Computer Networks :

Protocols and Practice

Part 5 : Datalink Layer

Olivier Bonaventure
http://inl.info.ucl.ac.be/

CNPP/2008.5. © O. Bonaventure 2008

These slides are licensed under the creative commons attribution share-alike license 3.0. You can obtain detailed
information
about this license at http://creativecommons.org/licenses/by-sa/3.0/
 Datalink layer

Point-to datalink layer

How to transmit and receive frames

Local area networks

Optimistic Medium access control
ALOHA, CSMA, CSMA/CD, CSMA/CA

Ethernet networks

WiFi networks

Deterministic Medium access control

Token Ring, FDDI

CNPP/2008.5. © O. Bonaventure, 2008

2
 Usage of the physical layer

Service provided by physical layer

Bit transmission between nodes attached to the
same physical transmission channel
cable, radio, optical fiber, ...

Better service for computers

Transmission/reception of short messages
Service provided by the datalink layer

2 Datalink Datalink

1 Physical Physical

CNPP/2008.5. © O. Bonaventure, 2008

3
 Datalink layer

Goals
Provide a reliable transfert of packets although
Physical layer sends/receives bits and not packets
Physical layer service is imperfect
transmission errors
Losses of bits
Creation of bits

Packet Packet

Datalink Datalink

CNPP/2008.5. © O. Bonaventure, 2008

4
 Frame delineation

Frame
Unit of information transfer between two entities of
the datalink layer
sequence of N bits
Datalink layer usually supports variable-length frames

Packet Packet

Datalink Frames Datalink

Physical Bits Physical

How can the receiver extract the frames from the

received bit stream ?

CNPP/2008.5. © O. Bonaventure, 2008

5
 Frame delineation

Naïve solutions
Use frame size to delineate frames
Insert frame size in frame header
Issue
What happens when errors affect frame payload and frame
header ?

Use special character/bitstring to mark beginning/

end of frame
Example
all frames start with #
Issue
What happens when the special character/bitstring appears
inside the frame payload ?

CNPP/2008.5. © O. Bonaventure, 2008

6
 Character stuffing

Character stuffing
Suitable for frames containing an integer number
of bytes
'DLE' 'STX' to indicate beginning of frame
'DLE' 'ETX' to indicate end of frame
When transmitting frame, sender replaces 'DLE' by
'DLE' 'DLE' if 'DLE' appears inside the frame
Receiver removes 'DLE' if followed by 'DLE'

Example
Packet : 1 2 3 'DLE' 4
Frame
'DLE' 'STX' 1 2 3 'DLE' 'DLE' 4 'DLE' 'ETX'

CNPP/2008.5. © O. Bonaventure, 2008

Packet 1 : 1 2 3 'DLE' 4
Frame 1 : 'DLE' 'STX' 1 2 3 'DLE' 'DLE' 4 'DLE' 'ETX'
Packet 2 : 'DLE' 'STX' 'DLE' 'ETX'
Frame 2 : 'DLE' 'STX' 'DLE' 'DLE' 'STX' 'DLE' 'DLE' 'DLE' 'ETX'
Packet 3 : 'STX' 'DLE'
Frame 3 : 'DLE' 'STX' 'STX' 'DLE' 'DLE' 'DLE' 'ETX'

'DLE' 'STX' 1 2 3 'DLE' 'DLE' 4 'DLE' 'ETX' 'DLE' 'STX' 'DLE' 'DLE' 'STX' 'DLE' 'DLE' 'DLE' 'ETX' 'DLE' 'STX' 'STX' 'DLE' 'DLE'
'DLE' 'ETX'
 Bit stuffing

Alternative to character stuffing

Suitable for frames composed of n bits
01111110 used as marker at beginning and end of
frame
Sender behaviour
If five bits set to ʻ1ʼ must be sent, sender adds a bit set to
ʻ0ʼ immediately after the fifth bit set to ʻ1ʼ
Receiver behaviour
Counts the number of successive bits set to 1
6 successive bits set to 1 followed by 0 : marker
5 successive bits set to 1 followed by 0 : remove bit set to 0

Example
Packet : 0110111111111111111110010

Frame
01111110011011111011111011111011001001111110

CNPP/2008.5. © O. Bonaventure, 2008

8
 Frame delineation

Co-operation with physical layer

Some physical layers are able to transmit special
physical codes that represent neither 0 nor 1
Example : Manchester coding

1 0 1 0 0 1 1 0
InvHInvB

invH (or N times invH) could be used to mark the beginning

of a frame and invB (or N times invB) to mark the end of a
frame

CNPP/2008.5. © O. Bonaventure, 2008

9
 Frame delineation in practice

Most datalink protocols use

Character stuffing or bit stuffing
Character stuffing is preferred by software
implementations
A length field in the frame header
A checksum or CRC in the header or trailer to
detect transmission errors

A receiver frame is considered valid if

the correct delimiter appears at the beginning
the length is correct
the CRC/checksum is vlaid
the correct delimiter appears at the beginning

CNPP/2008.5. © O. Bonaventure, 2008

10
 PPP : Point-to-Point Protocol

Goal
Allow the transmission of network layer (IP but
also other protocols) packets over serial lines
modems, leased lines, ISDN, ...
Architecture
PPP is composed of three different protocols
1. PPP
transmission of data frames (e.g. IP packets)
2. LCP : Link Control Protocol
• Negotiation of some options and authentication
(username, password) and end of connection
3. NCP : Network Control Protocol
• Negotiation of options related to the network layer
protocol used above PPP
(ex: IP address, IP address of DNS resolver, ...)

CNPP/2008.5. © O. Bonaventure, 2008

11

W. Simpson and Editor. The point-to-point protocol (PPP). Request for Comments 1661, Internet Engineering Task Force, July 1994.

W. Simpson and Editor. PPP in HDLC-like framing. Request for Comments 1662, Internet Engineering Task Force, July 1994.

There is an older protocol called SLIP

 PPP (2)

PPP frame format

Flag Address Control Checksum Flag
01111110 11111111 00000011
Protocol Packet 16bits/32bits 01111110

Identification of the network layer packet

transported in the PPP frame

Mechanisms used by PPP

character stuffing for asynchronous lines
bit stuffing for synchronous lines
CRC for error detection
16 bits default but 32 bits CRC can be negotiated
No error correction by default
a reliable protocol can be negotiated
Data compression option
content of PPP frames can be compressed. To be negotiated at
beginning of PPP connection

CNPP/2008.5. © O. Bonaventure, 2008

12
 DataLink layer

Point-to datalink layer

How to transmit and receive frames

Local area networks

Optimistic Medium access control
ALOHA, CSMA, CSMA/CD, CSMA/CA

Ethernet networks

WiFi networks

Deterministic Medium access control

Token Ring, FDDI

CNPP/2008.5. © O. Bonaventure, 2008

13
 Local area networks

How to efficiently connect N hosts together ?

Ideally we would like to have a single cable on
each host while being able to reach all the others

Network topologies
Star-shaped network

Ring-shaped network

Bus-shaped network

CNPP/2008.5. © O. Bonaventure, 2008

14
 Local area networks

Problems to be solved
How to identify the hosts attached to the LAN ?
The LAN is a shared resource
How to regulate access to this shared resource to
provide :
fairness
All hosts should be able to use a fair fraction of the shared
resource
performance
The shared resource should be used efficiently

CNPP/2008.5. © O. Bonaventure, 2008

15
 Static allocation

Time Division Multiplexing

A D

B C

T
A B C DA
DA B C D A B C D A B C D A B C D A B C D A B C D

No suitable for a computer network

Leads to low link utilisation and high delays
Computers generate bursty trafic
A more adaptive access control mechanism is
required
CNPP/2008.5. © O. Bonaventure, 2008

16
 Medium access control

Hypotheses
N stations need to share the same transmission
channel
A single transmission channel is available
Definition
Collision
If two stations transmit their frame at the same time, their
electrical signal appears on the channel and causes a collision

Options
Frame transmission
A station can transmit at any time
A station can only transmit at specific instants
Listening while transmitting
A station can listen while transmitting
A station cannot listen while transmitting

CNPP/2008.5. © O. Bonaventure, 2008

17
 Medium access control
How to regulate access to the shared
medium ?
Statistical or optimistic solutions
hosts can transmit frames at almost any time
if the low is low, the frames will arrive correctly at destination
if the low is high, frames may collide
distributed algorithm allows to recover from the
collisions
Deterministic or pessimistic solutions
Collisions are expensive and need to be avoided
Distributed algorithm distributes authorisations to
transmit to ensure that a single host is allowed to
transmit at any time
avoids collisions when load is high, but may delay transmission
when load is low

CNPP/2008.5. © O. Bonaventure, 2008

18
 DataLink layer

Point-to datalink layer

How to transmit and receive frames

Local area networks

Optimistic Medium access control
ALOHA, CSMA, CSMA/CD, CSMA/CA

Ethernet networks

WiFi networks

Deterministic Medium access control

Token Ring, FDDI

CNPP/2008.5. © O. Bonaventure, 2008

19
 ALOHA

Problem
terminals need to exchange data with computer
on main island
terminal

Computer

Solution
Geostationary satellite
satellite antenna on each island
island sends frame to satellite that relays them
towards earth
CNPP/2008.5. © O. Bonaventure, 2008

20

ALOHA is discussed in
N. Abramson, The ALOHA system – another alternative for Computer Communications, Proc. Fall Joint Computer Conference, AFIPCS
Conference 1970
N. Abramson, Development of the Alohanet, IEEE Transactions on Information Theory, Vol IT-31, No. 3, pp. 119-123
 ALOHA (2)
Data transmission
Frame forwarded by satellite Frame sent by terminal
Terminal

Acknowledgement Computer
Terminal can check that its frame was forwarded
by listening to satellite channel
Acknowledgement allows to confirm correct
reception of data frame

CNPP/2008.5. © O. Bonaventure, 2008

21
 ALOHA (3)
How to organise frame transmission ?
If a host is alone, no problem
If two hosts transmit at the same time, a collision
will occur and it will be impossible to decode their
transmission

A
B

collision
CNPP/2008.5. © O. Bonaventure, 2008

22
 ALOHA (3)

Medium access algorithm

First solution

N=1;
while ( N<= max) do
send frame;
wait for ack on return channel or timeout:
if ack on return channel
exit while;
else
/* timeout */
/* retransmission is needed */
N=N+1;
end do
/* too many attempts */

CNPP/2008.5. © O. Bonaventure, 2008

23
 ALOHA (4)

Drawback
When two stations enter in collision, they may
continue to collide after

T0 T3 T3+timeout

A
B Collision

T1 T2 T2+timeout

How to avoid this synchronisation among stations ?

CNPP/2008.5. © O. Bonaventure, 2008

24
 ALOHA (5)

Improved algorithm

N=1;
while ( N<= max) do
send frame;
wait for ack on return channel or timeout:
if ack on return channel
exit while;
else
/* timeout */
/* retransmission is needed */
wait for random time;
N=N+1;
end do
/* too many attempts */

CNPP/2008.5. © O. Bonaventure, 2008

25
 Slotted ALOHA

How to improve ALOHA

Idea
Divide time in slots and force stations to only transmit
frames at the beginning of slots
requires a common clock heard by all stations
decrease risk of collisions since a slot is either
empty (no frame)
full (a single frame was transmitted)
in collision (more than one frame was transmitted)

A
B

collision

CNPP/2008.5. © O. Bonaventure, 2008

26
 Carrier Sense Multiple Access

How to improve slotted Aloha ?

Idea
Stations should be polite
Listens to the transmission channel before transmitting
Wait until the channel becomes free to transmit

Limitations
Politeness is only possible if all stations can listen to
the transmission of all stations
true when all stations are attached to the same cable, but not in
wireless networks

CNPP/2008.5. © O. Bonaventure, 2008

27
 CSMA

CSMA
Carrier Sense Multiple Access

N=1;
while ( N<= max) do
wait until channel becomes free;
send frame immediately;
wait for ack or timeout:
if ack received
exit while;
else
/* timeout */
/* retransmission is needed */
N=N+1;
end do
/* too many attempts */

CNPP/2008.5. © O. Bonaventure, 2008

28
 non-persistent CSMA

Idea
Transmitting a frame immediately after the end of
the previous one is a very aggressive behaviour
If the channel is free, transmit
Otherwise wait some random time before listening again
N=1;
while ( N<= max) do
listen channel;
if channel is empty
send frame;
wait for ack or timeout
if ack received
exit while;
else /* retransmission is needed */
N=N+1:
else
wait for random time;
end do

CNPP/2008.5. © O. Bonaventure, 2008

29
 p-persistent CSMA
Tradeoff between CSMA and non-persistent
CSMA

N=1;
while ( N<= max) do
listen channel;
if channel is empty
with probability p
send frame;
wait for ack or timeout
if ack received
exit while;
else /* retransmission needed */
N=N+1;
else
wait for random time;
end do

CNPP/2008.5. © O. Bonaventure, 2008

30
 Improvements to CSMA

Problems with CSMA

If one bit of a frame is affected by a collision, the
entire frame is lost

Solution
Stop the transmission of a frame as soon as a
collision has been detected

How to detect collisions ?

Station listens to channel while transmitting
If there is no collision, it will hear the signal it transmits
If there is a collision, is will hear an incorrect signal

CSMA/CD
Carrier Sense Multiple Access with Collision
CNPP/2008.5.
Detection © O. Bonaventure, 2008

31

CSMA/CD is described in
R. Metcalfe and D. Boggs. Ethernet: Distributed packet-switching for local computer networks. Communications of the ACM, 19(7):395--404,
1976. available from
http://www.acm.org/pubs/citations/journals/cacm/1976-19-7/p395-metcalfe/
 CSMA/CD

Medium access control

N=1;
while ( N<= max) do
wait until channel becomes free;
send frame and listen;
wait until (end of frame) or (collision)
if collision detected
stop transmitting;
/* after a special jam signal */
else
/* no collision detected */
wait for interframe delay;
exit while;
N=N+1;
end do
/* too many attempts */

CNPP/2008.5. © O. Bonaventure, 2008

32

The interframe delay is used to ensure that the electronics of the receiver can be synchronised to the transmitted signal. A typical interframe
delay is 9.6µsec
 CSMA/CD : Example

A B

1
Start of frame

2
Frame is propagated on LAN (5 microsecond per kilometer)

Frame stops at left side and first bit reaches B

4
Frame leaves the LAN
CNPP/2008.5. © O. Bonaventure, 2008

33
 CSMA/CD : Collisions

A B

1
Frame starts at A and B almost at the same time
Collision : at this point on the
shared medium, it is impossible
to decode the signal

A detects the collision and

stops transmitting its frame

CNPP/2008.5. © O. Bonaventure, 2008

34
 CSMA/CD : Collisions (2)

Advantages
Improves channel utilisation as stations do not
transmit corrupted frames
a station can detect whether its frame was sent
without collision
implicit acknowledgement if destination is up
when a collision is detected, automatic retransmission

Is it possible for a station to detect all

collisions on all its frames ?

CNPP/2008.5. © O. Bonaventure, 2008

35
 CSMA/CD : Collisions (3)

A B

1
A and B send a small frame almost at the same time

2
The two frame collide in the middle of the medium

3
A did not notice any collision
B did not notice any collision
They both consider that their frames were received correctly

CNPP/2008.5. © O. Bonaventure, 2008

36
 CSMA/CD : Collisions (4)

How to ensure that all collisions are detected ?

Worst case scenario

A B

1 Start of the frame sent by A

2
After τ seconds, Aʼs frame reaches B
A time τ−ε, B starts to transmit its own frame
B notices the collision immediately and stops transmitting

3
A detects collision at time τ+τ−ε

CNPP/2008.5. © O. Bonaventure, 2008

37
 CSMA/CD : Collisions (5)
How can a station ensure that it will be able
to detect all the collisions affecting its
frames ?
Each frame must be transmitted for at least a
duration equal to the two way delay (2∗τ)
As the throughput on a bus is fixed, if the two way delay
is fixed, then all frames must be larger than a minimum
frame size
Improvement
To ensure that all stations detect collisions, a station that notices
a collision should send a jamming signal

CNPP/2008.5. © O. Bonaventure, 2008

38
 Exponential backoff

How to deal with collisions ?

If the stations that collide retransmit together, a
new collision will happen

Solution
Wait some random time after the collision
After collision, time is divided in slots
a slot = time required to send a minimum sized frame
After first collision, wait 0 or 1 slot before retransmitting
After first collision, wait 0, 1,2 or 3 slots before retransmitting
After first collision, wait 0..2i-1 slots before retransmitting

CNPP/2008.5. © O. Bonaventure, 2008

39
 CSMA/CD with exponential backoff

Medium access control

N=1;
while ( N<= max) do
wait until channel becomes free;
send frame and listen;
wait until (end of frame) or (collision)
if collision detected
stop transmitting;
/* after a special jam signal */
k = min (10, N);
r = random(0, 2k - 1) * slotTime;
wait for r time slots;
else
/* no collision detected */
wait for interframe delay;
exit while;
N=N+1;
end do
/* too many attempts */

CNPP/2008.5. © O. Bonaventure, 2008

40
 CSMA with Collision Avoidance

Goal
Design a medium access control method suitable for
wireless networks
on a wireless network, a sender cannot usually listen to its
transmission (and thus CSMA/CD cannot be used)

Improvements to CSMA
Initial delay before transmitting if channel is empty
Extended Inter Frame Space (EIFS)
Minimum delay between two successive frames
Distributed Coordination Function Inter Frame Space (DIFS)
Delay between frame reception and ack transmission
Short Inter Frame Spacing (SIFS, SIFS< DIFS < EIFS)

CNPP/2008.5. © O. Bonaventure, 2008

41

CSMA/CA is used by 802.11, see

LAN/MAN Standards Committee of the IEEE Computer Society. IEEE Standard for Information Technology - Telecommunications and
information exchange between systems - local and metropolitan area networks - specific requirements - Part 11 : Wireless LAN Medium
Access Control (MAC) and Physical Layer (PHY) Specifications. IEEE, 1999. available from
http://standards.ieee.org/getieee802/802.11.html.

A short but detailed description of CSMA/CA may be found in

M. Schwartz, Mobile Wireless Communications, Cambridge University Press, 2005
 CSMA/CA (1)

Sender
N=1;
while ( N<= max) do
if (channel is empty)
{ wait until channel free during t>=EIFS; }
else
{ wait until endofframe;
wait until channel free during t>=DIFS; }
send data frame ;
wait for ack or timeout:
if ack received
exit while;
else
/* timeout retransmission is needed */
N=N+1;
end do
/* too many attempts */

CNPP/2008.5. © O. Bonaventure, 2008

42
 CSMA/CA (2)

Receiver

While (true)
{
Wait for data frame;
if not(duplicate)
{ deliver (frame) }
wait during SIFS;
send ack (frame) ;
}

CNPP/2008.5. © O. Bonaventure, 2008

43
 CSMA/CA : Example

B C
A
EIFS

Data frame
Busy

SIFS
ACK frame

DIFS

Data frame

CNPP/2008.5. © O. Bonaventure, 2008

44
 CSMA/CA : Problem

B C
A
EIFS
EIFS

Busy Data frame Data frame

Busy

Timeout Timeout

Data frame
Data frame
Busy

CNPP/2008.5. © O. Bonaventure, 2008

45
 CSMA/CA
First improvement (2)
Sender
N=1;
while ( N<= max) do
if (channel is empty)
{ wait until channel free during t>=EIFS; }
else
{ wait until endofframe;
wait until channel free during t>=DIFS; }
backoff_time = int(random[0,min(255,7*2N-1)])*T
wait(backoff_time)
if (channel still free)
{ send data frame ;
wait for ack or timeout:
if ack received
exit while;
else /* timeout retransmission is needed */
N=N+1; }
end do

CNPP/2008.5. © O. Bonaventure, 2008

46

The value T is defined in the standard, but a detailed discussion of this value is outside the scope of this presentation.
 CSMA/CA : Example 2
A B C

EIFS
EIFS
Backoff[0,7]
Backoff[0,7]
Data frame
Channel busy!

SIFS Busy

Ack frame

DIFS
Backoff[0,7]

Data frame
CNPP/2008.5. © O. Bonaventure, 2008

47
 CSMA/CA
Hidden station problem
Often occurs in wireless networks

B
Hears A and C

C
A

Only hears B, not C Hears B, but not A

CNPP/2008.5. © O. Bonaventure, 2008

48
 CSMA/CA
Second improvement
Principle
Allow the sender to “reserve” some air time
Special (short) RTS frame indicates duration
Using a short RTS frame reduces the risk of collisions while
transmitting this frame

Allow the receiver to confirm the reservation

Special (short) CTS frame indicates reservation
Using a short CTS frame reduces the risk of collisions while
transmitting this frame
The stations that could collide with the
transmission will hear at least CTS

Frame contains an indication of transmission time

CNPP/2008.5. © O. Bonaventure, 2008

49
 CSMA/CA : Example 3
A B C

EIFS+Backoff

RTS [100 microsec]

SIFS
CTS[100microsec] Busy[
100microsec+
SIFS SIFS+
CTS+
Data [100 microsec] SIFS+
ACK
]
SIFS
ACK frame

CNPP/2008.5. © O. Bonaventure, 2008

50
 Datalink layer

Point-to datalink layer

Local area networks
Optimistic Medium access control
ALOHA, CSMA, CSMA/CD, CSMA/CA

Ethernet networks
Basics of Ethernet
IP over Ethernet
Interconnection of Ethernet networks

WiFi networks

Deterministic Medium access control

CNPP/2008.5. Token Ring, FDDI © O. Bonaventure, 2008

51
 Ethernet/802.3

Most widely used LAN

First developed by Digital, Intel and Xerox
Standardised by IEEE and ISO

Medium Access Control

CSMA/CD with exponential backoff
Characteristics
Bandwidth: 10 Mbps
Two ways delay
51.2 microsec on Ethernet/802.3
=> minimum frame size : 512 bits
Cabling
10Base5 : (thick) coaxial cable maximum 500 m,100 stations
10Base2 : (thin) coaxial 200 m maximum and 30 stations

CNPP/2008.5. © O. Bonaventure, 2008

52

LAN/MAN Standards Committee of the IEEE Computer Society. IEEE Standard for Information Technology - Telecommunications and
information exchange between systems - local and metropolitan area networks - specific requirements - Part 3 : Carrier Sense multiple access
with collision detection (CSMA/CD) access method and physical layer specification. IEEE, 2000. available from http://standards.ieee.org/
getieee802/802.3.html
 Ethernet/802.3

A B C D

Initial configuration
bus-shaped network

Remaining problems besides CSMA/CD

What is an Ethernet frame ?
How does station A sends a frame to station B ?
How does station B detects a frame
How to support broadcast ?
How to support multicast ?

CNPP/2008.5. © O. Bonaventure, 2008

53
 Ethernet Addresses (2)

Each Ethernet adapter has a unique

Ethernet address
ensures that two hosts on the same LAN will not
use the same Ethernet address
Ethernet Addressing format
48 bits addresses
Source Addresse
00

24 bits OUI 24 bits

(adapter manufacturer) (identifier of adapter)

Destination address
If high order bit is 0, host unicast address
If high order bit is 0, host multicast address
broadcast address = 111111..111

CNPP/2008.5. © O. Bonaventure, 2008

54

Ethernet addresses are usually printed as hexadecimal numbers, e.g.

alpha.infonet.fundp.ac.be (at 00:80:C8:FB:21:2B [ether] on eth0

cr1.info.fundp.ac.be at 00:50:BD:D0:E0:00 [ether] on eth0
backus.info.fundp.ac.be at 08:00:20:A6:62:8A [ether] on eth0
inspiron.infonet.fundp.ac.be at 00:50:04:8C:83:70 [ether] on eth0
corneille.info.fundp.ac.be at 00:20:AF:52:44:4B [ether] on eth0
See http://standards.ieee.org/regauth/oui/oui.txt for the list of allocations
 Ethernet Frames

DIX Format
proposed by Digital, Intel and Xerox

Preamble Used to mark the beginning of the frame

[8 bytes] Allows the receiver to synchronise its
clock to the senderʼs clock
Destination
address

Source address

Type Indication of the type of packet contained

[2 bytes] inside the frame

Data Upper layer protocol must ensure that

[46-1500 bytes the payload of the Ethernet frame is
at least 46 bytes and at most 1500 bytes

CRC [32 bits]

CNPP/2008.5. © O. Bonaventure, 2008

55

This is the most widely used format, it is notably used to carry IP packets.
 802.3 Frames

Ethernet 802.3
standardised by IEEE

Preamble Used to mark the beginning of the frame

[7 bytes] Allows the receiver to synchronise its
Delimiter[1byte] clock to the senderʼs clock
Destination
Address
Source
Address
Provides the real length of the network
Length layer packet placed inside the payload.
[2 bytes] 802.3 adds padding (bytes set to 0) to
ensure that the data field of the frame
Data contains at least 46 bytes
and padding
[ 46- 1500 bytes]

CRC [32 bits]

CNPP/2008.5. © O. Bonaventure, 2008

56
 Ethernet and 802.3 : details

How can the receiver identify the type of

network protocol packet inside the frame ?
Ethernet : thanks to Type field
802.3 : no Type field !

IEEE standard
Divide datalink layer in two sublayers
Medium Access Control (MAC)
lower sublayer responsible for the frame transmission and
medium access control (CSMA/CD)
interacts with but does not depend from the physical layer
example : 802.3
Logical Link Control (LLC)
higher sublayer responsible for the exchange of frames with the
higher layers
interacts with the higher layer
does not depend from the MAC layer
several variants of LLC exist
CNPP/2008.5. © O. Bonaventure, 2008

57
 802.2 : LLC

LLC Type 1
Unreliable connectionless service
Addition to 802.3
New LLC header allows to identify upper layer protocol
similar to Type field of Ethernet DIX

Packet

LLC Packet LLC

P DA SA L LLC Packet CRC MAC

LLC Type 2
Reliable transmission with acknowledgements
An example of a protocol developed by a
standardisation body but used by nobody...
CNPP/2008.5. © O. Bonaventure, 2008

58
 Ethernet Service

An Ethernet network provides a

connectionless unreliable service
Transmission modes
unicast
multicast
broadcast

Even if in theory the Ethernet service is

unreliable, a good Ethernet network should
deliver frames to their destination with a very hig
probability of delivery
not reorder the transmitted frames
reordering is obviously impossible on a bus

CNPP/2008.5. © O. Bonaventure, 2008

59
 Datalink layer

Point-to datalink layer

Local area networks
Optimistic Medium access control
ALOHA, CSMA, CSMA/CD, CSMA/CA

Ethernet networks
Basics of Ethernet
IP over Ethernet
Interconnection of Ethernet networks

WiFi networks

Deterministic Medium access control

CNPP/2008.5. Token Ring, FDDI © O. Bonaventure, 2008

60
 IPover PPP and
IP over Ethernet
IP: 10.0.0.1 Ethernet
PPP
IP: 10.0.1.9
IP: 10.0.0.2 R IP: 10.0.1.1
Eth : B
Eth : A

IP IP
PPP PPP Eth Eth
Modem Modem 10base5 10base5

PPP Header IP Header PPP Trailer

F Addr Ctrl Protocol S:10.0.0.1 D:10.0.1.9 CRC F

Ethernet header IP header Ethernet

0xFF 0x0021 IP
0x8021 NCP for IP P S:A D: B Type S:10.0.0.1 D:10.0.1.9 CRC

0x0800 IP
CNPP/2008.5. © O. Bonaventure, 2008

61
 IP on LANs

Problems to be solved
How to encapsulate IP packets in frames ?
How to find the LAN address of the IP destination ?
LAN efficiently supports broadcast/multicast
transmission
When a host needs to find the LAN address of
another IP host, it broadcasts a request
The owner of the destination IP address will reply and
provided its LAN address
LAN doesnʼt efficiently support broadcast/
multicast
Maintain a server storing IP address:MAC address pairs
Each host knows serverʼs MAC address and registers its
address pair
Each host sends request to server to map IP addresses

CNPP/2008.5. © O. Bonaventure, 2008

62
 Address Resolution Protocol

1
IP: 10.0.1.22 IP: 10.0.1.11 IP: 10.0.1.8 IP: 10.0.1.9
Eth : A Eth : E Eth : C Eth : B

10.0.1.22 needs to send an IP packet to 10.0.1.8

2
IP: 10.0.1.22 IP: 10.0.1.11 IP: 10.0.1.8 IP: 10.0.1.9
Eth : A Eth : E Eth : C Eth : B

ARP : broadcast frame Addr Eth 10.0.1.8 ?

3
IP: 10.0.1.22 IP: 10.0.1.11 IP: 10.0.1.8 IP: 10.0.1.9
Eth : A Eth : E Eth : C Eth : B

10.0.1.8 replies in an Ethernet frame and A knows the MAC address to send
its IP packet

CNPP/2008.5. © O. Bonaventure, 2008

63

D. C. Plummer. Ethernet address resolution protocol: Or converting network protocol addresses to 48.bits ethernet address for transmission on
ethernet hardware. Request for Comments 826, Internet Engineering Task Force, November 1982.
 ARP frame format

Preamble
[7 bytes]
Delimiter[1byte]

Destination
Broadcast : 111...111
Address
Source MAC address of the sender
Address
Common header for all ARP frames
Type: 0x806 - Hardware type Ethernet is 1
- Protocol type , IP is 0x0800.
Header
- Hardware length : length of MAC address
- Protocol length : length of network layer
address
Sender MAC - Operation : 1 for request, 2 for reply, 3 for
RARP request, and 4 for RARP reply.
Sender IP

Target MAC

Target IP

CNPP/2008.5. CRC [32 bits] © O. Bonaventure, 2008

64
 Optimisations

When should a host send ARP requests ?

Before sending each IP packet ?
No, each host/router maintains an ARP table that contains
the mapping between IP addresses and Ethernet
addresses. An ARP request is only sent when the ARP
table is empty

How to deal with hosts that change their

addresses ?
Expiration timer is associated to each entry in the
ARP table
Line of ARP table is removed upon timer expiration.
Some implementations send an ARP request to revalidate
it before removing the line
Some implementations remember when ARP lines have
been used to avoid removing an important entry
CNPP/2008.5. © O. Bonaventure, 2008

65
 IP over Ethernet : Example

IP: 10.0.1.9/24
IP: 10.0.1.8/24 Default R: 10.0.1.1
IP: 10.0.1.22/24 Default R: 10.0.1.1
Default R: 10.0.1.1 Eth : B
Eth : C ARP Table
Eth : A ARP Table
ARP Table empty
empty
empty

IP: 10.0.1.1/24
Eth : R
ARP Table
empty
R

Transmission of an IP packet from 10.0.1.22 to 10.0.1.9

Transmission of an IP packet from 10.0.1.22 to 10.0.2.9

CNPP/2008.5. © O. Bonaventure, 2008

66
 Datalink layer

Point-to datalink layer

Local area networks
Optimistic Medium access control
ALOHA, CSMA, CSMA/CD, CSMA/CA

Ethernet networks
Basics of Ethernet
IP over Ethernet
Interconnection of Ethernet networks

WiFi networks

Deterministic Medium access control

CNPP/2008.5. Token Ring, FDDI © O. Bonaventure, 2008

67
 Ethernet today

The coaxial cable is not used anymore

Ethernet cabling today

Structured twisted pair cabling
Optical fiber for some point-to-point links

Ethernet organisation
Not anymore a bus
Ethernet is now a star-shaped network !

CNPP/2008.5. © O. Bonaventure, 2008

68
 Ethernet with structured cabling
How to perform CSMA/CD in a star-shaped
network ?

Hub :
receives electrical signal on one port,
regenerates this signal and forwards it over all
other ports besides the port from which it
received it

Collision domain : set of stations that could be in collision

CNPP/2008.5. © O. Bonaventure, 2008

69
 Hub and the reference model

A hub is a relay operating the physical layer

Datalink Datalink
Physical Physical

Host A Hub Host B

CNPP/2008.5. © O. Bonaventure, 2008

70
 Ethernet with structured cabling (2)

How to build a larger Ethernet network ?

Interconnect hubs together
Hub

Hub Hub

Hub

Issue : maximum 51.2 microsec

delay between any pair of stations

CNPP/2008.5. Collision domain : entire network © O. Bonaventure, 2008

71
 Ethernet Switch
Can we improve the performance of hubs ?
Ethernet switch
Operates in the datalink layer
understands MAC address and filters frames based on
their addresses

Eth : C
Eth : A

Address Port
Src:A Dst:B
A West

B South

Eth : B C East
CNPP/2008.5. © O. Bonaventure, 2008

72

A good reference on Ethernet switches is

R. Seifert, J. Edwards, The All-New Switch Book, Wiley, 2008

 Switch in the reference model

A switch is a relay that operates in the

datalink layer

Network Network
Datalink Datalink
Physical Phys. Phys. Physical

Host A Switch Host B

CNPP/2008.5. © O. Bonaventure, 2008

73
 Port-address table

How to build the port-address table used by

Ethernet switches ?
Manually
Works in a lab, but Ethernet must be plug and
play
Automatically
Frame source address allows switch to learn the
location of hosts
What happens when a destination address
cannot be found in the port-address table ?
But be careful to age the information inside tables
as some hosts move from one port to another

How to forward broadcast frames ?

How to forward multicast frames ?
CNPP/2008.5. © O. Bonaventure, 2008

74
 Frame processing

Basic operation of an Ethernet switch

Arrival of frame F on port P
src=F.Source_Address;
dst=F.Destination_Address;
UpdateTable(src, P); // src heard on port P
if (dst==broadcast) || (dst is multicast)
{
for(Port p!=P) // forward all ports
ForwardFrame(F,p);
}
else
{
if(dst isin AddressPortTable)
{
ForwardFrame(F,AddressPortTable(dst));
}
else
{
for(Port p!=P) // forward all ports
ForwardFrame(F,p);
}
}
CNPP/2008.5. © O. Bonaventure, 2008

75
 Network redundancy
How to design networks that survive link and
node failures ?
Add redundant switches
Address Port
A West
B South Eth : C

C East
Eth : A

Src:A Dst:C

Address Port
A West
Address Port B South
A North
C North
B South Eth : B
CNPP/2008.5. C East © O. Bonaventure, 2008

76
 Network redundancy (2)

Does this always work ?

Assume all switches have rebooted
Address Port

Eth : C

Eth : A

Src:A Dst:C

Address Port

Address Port

Eth : B
CNPP/2008.5. © O. Bonaventure, 2008

77
 How to solve this problem ?

The lawyerʼs way

Add a sticker on all switches to indicate that they
must only be used in tree shaped networks and
should never ever be interconnected with loops

The computer scientistʼs way

Define a distributed algorithm that allows
switches to automatically discover the llinks
causing loops and remove them from the
topology

CNPP/2008.5. © O. Bonaventure, 2008

78
 Principle of the solution

Build a spanning tree inside network

Each switch has a unique identifier
The switch with the lowest id is the root
Disable all links that do not belong to spanning

Switch 2

Switch 1
Switch 9

Switch 44

CNPP/2008.5. © O. Bonaventure, 2008

Switch 7 Switch 22

79
 How to build the spanning tree

Distributed algorithm run by switches

Goals of the spanning tree protocol
Elect the root of the spanning tree
In practice, this will be the switch with the lowest id
Compute the distance between each switch and
the root
When several switches are attached to the same
LAN elect one forwarder and disable the others
determine which ports/links should belong to the
spanning

CNPP/2008.5. © O. Bonaventure, 2008

80
 Root and Designated Switches

Root switch
The Root Switch is the root of the spanning tree
The Root switch may change upon the arrival of
new switches in the network
Switch 1

Designated switch
to avoid loops, only
one switch should
be responsible for Switch 9
Switch 4
forwarding frames
from the root on
any link
Root switch is always
designated switch
for all its links
CNPP/2008.5. © O. Bonaventure, 2008

81
 The switch identifiers

Switch identifiers must be unique

The easiest solution is to ask each manufacturer
to embed a unique Ethernet address on each
switch

But since the switch with the lowest identifier

is the network root, network operators need
to influence the selection of the root switch
64 bits switch identifier
Upper 16 bits
Priority defined by operator (default value : 32768)
Lower 48 bits
Unique Ethernet address assigned by manufacturer
CNPP/2008.5. © O. Bonaventure, 2008

82
 The link costs

Each switch port is attached to a link

The costs of the links can be configured on
each link by the network operator
Common guideline : Cost = 1000 / bandwidth

Recommended values of link costs

Bandwidth Recommended Recommended

link cost range link cost value
10 Mbps 50-600 100
100 Mbps 10-60 19
1000 Mbps 3-10 4
CNPP/2008.5. © O. Bonaventure, 2008

83
 Building the spanning tree

802.1d protocol
802.1d uses Bridge PDUs (BPDUs) containing
Root ID : identifier of the current root switch
Cost : Cost of the shortest path between the switch
transmitting the BPDU and the root switch
Transmitting ID : identifier of the switch that transmits
the BPDU

The BPDUs are sent by switches over their

attached LANs as multicast frames but they are
never forwarded
switches that implement 802.1d listen to a special
Ethernet multicast group

CNPP/2008.5. © O. Bonaventure, 2008

84
 Ordering of BPDUs

BPDUs can be strictly ordered

BPDU11[R=R1,C=C1, T=T1] is better than
BPDU2 [R=R2,C=C2, T=T2] if
R1<R2
R1=R2 and C1<C2
R1=R2 and C1=C2 and T1<T2

Example
BPDU1 BPDU2
R1 C1 T1 R2 C2 T2
29 15 35 31 12 32
35 80 39 35 80 40
35 15 80 35 18 38

CNPP/2008.5. © O. Bonaventure, 2008

85
 Building the spanning tree (2)

Behaviour of 802.1d protocol

The root switch sends regularly BPDUs on all its
ports
R=Root switch id, C=0, T= Root switch id
Bootstrap
If a switch does not receive BPDUs, it considers itself as root
and sends BPDUs
On each port, a switch parses all the received
BPDUs and stores the best BPDU received on
each port
Each switch can easily determiner the current root by
analysing all the BPDUs stored in its tables
A switch stops sending BPDUs on a port if it
received a better BPDU on this port
802.1d stabilises when a single switch sends a
BPDU over each LAN
CNPP/2008.5. © O. Bonaventure, 2008

86
 802.1d port states
802.1d port state based on received BPDUs
Root port
port on which the best 802.1d BPDU was received
port used to receive the BPDUs sent by the root form
the shortest path
A root port does not transmit BPDUs
Only one root port on each switch
Designated port
port(s) used to send switchʼs BPDU upon reception of a
BPDU from the root via the Root port
Switchʼs BPDU is
current root, cost to reach root, switch identifier
0, one or more designated ports on each switch
a port is designated if the switchʼs BPDU is better than
the best BPDU received on this port
Blocked port (only receives 802.1d BPDUs)
CNPP/2008.5. © O. Bonaventure, 2008

87
 802.1d port states (2)

Example
BPDUs received by switch 18
Root Cost Transmitter
port1 12 93 51
port2 12 85 47
port3 81 0 81
port4 15 31 27

Root : switch 12
port2 is the root port
Switchʼs BPDU
R=12, C=86, T=18
This BPDU is better than the BPDUs received on
the other ports. They are thus designated

CNPP/2008.5. © O. Bonaventure, 2008

88
 802.1d port states (3)

Example
BPDUs received by switch 92
Root Cost Transmitter
port1 81 0 81
port2 41 19 125
port3 41 12 315
port4 41 12 111
port5 41 13 90

root : 41
root port : port4
Switchʼs BPDU
R=41,C=13, T=92
Port state
port1 and port 2 : designated
port 3 and port 5 : blocked
CNPP/2008.5. © O. Bonaventure, 2008

89
 Port activity

A port can be either active or inactive for

data frames
Active port
The switch captures Ethernet frames on its active ports
and forwards them over other ports (based on its own
port/address tables)
The switch updates its port/address table based on the
frames received on this port
Inactive port
The switch does not listen to frames neither forward
frames on this port
The port activity is fixed once the spanning
tree has converged
Root and designated ports become active
Blocked ports become inactive
Duration spanning tree computation, all ports are
CNPP/2008.5. inactive © O. Bonaventure, 2008

90
 Port states and activity

Receive Transmit
BPDUs BPDUs
Blocked yes no
Root yes no
Designated yes yes

Learn Forward Data

Addresses Frames

Inactive no no
Active yes yes

CNPP/2008.5. © O. Bonaventure, 2008

91
 Example network

Compute the spanning tree in this network

by using 802.1d Switch 7

Switch 9
Switch 12

Assume that 9 boots and then 12

CNPP/2008.5. and eventually 7 boots © O. Bonaventure, 2008

92
 Impact of failures

What kind of failures should be considered ?

Failure (power-off) of the root switch
A new root needs to be elected

Failure of a designated switch

Another switch should replace the designated one

Failure of a link
If the network is redundant, a disabled link should be
enabled to cope with the failure

Failure of a link that disconnects the network

We now have two different networks and a root switch
must be elected in each network

CNPP/2008.5. © O. Bonaventure, 2008

93
 How to deal with failures ?

Failure detection mechanisms

Root switch sends its BPDU every Hello timer
and designated switches generate their own
BPDUs upon reception of this BPDU
Default Hello timer is two seconds

BPDUs stored in the switches age and are

removed when they timeout

Failure notification mechanism

When a switch detects an important failure, it
sends a topology change (TC) BPDU to the Root
Upon reception of a TC BPDU all switches stop
forwarding data frames and recompute spanning
CNPP/2008.5.
tree © O. Bonaventure, 2008

94
 IP over Ethernet
Detailed example
IP: 10.0.1.8/24 IP: 10.0.2.9/24
R default: 10.0.1.1 Eth : B
Eth : C 10.0.1.0/24 via 10.0.2.1
ARP table 10.0.3.0/24 via 10.0.2.2
Empty ARP table
Empty

Router Switch Router

H1 R1 S2 R2
IP: 10.0.2.2/24
IP: 10.0.1.22/24 Hub IP: 10.0.1.1/24 IP: 10.0.3.2/24
IP: 10.0.3.11/24
R default: 10.0.1.1 IP: 10.0.2.1/24 Eth : F
Eth : R2-West R default:10.0.3.2
Eth : A Eth : R1-West Eth : R2-East
ARP table Eth : R1-East ARP table
ARP table Empty
Empty ARP table Empty
Empty

Examples
IP packet from 10.0.1.22 to 10.0.3.11
IP packet from 10.0.2.9 to 10.0.1.22
IP packet from 10.0.3.11 to 10.0.1.22

CNPP/2008.5. © O. Bonaventure, 2008

95
 Ethernet Evolution

Networks require higher bandwidth

Fast Ethernet
Physical layer
bandwidth : 100 Mbps
twisted pair or optical fiber
No coaxial cable anymore

MAC sublayer
CSMA/CD unchanged
minimum frame size : 512 bits
slot time : 5.12 micro seconds
Maximum distance : shorter than Ethernet 10 Mbps
Same frame format as 10 Mbps Ethernet

CNPP/2008.5. © O. Bonaventure, 2008

96

R. Seifert. Gigabit Ethernet : Technology and Applications for High-Speed LANs. Addison Wesley, 1998. ISBN 0201185539.
http://www.ethermanage.com/ethernet/ethernet.html
 Ethernet Evolution (2)

Gigabit Ethernet
Physical layer
Bandwidth 1 Gbps
Optical fiber or twisted pair

MAC sublayer
CSMA/CD still supported
How was this achieved ?
Two options
Increase minimum frame size : not backward compatible with
Ethernet
Reduce the maximum distance as for FastEthernet : but then
networks would have a diameter of 10 m
Gigabit CSMA/CD hack
minimum frame size is still 512 bits but the sender must
continue to send an electrical signal during the equivalent o
4096 bits
same frame format as Ethernet
but extensions allow to transmit Jumbo frames of up to 9KBytes
CNPP/2008.5. © O. Bonaventure, 2008

97
 The Ethernet zoo

10BASE5 Thick coaxial cable, 500m

10BASE2 Thin coaxial cable, 185m

10BASE-T Two pairs of category 3+ UTP

10BASE-F 10 Mb/s over optical fiber

100BASE-TX Category 5 UTP or STP, 100 m maximum

100BASE-FX Two multimode optical fiber, 2 km maximum

1000BASE-CX Two pairs shielded twisted pair, 25m maximum

1000BASE-SX Two multimode or single mode optical fibers with lasers

10 Gbps optical fiber but also cat 6 twisted pair

40-100 Gbps being developed, standard expected in 2010, 40Gbps

one meter long for switch backplanes, 10 meters for copper
cable and 100 meters for fiber optics
CNPP/2008.5. © O. Bonaventure, 2008

98

The 10 Gbps zoo is much larger than this, see e.g. http://en.wikipedia.org/wiki/10_gigabit_Ethernet
 Full duplex Ethernet

Observations
In many networks, Ethernet is a often a point-to-
point technology
host-to-switch
switch to switch

S1 S2

HUB
HUB

Twisted-pairs and fiber-based physical layers

allow to send and receive at the same time
CNPP/2008.5. © O. Bonaventure, 2008

99
 Ethernet full duplex (2)

No collision is possible on a full duplex

Ethernet/FastEthernet/GigabitEthernet link
Disable CSMA/CD on such links

Advantages
Improves bandwidth
Both endpoints can transmit frames at the same time

CSMA/CD is disabled
No constraint on propagation delay anymore
Ethernet network can be as large as we want !

No constraint on minimum frame size anymore

We do not need the frame extension hack for Gigabit Ethernet!

CNPP/2008.5. © O. Bonaventure, 2008

100
 Full duplex Ethernet (3)

Drawback
If CSMA/CD is disabled, access control is
disabled and congestion can occur

Server
S1 S2
Client

FastEthernet (100 Mbps) Ethernet (10 Mbps)

How to solve this problem inside Ethernet ?

Add buffers to switches
but infinite buffers are impossible and useless anyway
Cause collisions (e.g. jamming) to force collisions on the inter-
switch link and uplink is server is too fast
Drawback : interswitch link could be entirely blocked
Develop a new flow control mechanism inside MAC layer
Pause frame to slowdown transmission
CNPP/2008.5. © O. Bonaventure, 2008

101
 Ethernet flow control
server
Client

S1
FastEthernet Ethernet
(100 Mbps) (10 Mbps)

100 nsec Frame1 [10000 bits]

Frame2 [10000 bits] Frame1 [10000 bits]
Frame3 [10000 bits]
PAUSE [2msec] 1 microsec

Sender blocked

Frame2 [10000 bits]

PAUSE frame indicates how much time the

upstream should wait before transmitting next frame
CNPP/2008.5. © O. Bonaventure, 2008

102
 Virtual LANs
Allows to build several logical networks on
top of a single physical network
Each port on each switch is
A B associated to a particular VLAN
All the hosts that reside on the same
VLAN can exchange Ethernet
frames
A host on VLAN1 cannot send an
C S F
Ethernet frame towards another
host that belongs to VLAN2
Broadcast and multicast frames are
D E only sent to the members of the
VLAN

VLAN1 : A,E,F
VLAN2 : B,C,D
CNPP/2008.5. © O. Bonaventure, 2008

103
 VLANs in campus networks

How to support VLANs in a campus network

Possible solutions
A
Place on each switch a table
B
that maps each MAC address
on a VLAN id
difficult to manage this table

C S1 S2 F Change frame format used on

inter-switch links to include a
VLAN identifier
new header added by first switch
D E new header removed by last switch

VLAN1 : A,E,F
VLAN2 : B,C,D
CNPP/2008.5. © O. Bonaventure, 2008

104

See

[IEEE802Q] "IEEE Standards for Local and Metropolitan Area

Networks: Virtual Bridged Local Area Networks", Draft Standard, P802.1Q/D9, February 20, 1998.
 VLAN frame format

Used on inter-switch links

Destination
Address
Source
Address
VLAN Identifies the frame as containing VLANtag
Type Id
Protocol
0x8100 Tag control information contains two types
of information :
Tag Payload
Control Info
- VLAN identifier (12 bits) : up to 4094
different VLANs can be defined
CRC [32 bits] - Priority (3 bits) : indicates the importance
of the frame and can be used by switches
to provide a better service for some frames
(e.g. Voice)

Can also be used by trusted hosts (e.g. servers)

CNPP/2008.5. or routers © O. Bonaventure, 2008

105
 Datalink layer

Point-to datalink layer

Local area networks
Optimistic Medium access control
ALOHA, CSMA, CSMA/CD, CSMA/CA

Ethernet networks
Basics of Ethernet
IP over Ethernet
Interconnection of Ethernet networks

WiFi networks

Deterministic Medium access control

CNPP/2008.5. Token Ring, FDDI © O. Bonaventure, 2008

106
 Modes of operation

Ad hoc
or Independent
Basic Service Set

Infrastructure mode
DS

BSS
BSS
CNPP/2008.5. ESS © O. Bonaventure, 2008

107
 The WiFi zoo

Standard Frequency Typical Raw Range in/out

throughput bandwidth (m)

802 .11 2.4 GHz 0.9 Mbps 2 Mbps 20 / 100

802 .11a 5 GHz 23 Mbps 54 Mbps 35 / 120

802 .11b 2.4 GHz 4.3 Mbps 11 Mbps 38 / 140

802 .11g 2.4 GHz 19 Mbps 54 Mbps 38 / 140

802 .11n 2.4 / 5 GHz 74 Mbps up to 600 70 / 250

Mbps

CNPP/2008.5. © O. Bonaventure, 2008

108

Source http://en.wikipedia.org/wiki/IEEE_802.11n
 WiFi zoo and performance

Performance issues with the multiple WiFi

transmission rates
802.11, 802.11b and 802.11g operate on 2.4 GHz
frequency bands
Many access points are multi-standard

1 Mbps

11 Mbps
Laptop far away
54 Mbps

Laptop supporting
only 802.11b
CNPP/2008.5. © O. Bonaventure, 2008

109
 Practical issues
with WLAN deployments
Home environment

A WLAN can interfere with the neighbourʼs

CNPP/2008.5. WLAN © O. Bonaventure, 2008

110
 Practical issues
with WLAN deployments
Enterprise networks

One access points can interfere with many other

CNPP/2008.5. access points © O. Bonaventure, 2008

111
 The WiFi channel frequencies

WiFi standards operate on several

frequencies called channels
Usually about a dozen channels

Why multiple channels ?

Some channels my be affected by interference
and have a lower performance
Some frequencies are reserved for specific
usage in some countries
Allows frequency reuse when there are multiple
WiFi networks in the same area
Unfortunately, many home access points operate by
default on the same factory set channel which causes
interference and reduced bandwidth

CNPP/2008.5. © O. Bonaventure, 2008

112

Example

802.11b channel frequencies

Channel Lower frequency Central frequency Upper frequency
1 2.401 2.412 2.423
2 2.404 2.417 2.428
3 2.411 2.422 2.433
4 2.416 2.427 2.438
5 2.421 2.432 2.443
6 2.426 2.437 2.448
7 2.431 2.442 2.453
8 2.436 2.447 2.458
9 2.441 2.452 2.463
10 2.446 2.457 2.468
11 2.451 2.462 2.473
 WLAN in enterprise
environments
What could be done to improve the
performance of WLANs ?
Reduce interference as much as possible
Tune channel frequencies
Reduce transmission power
Similar to techniques used in GSM networks

Recent deployments rely on centralised controllers

CNPP/2008.5. and thin access points © O. Bonaventure, 2008

113
 Datalink layer

Point-to datalink layer

Local area networks
Optimistic Medium access control
ALOHA, CSMA, CSMA/CD, CSMA/CA

Ethernet networks
Basics of Ethernet
IP over Ethernet
Interconnection of Ethernet networks

WiFi networks

Deterministic Medium access control

CNPP/2008.5. Token Ring, FDDI © O. Bonaventure, 2008

114
 Ring networks

S2
S1

unidirectional ring

S3
Interface

S5

S4
Problem to be solved
How to share fairly ring transmission capacity
among all devices attached to the ring ?
CNPP/2008.5. © O. Bonaventure, 2008

115
 Ring networks (2)

How to share transmission capacity ?

To avoid collisions, only one station should be
able to transmit a frame at any time

The station that has the right to transmit must

own a special frame called token

How can stations exchange token ?

Token is a special frame that can be sent over
the ring network
A station that needs to transmit a data frame can
capture the token and remove it from the ring
send one or more data frames
send the token back on the ring to allow other stations
to capture it and transmit

CNPP/2008.5. © O. Bonaventure, 2008

116
 Ring networks (3)

Consequence
When there are no data frames sent, stations
should continuously exchange the token
S1 S2
How to achieve this ?
A station must relay the
electrical signal it receives Unidirectional
ring
upstream when not S3
transmitting Interface
it introducing a delay of one S5
bit transmission time
If all stations behave so, S4
and token is small,
token will travel permanently
If token is not small, increase the delay on the
token ring network
CNPP/2008.5. © O. Bonaventure, 2008

117
 Ring networks (4)

Data frame transmission

A data frame requires a longer transmission time
than the ring delay

S1 S2

Sender behaviour
Captures token
Sends data frame
Removes data frame S3
from ring
Sends token S5

S4

CNPP/2008.5. © O. Bonaventure, 2008

118
 Ring networks in practice

Two types of ring LANs

Token Ring
Invented by IBM
Standardised by IEEE/ISO (802.5)
Ring build with point-to-point twisted pair links
4 Mbps
16 Mbps
Some work for 100 Mbps Token Ring
Fiber Distributed Data Interface (FDDI)
First data networks built with optical fiber
standardised by ANSI
100 Mbps
up to 200 km and 1000 stations

Other ring technologies exist and are used

SONET/SDH
DPT
CNPP/2008.5. © O. Bonaventure, 2008

119
 Token Ring (1)

Token
travels permanently on ring when stations are
idle
Size 24 bits
Minimum delay on ring
24 bits transmission times
Actual ring delay
Each station introduces a one-bit transmission time
delay
Physical links have a propagation delay
Each ring contains a monitor station that measures
delay during ring initialisation and adds delay if needed
Interfaces
Two modes of operation
Listen : interface adds a one bit transmission delay
Transmit : only if station owns the token
CNPP/2008.5. © O. Bonaventure, 2008

120

Token Ring is defined in :

LAN/MAN Standards Committee of the IEEE Computer Society. IEEE Standard for Information technology--Telecommunications and
information exchange between systems--Local and metropolitan area networks--Specific requirements--Part 5: Token Ring Access Method and
Physical Layer Specification. IEEE, 1998. available from http://standards.ieee.org/getieee802/802.5.html
 Token Ring (2)

Frame format
Token (24 bits)
SD : starting delimiter
invalid physical layer symbol with
Manchester coding 1 1 1
AC : Access control SD AC ED
ED : ending de fin
invalid physical layer symbol with
Manchester coding
Data frame
1 1 1 (2or)6 (2 or)6 0... (nolimit) 4 1 1
SD AC FC Dest Source Data CRC ED FS

FC : Frame control
Allows to distinguish between control frames and data frames
FS : Frame status
CNPP/2008.5. © O. Bonaventure, 2008

121
 Token Ring (3)

Frame transmission
S2
Delay 1 bit S1
Listen mode

Unidirectional Ring

S3
Transmit mode
Interface

To From S5
Station Station

S4
How to capture the token ?
Rely on Token bit of AC field and one bit delay
CNPP/2008.5. © O. Bonaventure, 2008

122
 Token Ring (4)

Whatʼs special about Token Ring

Can efficiently support acknowledgements
Frame Status contains two bits : A and C
A and C are set to 0 when transmitting a frame
When a receiver sees one frame destined to itself, it
sets A to 1
When a receiver copies one frame destined to itself
inside its buffers, it sets C to 1

Data frame (and FS) return to sender. By

checking A and C, it knows that :
if A=0 and C=0, destination is down
if A=1 and C=0, destination is up, but congested
if A=1 and C=1, frame was received by destination

CNPP/2008.5. © O. Bonaventure, 2008

123
 Token Ring (5)

Issues with Token Ring

How to ensure fairness ?
A station should not be allowed to transmit indefinitely
Token Holding Time
Maximum time during which a station can own the token and
transmit data frames without releasing the token
Default : 10 milliseconds

How to bootstrap the Token Ring ?

Which station sends the first token ?

How to ensure that the Ring delay is long enough ?

What happens when a station fails ?
If it did not own the token, no issue
If it owned the token while failing, then
Which station will remove the current data frame from the
ring ?
Which station will send the token on the ring ?
CNPP/2008.5. © O. Bonaventure, 2008

124
 Token Ring (6)

How to bootstrap a Token Ring ?

Complex problem
Main idea
One station should send the token
The first station on the ring hears nothing and notices
that there is a problem. It sends a special frame called
CLAIM_TOKEN
If it receives the frame back, it becomes the monitor
Each station must be able to become monitor

Monitorʼs responsibilities
Ensure that token is never lost or corrupted
Insert an artificial delay of 24 bit transmission
times on the ring
Remove orphan and looping frames
If the monitor fails, the ring must be bootstrapped
again © O. Bonaventure, 2008
CNPP/2008.5.

125
 Token Ring (7)

Token surveillance
Monitor checks how often its sees the token
If there are N stations on the ring, then the monitor should
see the token at worst every N*THT seconds
If token is lost, monitor cuts ring, removes electrical signal
and resend a new token
Orphan frames
Frame with invalid coding or incomplete frame
monitor cuts ring, removes electrical signal and resend a
new token
Looping frames
Every time monitor sees a frame, it sets its Monitor
bit of the AC field to 1
All stations send their frames with Monitor=0
If a frame is seen twice by the monitor, it cuts ring,
removes electrical signal and resend a new token
CNPP/2008.5. © O. Bonaventure, 2008

126
 FDDI

Network topology FDDI

Single ring like Token Ring
Two counter rotating rings to deal with failures

Bypass to reroute
around failure Failure

S2
S1

S3

S5

Usually, only one ring is used

S4

CNPP/2008.5. © O. Bonaventure, 2008

127

FDDI is defined in
ANSI. Information systems - fiber distributed data interface (FDDI) - token ring media access control (mac). ANSI X3.139-1987 (R1997), 1997
 FDDI (2)

Medium access control

Token based access control
A station can only transmit a data frame provided that it
owns the token
Token Holding Time (THT)
maximum duration of transmission
Token Rotation Time (TRT)
maximal delay for a token to rotate around the entire ring
TRT ≤ Actives_Stations * THT + Ring_Latency

When should the Token be released

Immediately after removal of the data frame sent
as in Token Ring
Immediately after transmission of the data transfer,
without waiting for it to come back
solution chosen for FDDI due to the high bandwidth and long
latency of the FDDI ring

CNPP/2008.5. © O. Bonaventure, 2008

128
 FDDI (3)

Delay sensitive service

How to support two types of frames in FDDI ?
normal data frames (asynchronous frames)
example : file transfer, email, www
delay sensitive data frames (synchronous frames)
example : telephone, videoconference

Solution
Delay sensitive frames can be supported
provided that a FDDI ring can bound the
transmission delay of such a frame
synchronous frames should be transmitted earlier than
normal frames on each station
Since a station can always transmit when it captures
the token, a solution should bound the Token Rotation
Time to provide strict guarantees to delay sensitive
frames
CNPP/2008.5. © O. Bonaventure, 2008

129
 FDDI (4)

How to bound the TRT ?

Target Token Rotation Time (TTRT)
At ring initialisation, all stations propose their expected
TTRT and the smallest proposed value is chosen
All stations must control their transmissions such that
the token rotation time is always smaller than TTRT
each station measures the current TRT
When a station captures the token, it can send its synchronous
frames
there is a maximum amount of synchronous frames that can
be sent by each station. This maximum is negotiated by using
control frames.
If after having sent synchronous frames TRT < TTRT, this
means that the token is circulating quickly and the station can
send asynchronous frames
Otherwise the token must be released

CNPP/2008.5. © O. Bonaventure, 2008

130
 FDDI (5)

Frame format
Start of Frame
[8 bits]

Control [8 bits] Frame type (1 bit)

synchronous or asynchronous
48 bits addresses Destination Type of addresses
16 bits or 48 bits
Protocol Type information
Source 6 bits, similar to Type Ethernet DIX
special types for control frames and
token

Maximum 4478 bytes Data

CRC [32 bits]

End of Frame
[8 bits]
Similar to A and C bits in Token Ring
Status [24 bits]
CNPP/2008.5. © O. Bonaventure, 2008

131
 Interconnection of Token Rings
How to interconnect Token Ring networks ?

S1 S2 S7

S5
Bridge/Switch
S4 S9

Possible solutions
Use the spanning tree designed for Ethernet
CNPP/2008.5.
Invent a new protocol © O. Bonaventure, 2008
solution chosen by IBM for Token Ring

132
 Interconnection of Token Rings (2)
S1
S7

S5

S9

Basic idea
Use source routing

Problems
How to identify the paths
How to discover the paths ?
CNPP/2008.5. © O. Bonaventure, 2008

133
 Interconnection of Token Rings (3)
S1 S7

LAN1 LAN5
B3

S5
S9
B6
B1
LAN3
B4

Identification of paths
Each LAN has one unique identifier
Each bridge has one identifier
Each path is a list of pairs LAN#,bridge#

CNPP/2008.5. © O. Bonaventure, 2008

134
 Interconnection of Token Rings (4)

S1 S7

LAN1 B3
LAN5

S5

B6 S9

B1

B4 LAN3

How to discover the path ?

Control frame : all paths explorer
Sent by source towards destination
Forwarded by all bridges that add their identifier and LAN
identifier
Destination sends back the ape frame to source by using
reverse path
Each station caches the recent paths
CNPP/2008.5. © O. Bonaventure, 2008

135
 Spanning Tree versus
Source Routing
Spanning tree Source routing
complexity in complexity in all
switches/bridges stations
only a subset of the the entire network is
network is used used
entirely transparent requires support on
stations
multicast natively spanning tree
supported required for multicast
few control frames many control frames
(802.1d) can be required

CNPP/2008.5. © O. Bonaventure, 2008

136