Media Access Control

1
Media Access Control

• Basics
• Physical Structures
• Protocol structure (LLC and MAC)
• MAC access procedures
• Random access (Aloha, CSMA, CSMA-CD)
• Scheduled (reservation, polling, token passing)
• LAN Standards
• Ethernet, Rings, wireless
• LAN Bridges

2
Shared Medium LAN
3
2
4
1
Shared Multiple Access Medium
5
M
?

• Stations share the physical medium
• Broadcasting is possible
• No explicit routing
• Access to the medium must be controlled

Figure 6.1
3
Techniques for medium sharing
Medium Sharing Techniques
Static Channelization
Dynamic Medium Access Control
CDMA Wireless FDMA TDMA
Scheduling
Random Access
Token Passing Rings Polled systems
Ethernet and variants Wireless LANs
Figure 6.2
4
Collisions in Random Access
Distance d meters tprop d / ? seconds
A transmits at t 0
A
B
B transmits before t tprop and
detectscollision shortly thereafter
A
B
A detects collision at t 2 tprop
A
B

• Overlapping messages collide on the medium
• Not all stations see the collision at the same
  time
• Detection of Collisions can take 2 tprop,
  generally wasted time
• Delay bandwidth product (2tpropR) is wasted
  bandwidth.
• Efficiency L/(L2tpropR) 1/(12a), a tpropR/L

Figure 6.7
5
Performance of random access

• tpropR/L is a critical parameter
• The delay bandwidth product in frames
• Measures bandwidth lost due to the need to
  coordinate
• Results in lower maximum throughput than R
• Frames will in general queue for access to the
  medium
• Ideally an M/M/1 or M/D/1 queue (no losses)
• Delay increases dramatically as load approaches
  capacity.

6
Delay (in frames) versus load
Capacity lost due to coordination
ET/EX
Transfer Delay
1
r
rmax
1
Load
Figure 6.8
7
Dependence of throughput on aRtprop/L
a? gt a
ET/EX
a
a?
Transfer Delay
1
r
rmax
r?max
1
Load
Figure 6.9
8
Random Access Aloha

• Originally done for radio network on Hawaiian
  Islands
• Each station broadcasts a data packet when it has
  one to send.
• If transmission doesnt overlap any other it
  succeeds
• If transmission overlaps another, both will be
  retransmitted
• Retransmission time must be randomized to make
  subsequent collisions less likely.

9
Analyzing Aloha
First transmission
Retransmission
t
t0
t0X
t0-X
t0X2tprop??
t0X2tprop
Vulnerable period
Backoff period
Time-out
Retransmission if necessary

• Let S be the arrival rate of new packets (same as
  successful transmissions)
• Let G be the total arrival rate (new and
  retransmitted)
• S G?Pno collision
• Transmission is successful if no arrivals in a
  period of 2X
• Pk transmissions in 2X (2G)k/k! ? e-2G
• ? S G ? (2G)0/0! ? e-2G Ge-2G

Figure 6.16
10
Throughput of Aloha
0.368
Ge-G
S
(Slotted Aloha)
0.184
(Aloha)
Ge-2G
G

• Peak throughput is 1/(2e)
• Throughput decreases at increasing load (more
  retransmissions)
• Unstable (if offered load greater than 1/(2e)
  system cant carry it and throughput goes down)

Figure 6.17
11
Slotted Aloha
t
(k1)X
t0 X2tprop
kX
t0 X2tpropB
Backoff period
Time-out
Retransmission if necessary
Vulnerable period

• Stations start transmissions only at fixed
  timeslots
• Period where collisions occur is reduced to one
  slot time (X)
• Performance characterized by SGe-G
• Peak performance is 1/e (0.36), still unstable
• Difficult to achieve in any large network

Figure 6.18
12
Packet delay in Aloha

• Successful packets are delayed by Xtprop
• Each subsequent transmission delays by
  X2tpropB, where B is backoff interval
• ET Xtprop(e2G-1)(X2tpropB)
• ET/X 1a (e2G-1)(12aB/X) where atprop/X
• For Slotted Aloha
• ET/X 1a (eG-1)(12aB/X)

13
Carrier Sense MA (CSMA)

• Concept -gt limit probability of collision by
  sensing whether a transmission is in progress
• Vulnerable time is now only the length of one
  propagation delay
• Lower vulnerability may mean lower collision
  probability and less wasted bandwidth
• Different variants depending on what a station
  sensing busy does

Figure 6.19
14
CSMA Variants

• 1-Persistent Station sensing busy transmits
  immediately when transmission is done
• collides with anyone else with a packet arriving
  during the transmission (bad)
• non-persistent station backs off immediately
• Collision probability is random, but may take
  extra delay
• p-persistent Station tries to transmit after
  each propagation time interval with probability p

B
sensing
Figure 6.20
15
non-persistent CSMA Performance
S
0.81
Non-Persistent CSMA
0.01
a (tprop/X)
0.51
0.14
0.1
G
1

• Maximum throughput strongly depends on a
• High throughput possible (but at high collision
  level)

Figure 6.21 - Part 1
16
1-persistent CSMA Performance
S
0.53
1-Persistent CSMA
a (tprop/X)
0.01
0.45
0.16
0.1
G
1

• Lower peak performance, faster fall off in
  performance
• peak occurs with fewer collisions

Figure 6.21 - Part 2
17
CSMA with Collection Detection
It takes 2 tprop to find out if channel has been
captured

• Stations look for collisions when transmitting
• In case of collision, station aborts transmission
• Saves bandwidth wasted during collision

Figure 6.22
18
Analyzing p-persistent CSMA/CD
frame
contention
frame
idle
Contention period is sequence of mini-slots
(2tprop)

• Probability of 1 successful transmission

With n stations transmitting with probability p
Psuccess is maximized at p1/n
For large n
Figure 6.23
19
CSMA/CD Performance

• Average number of mini-timeslots
• EJ ?j(1-Pmax)j-1Pmax 1/Pmax e
• Maximum throughput occurs when there is no idle
  time.
• Each transmission is followed by a contention
  interval
• Fraction of bandwidth EX/(EXtprope
  2tprop)
• 1/(1(2e1)a) 1/(16.44a)

20
Throughput for Random Access
CSMA/CD
1-P CSMA
Non-P CSMA
?max
Slotted Aloha
Aloha
a

• High throughput if propagation delay is much less
  than frame time.
• CSMA/CD and CSMA have better throughput with low
  propagation delay (e.g. ethernet)
• Slotted Aloha does better with high delay (e.g.
  satellite)

Figure 6.24
21
Reservation Systems
Reservation interval
Data Transmissions
r
d
r
d
d
d
d
d
time
1 frame
1 frame
Each station has own minislot for making
reservations
r

3
M
2
1

• Stations take turns transmitting during a
  variable length frame
• Each frame preceded by reservation interval.
• With long propagation delays reservations apply
  to future frames
• If each minislot is v frame times, maximum
  throughput is 1/(1v) and occurs when every
  station is transmitting.
• Reservation system can reserve more than one slot
  at once
• 1/(1v/k) for k slots at once.

Figure 6.25
22
Aloha based reservations

• Reservation interval overhead grows with the
  number of stations even if each rarely uses it.
• Solution Instead of dedicated slots, contend
  for slots (slotted aloha)
• Slotted aloha means only 1/e of reservation slots
  are used, so e are needed.
• Efficiency is thus 1/(1e v). -- gt88 for v5.
• Scheme is used in GPRS cellular.

23
Polling Systems
(a)
Central Controller
(b)
(c)
Distributed Algorithm

• Stations take turns sending if they have material
  to send
• Turns coordinated by a controller or via a
  distributed algorithm
• No collisions and contention, but a station has
  to wait for its turn

Figure 6.27
24
Polling Performance
polling messages

2
1
4
5
3
1
2
M
t
packet transmissions

• walk time is time spent polling before the next
  station transmits.
• ? is total walk time during one cycle
• Tc is total time for one cycle
• X is the time to transmit a packet
• With ? arrivals, (load ? ? EX)
• ETc ?/(1-?)
• Dominated by walk time for light load, dominated
  by M/M/1 queuing for heavy load.

Figure 6.28
25
Packet Delay for Polling
10
5
1
ET/EX
0.5
0
?

• For constant length packets (M/D/1)
• ET/EX ?/(2(1-?) a(1-?/M)/(2(1-?))1
  ?avg/EX
• For alt1, performance is like ideal M/D/1 queue
• Large a increases delay even at idle and reduces
  capacity.
• a depends on Number of stations, Time to send
  poll, propagation delay, and time for nodes to
  respond.

Figure 6.29
26
Ring Networks
listen mode
transmit mode
input from ring
output to ring
delay
delay
to device
from device

• Stations linked in a ring topology
• Each station can either retransmit incoming
  signal or insert its own.
• Contention resolved by token passing

Figure 6.30
27
Token Passing Schemes
Single Packet
Multi-token
Single Token
d
a)
b)
c)
d
d
d
d
d
d
d
Busy token
d
d
Free token
Token inserted after last data bit is sent
Token inserted after first data bit returns
Token inserted after last data bit returns

• Ring Latency time for a bit to make a
  complete loop
• M?bits/stationpropagation delay
• a ring latency/frame transmit time

Figure 6.31
28
Maximum utilization
M50
Multiple Token Operation
M10
M50
M10
Maximum Throughput
Single Packet Operation
Single Token Operation
a ?

• Multi-Token ?max1/(1a/M)
• Single Token ?max1/(MAX(1,a) a/M)
• Single Packet ?max1/(1a(11/M))
• Multi-token provides better performance for high
  latency.

Figure 6.32
29
Waiting Time, unlimited transmission/token

• Like Polling.
• Two components
• waiting for token
• waiting for other packets ahead
• For low a, like M/D/1 Queue.
• high a increases wait time but doesnt reduce
  maximum throughput

Figure 6.33
30
Waiting time, 1 packet/token
10
1
0
0.1

• Curves for M32, multi-token ring
• Need to wait for a token for each packet
• More complex mathematically
• Note impact of reduced throughput at high load

Figure 6.34
31
Waiting Time, single-packet ring
1
0
0.1

• M32.
• Throughput significantly limited by value of a
• Limited throughput make this mode impractical for
  rings with high latency.

Figure 6.35
32
Summary Scheduled Approaches

• Reservation System
• Simple model, sensitive to number of stations and
  latency
• Polling Systems
• Performance depends on walk time and number of
  nodes
• maximum utilization near 1with high delay
• Rings
• performance critically depends on ring latency
• with limited packets/turn utilization reduced
• multi-token operation gives best utilization with
  high latency

33
Summary of medium access
34
Channelization

• Frequency division (FDMA)
• Some bandwidth wasted on guard bands
• Assigned channels are low speed (higher delay)
• Time division (TDMA)
• Some bandwidth wasted on synchronization
  intervals
• Assigned channels are high speed (lower delay)
• Code division (CDMA) (spread spectrum)
• channels overlap in time and frequency, separated
  by redundancy and orthogonal codes
• high speed, accommodates variable rates, number
  of nodes.

35
For Next week

• Problem set
• Read Chapter 7 sections 7.1-7.3