Evolution of Ethernet (Toward Gigabit and above)

1
Evolution of Ethernet (Toward Gigabit and above)
2
0. Contents

• Materials to cover
• 1. Ethernet history
• Part1. Ethernet Evolution in LAN
• 2. Media (From Shared to Dedicated)
• 3. Bandwidth (From Shared to Dedicated)
• 4. shift from half duplex to full duplex links
• 5. flow control
• Part 2. Gigabit Ethernet
• 6.1. Full-duplex Gigabit Ethernet
• 6.2. Half-duplex Gigabit Ethernet
• Part 3. 10Gigabit Ethernet
• 7. 10G Ethernet
• Reference (in Part 1 and 2)
• Gigabit Ethernet, written by Rich Seifert
  published by Addison Wesley

3
0. Contents

• Questions.
• How the LAN environment has changed?
• What are half-duplex and full-duplex Ethernet?
• Why they gave up half-duplex?
• Performance
• Cost
• How the full-duplex Ethernet evolves to 10Gbps?

4
1. Ethernet History

• 1973-82 the creation of Ethernet
• 1973 Ethernet at 3Mbps first implemented in a
  lab at Xerox Co. by Dr. Robert Metcalfe
• 1980 DIX(Dec, Intel and Xerox) cartel developed
  and published the standard for a 10Mbps Ethernet.
• for thick multi-drop coaxial cable only
• 1982-1990 10Mbps Ethernet reaches maturity.
• Widespread adoption of structured wiring systems
  using unshielded twisted-pair cabling.
• 1983 June IEEE Standards Board approved
  IEEE802.3
• 10BASE-T(Ethernet over twisted pair) approved.

5
1. Ethernet History

• 1983-1997 LAN bridging and switching
• 1984 First transparent bridge LAN were
  developed by DEC.
• 1990 IEEE802.1D standard for vendor-interoperabl
  e interconnection of LANs using bridges
• 1991 Kalpana Corporation developed a new class
  of Ethernet bridge, called LAN switch which a
  number of attachment can be attached at the full
  capacity of each link simultaneously.
• 1992-1997 Fast Ethernet
• 1995 Fast Ethernet - high speed version of
  Ethernet - standard spawned.
• 1996- Gigabit Ethernet
• 1998 standard for 1000Mbps Ethernet was
  approved.

6
1. Ethernet History

• 2000- 10G Ethernet
• March 1999 IEEE 802.3ae task force organized
• Sept. 2000 First draft released
• The first half of 2002 pre-standard product
  expected
• 2002 publish of standard expected

7
1. Ethernet History

• Why Ethernet has been so popular ?
• the first LAN announced
• most common
• low cost (cost/bit) In LAN, the cheapest
  approach will usually wins.
• due to simple structure the least number of
  states involved ? low implementation cost
• no priority control(all communication is fair)
• no central control station required in Ethernet
• no legal barrier At the middle of 1970s the
  Ethernet inventor Dr. Robert Metcalfe, the
  original inventor and Xerox effectively
  relinquished its patent right and the name
  Ethernet.

8
1. Ethernet History

• Nomenclature of Ethernet media n-signal-phy
• n data rate in Mbps e.g., 10,100,1000
• signal BASE meaning baseband or BROAD meaning
  broadband
• BASE physical medium is dedicated to the
  Ethernet
• BROAD physical medium can simultaneously
  support Ethernet and other possible non-Ethernet
  services.
• phy
• for the first few systems (This is exceptions)
  maximum length of cable segment in meters
• 10BASE2, 10BASE5, 10BROAD36
• for the later systems the nature of physical
  medium
• T twisted pair, T4 4 pairs of T, T2
  2 pairs of T
• F fiber FL asynchronous active hub,
  FP passive hub, FB synchronous active hub
• (continue)

9
1. Ethernet History

• phy (continued)
• 1000Mbps Ethernet
• 1000BASE-X generic designation of 1000Mbps
  system using 8B/10B encoding
• 1000BASE-CX 25m maximum 2 pairs of 150 ? STP
• 1000BASE-SX optical fibers for
  short-wave(850nm, multi-mode) lasers
• 1000BASE-LX optical fibers for
  long-wave(1300nm, multi and single mode) lasers.

10
1. Ethernet History

• Getting to 1 Gigabit (very important)
• Media (From Shared to Dedicated)
• Bandwidth (From Shared to Dedicated)
• shift from half duplex to full duplex links
• new explicit flow control
• To reduce buffer overflow on dedicated channels
  and full-duplex operation
• Buffer overflow increases with speed of
  communication channel.
• medium-independent interfacing
• At first, it was needed due to hardware limit
• Drivers(transceivers) need to be close to network
  Interface (links),
• Logical circuits needs to close to computer in
  the comfortable environment
• to support variety of underlying physical media
• (Drivers tends to be unified, logical circuits
  changes to software.)
• automatic link configuration
• Determine and configure device capabilities
  automatically.

11
Part I Ethernet Evolution in LAN
12
2. Media (From Shared to Dedicated)

• Why coaxial cable in the first place ?
• It is easy to design a high-speed LAN using
  coaxial cable than twisted-pair.
• It has less susceptible to noise ingress (EMI
  Electro-Magnetic Interference susceptibility).
• Bandwidth capability far exceeds to UTPs.
• Impedance control is much easier, thus it is much
  easier to design transceiver circuitry.
• A shared medium is perfectly suitable for a
  shared-bandwidth LAN.
• (In 1980, there was little pre-installed building
  that provides with data-grade twisted-pairs)

13
2. Media (From Shared to Dedicated)
14
2. Media (From Shared to Dedicated)

• Structured wiring
• objective
• Allow a building to be wired in a generic manner,
  without consideration of what equipment would be
  later selected and installed
• direct motif (advanced) demand for new
  regulation about the cabling
• Bell System telephone monopoly was broken.
• All telephone equipment right up to the desk set,
  was owned and controlled by Bell.
• The deregulation(breaking monopoly) created new
  market of PBX(private branch exchange). The new
  PBX vendors may install links in various ways.

15
2. Media (From Shared to Dedicated)

• Structure of structured wiring
• horizontal distribution cable between work
  position and hub in wiring closet on each floor
• using category 5 UTP(Unshielded Twisted Pair) or
  legacy category 3 UTP.
• star topology Each work position is provided
  with at least one dedicated cable that runs from
  the work positions to a wiring closet. It is not
  shared or daisy-chained.
• vertical distribution cable connecting hubs
  using optical fiber(2km) or UTP(100m) ...

16
2. Media (From Shared to Dedicated)
17
2. Media (From Shared to Dedicated)

• Advantage of structuring-wiring from star-wired
  topology
• ease of re-configuration executing moves,
  addition and change to configuration.
• fault isolation with unitary granularity
• A fault on a segment(horizontal distribution
  cable) damages only a single user.
• easy of network management
• Most necessary facilities for management is in
  the control box.
• Disadvantage of star-wired topology
• expensive
• extended cabling length dedicated medium
• Cannot use coaxial cable
• extra equipments required like hubs, switches,
  amplifiers
• UTP is a terrible medium for high-speed
  communication systems.

18
2. Media (From Shared to Dedicated)

• As a result of de facto standard of structured
  wiring
• We take the unique topology of LAN as star.
• LAN technology independence The same
  infra-structure can support Ethernet, Token Ring,
  LocalTalk, at different speed.

19
2. Media (From Shared to Dedicated)

• 10BASE-T/100BASE-T revolution
• The advantages of the structured wiring system
  far outweigh the cost penalty.
• Fast Ethernet offered no shared-media option.
• Since then, bandwidth of LAN is shared, but the
  transmission medium is dedicated.
• Category 5 UTP in Gigabit Ethernet
• Currently the de-facto standard of wiring work
  locations(desktops) is UTP wires.
• Category 5 UTP can support 100Mbps with maximum
  allowed 100m radius.
• Extending Category 5 UTP with four
  pairs(1pair/250Mbps) at 1000Mbps.

20
2. Media (From Shared to Dedicated)

• Building and campus backbone is Optical fiber.
• The major medium of choice for the vertical
  distribution today is 62.5/125 ?m multi-mode
  optical fibers.
• It can support 2km of 62.5?m multi-mode in 10M,
  100M Ethernet.
• If longer distance is required, use single-mode
  fiber.
• Coaxial cable becomes historical artifact in
  structured wiring system.
• Since 100Mbps LAN, coaxial cable is not used any
  more!
• Coaxial cable is (only) used in cable network in
  data communication networks.

21
3. Bandwidth (From Shared to Dedicated)

• Increasing throughput(LAN capacity)
• 1. Bandwidth dedicated LAN,
• 2. Increased transmission speed.
• LAN segmentation

22
3. Bandwidth (From Shared to Dedicated)

• collision domain
• Definition a set of Ethernet stations to create
  collision contending for access to a shared LAN
• segmentation a segment is collapsed into many
  sub-segments.
• Micro-segmentation an extreme segmentation with
  each segment attaching a single
• end
  station in full duplex mode,
• there is no collision between end stations, and
• each end station has dedicated bandwidth
  (transmission and reception simultaneously).

23
3. Bandwidth (From Shared to Dedicated)

• Advantage of switched LAN
• extended distance limitation total limit
  limit1 limit2
• increased aggregate capacity ?port1,n
  DataRateport
• increased data rate flexibility
• Links of different data rate can be merged with a
  switched LAN.
• Deployment of gigabit Ethernet.
• Shared Gigabit Ethernet are locally used(used as
  a subset of LAN).
• Usage for primarily in MAN(Metropolitan Area
  Network) or in campus backbone.

24
4. Full Duplex Ethernet

• Full duplex operation in CSMA/CD
• Full duplex operation frees the Ethernet MAC in a
  collision domain.
• no more CSMA/CD
• no more MA(multi-access)
• No more shared medium, multiplexing is
  unnecessary.
• no collision occurs in full duplex mode.
• No more CS(carrier sense) required
• No more CD(collision detection) required

25
4. Full Duplex Ethernet

• How full-duplex is possible?
• Shift to dedicated media by the use of structured
  wiring systems
• Micro-segmentation provided by switches
• three conditions to be met to use an Ethernet in
  full-duplex mode
• micro-segmentation There can be only two
  devices on a segment.
• Each node must be capable of supporting
  simultaneous transmission and reception without
  interference.
• The network interfaces must be capable of and
  configured to use full duplex mode.

26
4. Full Duplex Ethernet
Fig 4.3. An implementation example of full-duplex
switched LAN
27
4. Full Duplex Ethernet

• Transmitter operation in full duplex mode
• Transmit packets without worrying about collision
  
• minimum spacing between frames. breathing room
  to perform necessary housekeeping chores like
  buffer management, interrupts handling, updating
  network management statistics...
• Receiver operation in full duplex mode (read)
• same as the half duplex mode
• Check 1. Start-of-Frame, 2. Destination Address,
  3. CRC, 4. minimum valid frame length, ..
• Implication of full duplex operation
• It eliminates the function of MAC(CSMA/CD)
• thus the link length restrictions of CSMA/CD.
• It increases the aggregate channel capacity.
• with the help of (central) switch.

28
4. Full Duplex Ethernet
switching hub
Fig 4-8. An example of switch-to-switch
connection (a mixture of full-duplex and
half-duplex links)
29
4. Full Duplex Ethernet

• Minimum frame size constraint (512bits) in half
  duplex mode
• Transmitting node must notice collision of its
  own frame during transmitting if it occurs.
• If it does not notice, the collided frame is
  mistakenly regarded as successful in sending
  node.
• This results in the same effect as packet
  transmission loss.
• To prevent from this effect (late collision),
• Frame size should be at least as long as the
  worst-case round-trip propagation delay of the
  network, plus the time to transmit the 32-bit jam
  signal.

30
4. Full Duplex Ethernet

• Minimum frame size constraint in full duplex
  mode
• In full duplex mode, we do not need minimum frame
  size for proper CSMA/CD operation.
• But sustaining this constraint enables unified
  standard.
• no further options or specification seamless
  bridging between full duplex and half duplex mode
  Ethernet, and
• backward compatibility no modification of
  software in upper layers

31
Ethernet Frame Format

• IEEE 802.3 frame format (read)
• bit ordering LSB first
• byte ordering from first byte to last
• preamble(8 octets) 5516 5516 5516 5516 5516
  5516 5516
• SFD D516
• Converting 9 octets above to streaming bits,
  1010101 010 1010 10101011
• A sequence of 102 repeat until 112 appears at the
  last two bits.
• DA unicast address or multicast address
• SA unicast address only
• type indicates the higher-layer protocols,
  enabling upward multiplexing (DIX format)
• length size of data field in byte (1983-1996
  format)
• type/length the value of type between 0 to 1500
  is reserved, which are valid range for length
• FCS

32
5. Ethernet Flow Control

• Transmission loss versus buffer overflow
• Frame loss due to buffer congestion is much
  greater than transmission loss.
• Ethernet specifies a bit error rate(BER) of 10-8
  in worst case.
• This corresponds to frame loss rate of 10-6 or
  10-5.
• Why flow control in full duplex mode?
• increased link speed As link speed increases,
  buffer overflow increases.
• unlimited transmission freedom because MAC
  disappears.
• In half-duplex mode, flow rate of each node is
  controlled by MAC
• In full-duplex mode, each node can transmit data
  as fast as it wishes.
• Switch environment limited link capacity

33
5. Ethernet Flow Control

• How to solve Buffer overflow problem ?
• indirect solution end-to-end flow control
• direct solution link flow control at the link
  layer
• need new protocol (new protocol stack)
• How to prevent buffer overflow problem in
  CSMA/CD LAN? (advanced)
• backpressure to forestall incoming traffic
  utilizing MAC algorithm
• backpressure To achieve a goal, do something
  legal that is not provided for that purpose.
• Force collisions with incoming frames
• not recommended because it confuses the
  management layer.
• False Carrier Sense Make appear as if the
  channel is busy.
• A congested node scan preambling signal until
  congestion resolves.
• As BOF(beginning of frame byte) does not follow,
  all the other nodes receive nothing.

34
5. Ethernet Flow Control

• Explicit flow control in full-duplex networks
• Full duplex Ethernet uses explicit flow control
  PAUSE
• False Carrier sense cannot be used because no CS
  nor CD function is used in full-duplex Ethernet.

35
5. Ethernet Flow Control

• MAC control sub-layer
• IEEE 802.3x committee defines an optional layer
  MAC control to specify a more generic
  architectural framework for the control for the
  Ethernet MAC.
• PAUSE one of functions defined in MAC control
  sub-layer
• This is very important function in Distributed
  buffer, which will be mentioned later.

36
5. Ethernet Flow Control

• PAUSE function
• Purpose Prevent switches or receiving nodes
  from unnecessary discarding frames due to
  short-term transient buffer overflow conditions
• Remarks
• It only provides stop-start form of flow
  control.
• It is defined across full-duplex link only. (It
  does not provide end-to-end flow control)
• PAUSE command can be issued during data transfer
  on the opposite direction
• The central switch issues(requests) and confirmed
  by the terminals(e.g., PC)

All link capacity is 1G
1.5G 0G 0G
0.5G 0G 0G
0.5G 0.5G 0.5G
PAUSE state PUASE state 0.5G
PAUSE_req PAUSE_req
switch
switch
switch
(a) Flow control required.
(b) PAUSE request issued
(c) PAUSE state
37
5. Ethernet Flow Control

• PAUSE operation
• Syntax PAUSE (term)
• The station that receives a PAUSE frame, stops
  sending data frames for the period (term).
• PAUSE operation is done in frame basis.
• So the frame en route is not affected, but later
  frame will be paused.
• regulation of pause time overriding
• The latest PAUSE command overrides all previous
  PAUSEs.
• The command PAUSE can selectively pause a
  designated sender.
• Examples
• Case 1 PAUSE 0
• Cancel PAUSE state by issuing another PAUSE
  command with (term0).
• Case 2 PAUSE 10, then immediately PAUSE 100
• Prolong PAUSE state by issuing another PAUSE with
  enough term.
• PAUSE 10 is useless in this example.

38
Part II Gigabit Ethernet
39
6. Gigabit Ethernet

• Higher layer software and interfaces
• Gigabit Ethernet is a data link and physical
  layer technology only.
• Gigabit Ethernet requires no changes to
  higher-layer protocols.
• All types of Ethernet provides connectionless
  unacknowledged service.
• LLC on top of Ethernet uses connectionless
  unacknowledged service.

40
6.1. Full-Duplex Gigabit Ethernet

• Background of LAN switches
• The core technology in ATM is switch, which was
  developed at the beginning of 1990s.
• The developed technology were applied to LAN
  switches on the middle of 1990s.
• The LAN switch cost sharply went down, and LAN
  switch becomes in common after that.

41
6.1. Full-Duplex Gigabit Ethernet

• Full duplex Gigabit Ethernet
• for LAN campus backbone
• dedicated bandwidth high throughput and
  expensive
• based on LAN switches technology
• Performance enhancement comes from switches.
• A switching hub uses dedicated medium and
  simultaneous forwarding(switching) fabrics.
• MAC virtually disappears.
• There is no need of CS nor CD, thus no Carrier
  Extension nor Frame Burst.
• very simple and easy to understand
• Instead, it should employ a stringent buffer
  control to handle bursty traffic.
• Cost Switching hubs are more expensive than
  repeating hubs.
• Distributed Buffer a special(limited) switch
  between full and half-duplex (advanced)

42
6.2. Preliminaries of Half-Duplex MAC

• Advantage of half duplex disappears in 1000Mbps
  LAN. (important)
• Time loss reduces significantly due to early
  interruption of transmission. In other words, if
  transmitter notices collision during transmitting
  the early part of packet.
• At 10, or 100 Mbps, most of collisions occurs at
  the early part of transmission.
• At 1000Mbps, collision may happen at the middle
  or even later part of transmission.
• (See next slide)

43
6.2. Preliminaries of Half-Duplex MAC
44
6.2. Preliminaries of Half-Duplex MAC

• Half-duplex Ethernet MAC flow (transmission)
  (read)

Transmit Process
Assemble frame.
Send jam.
channel busy
Increment attempts.
yes
no
yes
Wait interframe gap.
too many attempts
no
Start transmission.
Compute backoff.
no
yes
collision detected
Wait backoff time.
no
Tx done
yes
unsuccessful transmission (excessive collisions)
successful transmission
45
6.2. Preliminaries of Half-Duplex MAC

• Half-duplex Ethernet MAC flow (reception) (read)

Receive Process
channel active
no
yes
Start receiving.
channel still active
valid frame check sequence
yes
no
yes
no
Received frame too small? (collided fragment)
yes
extra bits
no
yes
no
Recognize address
no
yes
successful transmission
Receive alignment error
Receive frame check error
46
6.2. Preliminaries of Half-Duplex MAC

• Transmission in half duplex mode (advanced)
• Carrier Sense(CS) Before sending a frame in its
  own Tx buffer, check any other node is using the
  shared medium.
• preamble For easy and safe reception, a fixed
  sequence is transferred before main transmission
• inter-frame gap When transmission of a frame is
  successfully done, wait a little time to perform
  necessary housekeeping functions such as 1.
  adjusting buffer pointers, updating management
  counters, interrupting a host CPU, ..
• Collision Detection(CD) collision arises when
  two nodes try to send their frames, then the
  transmitting nodes notice collision and stop
  transmission of frame.
• jamming To completely destroy the collided
  frame, jamming signal is sent after collision.
• (continue)

47
6.2. Preliminaries of Half-Duplex MAC

• (continued)
• back-off the stations which aborts the
  remainder of its frame wait for a random period
  of time.
• exponential back-off at the jth trial (after
  (j-1) collisions) random variable r is chosen.
• 0 lt r lt 2 min(j,10)
• If two frames collides, and two nodes choose
  different rs, the contention resolves.

48
6.2. Preliminaries of Half-Duplex MAC

• Minimum time to detect collision in half duplex
  Ethernet
• b propagation delay between two points.
• a maximum propagation delay in network
• maximum time to detect collision between two
  points in bus structure, or dual cable structure
  2b
• maximum time required to detect collision in bus
  structure, dual cable structure
• (round-trip propagation delay) 2a
• In unidirectional dual cable structure, 2a means
  four times the propagation delay from the headend
  to the termination resistor.

t1.
t1?.
t12?.
minimum time to detect collision in bus
49
6.2. Preliminaries of Half-Duplex MAC

• Change of maximum LAN size in half duplex
  Ethernet (important)
• All Ethernet defines the minimum frame size
  64bytes (all including except preamble signal)
• To calculate the maximum LAN size for different
  link speed, we get the following.
• 10M Ethernet 2.8km ? sufficient!
• 100M Ethernet 200m
• Due to structured cabling, length between a
  terminal and a central hub is much shorter than
  multi-drop 10M Ethernet.
• ? not sufficient but still admissible
• 1000M Ethernet 20m (?) ? no more admissible
• If minimum size increases to 512bytes size limit
  becomes 160m(admissible)

50
6.3. Half-Duplex Gigabit Ethernet

• Late collision detection
• Definition A transmitting node receives
  collision signal after it finishes Tx.
• The node A transmits a packet PA.
• PA collides with PB sent from other node B.
• A finishes transmission of PA.
• After finishing transmission A receives collision
  signal caused from its packet PB.
• Why late collision is harmful?
• Sender thinks transmission is successful, but
  receiver cannot receive the frame.
• This results in loss of a packet.
• How to prevent from late collision ?
• Satisfy the next condition
• (transmission time of minimum length of a frame)
  gt (maximum round-trip propagation delay of
  network)

51
6.3. Half-Duplex Gigabit Ethernet

• How to avoid late collision?
• Increase the minimum frame size.
• If there are many short frames, throughput
  decreases sharply.
• Decrease network size limit cases to apply to.
• Slow down network speed out-of-date concept !
• Solutions in Gigabit Ethernet
• carrier extension, and
• frame burst (optional)
• Common rules for two solutions
• Let the upper layers do not notice change of MAC
  layer.
• Do not change minimum packet length ? backward
  transparency or software transparency.
• minimum frame size still 64 octets (This
  constraint applies to 10, 100Mbps Ethernet)

52
6.3. Half-Duplex Gigabit Ethernet

• Carrier extension
• objective
• Add the encoding signal(that is not known to the
  upper layers) to satisfy the minimum transmission
  time without increasing minimum packet length.
• minimum frame size still 64 octets (this is
  rule of 10, 100Mbps Ethernet)
• Minimum frame transmission time 512-octet time.
• action
• If the frame size is less than 512 octets, attach
  carrier extension after transmitting the frame.
• performance
• Transmission of small frames causes major
  degradation of LAN performance

53
6.3. Half-Duplex Gigabit Ethernet

• Frame burst (optional)
• objective Increase transmission packet length
  by using a special frame format called
  multi-frame without increasing minimum packet
  size.
• minimum frame size still 64 octets (this is
  rule of 10, 100Mbps Ethernet)
• Minimum multi-frame length 512-octet time.
• Action
• If short frame arrive at the transmitter, store
  it in the transmitting buffer until some other
  frame arrives.
• Then send them in a multi-frame format at once.
• Multi-frame format that can store up to six
  frames.
• Performance Delay increases.

54
6.3. Half-Duplex Gigabit Ethernet

• Format (carrier extension/frame bursting)
  (read)

8
6
6
2
46-493
4
448-1 (bytes)
preamble /SFD
destination address
source address
length /type
data
FCS
carrier extension
minimum 64 bytes
Carrier extended frame (64-511bytes)
8
6
6
2
494-1500
4 (bytes)
preamble /SFD
destination address
source address
length /type
data
FCS
Carrier non-extended frame (?512bytes)
preamble /SFD
MAC frame 1
extension (if needed)
IFG
preamble /SFD
MAC frame 2
IFG
MAC frame n
preamble /SFD
IFG
n?6
8192 bytes (maximum time to start of last frame
in burst)
Frame bursting
55
6.3. Half-Duplex Ethernet MAC

• Why half duplex was proposed ?
• for quick standardization and quick market
  delivery and large demand
• Committee members know the major part of Gigabit
  Ethernet is full-duplex.
• But they want to utilize
• IEEE802 committee (the most well-known authority
  in LAN), and
• use the acquainted name CSMA/CD(Ethernet)

56
Part III 10Gigabit Ethernet
57
7.1. why 10GE ?

• Why Ethernets in 10Gbps market?
• 10Gbps Ethernet keeps the tradition of Ethernet
  and we expect the same success as Ethernet series
  did.
• Easy migration to higher performance
• How to increase bandwidth without disrupting the
  existing network?
• 10M, 100M, 1G Ethernet then 10G Ethernet is
  natural.
• Low cost of ownership including both
  acquisition and support costs
• cheaper than the most strong competitor
  POS(Packet over SONET)
• Familiar management tools and common skills base
• Full ranges of Ethernet uses common management
  protocols like SNMP, RMON
• Ability to support new applications and data
  types
• Flexibility in network design

58
7.1. why 10GE ?

• 1. Easy migration to higher performance
• Easy and straightforward migration to higher
  performance levels without disruption.
• 10M,100M, 1G Ethernet has the same minimum and
  maximum length and the same format.
• 2. Low cost of ownership
• popularity of Ethernet
• High volume market guaranteed.
• supported by an aggressive merchant chip market
  that provides highly integrated silicon
  solutions.
• The Ethernet market tends to spawn highly
  competitive start-ups with each new generation of
  technology.
• design philosophy of the Ethernet industry
  low-cost design.
• Use low cost transmitter(e.g., vertical cavity
  surface emitting lasers VCSEL) in place of
  expensive10-Gbps telecommunication lasers
• Use inexpensive synchronous system rather than
  synchronous system
• Can use installed OC-192(9.95Gbps) infrastructure

59
7.1. why 10GE ?

• 3. Familiar management tools and common skills
  base
• Can use MPLS, SNMP, RMON, 802.xxx
• 4. Ability to support new applications and data
  types
• Direct result from above protocols like
• QoS(Quality of Service), CoS(Class of Service),
  caching, server load balancing, security,
  policy-based networking
• 5. Flexibility in network design
• Not only the speed of network increases, but also
  the network size increases.
• Two PHY types provided.
• LAN PHY
• WAN PHY (Optional extended operating feature
  added to LAN PHY.)

60
7.2. General Properties

• Standardization of 10Gbps Ethernet
• IEEEE 802.3ae WG starting from March 1999
  March 2002.
• Importance of 10Gbps
• overlapped technology in both LAN and MAN
• If LAN extends to 10G, it becomes MAN virtually.
• close speed to OC192(155.5 64 9.95Gbps)
• direct comparison between Ethernet and SONET
• speed bottleneck in electronic-based techniques
• It is known that it is economic to use DWDM than
  speedup in increasing link capacity.

61
7.2. General Properties

• Half/full duplex

LAN speed half duplex full duplex
10M O -
100M At first, only half duplex announced. Later full duplex added
1G O (but not practically used) O
10G - O
62
7.2. General Properties

• Comparison of Gigabit Ethernet and 10Gigabit
  Ethernet

attribute Gigabit Ethernet 10Gigabit Ethernet
typical usage campus LAN MAN
duplex half ? full duplex full duplex only
used medium optical copper optical only
coding 8B/10B 64B/66B
63
7.3. Structure

• Types of 10GE
• LAN PHY
• X 8b/10b Coding only for LAN
• R 64b/66b Coding, (without WIS)
• for UniPHY that unifies LAN PHY and WAN PHY
• WAN PHY
• W 64b/66b SONET scrambling, with WIS
• for WAN

64
7.3. Structure

• Structures of 10GE

65
7.3. Structure

• Structures of PHY (physical layer)
• PCS (Physical Coding Sublayer)
• 64b/66b Coding or Decoding
• packet delineation finding the boundary of
  packets
• WIS (WAN Interface Sublayer)
• conversion between 10GE and SONET 0C-192c
• PMA (Physical Media Attachment)
• SerDes(serialization/de-serialization)
• Adjust the length of Inter-frame Gap to
  compensate the speed difference between MAC and
  PMD in case of WAN.
• PMD (Physical Media Dependent interfaces)
• providing physical connections to the wire
• tranceiving and clock recovery
• consideration of DWDM (multiplexing)
• o/e (optical/electric) or e/o conversion

66
7.4. Added Requirements for MAN

• MAN requires more than LAN such as the following
  things.
• Packet classification
• Classify the input packet into one of classes by
  reading header (IP and port address, TOS(type of
  service) field, )
• used in connectionless service (like in Ethernet)
• Flow control, QoS and tariff base on packet
  classification.
• Per-port flow control
• the way to keep the user access rate under the
  level in contract
• The rate a port can use is usually less than the
  link speed.
• MAN needs to define(make a contract) the flow
  rate that a user can allocate.
• can be defined by SLA(Service Level Agreement) in
  Differentiated Service(Diff-Serv).
• QoS (Quality of Control)
• One way is to attach IEEE802.1p tag and check
  three bit priority field in it.

67
7.4. Added Requirements for MAN

• Restoration
• the ability that network provides with a new path
  in case that a part of network equipment under
  service is damaged
• generally required transit time lt 50msec
• Currently there is no restoration in 10GE.
• This can be provided MAN level MAC called
  RPR(resilient packet ring).

68
7.4. Added Requirements for MAN

• Asynchronous/synchronous system comparison
• Depending on whether the system maintains(has) a
  global clock
• Synchronous system SONET/SDH
• Asynchronous system LAN(Ethernet), PDH
• Synchronous system
• Can avoid timing drift between transmission and
  reception equipments
• Can reduce transmission error due to stable
  clocking.
• expensive
• suitable for synchronous multiplexing
• Asynchronous system
• Suitable for asynchronous multiplexing like LANs.
• cheap
• (There are many views of asynchronous/synchrono
  us system. This is only one of them.)
• Another example of this classification two areas
  of IMT-2000.
• Synchronous system using satellite leaded by USA
  and
• asynchronous system leaded by Europe.

69
7.5. Usage

• Usage
• as a more powerful wide-range LAN
• to cover a office building or to cover a campus
• as a fast link in the MAN over dark fibers
• dark fibers fibers installed but not used yet
• expense(rate) for using dark fiber much cheaper
  than using newly installed fibers
• topology mesh using point-to-point connection
• as a fast link to connect to the MAN CWDM ring
• CWDM Course WDM(Wavelength Division
  Multiplexing)
• similar to FDM(Frequency Div. Muxing) except
  frequency range and signal speed
• as a fast link to connect to the backbone(core)
  DWDM in WAN
• DWDM Dense WDM
• currently available up to OC-192(9.95Gbps) for a
  wavelength and 100 wavelengths for a fiber

70
7.5. Usage

• How and which way 10-G Ethernet is implemented.

10GbE over xxx properties over (non-DWDM) Fiber over DWDM over SONET over DWDM over SONET
network reliability unproven unproven high high
cost (cents/bit) normal depending upon DWDM devices high depending upon DWDM devices
capacity expandability lt10G expandable lt10G expandable
application - Traffic does not increase much. - Traffic increases much. Traffic does not increase much. - SONET is available. - for mission critical job (when reliability is more important than cost.) - if long(gt70km) segment requires.
71
8. Conclusion

• We reviewed the evolution of LAN from the point
  of Ethernet
• and additional requirements as 10GE in Metro.
• Also we examined how Ethernet has evolved and
  adapted to the changes of decades the modern
  technologies.
• Can you figure out the outline of 100GE?