Principles of Computer Networks, COMP 3203.

Evangelos Kranakis, Carleton University, SCS 1

LANS: Ethernet

October 5, 2019
Principles of Computer Networks, COMP 3203. Evangelos Kranakis, Carleton University, SCS 2

MAC and Contention

• Multiple Access Control requires contention resolution,
– i.e., prioritize the use of the medium so as to ensure good
performance.
• There are two types of contention.
– Uncoordinated
– Coordinated
• In the former, no scheduling is required, while in the latter a
coordination algorithm must be performed by all users.

October 5, 2019
Principles of Computer Networks, COMP 3203. Evangelos Kranakis, Carleton University, SCS 3

Uncoordinated Contention (1/3)

• Suppose n nodes are contending for a channel.
• A node transmits during a contention with probability p.
• Probability of a successful transmission is equal to the
probability that exactly one node transmits:

Pr[Success] = np(1 − p)n−1 .

• Pr[Success] is maximized (as a function of p) when p = 1/n.

• Why?

October 5, 2019
Principles of Computer Networks, COMP 3203. Evangelos Kranakis, Carleton University, SCS 4

Uncoordinated Contention (2/3)

• Take the derivative of the function

f (p) := np(1 − p)n−1

with respect to p.
d
f (p) = n(1 − p)n−1 + np(n − 1)(−1)(1 − p)n−2
dp
= n(1 − p)n−2 ((1 − p) − (n − 1)p)
d
• If you set dp f (p) = 0 we see that

1 − p − (n − 1)p = 0.

• Hence, p = 1/n.

October 5, 2019
Principles of Computer Networks, COMP 3203. Evangelos Kranakis, Carleton University, SCS 5

Uncoordinated Contention (3/3)

• Hence, f (p) obtains its maximum value when p = 1/n.
• In which case

Pmax := Pr[Success]
 n−1
1 1
= n 1−
n n
 n−1
1
= 1−
n
≈ 1/e.

• The average number of contentions until success occurs is

measured as the mean of a geometric distribution and is equal
to
1
= e.
1/e

October 5, 2019
Principles of Computer Networks, COMP 3203. Evangelos Kranakis, Carleton University, SCS 6

Limits of Contention
The only way to improve the performance of contention is by
limiting the number of users.
Pr [Success]

1/e

0
Number of contenders

October 5, 2019
Principles of Computer Networks, COMP 3203. Evangelos Kranakis, Carleton University, SCS 7

Access Method: Limited Contention Algorithms

• The problem with the previous approach is that it assumes
that each of the contenders wants to talk all the time.
• In practice this is not the case.

October 5, 2019
Principles of Computer Networks, COMP 3203. Evangelos Kranakis, Carleton University, SCS 8

Access Method: Limited Contention Algorithms

• After a collision if some way could be found of resolving the
collision quickly it may be possible to increase the throughput
• Algorithms for resolving collisions after they occur are
sometimes called limited contention algorithms
• Periods of normal operation (with no collisions) are punctuated
by periods where a different algorithm is used to resolve a
collision
Slots
Frame Contention Frame Idle Frame

Transmission Contention

October 5, 2019
Principles of Computer Networks, COMP 3203. Evangelos Kranakis, Carleton University, SCS 9

Limited Contention Algorithms

• Limited Contention protocols combine best properties of
1. contention (contention at low load provides low delay) and
2. collision-free (good channel efficiency at high loads).

October 5, 2019
Principles of Computer Networks, COMP 3203. Evangelos Kranakis, Carleton University, SCS 10

Types of MAC Protocols

Two basic classes of MAC protocols.
1. Random Access
(a) Slotted Aloha
(b) Unslotted (or Pure) Aloha
(c) CSMA
(d) CSMA-CD
2. Scheduling
(a) Reservation
(b) Polling
(c) Token-Passing

October 5, 2019
Principles of Computer Networks, COMP 3203. Evangelos Kranakis, Carleton University, SCS 11

Aloha Ethernet: In two Varieties

• Devised in 60’s by Abramson at U of Hawaii to interconnect
terminals located at campuses of different islands.
• A radio transmitter was attached to the terminals; a message is
transmitted as soon as it becomes available.
• If packets collide they are retransmitted.
• Adapted by Boggs and Metcalf at Xerox in 1973 to get
Ethernet
• Aloha Ethernet come in two varieties: Slotted and Uslotted.
– Slotted: time divided into slots that handle fixed length
packets with transmissions only at slot boundary
– Unslotted: no restrictions on packet size or time of
transmission

October 5, 2019
Principles of Computer Networks, COMP 3203. Evangelos Kranakis, Carleton University, SCS 12

Backoff Protocols
• Based on a probability distribution and a queue.
• There is a queue of stations (nodes) waiting to transmit.
• Each station (node) keeps track of the number of attempts to
transmit
• A function p(x) (known to all hosts) is defined in advance:
– p(x) is the probability you transmit in the x-th attempt
– same function is used by all nodes
– p(x) decreasing in x

October 5, 2019
Principles of Computer Networks, COMP 3203. Evangelos Kranakis, Carleton University, SCS 13

Backoff Algorithms
• Implementation is done as follows.
– Station i has a variable bcki and uses the following protocol:
1. Initialize bcki to 0;
2. Station attempts transmission if queue is not empty with
probability p(bcki );
(a) if it fails to transmit due to collision it increments the
variable bcki ;
(b) Else it assigns 0 to the variable bcki ;
3. If queue is empty and no transmission is made the
variable bcki remains unchanged.
• Major concern:
how stable is traffic under various functions p(x)?

October 5, 2019
Principles of Computer Networks, COMP 3203. Evangelos Kranakis, Carleton University, SCS 14

Aggressiveness: Binary Exponential Backoff

• The function p(x) is important:
– it tells you how aggressive you should be in your x-th
attempt to transmit.
• Backlogged packet attempts with probability p(x) where x is
the # of unsuccessful attempts.
1. Exponential backoff: p(x) = 2−x .
1
2. Polynomial backoff (k is constant): p(x) = (x+1)k
.
– k = 0: constant backoff
– k = 1: linear backoff
– k = 2: quadratic backoff
• Polynomial backoff has been shown to be stable for k > 1.

October 5, 2019
Principles of Computer Networks, COMP 3203. Evangelos Kranakis, Carleton University, SCS 15

Carrier Sense Multiple Access/Collision Detection

There are two important aspects to Ethernet:
• CSMA
– whereby the carrier is sensed for the presence of signals, and
• CD
– whereby the hosts attempt detection of collisions.
– This is used to cut down on unnecessary collisions!

October 5, 2019
Principles of Computer Networks, COMP 3203. Evangelos Kranakis, Carleton University, SCS 16

Ethernet CSMA/CD
• CSMA (Carrier Sense Multiple Access)
1. A station wishing to transmit must first listen for existing
traffic on the line.
2. If no voltage detected, the line is considered to be idle and a
transmission is initiated.
• CD (Collision Detection)
1. If no voltage detected, a transmission is initiated. During
transmission, station checks line for extremely high voltages
that indicate collisions.
2. If collision detected station waits a predetermined amount
of time for the line to clear (backoff) and sends data again.

October 5, 2019
Principles of Computer Networks, COMP 3203. Evangelos Kranakis, Carleton University, SCS 17

What Do I Do after a Collision? Exponential Backoff !

A B
Collision!
1. After 1st collision, each station “waits at random” either 0 or 1
time slots and tries again.
2. If they pick same random value they collide again. After 2nd
collision, each station “waits at random” either 0 or 1 or 2 or 3
time slots and tries again.
3. If they pick same random value they collide again. After 3rd
collision, each station “waits at random” either 0 or 1 or 2 or 3
or 4 or 5 or 6 or 7 time slots and tries again.
4. And so on...

October 5, 2019
Principles of Computer Networks, COMP 3203. Evangelos Kranakis, Carleton University, SCS 18

How Long Does it Take before a Successful Transmission?

• Remember, collisions are random events!
• It makes no sense to talk about deterministic time!
• It is more precise to talk about expected time!
• The expected time depends on the backoff protocol being used!

October 5, 2019
Principles of Computer Networks, COMP 3203. Evangelos Kranakis, Carleton University, SCS 19

Expected Time
• Suppose that two synchronous ethernet stations A, B are at
contention period t:
• This means, they have failed in previous contention periods
0, 1, . . . , t − 1, and are now in contention period t.
• They each pick a random number in the interval 0..2t − 1.
– A picks random value a in 0..2t − 1.
– B picks random value b in 0..2t − 1
• The expected number of steps to reach contention period t is
cumulative,
– i.e., you must add the “waiting times”, for all the trials
before t.

October 5, 2019
Principles of Computer Networks, COMP 3203. Evangelos Kranakis, Carleton University, SCS 20

How Persistent Should I Be?

• Successful transmission (i.e., no collision) can occur only when
the two stations select different values at random, i.e., at 6= bt .
• Hence, if that happens then
2t (2t − 1) 1
Pr[no collision] = = 1 −
(2t )2 2t
• The waiting time for a success in t-th attempt will be

min{at , bt },

where at , bt are the values chosen above!

• NB: The bigger the t is the higher (closer to 1) is the
probability that you will succeed next time (t + 1).
• NB: Don’t forget that there are many stations (not just two)
contending for the medium!

October 5, 2019
Principles of Computer Networks, COMP 3203. Evangelos Kranakis, Carleton University, SCS 21

How Persistent Should I Be?

idle
A B

A B
What do I do now?

How persistent should I be?

October 5, 2019
Principles of Computer Networks, COMP 3203. Evangelos Kranakis, Carleton University, SCS 22

Variants of CSMA: Smart Transmitters

Slots
Frame Contention Frame Idle Frame

Transmission Contention
persistence

• When a station has data to send and line is busy it waits for
the line to go idle.
• How persistent is a transmitter in capturing an idle channel?

October 5, 2019
Principles of Computer Networks, COMP 3203. Evangelos Kranakis, Carleton University, SCS 23

Avoid transmissions that are certain to cause collisions!

• In most LAN systems, a node can sense if another node is
using the channel after some delay for the signal to reach it
• CSMA senses the medium for the presence of carrier signal.
• In this case, a node can wait until the channel becomes idle
before transmitting
• Schemes that take advantage of this are called Carrier Sense
Multiple Access (CSMA)
1. 1-Persistent CSMA
2. Non-persistent CSMA
3. p-persistent CSMA

October 5, 2019
Principles of Computer Networks, COMP 3203. Evangelos Kranakis, Carleton University, SCS 24

Slotted CSMA: 1-Persistence

When node has data to send.
1. First listens to channel to see if anybody else is transmitting.
2. If channel busy the node waits until channel becomes idle.
3. When node detects idle channel it transmits frame.
4. If collision occurs node waits a random amount of time and
starts over again.
1-persistent means that a transmitter with a packet to send
transmits with probability 1 whenever a busy line goes idle.

October 5, 2019
Principles of Computer Networks, COMP 3203. Evangelos Kranakis, Carleton University, SCS 25

Slotted CSMA: Non-Persistence

Nonpersistent: Do not persist! Act like a collision and wait a
random number of slots before retrying.
When node has data to send.
1. First listens to channel to see if anybody else is transmitting.
2. When node detects idle channel it transmits frame.
3. If channel is busy the node does not continually listen in order
to seize it for transmission when idle. Instead it waits a
random period of time and then repeats the algorithm.
Non-Persistenxe behaves better than 1-persistence.

October 5, 2019
Principles of Computer Networks, COMP 3203. Evangelos Kranakis, Carleton University, SCS 26

Slotted CSMA: p-Persistence

p-Persistent: send with probability p in next available slot
When node has data to send.
1. First listens to channel to see if anybody else is transmitting.
2. If channel busy wait until next slot and try again.
3. If channel idle then Repeat until either sucess or another node
begins transmitting.
(a) it transmits with probability p.
(b) it defers until next slot with probability 1 − p.
4. if another node begins transmitting node acts as if there is
collision: waits a random amount of time and starts again.

October 5, 2019
Principles of Computer Networks, COMP 3203. Evangelos Kranakis, Carleton University, SCS 27

(Example) Slotted CSMA: p-Persistence

• A p-persistent protocol transmits with probability p, where
0 ≤ p ≤ 1, after a line becomes idle.
• E.g., Assume p = 1/4 and 100 stations are waiting for the busy
line to become idle.
• Each of these stations transmits with probability 1/4.
• Hence only ≈ 25 stations will attempt to transmit and the
remaining ≈ 75 stations defer until next time slot.

October 5, 2019
Principles of Computer Networks, COMP 3203. Evangelos Kranakis, Carleton University, SCS 28

CSMA Performance

0.01 persistent

Nonpersistent
Throughput per packet time

0.5−persistent
CSMA
1−perstent
CSMA
Slotted Aloha

Pure Aloha

0 Attempts per packet time

October 5, 2019
Principles of Computer Networks, COMP 3203. Evangelos Kranakis, Carleton University, SCS 29

Ethernet CSMA-CD
CSMA improves over Aloha because it senses the carrier.
Still, however, collisions involve entire packets!
An obvious improvement is to abort transmission when collision is
detected!
1. Station senses channel
2. if channel idle then transmit
3. else use a CSMA strategy (p-, non-persistent)
4. if collision detected during transmision
5. then transmit jamming signal and abort transmission
6. schedule future transmission with backoff.

October 5, 2019
Principles of Computer Networks, COMP 3203. Evangelos Kranakis, Carleton University, SCS 30

Collision-Free Reservation Protocols: Basic Bit-Map

Assume there are exactly N stations labeled 0 to N − 1. Let the
propagation delay be negligible.
Basic Bit-Map Method:
1. Each contention period is exactly N slots.
2. If station i has a packet to send then it inserts a 1 bit in the
i-th slot. No other station is allowed to transmit during this
slot.
3. After N slots have passed by, each station has complete
knowledge of which stations wish to transmit.
4. Stations now begin transmitting in numerical order.

October 5, 2019
Principles of Computer Networks, COMP 3203. Evangelos Kranakis, Carleton University, SCS 31

Analysis of Basic Bit-Map

Contention Resolution FRAMES Contention Resolution FRAMES
1 1 Frame Frame 1 Frame
1 2 3 4 5 6 2 5 4 4

Time

A station is out of luck if it becomes ready just after its bit slot
passes by.

October 5, 2019
Principles of Computer Networks, COMP 3203. Evangelos Kranakis, Carleton University, SCS 32

Collision-Free Reservation Protocols: Binary Countdown

Here we assume all addresses are in binary and of the same length.
Binary
Bidding Cycle
Address
0 0 1 0 0 − − −

0 1 0 0 0 − − −

1 0 0 1 1 0 0 −

1 0 1 0 1 0 1 0 Winning Node

1 0 1 0 Result

Nodes write their bit starting from highest to lowest order to a

register. They drop out as long as another node has written a 1 in
current position, but continue otherwise.

October 5, 2019
Principles of Computer Networks, COMP 3203. Evangelos Kranakis, Carleton University, SCS 33

Splitting Algorithms
• Idea is to partition set of nodes into “groups”.
• Each group is assigned to a slot.
• Thus a collision can only occur within a group.
• Example 1: Assume the nodes have unique (fixed length)
identifiers (e.g., 1, 2, . . . , m)
– Divide nodes into pairs.
– Node 2i or 2i + 1 goes in slot i
– If collision occurs they go in order

October 5, 2019
Principles of Computer Networks, COMP 3203. Evangelos Kranakis, Carleton University, SCS 34

Paradigm
• Similar to US Army testing for syphilis. Take blood test from
n soldiers.
• A portion of each sample is poured into a single test tube. This
is tested for antibodies.
• If none found all soldiers in group declared healthy.
• If antibodies were found then split soldiers into two groups:
1..n/2 and n/2 + 1..n and iterate.
• Eventually all infected soldiers are found!

October 5, 2019
Principles of Computer Networks, COMP 3203. Evangelos Kranakis, Carleton University, SCS 35

Tree Splitting Algorithm

Nodes are leaves of a tree.
1

2 3

4 5 6 7

a b c d e f g h

During first slot any node of the tree rooted at 1 may contend.
If collision occurs, in next slot only nodes in left subtree (rooted at
2) may contend.
Again, if collision occurs, in next slot only nodes in left subtree
(rooted at 4) may contend.

October 5, 2019
Principles of Computer Networks, COMP 3203. Evangelos Kranakis, Carleton University, SCS 36

Tree Algorithm: Example

Eight stations a, b, . . . , h contend for a shared channel. Stations
a, c, d, f, g suddenly become ready at once!
Slot 1: a, c, d, f, g; (collision)
Slot 2: a, c, d; (collision)
Slot 3: a; (success)
Slot 4: c, d; (collision)
Slot 5: c; (success)
Slot 6: d; (success)
Slot 7: f, g; (collision)
Slot 8: f ; (success)
Slot 9: g; (success)

October 5, 2019
Principles of Computer Networks, COMP 3203. Evangelos Kranakis, Carleton University, SCS 37

Exercisesa
1. What is the difference between coordinated and uncoordinated
access?
2. Suppose that n nodes in uncoordinated contention select
1
p = kn , where k ≥ 1 is a given integer. What is the Pr[Success]
as a function of k? Derive the formula.
3. Suppose that in uncoordinated contention at any given time at
√
most n of the hosts contend for the channel. What is the
Pr[Success]?
4. What makes unslotted Aloha less efficient than slotted Aloha?
What is the tradeoff?
5. In what ways do exponential and polynomial backoff
algorithms differ?
a Do not submit.

October 5, 2019
Principles of Computer Networks, COMP 3203. Evangelos Kranakis, Carleton University, SCS 38

6. Eight stations a, b, . . . , h contend for a shared channel. Stations

c, d, f, g suddenly become ready at once!. Describe how the
splitting algorithm allocates slots.
7. Use binary countdown to resolve contention between hosts with
the following addresses:

01001011, 01001010, 00000011

October 5, 2019