Where Are We?
Layered Architecture
Basics:
Network Classification
Network Architecture
Delay Models
Application
Transport
Implementation:
Protocol Design
Network
Data Link
Physical
1
2
Data Link Layer
Multiaccess Media
Functionality
• Reliable Delivery of Frames
A
• Flow Control
B
C
D
E
F
• Error Detection
• Error Correction
3
4
Multiaccess Media
Multiaccess Media
Rules
• “Don’t interrupt when someone else is speaking”
• “Raise your hand if you have a question”
• “Give everyone a chance to speak”
Wavelan
Cocktail Party
6
5
Multiaccess Protocols
Protocols for Multiaccess Networks
• Channel Partitioning (Wireless Communication)
• Random Access (Ethernet)
A
B
C
D
E
F
• Taking Turns (Token Ring)
• Hosts broadcast packets
• When a collision occurs, all transmitted packets are lost
• Lost packets have to be retransmitted
=> Need Multiaccess Protocol
7
8
Protocols for Multiaccess Networks
Model - Slotted Aloha
Goal:
• Time is divided into slots:
• Understand Multiaccess Protocols
• Understand Ethernet Protocol
L
C
Issues:
unit time =
• How to deal with collisions?
• Maximal traffic load?
L
C
seconds
• Packet arrival rate (over all hosts) of λ packets/time unit
(− > Protocol design)
• Collision or Perfect Reception
(− > Protocol performance)
• Immediate Feedback: 0, 1, e
• Retransmission Probability: qr
• Infinite number of hosts i.e. each node has at most one packet
to transmit)
9
10
Model - Slotted Aloha
Model - Slotted Aloha
Questions
Notation
• Throughput?
• λ: aggregated arrival rate
• n: number of backlogged packets
• G(n) = nqr : average number of arrivals per time slot
• How to choose qr ?
Want to compute
Would qr = 1 work?
Probability for retransmission after k − 1 slots:
Average time until retransmission:
• Psucc : probability of successful transmission in a time slot (as a
function of G(n))
• Throughput =
11
Psucc
1
12
Model - Slotted Aloha
Psucc
Model - Slotted Aloha
n
n−1
nqr 1 − qr
= nqr 1 − qr
=
1 − qr
For qr small, we have
n
1 − qr ≈ e−nqr
and
G e -G
e-1
nqr
≈ nqr ,
1 − qr
and we obtain
G
G=1
Psucc
≈ nqr e−nqr = G(n)e−G(n)
• If G(n)e−G(n) > λ:
where G(n) = nqr
• If G(n)e−G(n) < λ:
Note: Poisson distribution with parameter G:
• Optimal G(n) = nqr = 1, or
k
pk =
G −G
e
k!
qr =
13
Model - Slotted Aloha
What did we learn?
• λmax = e−1 ≈ 0.368
• qr should dynamically change
Binary Exponential Backoff
• qr = 2−k
15
1
n
14