology had been created by HP and AT&T in 1992 to
support deterministic medium access for realtime applications. This technology
was standardized by the IEEE 802.12 working group. The access method is
called "demand priority". 100VG-AnyLAN supports voice grade cables (VG) but
requires special hub hardware. The 802.12 working group is no longer active.

12

4B/5B Coding

MII

4 x 25

Mbit/s

PCS

1000

16 code
groups

32 code
groups

13

125 MBaud

(C) Herbert Haas

4B/5B Encoder/Decoder

010 10

PMA

2005/03/11

The diagram above shows the basic principle of the 4B5B block coding principle,
which is used by 802.3u and also by FDDI. The basic idea is to transform any
arbitrary 4 bit word into a (relatively) balanced 5 bit word. This is done by a fast
table lookup.

Balancing the code has many advantages: better bandwidth utilization, better
laser efficiency (constant temperature), better bit-synchronization (PLL), etc.

Note that the signaling overhead is 5/4  12.5 %.

13

Gigabit Ethernet

Media Access Control (MAC)

Gigabit Media Independent Interface (GMII)

1000Base-LX

Fiber Optic

(C) Herbert Haas

1000Base-X

8B/10B encoder/decoder

1000Base-SX

LWL

IEEE 802.3z physical layer

2005/03/11

SWL

Fiber Optic

1000Base-CX

Shielded
Balanced

Copper

1000Base-T

encoder/decoder

1000Base-T

UTP

Cat 5e

IEEE 802.3ab

physical layer

14

Gigabit Ethernet has been defined in March 1996 by the working group IEEE

802.3z. The GMII represents a abstract interface between the common Ethernet
layer 2 and different signaling layers below. Two important signaling techniques
had been defines: The standard 802.3z defines 1000Base-X signaling which uses
8B10B block coding and the 802.3ab standard uses 1000Base-T signaling. The
latter is only used over twisted pair cables (UTP Cat 5 or better), while
1000BaseX is only used over fiber, with one exception, the twinax cable
(1000BaseCX), which is basically a shielded twisted pair cable.

BTW: The "X" stands for block coding.

14

GE Signaling

IEEE 802.3

Ethernet

802.2 LLC

802.3 CSMA/CD

802.3 PHY

(C) Herbert Haas

PHY

2005/03/11

IEEE 802.3z

Gigabit Ethernet

IEEE 802.2 LLC

CSMA/CD

or full duplex MAC

Reconciliation Sublayer

PCS

PMA

PMD

ANSI X3T11

Fibre Channel

FC-4

upper layer mapping

FC-3

common services

FC-2

signalling

FC-1

encoder/decoder

FC-0

interface and media

Gigabit Ethernet layers have been defined by adaptation of the LLC and MAC
layers of classical Ethernet and the physical layers of the ANSI Fiber Channel
technology. A so-called reconciliation layer is used in between for seamless
interoperation. The physical layer of the Fiber Channel technology uses 8B10B
block coding.

15

15

GE 8B/10B Coding

GMII

256 code groups

8 x 125 Mbit/s

PCS

1

1111111

Only used

by

1000BaseX

1024 code groups

8B/10B Encoder/Decoder

125 million code

groups per

second

11111111 11

1250 Mbaud

(C) Herbert Haas

2005/03/11

PMA

16

8B10B block coding is very similar to 4B5B block coding but allows fully
balanced 10-bit codewords. Actually, there are not enough balanced 10-bit
codewords available. Note that there are 256 8-bit codewords which need to be
mapped on 1024 10-bit codewords. But instead of using a fully balanced 10-bit
codeword for each 8-bit codeword, some 8-bit codewords are represented by two
10-bit codewords, which are sent in an alternating manner. That is, both
associated 10-bit words are bit-complementary.

Again, the signaling overhead is 12.5%, that is 1250 Mbaud is necessary to
transmit a bit stream of 1000 Mbit/s.

16

1000BaseX

 Two different wavelengths supported
 Full duplex only

 1000Base-SX: short wave, 850 nm MMF
 1000Base-LX: long wave, 1300 nm MMF or SMF

 1000Base-CX:

 Twinax Cable (high quality 150 Ohm balanced

shielded copper cable)

 About 25 m distance limit, DB-9 or the newer

HSSDC connector

(C) Herbert Haas

Gigabit Ethernet can be transmitted over various types of fiber. Currently (at
least) two types are specified, short and long wave transmissions, using 850 nm
and 1300 nm respectively. The long wave can be used with both single mode
(SMF) and multimode fibers (MMF). Only SMF can be used for WAN
transmissions because of the much lower dispersion effects.

Note that there are several other implementations offered by different vendors,
such as using very long wavelengths at 1550 nm together with DWDM
configurations.

The twinax cable is basically a shielded twisted pair cable.

2005/03/11

17

17

1000BaseT

 Defined by 802.3ab task force
 UTP

 Uses all 4 line pairs simultaneously for duplex

transmission! (echo cancellation)

 5 level PAM coding

• 4 levels encode 2 bits + extra level used for Forward

Error Correction (FEC)

 Signal rate: 4 x 125 Mbaud = 4 x 250Mbit/s data

rate

• Cat. 5 links, max 100 m; all 4pairs, cable must

conform to the requirements of ANSI/TIA/EIA-568-A

 Only 1 CSMA/CD repeater allowed in a

collision domain

(C) Herbert Haas

It is very difficult to transmit Gigabit speeds over unshielded twisted pair cables.
Only a mix of multiple transmission techniques ensure that this high data rate can
be transmitted over a UTP Cat5 cable. For example all 4 pairs are used together
for both directions. Echo cancellation ensures that the sending signal does not
confuse the received signal. 5 level PAM is used for encoding instead of 8B10B
because of its much lower symbol rate. Now we have only 125 Mbaud x 4
instead of 1250 Mbaud.

The interface design is very complicated and therefore relatively expensive.
Using Cat 6 or Cat 7 cables allow 500 Mbaud x 2 pairs, that is 2 pairs are
designated for TX and the other 2 pairs are used for RX. This dramatically
reduces the price but requires better cables, which are not really expensive but
slightly thicker. Legacy cable ducts might be too small in diameter.

2005/03/11

18

18

Several Physical Media Supported

Logical Link Control LLC

MAC Control (optional)

Media Access Control MAC

PLS

AUI

PMA (MAU)

MDI

Medium

1-10 Mbit/s 10 Mbit/s 100 Mbit/s 1000 Mbit/s

AUI Attachment Unit Interface, PLS Physical Layer Signaling, MDI Medium Dependent Interface
PCS Physical Coding Sublayer, MII Media Independent Interface, GMII Gigabit Media Independent

Interface, PMA Physical Medium Attachment, MAU Medium Attachment Unit, PMD Physical Medium
Dependent

Reconciliation Reconciliation Reconciliation

MII

PLS

AUI

PMA

MDI

Medium Medium Medium

MII

PCS

PMA

PMD

MDI

GMII

PCS

PMA

PMD

MDI

Data Link Layer

PHY

(C) Herbert Haas

2005/03/11

The diagram above shows various physical media designs supported by the
official GE standard. Each modern GE card could theoretically support the old
10 Mbit/s standard as well. However many vendors create GE NICs that only
support GE or GE and FE—who would connect a precious GE interface with
another interface, which is 100 times slower?

19

19

10 Gigabit Ethernet / IEEE 802.3ae

 Only optical support

 850nm (MM) / 1310nm /1550 nm (SM only)
 No copper PHY anymore !

 Different implementations at the

moment – standardization not finished!

 8B/10B (IBM), SONET/SDH support, …
 XAUI ("Zowie") instead of GMII

(C) Herbert Haas

10 GE only supports optical links. Note that GE is actually a synchronous
protocol! There is no statistical multiplexing done at the physical layer anymore,
because optical switching at that bit rate only allows synchronous transmissions.

The GMII has been replaced (or enhanced) by the so-called XAUI, known as
"Zowie".

Note: At the time of writing this module, the 10 GE standard was not fully
finished. Though, some vendors already offer 10 GE interface cards for their
switches.

These interfaces are very expensive but the investment ensures backward
compatibility to lower Ethernet rates and at the same time provides a very high
speed WAN interface.

An alternative technology would be OC192, which requires a very expensive and
complex SONET/SDH environment.

2005/03/11

20

20

Note

 GE and 10GE use synchronous

 Recommendation: Don't use GE over

physical sublayer !!!

copper wires

 Radiation/EMI
 Grounding problems
 High BER
 Thick cable bundles (especially Cat-7)

(C) Herbert Haas

Both GE and 10GE are synchronous physical technologies on fiber. It not
recommended to use GE over copper wires anymore although 802.3ab would
specify it. This is because the whole electrical hardware (cables and connectors)
are re-used from older Ethernet technologies and have not been designed to
support such high frequencies.

For example the RJ45 connector is not HF proof. Furthermore, shielded twisted
pair cables require a very good grounding, seldom found in reality. The Bit Error
Rate (BER) is typically so high that the effective data rate is much lower than
GE, for example 30% only.

2005/03/11

21

21

Summary

 Ethernet evolved in the opposite direction:

 Collision free
 WAN qualified
 Switched

 Several coding styles  Complex PHY

architecture

 Plug & play through autonegotiation
 Much simpler than ATM but no BISDN

solution – might change!

(C) Herbert Haas

2005/03/11

22

22

Quizz

 Why tends high-speed Ethernet to

synchronous PHY?

 Can I attach a 100 Mbit/s port to a

1000 Mbit/s port via fiber?

 What is the idea of Etherchannels?

(Maximum bit rate, difference to
multiple parallel links)

(C) Herbert Haas

2005/03/11

23

23

Hints

 Q1: On fiber its difficult to deal with

 Q2: No, autonegotiation on fiber

 Q3: "normal" parallel links would be

asynchronous transmission, photons
cannot be buffered easily, store and
forward problems

does not care for data rates

disabled by STP, Etherchannel
supports up to 8 links

(C) Herbert Haas

2005/03/11

24

24