Ethernet
Carrier Sense Multiple Access
with Collision Detection
Giuseppe Bianchi
Ethernet Ancestors
 Late 1960: ALOHA network
 Norman Abramson, University of Hawaii
 Application: radio network among islands
Distributed, uncoordinated network!
 First random
access mechanism
(Pure aloha / Slotted aloha)
Giuseppe Bianchi
Birth of Ethernet
May 22, 1973: Ethernet memo
Bob Metcalfe (Xerox Palo Alto Research Center)
Carrier Sense Multiple Access with Collision
Detection and expo backoff
3 mbps speed
US Patent 4.063.220
“Multipoint Data Communication
System with Collision Detection”
end 1977
Giuseppe Bianchi
On the birth of Ethernet
 [quoting from Ethernet – The definitive guide”]:
 In late 1972, Metcalfe and his Xerox PARC colleagues
developed the first experimental Ethernet system to
interconnect the Xerox Alto, a personal workstation with a
graphical user interface. The experimental Ethernet was
used to link Altos to one another, and to servers and laser
printers. The signal clock for the experimental Ethernet
interface was derived from the Alto's system clock, which
resulted in a data transmission rate on the experimental
Ethernet of 2.94 Mbps. Metcalfe's first experimental
network was called the Alto Aloha Network. In 1973
Metcalfe changed the name to "Ethernet," to make it clear
that the system could support any computer--not just
Altos--and to point out that his new network mechanisms
had evolved well beyond the Aloha system. He chose to
base the name on the word "ether" as a way of describing
an essential feature of the system: the physical medium
(i.e., a cable) carries bits to all stations, much the same way
that the old "luminiferous ether" was once thought to
propagate electromagnetic waves through space. Thus,
Ethernet was born.
Giuseppe Bianchi
Ethernet Standardization:
from DIX to 802.3
 1980: DIX Ethernet Standard
 DIX = Digital-Intel-Xerox vendor consortium
 Interoperable products from the three founding companies
 1982: Xerox relinquish “Ethernet” trademark
 1985: IEEE 802.3
 Ethernet becomes an IEEE 802 standard
Minor modifications vs DIX standard
Path towards worldwide interoperability
 Speed: 10 mbps
 Medium:
Thick coaxial, 500 mt max cable (10BASE5)
Thin coaxial, 185mt max cable (10BASE2)
Network extension via repeaters (up to a max limit)
Thick-RG213
Thin-RG58
Ethernet standard: the world’s FIRST open, multi-vendor standard!
Quoting Metcalfe: “the invention of Ethernet as an open, non-proprietary,
industry-standard local network was perhaps even more significant
than the invention of Ethernet technology itself”
Giuseppe Bianchi
A note on Ethernet terminology
speed
Signal method
medium
EXAMPLE: 100Base-T, 1000Base-LX, …
 Speed
 10, 100, 1000, 10G
 Signal method
 Base, Broad
Broad = RF modulated on coax
» only one case: 10BROAD36, now obsolete
 Medium
 Old notation: 2,5 = 200/500 mt (thin/thick coax)
 More recent notation: T, Tx, T4, T2, FX, X, CX, SX, LX
Depends on which specific twisted pair category & fibre category;
Different labels (e.g. T, TX, T4, T2) account for different encoding
details
Giuseppe Bianchi
Historical Ethernet topology: bus
Multiple Access shared transmission medium
 thick / thin coaxial cable
A
B
C
A
Giuseppe Bianchi
D
B
C
E
D
F
E
F
Ethernet transceiver (10base-5)
Giuseppe Bianchi
Ethernet transceiver (10Base-2)
Giuseppe Bianchi
Breakthrough idea (1990)
A
B
C
D
E
Multi-Port
Repeater
(HUB)
A
B
Giuseppe Bianchi
F
Collapsed Backbone
From Bus
To Star topology
C
D
E
F
Twisted Pair revolution
 1990: 802.3i
 10BASE-T twisted pair
 Invented by SynOptics Communications
 Alternatives
 Reuse structured cabling
 UTP (Unshielded)
system standards
 FTP (Foiled)
 Overcomes management and installation
1 shield for all the cable
problems from coaxial cabling
 STP (Shielded):
 Ethernet market takes-off!!
One shield per pair
Giuseppe Bianchi
Ethernet RJ45 connector
Giuseppe Bianchi
Signal transmission & reception on
twisted pair
 Pair 3 used for transmission
 Pair 2 used for reception
 Pairs 1 & 4 unused
 In 10Base-T & 100Base-T(x) Ethernet
 Used in 1000Base-T
Giuseppe Bianchi
Straight & crossover
 RJ45 meant to connect PC to Hub
 What About PC to PC or Hub to Hub connection?
Giuseppe Bianchi
Ethernet Frame
64-1518 bytes
8 bytes
preamble
14 bytes
Ethernet header
6 bytes
Destination address
46-1500 bytes
4 bytes
DATA
FCS
6 bytes
2 bytes
Source address
Type
 Preamble:
 For bit-level and frame-level synchronization
 Destination address:
 Who’s this frame for (who is the intended receiver)
 Hardware address (48 bits address of the network interface card)
 Source address:
 Who’s this frame from (who is the transmitter)
 Hardware address (48 bits address of the network interface card)
 Type:
 Which type of information is carried by the frame (which upper layer protocol)
 Frame Check Sequence
 Verifies if the frame has been received properly (not corrupted)
Giuseppe Bianchi
Transmission of bits
Idea:
send data + clock in the same signal!
10 mbps  Manchester encoding
1
0
1
1
0
0
When a frame is being transmitted, a signal level transition occurs every 0,2 ms
Allows to detect when transmissions are NOT occurring on the channel (IDLE channel)
Giuseppe Bianchi
Preamble (10 mbps)
8 bytes
7 bytes preamble
 7 x (10101010)  square wave @ 5 MHz
 for clock recovery
Last byte: SFD (Start Frame Delimiter)
 1 x (10101011)
 signals the start of the frame
SFD
bit sequence
1
0
Manchester
Encoding
Giuseppe Bianchi
1
0
1
0
1
1
48 bit addresses
 Typically referred to as
 Interface address
 Hardware address
 MAC address
 First bit:
 0 = physical address of an interface
Unicast address
 1 = group address
 Second bit:
 0 = globally administered address
Assigned by the manifacturer
 1 = locally administered address
First 24 bits: OUI
(Organization Unique Identifier)
(unique for each vendor)
Typically written in hex
e.g.: F0-11-00-4F-A2-1C
Each byte transmitted
from LSB to MSB
0000.1111.1000.1000.0000.0000.
1111.0010.0100.0101.1000.0011
Can be configured through driver
mcast addresses: start with 1
(first octet LSB!)
Why destination first? Station who does not match dest may ignore rest of frame!
Giuseppe Bianchi
Examples
Individual unicast: xy-xx-xx-xx-xx-xx  y multiple of 4
802.3 & 802.4: transmitted from LSB to MSB
802.5 & FDDI: transmitted from MSB to LSB
Giuseppe Bianchi
Multiple Access: Ethernet vs Aloha
Aloha:
Transmit without
listening
Ethernet:
Listen before
transmitting
Regardless of
channel activity
Transmit only if
channel idle
Transmit all the frame
Listen while
transmitting, and stop
transmission if
collision is detected
Giuseppe Bianchi
Ethernet vs Aloha: listen before tx
A
ALOHA
B
A
ETHERNET
B
Listen before transmit
Giuseppe Bianchi
Ethernet vs Aloha:
collision handling
A
ALOHA
B
Collision lasts until the end of the transmitted frames;
waste of channel capacity
A
ETHERNET
B
Detect collision; enforce collision with JAM sequence; end transmission
Giuseppe Bianchi
Carrier Sense Multiple Access
1. Listen before talking
1. NIC ready to
transmit a frame
3. Transmit frame
IFS
2. Listen for at least
an Inter Frame Spacing
(channel must be
idle meanwhile)
Giuseppe Bianchi
Ethernet Notation = Inter Packet Gap (IPG)
802.3 Notation = Inter Frame Spacing (IFS)
Minimum: 96 bits (@ 10 Mbps = 9.6 ms)
Carrier Sense Multiple Access
2. If channel detected busy: defer
1. Frame
ready
3. Busy
Detect
4. Defer
5. Listen
for ≥ IFS
2. Listen
(similar defer situation if channel immediately busy)
Giuseppe Bianchi
Collision Detection
3. Listen while talking
If collision detected:
Continue to transmit other 32 bits of signal
(Collision Enforcement Jam Signal)
Collision detection: media-dependent
On coax:
Monitor average DC voltage
When more than 1 station transmits, voltage gets greater than
given threshold
On point-to-point fiber or twisted pair links:
Collision detected by the simultaneous occurrence of activity on
both transmit and receive paths
Giuseppe Bianchi
Why collisions occur?
distance d (m)
prop delay d/200 ms
Start tx
IFS
IFS
Start tx
Detect collision
Jam sequence
Detect coll.
Jam seq.
Collision occurs if stations access the channel
in instants of time which differ for less than their
mutual propagation delay
time
Speed of EM signal in cable: ~200.000 km/s  200m/ms
Giuseppe Bianchi
When a frame is “too short”…
prop delay d/200 ms
Start tx
IFS
IFS
Can’t
detect
collision!!
Thinks
Tx was
OK
Start tx
Collision!
Detect collision
time
Transmitting
Station
Giuseppe Bianchi
Destination
station
Interferer
Network diameter
Stations placed at opposite network edges
 Essential condition:
 a station MUST be able to
detect a collision
 Shortest possible frame:
 6+6+2+46+4 = 64 bytes =
= 512 bits (excl preamble) 64 bytes
 Condition on network
diameter:
 a collision MUST be detected
on shortest possible frame
 Bound on maximum RTT
512 bits
frame
RTT=2 x prop
must be lower
than 512 bits
tx time
@ 10 Mbps: 512/10 [ms] > 2d [m] / 200 [m/ms]  d < 5120 [m]
@ 100 Mbps: 512/100 [ms] > 2d [m] / 200 [m/ms]  d < 512 [m]
NETWORK DIAMETER REDUCES AS SPEED INCREASES!!
Giuseppe Bianchi
Network diameter in practice
 Max RTT lower than 51,2 ms
 802.3 standard  max RTT = 46,38 ms
 Max link length
 Thick coax  up to 500 mt
 Fiber link betw. repeaters  up to 2 km
 Transceiver cable  up to 50 mt
 Topological limits
 5-4-3 rule: between any two nodes on
the network, there can only be
 a maximum of five segments
 connected through four repeaters
 only three of the five segments may
contain user connections.
 Repeaters add delay!
 Must account for repeater delays
 Min_TX_time >
2 x d/200 + N x Repeater_delay
MAX network diameter for 10Base-5: 2800 mt
Giuseppe Bianchi
The need for random backoff
A
IFS
IFS
IFS
IFS
B
Retransmission after collision must occur at
different instant of times
Otherwise collision will repeat forever
Deterministic rule?
Hard, for a fully distributed mechanism
Idea: generate a random time after a
collision
Different stations will extract (sooner or later) different
backoff times
Giuseppe Bianchi
Rationale for backoff “slots”
Detect coll & Jam
Detect coll & Jam
Max
RTT delay:
< 51,2 ms
Retransmit immediately
(neglect IFS for simplicity)
Schedule Retransm
After random time T…
… busy channel: defer!
T greater or equal than 51,2 us  NO COLLISION with another STA that has immediately rtx-ed
T greater or equal than N x 51,2 us  NO COLLISION with another STA that rtx-ed at (N-1) x 51,2
Giuseppe Bianchi
Backoff slot-time
First 64 bytes
Remaining frame bytes
(up to 1518 minus 64)
 Set to 512 bits
 As minimum frame size
Collision possible Collision impossible
 As maximum RTT
(Late Collision)
 guarantees that
1) A station transmitting at the beginning of the previous
backoff slot will be ALWAYS detected;
2) A station transmitting for more than 512 bytes will acquire
for sure the channel
o
o
No “late collision” possible
If they occur  misconfiguration or hardware failures
 Ethernet Backoff rule: extract a DISCRETE
value B, and schedule transmission at time
B x 51,2 us
Giuseppe Bianchi
Ethernet (truncated)
Exponential backoff rules
Once a frame collides:
End transmission with Jam signal
Let N be the retry index
N=0 for a new frame; N=1 for first rtx; N=2 for second rtx, etc
Abort transmission after N=16 unsuccessful retries
Let K=min(10,N)
Extract a random integer number R in the range:
0 ≤ R < 2K
Example: if K=6, extract R in (0, 31)
Max range: when K=10  R in (0, 1023)
Schedule retransmission at time:
Backoff = R x 51,2 us (10 mbps case)
» Where 51,2 us = tx time of 512 bits @ 10 Mbps
» 100 Mbps Ethernet: slot-time = 5,12 us
Retransmit at given time only if channel idle (otherwise defer
until channel becomes idle for an IFS)
Giuseppe Bianchi
Ethernet (R)Evolution
From multiple access to
point-to-point collision-free networks
Giuseppe Bianchi
Ethernet Repeaters
 Physical layer device
 3R functions
 Re-Shaping
Restores the proper signal waveform
 Re-Timing
Restores the proper impulse duration (clock recovery)
 Re-Transmitting
Retransmits signal on “other” port
Regardless of who’s the signal intended for
D
C
R
B
A sends frame to B
C & D receive signal, too
Giuseppe Bianchi
A
Repeaters & collisions
Retransmits collisions, too
Actually, regenerates (extending them to 96 bits) 010101…
jam sequences
And generates collisions when signal is received
simultaneously on both ends
D
C
R
B
A
A sends frame to B
C sends frame to D
Repeater generates collision (Jam signal) on both links
Giuseppe Bianchi
Multi-port repeater (hub)
 Signal regenerated & retransmitted on all repeater
ports except the port from which the signal has
arrived
R
Giuseppe Bianchi
Multiport repeaters  star topology
 Coax &
transceivers
no more
necessary!
 1990:
emergence of
10Base-T
technology
 Transmission
medium = twisted
pair point-to-point
link
 Hub distributes
signal across all
stations
Giuseppe Bianchi
Hub-based star topologies:
Why collision detection?
How collision is detected?
D
AC
BD
A
RX
TX
TX
HUB
RX
A&B detect collision by
listening on RX
RX
TX
TX
B
Giuseppe Bianchi
Signal-level collision occurs
on C&D RX links
RX
C
Ethernet bridge
 Data-link layer device
 Signal retransmission takes into account destination address
D
C
B
B
A
A sends frame to B; Bridge recognizes that station B is on the
same side the signal has been received and DOES NOT retransmit
D
C
B
B
A
A sends frame to D; Bridge recognizes that station D is on the other side: retransmits
Giuseppe Bianchi
Multiport bridge  Switch
 HUB
 broadcasts signal on all links
 Logically behaves as a bus
 Only one tx at a time
A
HUB
B
 SWITCH
 Repeats signal on specifically
addressed link
 Bridging function
 Many tx-rx pairs at a time
More bw!
Giuseppe Bianchi
A
C
D
E
F
E
F
SWITCH
B
C
D
Switches & collisions
 A sends frame to C;
 D sends frame to E
 No collision
 Simultaneous transmission
occurs
SWITCH
A
B
C
D
E
F
E
F
 A sends frame to C;
 E sends frame to C
SWITCH
 Switch understand that
two frames are
destined to a same link
 “Collision” solved by
buffering one frame &
transmitting the other
Giuseppe Bianchi
A
B
C
D
Broadcast domain vs collision domain
HUB
Switch
HUB
Without Switching
With switching
Collision
Domain
LAN
Collision
Domain
Broadcast Domain
Giuseppe Bianchi
Collision
Domain
Switch
Collision
Domain
Collision
Domain
Broadcast Domain
Full-duplexing
 Switched star network: only possible collision is on a same link
 When both station and switch simultaneously start transmission
 But rx and tx are carried out on different pairs  collision is
fictitious
 detected by the network interface card (activity sensed on Rx pair while NIC is transmitting)
 But it is not a real collision between the signals!
 it would be with a central hub; it is not with central switch
 1997: full duplex standard (802.3x)
 Simultaneously transmit and receive (2x speed increase)
 In a switched star network collision is no more possible!!
 CSMA/CD no more necessary
 Network diameter no more a concern
 Arbitrary increase of the link speed is now possible!




1998-1999: Gigabit Ethernet
2002 (july): 10 GigaEthernet
Ethernet in the metropolitan area (e.g. Fastweb, Telecom IT)
Future: 40 Giga-Ethernet, 160 Giga-Ethernet
Giuseppe Bianchi
Buffering
A
B
A
A
A
B
C
D
D
C
A
D
A
Giuseppe Bianchi
X
B
C
D
buffering
A
B
A
A
A
B
C
D
D
A
D
Giuseppe Bianchi
A
X
B
C
C
D
buffering
A
B
A
A
A
B
C
D
D
A
D
Giuseppe Bianchi
A
X
B
C
C
D
buffering
A
B
A
A
A
A
B
C
X
D
D
D
Giuseppe Bianchi
A
B
C
C
D
buffering
A
B
A
A
A
A
B
C
D
D
Giuseppe Bianchi
D
X
A
B
C
D
C
Buffers
Solve output port contention
1 packet transmitted
Other stored in waiting line
May have finite capacity
Max number of stored bytes or packets
Introduce loss!!
Packet arriving when buffer full gets lost
Introduce latency!!
Latency is NOT deterministic, but statistical
QUEUEING THEORY: we need a mathematical tool to quantitatively
evaluate loss/delay introduced by networking devices
…. and much more 
Giuseppe Bianchi
Bridge/Switch operation
Preamble +
SFD
DEST
lookup
SRC
LEN or
type
Data
PAD
FCS
 Store & Forward:
 read frame (memorize into onboard buffer)
 Check CRC
Discard frame if
» CRC fails
» too short (<64 bytes, “runt”)
» too long (>1518 bytes, “giant”)
 Look up destination into forwarding (switching)
table
 Forward packet to outgoing port
 Cut-through
 Just read first few bytes (until destination
address)
 Don’t do any check
 Look up forwarding table and select destination
 forward frame while receiving it
Giuseppe Bianchi
Store & forward vs cut-through
latency
1518 bytes frame
Assume full 8 bytes preamble received
S&F @ 10 mbps ≥ 1526*8/10 ms = 1222 ms
C-T @ 10 mbps ≥ 14*8/10 ms = 11.2 ms
S&F @ 100 mbps ≥ 122 ms
C-T @ 100 mbps ≥ 1.1 ms
Not a real problem at high rate
Giuseppe Bianchi
S&F vs CT: Fragment-free mode
 Compromise between cut-through and store-andforward
 Reads first 64 bytes
includes the frame (+LLC) header
 Then starts send packet
before the entire data field is read and the FCS is checked.
 Advantages:
 Verify reliability of header information (addresses, frame type, LLC header
information)
 Detects & discards runts & collisions
Preamble +
SFD
DEST
SRC
Cut-through
Giuseppe Bianchi
LEN or
type
Data
Fragment-free
PAD
FCS
Store & Forward
Further issues with C-T
Cut-through possible only if source
and destination ports have same bit
rate
Symmetric switching.
Different rates  buffering necessary
 S&F only
Asymmetric switching
Asymmetric switching typical in
client/server environments
More bandwidth dedicated to the server port to
prevent a bottleneck
Giuseppe Bianchi
Forwarding database
 Mapping between MAC
addresses and ports
 Ports: module/port-#
 Static entries:
 Configured by sysadmin
 Permanent database
 Dinamic entries:
 “Learned”
 Expire after ageing process
reaches upper value
 E.g. 300 seconds
 configurable
Giuseppe Bianchi
Dest MAC Address
----------------00-00-08-11-aa-01
00-b0-8d-13-1a-f1
a8-11-06-00-0b-b4
08-01-00-00-a7-64
00-ff-08-10-44-01
Ports
----1/1
1/7
2/3
2/4
2/6
Age
--1
4
0
1
5
Address Learning /1
STA 1
00-11-22-33-44-01
P1
00-11-22-33-44-01
08-55-66-77-88-02
08-aa-bb-cc-dd-03
08-01-02-f1-f2-04
P1
P1
P2
P3
P3
STA 4
08-01-02-f1-f2-04
STA 2
08-55-66-77-88-02
P2
Frame arrives at
port X
Hence it has come
from LAN attached to
port X
Giuseppe Bianchi
STA 3
08-aa-bb-cc-dd-03
SRC address
used to update
forwarding DB
SRC MAC  Port
Address Learning /2
STA 1
00-11-22-33-44-01
P1
STA 2
08-55-66-77-88-02
00-11-22-33-44-01
08-55-66-77-88-02
08-aa-bb-cc-dd-03
08-01-02-f1-f2-04
08-00-0f-cc-cc-a2
P1
P1
P2
P3
P1
5
7
0
6
0
P3
STA 4
08-01-02-f1-f2-04
P2
08-00-0f-cc-cc-a2
Incoming frame
whose SRCaddr not
in forwarding DB:
Create new entry
Ageing-time=0
Giuseppe Bianchi
STA 3
08-aa-bb-cc-dd-03
Incoming frame
whose SRCaddr
already in forwarding
DB:
Refresh ageing-time
Ageint-time=0
Address Learning /3
STA 1
00-11-22-33-44-01
P1
STA 2
08-55-66-77-88-02
00-11-22-33-44-01
08-55-66-77-88-02
08-aa-bb-cc-dd-03
08-01-02-f1-f2-04
08-00-0f-cc-cc-a2
08-00-0f-cc-cc-a2
P1
P1
P2
P3
P1
P2
5
7
4
6
2
0
P3
STA 4
08-01-02-f1-f2-04
P2
08-00-0f-cc-cc-a2
STA 3
08-aa-bb-cc-dd-03
Giuseppe Bianchi
Incoming frame
whose SRCaddr
already in forwarding
DB but associated to
different port:
Update associated port
Refresh ageing time
Frame forwarding
 Very first operation performed by the bridge/switch upon
frame reception
 Before learning
Preamble +
LEN or
DEST SRC
SFD
type
Data
PAD
FCS
1. Frame OK?
Port X
 CRC check
 Only for Store & Forward
2. Incoming port enabled
(in forwarding state)?
 Switch port may be disabled
 e.g. to isolate malfunctioning stations/LANs
Port Y
3. If DEST is NOT in forwarding DB
 broadcast frame (flooding)
 forward frame to all ports EXCEPT incoming one
4. If DEST is in forwarding DB
 Check whether DEST port = incoming port
 If YES, discard packet (dest on same LAN of src)
 If NO, forwards packet to output port
» Unless output port blocked
Flooding occurs also for broadcast frames (obvious) and for multicast frames (unless more sophisticated policies are set)
Giuseppe Bianchi
Example / 1
start-up
P1
P3
P2
Initial state: forwarding DB = empty
Giuseppe Bianchi
Example / 2
STA 1  STA 2
STA 1
00-11-22-33-44-01
00-11-22-33-44-01
P1
P1
0
P3
P2
STA 1 transmits frame to STA 2
Flooding occurs (STA2 not registered in DB)
Bridge learns STA1=P1
Giuseppe Bianchi
Example / 3
STA 2  STA 1
STA 1
00-11-22-33-44-01
00-11-22-33-44-01
00-aa-bb-cc-dd-02
P1
P1
P3
STA 2
00-aa-bb-cc-dd-02
2
0
P3
P2
 STA 2 may respond
 depends on involved protocol/app rules (e.g. TCP handshake)
 transmits frame to STA 1
 Destination selected
 Bridge learns STA2=P3
Giuseppe Bianchi
Example / 4
STA 3  STA 1
STA 1
00-11-22-33-44-01
P1
00-11-22-33-44-01
00-aa-bb-cc-dd-02
08-80-f0-00-ff-03
P1
P3
P1
12
10
0
STA 2
00-aa-bb-cc-dd-02
P3
STA 3
08-80-f0-00-ff-03
P2
 STA 3 on LAN 1 transmits to STA 1
 Frame arrives to STA1 on LAN 1
 But arrives also to Bridge
 Bridge discards frame (STA1 on same port of incoming
frame)
 This operation is referred to as FILTERING FUNCTION
 Bridge learns STA3=P1
Giuseppe Bianchi
Example / 5
STA 1 moves; STA 1  STA 3
P1
00-11-22-33-44-01
00-aa-bb-cc-dd-02
08-80-f0-00-ff-03
00-11-22-33-44-01
P1
P3
P1
P2
13
11
1
0
STA 2
00-aa-bb-cc-dd-02
P3
STA 3
08-80-f0-00-ff-03
P2
STA 1
00-11-22-33-44-01
 STA 1 moves on LAN 2
 Then transmits to STA 3
 Frame arrives to Bridge on P2, and forwarded to P1
 According to forwarding DB information
 Bridge learns that STA 1 moved
 Deletes previous entry with P1
 Adds new entry with P2
Giuseppe Bianchi
Example / 6
STA 2 moves; STA 1  STA 2
???
STA 2
00-aa-bb-cc-dd-02
P1
00-aa-bb-cc-dd-02
08-80-f0-00-ff-03
00-11-22-33-44-01
P3
P1
P2
13
3
2
P3
STA 3
08-80-f0-00-ff-03
P2
STA 1
00-11-22-33-44-01
 STA 2 moves on LAN 1
 STA 1 transmit frame to STA 2
 Frame forwarded on old port P3!!
 Bridge will learn only when STA2 will transmit first frame
 OR when ageing time will expire
and STA2  P3 entry will be removed from forwarding DB
Giuseppe Bianchi
Why a station should move?
Fault-tolerant architectures!
P1
P2
Link 1
Link 2
As link 1 fails, server switches on link 2  server MOVES from original port P2 to new port P1 !!
(need to reduce ageing time – but trade-off required: too short ageing time, too much burden on switch)
(effective solution: i) periodically send “advertisement” frames ii) send frame after switching to link 2)
Giuseppe Bianchi
Wireless LANs (Wi-Fi)
Just a very basic overview
Giuseppe Bianchi
WLAN history
 Original goal:
 Deploy “wireless Ethernet”
 First generation proprietary solutions (end ’80, begin ’90):
 WaveLAN (AT&T)
 HomeRF (Proxim)
 Abandoned by major chip makers (e.g. Intel: dismissed HomeRF in april 2001)
 IEEE 802.11 Committee formed in 1990
 Charter: specification of MAC and PHY for WLAN
 First standard: june 1997
 1 and 2 Mbps operation
 Reference standard: september 1999
 Multiple Physical Layers
 Two operative Industrial, Scientific & Medical (ISM) shared unlicensed band
»
»
2.4 GHz: Legacy; 802.11b/g
5 GHz: 802.11a
 1999: Wireless Ethernet Compatibility Alliance (WECA)
certification
 Later on named Wi-Fi
 Boosted 802.11 deployment!!
Giuseppe Bianchi
WLAN data rates
 Legacy 802.11
 Work started in 1990; standardized in 1997
 1 mbps & 2 mbps
 The 1999 revolution: PHY layer
impressive achievements
 802.11a: PHY for 5 GHz
 published in 1999
 Products available since early 2002
 802.11b: higher rate PHY for 2.4 GHz
 Published in 1999
 Products available since 1999
 Interoperability tested (wifi)
 2003: extend 802.11b
 802.11g: OFDM for 2.4 GHz
 Published in june 2003
 Products available, though no extensive
interoperability testing yet
 Backward compatibility with 802.11b Wi-Fi
 Ongoing standardization effort: 802.11n
 Launched in september 2003
 Minimum goal: 108 Mbps (but higher numbers
considered)
Giuseppe Bianchi
Standard
Transfer
Method
Freq.
Band
Data
Rates
Mbps
802.11 legacy
FHSS,
DSSS,
IR
2.4 GHz,
IR
1, 2
802.11b
DSSS,
HRDSSS
2.4 GHz
1, 2, 5.5,
11
"802.11b+"
non-standard
DSSS,
HRDSSS,
(PBCC)
2.4 GHz
1, 2, 5.5,
11,
22, 33,
44
802.11a
OFDM
5.2, 5.5
GHz
6, 9, 12,
18, 24,
36, 48,
54
802.11g
DSSS,
HRDSSS,
OFDM
2.4 GHz
1, 2, 5.5,
11; 6, 9,
12, 18,
24, 36,
48, 54
Why multiple rates?
“Adaptive” (?) coding/modulation
Example: 802.11a case
Giuseppe Bianchi
Wireless Medium Unreliability
11 Mbps 802.11b outdoor measurements - Roma 2 Campus - roof nodes
Giuseppe Bianchi
Must rely on explicit ACKs
Successful DATA transmission:
ONLY IF an ACK is received
ACK transmission provided by
MAC layer
Immediate retransmission
» Don’t get confused with higher layer rtx
DATA-ACK exchange:
Also called two-way handshake
Or Basic Access Mechanism
Giuseppe Bianchi
SENDER
RECEIVER
DATA
ACK
Must forget Collision Detection!
 One single RF circuitry
 Either TX or RX…
 Half-duplex
 Even if two simultaneous TX+RX:
large difference (100+ dB!) in TX/RX
signal power
 Impossible to receive while transmitting
 On a same channel, of course
tx
STA
rx
 Collision detection at sender:
meaningless in wireless!
 Ethernet = collision detection at sender
 Wireless = large difference in the interference
power between sender & receiver!
 Collision OCCURS AT THE RECEIVER
A detects a very low interference
(C is far)
no “collision”
A
B
Giuseppe Bianchi
C
B detects a disructive interference
(C is near)
collision occurs
Carrier Sense Multiple Access
 Station may transmit ONLY IF senses channel IDLE for a
DIFS time
 DIFS = Distributed Inter Frame Space
 Key idea: ACK replied after a SIFS < DIFS
 SIFS = Short Inter Frame Space
 Other stations will NOT be able to access the channel during
the handshake
 Provides an atomic DATA-ACK transaction
Packet
arrival
TX
DIFS
DATA
SIFS ACK
RX
OTHER
STA
Packet
arrival
DIFS
Must measure
a whole DIFS
Giuseppe Bianchi
OK!
Grasping wi-fi (802.11b) numbers
 DIFS = 50 ms
 Rationale: 1 SIFS + 2 slot-times
Slot time = 20 ms, more later
 SIFS = 10 ms
 Rationale: RX_TX turnaround time
The shortest possible!
(28+payload) [bytes] x 8 / TX_rate [mbps] = ms
DATA
PHY
MAC header 24 (30)
16
128
Preamble
48
SFD
PLCP hdr
1 mbps DBPSK
192 ms
Payload
FCS
 DATA frame: TX time = f(rate)
 Impressive PHY overhead!
 192 ms per every single frame
 Total data frame time (1500 bytes)
 @1 Mbps: 192+12288= 12480 ms
»
PHY+MAC overhead = 3.3%
»
PHY+MAC overhead = 16.%
 @11 Mbps: 192+ 1117.1 = 1309.1 ms
ACK
PHY
ACK 14
192 ms
112 ms
Giuseppe Bianchi
 Overhead increases for small frames!
 ACK frame: TX at basic rate
 Typically 1 mbps but 2 mbps possible…
 ACK frame duration (1mbps): 304 ms
And when an ACK is “hidden”?
1)
2)
3)
Sender TX
Receiver RX
STA defers
STA
SENDER
RECEIVER
STA
SENDER
RECEIVER
STA
SENDER
RECEIVER
Receiver ACKs
(after SIFS)
STA cannot hear…
STA tranmits
And destroys ACK!
STA
BUSY DETECT (DATA)
DIFS
STA TX!
SIFS
ACK
Giuseppe Bianchi
The Duration Field
Frame
Control
Duration
/ ID
Address 1
Address 2
6
6
2
2
# microseconds
0
1
2
3
4
5
6
7
8
Address 3
6
0
Sequence
Control
Address 4
Data
Frame
check
sequence
6
0-2312
4
2
When bit 15 = 1  NOT used as duration
(used by power-saving frames to specify station ID)
9 10 11 12 13 14 15
 Allows “Virtual Carrier Sensing”
 Other than physically sensing the channel, each station keeps a Network
Allocation Vector (NAV)
 Continuously updates the NAV according to information read in the
duration field of other frames
DIFS
DATA
SIFS
OTHER
STA
Giuseppe Bianchi
Physical carrier sensing
ACK
Virtual carrier sensing
NAV (data)
And when a terminal is “hidden”?
STA
RECEIVER
SENDER
… this can be “solved” by increasing the sensitiveness of the Carrier Sense…
Quite stupid, though (LOTS of side effects – out of the goals of this lecture)
… this can’s be “solved”
by any means!
 The Hidden Terminal
Problem
 SENDER and STA cannot hear each
other
 SENDER transmits to RECEIVER
 STA wants to send a frame
RECEIVER
 Not necessarily to
RECEIVER…
 STA senses the channel IDLE
 Carrier Sense failure
SENDER
STA
Giuseppe Bianchi
 Collision occurs at RECEIVER
 Destroys a possibly very long
TX!!
The RTS/CTS solution
Packet
arrival
TX
DIFS
RTS
DATA
SIFS CTS SIFS
RX
others
SIFS
ACK
NAV (RTS)
NAV (CTS)
RX
RTS
CTS
TX
CTS
hidden
(Update NAV)
Giuseppe Bianchi
RTS/CTS: carry the amount of time the channel
will be BUSY. Other stations may update a
Network Allocation Vector, and defer TX
even if they sense the channel idle
(Virtual Carrier Sensing)
Why backoff?
DIFS
DATA
STA1
SIFS ACK
DIFS
STA2
STA3
Collision!
RULE: when the channel is initially sensed BUSY, station defers transmission;
THEN,when channel sensed IDLE again for a DIFS, defer transmission of a
further random time (Collision Avoidance)
Giuseppe Bianchi
Slotted Backoff
STA2
DIFS
w=7
Extract random number
in range (0, W-1)
Decrement every slot-time s
w=5
STA3
Note: slot times are not physically delimited on the channel!
Rather, they are logically identified by every STA
Slot-time values: 20ms for DSSS (wi-fi)
Accounts for:
1) RX_TX turnaround time
2) busy detect time
3) propagation delay
Giuseppe Bianchi
Backoff freezing
When STA is in backoff stage:
It freezes the backoff counter as long as the channel is
sensed BUSY
It restarts decrementing the backoff as the channel is sensed
IDLE for a DIFS period
STATION 1
DIFS
DATA
DIFS
SIFS ACK
STATION 2
DIFS
SIFS ACK
6
5
Giuseppe Bianchi
BUSY medium
Frozen slot-time 4
DIFS
3
2
1
Why backoff between
consecutive tx?
 A listening station would never find a slot-time after the
DIFS (necessary to decrement the backoff counter)
 Thus, it would remain stuck to the current backoff counter
value forever!!
S1
DIFS
DATA
DIFS
SIFS ACK
DATA
DIFS
SIFS ACK
S2
BUSY medium
DIFS
6
5
DIFS
Frozen slot-time 4
Giuseppe Bianchi
BUSY medium
DIFS
3
Backoff rules
First backoff value:
Extract a uniform random number in range (0,CWmin)
If unsuccessful TX:
Extract a uniform random number in range (0,2×(CWmin+1)-1)
If unsuccessful TX:
Extract a uniform random number in range (0,22×(CWmin+1)-1)
Etc up to 2m×(CWmin+1)-1
Exponential Backoff!
For 802.11b:
CWmin = 31
CWmax = 1023 (m=5)
Giuseppe Bianchi
Further backoff rules
 Truncated exponential backoff
 After a number of attempts, transmission fails and frame is dropped
 Backoff process for new frame restarts from CWmin
 Protects against cannel capture
unlikely when stations are in visibility, but may occur in the case of
hidden stations
 Two retry limits suggested:
 Short retry limit (4), apply to frames below a given threshold
 Long retry limit (7), apply to frames above given threshold
 (loose) rationale: short frames are most likely generated bu realk time
stations
Of course not true in general; e.g. what about 40 bytes TCP ACKs?
Giuseppe Bianchi
DCF overhead
Sstation 
E[ payload ]
E[TFrame _ Tx ]  DIFS  CWmin / 2
TFrame _ Tx  TMPDU  SIFS  TACK
TFrame _ Tx  TRTS  SIFS  TCTS  SIFS  TMPDU  SIFS  TACK
TMPDU  TPLCP  8  (28  L) / RMPDU _ Tx
TACK  TPLCP  8  14 / RACK _ Tx
TRTS  TPLCP  8  20 / RRTS _ Tx
TCTS  TPLCP  8  14 / RCTS _ Tx
Giuseppe Bianchi
Example: maximum achievable
throughput for 802.11b
backoff
DATA
SIFS
DATA
DIFS
ACK
Cycle time


Data Rate = 11 mbps; ACK rate = 1 mbps
Payload = 1500 bytes


Data Rate = 11 mbps; ACK rate = 1 mbps
Payload = 576 bytes
TMPDU  192  8  ( 28  1500) / 11  1303
TMPDU  192  8  ( 28  576) / 11  631
TACK  192  8  14 / 1  304
TACK  192  8  14 / 1  304
SIFS  10; DIFS  50
31
E[ Backoff ]   20  310
2
1500  8
Thr 
 6.07 Mbps
1303  10  304  50  310
SIFS  10; DIFS  50
31
E[ Backoff ]   20  310
2
576  8
Thr 
 3.53Mbps
631  10  304  50  310
REPEAT RESULTS FOR RTS/CTS  Not viable (way too much overhead) at high rates!
Giuseppe Bianchi
DCF overhead (802.11b)
Giuseppe Bianchi
DCF overhead (802.11b)
1
0,8
0,7
0,6
0,5
0,4
0,3
BAS-2M bps
0,2
RTS-2M bps
0,1
BAS-11M bps
0
RTS-11M bps
10
0
30
0
50
0
70
0
90
0
11
00
13
00
15
00
17
00
19
00
21
00
23
00
Nor malized T hr oughput
0,9
Payload Size (B ytes)
Giuseppe Bianchi
Performance Anomaly
 Question 1:
 Assume that throughput measured for a single 11 mbps greedy station is approx 6 mbps.
What is per-STA throughput when two 11 mbps greedy stations compete?
 Answer 1:
 Approx 3 mbps (easy )
 Question 2:
 Assume that throughput measured for a single 2 mbps greedy station is approx 1.7 mbps.
What is per-STA throughput when two 2 mbps greedy stations compete?
 Answer 2:
 Approx 0.85 mbps (easy )
 Question 3:
 What is per-STA throughput when one 11 mbps greedy station compete with one 2 mbps
greedy station?
 Answer 3:
 ...
Giuseppe Bianchi
Understanding Answers 1&2
(neclect collision – indeed rare – just slightly reduce computed value)
backoff
Frozen backoff
STA 1
SIFS
DIFS
STA 2
ACK
SIFS
DIFS
ACK
Cycle time
Thr[1]  Thr[2] 


E[ payload ]
1500  8

E[cycle time] TMPDU [1]  SIFS  ACK  DIFS  TMPDU [2]  SIFS  ACK  DIFS  E[backoff ]
Data Rate = 11 mbps; ACK rate = 1 mbps
Payload = 1500 bytes


Data Rate = 2 mbps; ACK rate = 1 mbps
Payload = 1500 bytes
TMPDU  192  8  ( 28  1500) / 11  1303
TMPDU  192  8  ( 28  1500) / 2  6304
TACK  192  8  14 / 1  304
TACK  192  8  14 / 1  304
SIFS  10; DIFS  50
31
E[ Backoff ]   20  310
2
1500  8
Thr 
 3.3Mbps
2  (1303  10  304  50)  310
SIFS  10; DIFS  50
31
E[ Backoff ]   20  310
2
1500  8
Thr 
 0.88 Mbps
2  (6304  10  304  50)  310
Giuseppe Bianchi
Emerging “problem”:
long-term fairness!
 If you have understood the previous example, you easily
realize that
802.11 provides FAIR access to
stations
in terms of EQUAL NUMBER of
transmission opportunities in the long
term!
But
this isSTA2
INDEPENDENT
OF
STA1 STA2
STA1
STA2
STA1
transmission speed!
Giuseppe Bianchi
Computing answer 3
Frozen backoff
STA (2mbps)
SIFS
DIFS
STA 11 SIFS
ACK
DIFS
ACK
Cycle time
RESULT: SAME THROUGHPUT (in the long term)!!
Thr[1]  Thr[2] 
E[ payload ]

E[cycle time]

1500  8

TMPDU [1]  SIFS  ACK  DIFS  TMPDU [2]  SIFS  ACK  DIFS  E[backoff ]

1500  8
 1.39 Mbps!!!!!!
6304  1303  2(10  304  50)  310
DRAMATIC CONSEQUENCE: throughput is limited by
STA with slowest rate (lower that the maximum throughput
achievable by the slow station)!!
Giuseppe Bianchi
Performance anomaly into action
Why the network is
soooo slow today? We’re so
Close, we have a 54 mbps and
“excellent” channel, and we get
Less than 1 mbps …
Hahahahahah!!
Poor channel, Rate-fallbacked @ 1mbps 
Giuseppe Bianchi