Lecture 6:

Media Access Control

CSE 123: Computer Networks
Stefan Savage
Today: Media access
 How to share a channel among different hosts?

 Channel partitioning
 FDMA (frequency division multiple access)
 TDMA (time division multiple access)
 CDMA (code division multiple access)

 Random access
 Contention-based
» Aloha
» CSMA, CSMA/CD, CSMA/CA
» Ethernet, 802.11
 Contention-free
» Token-ring, FDDI
Channel Partitioning
 Need to share media with multiple nodes (n)
 Multiple simultaneous conversations

 A simple solution
 Divide the channel into multiple, separate channels

 Channels are completely separate

 Bitrate of the channel is split across channels
 Nodes can only send/receive on their assigned channel

 Several different ways to do it…

3
Frequency Division (FDMA)
 Divide bandwidth of f Hz into n channels each with
bandwidth f/n Hz
 Easy to implement, but unused subchannels go idle
» Also need “guard bands” between channels to prevent interference
 Used by traditional analog cell phone service, radio, TV
Amplitude

Frequency
Amplitude

Frequency
4
Time Division (TDMA)
 Divide channel into rounds of n time slots each
 Assign different hosts to different time slots within a round
 Unused time slots are idle
 Used in GSM cell phones & digital cordless phones

 Example with 1-second rounds

 n=4 timeslots (250ms each) per round
Host #

1 2 3 1 2 3 4 2 4

1 sec 1 sec 1 sec

5
Code Division (CDMA)
 Do nothing to physically separate the channels
 All stations transmit at same time in same frequency bands
 One of so-called spread-spectrum techniques

 Sender modulates their signal on top of unique code

 Sort of like the way Manchester modulates on top of clock
 The bit rate of resulting signal much lower than entire channel

 Receiver applies code filter to extract desired sender

 All other senders seem like noise with respect to signal

 Used in some newer digital cellular technologies

(Verizon/Sprint US Cellular)
6
Partitioning Visualization
FDMA
power

TDMA
power

CDMA

power

Courtesy Takashi Inoue

7
Problem w/Channel partitioning
 Not terribly well suited for random access usage
 Why?

 Instead, design schemes for more common situations

 Not all nodes want to send all the time
 Don’t have a fixed number of nodes

 Potentially higher throughput for transmissions

 Active nodes get full channel bandwidth

8
Aloha
 Designed in 1970 to support wireless data connectivity
 Between Hawaiian Islands—rough!

 Goal: distributed access control (no central arbitrator)

 Over a shared broadcast channel

 Aloha protocol in a nutshell:

 When you have data send it
 If data doesn’t get through (receiver sends acknowledgement)
then retransmit after a random delay
 Why not a fixed delay?

9
Collisions
 Frame sent at t0 collides with frames sent in [t0-1, t0+1]
 Assuming unit-length frames
 Ignores propagation delay

10
Slotted Aloha
 Time is divided into equal size slots (frame size)
 Host wanting to transmit starts at start of next slot
 Retransmit like w/Aloha, but quantize to nearest next slot
 Requires time synchronization between hosts

Success (S), Collision (C), Empty (E) slots

11
Channel Efficiency
Q: What is max fraction slots successful?
A: Suppose n stations have packets to send
 Each transmits in slot with probability p
At best: channel
 Prob[successful transmission], S, is: used for useful
transmissions 37%
S = p (1-p)(n-1) of time!
0.4

 any of n nodes: 0.3

Slotted Aloha
0.2
S = Prob[one transmits] = np(1-p)(n-1) 0.1
Pure Aloha
(optimal p as n->infinity = 1/n)
0.5 1.0 1.5 2.0
= 1/e = .37 offered load = n X p

12
Carrier Sense (CSMA)
 Aloha transmits even if another host is transmitting
 Thus guaranteeing a collision

 Instead, listen first to make sure channel is idle

 Useful only if channel is frequently idle
 Why?

 Listen how long to be confident channel is idle?

 Depends on maximum propagation delay
 Small (<<1 frame length) for LANs
 Large (>>1 frame length) for satellites

13
Retransmission Options
 non-persistent CSMA
 Give up, or send after some random delay
 Problem: may incur larger delay when channel is idle

 1-persistent CSMA
 Send as soon as channel is idle
 Problem: blocked senders all try to send at once

 P-persistent CSMA
 If idle, send packet with probability p; repeat
 Make sure (p * n) < 1

CSE 123 – Lecture 6: Media Access Control 14

Jamming
 Even with CSMA there can still be collisions. Why?

Time for B to detect A’s transmission

X (wire)
collision
A B
 If nodes can detect collisions, abort! (CSMA/CD)
 Requires a minimum frame size (“acquiring the medium”)
 B must continue sending (“jam”) until A detects collision

 Requires a full duplex channel

 Aside: wireless is typically half duplex; need an alternative for
wireless channels (we’ll return to this)
15
Collision Detection
 How can A know that a collision has taken place?
 Worst case:
» Latency between nodes A & B is d
» A sends a message at time t and B sends a message at t + d – epsilon
(i.e., just before receiving A’s message)
 B knows there is a collision, but not A… B must keep transmitting so A
knows that its packet has collided
 How long? 2 * d (keep talking until you know everyone has heard it)

 IEEE 802.3 Ethernet specifies max value of 2d to be 51.2us

 This relates to maximum distance of 2500m between hosts
 At 10Mbps it takes 0.1us to transmit one bit so 512 bits take 51.2us to send
 So, Ethernet frames must be at least 64B (512 bits) long
» Padding is used if data is too small

 Transmit jamming signal to insure all hosts see collision

 48 bit times sufficient
16
Ethernet
 First local area network (LAN)
 Developed in early ’70s by Metcalfe and Boggs at PARC
 Originally 1Mbps, now supports 10Mbps, 100Mbps, 1Gbps
and 10Gbps flavors (40/100G in development)
 Currently the dominant LAN technology
 Becoming the dominant WAN technology

17
Classic Ethernet
 IEEE 802.3 standard wired LAN
(modified 1-persistent CSMA/CD)
 Classic Ethernet: 10 Mbps over coaxial cable
 All nodes share same physical wire
 Max length 2.5km, max between stations 500m
(wire)
nodes
 Framing
 Preamble, 32-bit CRC, variable length data
 Unique 48-bit address per host (bcast & multicast addrs too)
Preamble (8) Source (6) Dest (6) Len (2) Payload (var) Pad (var) CRC (4)

18
Ethernet improvements

 Problems with random delay with fixed mean

 Few senders = unnecessary delay
 Many senders = unnecessary collisions

 Binary exponential back-off balances delay w/load

 First collision: wait 0 or 1 min frame times at random, retry
 Second time: wait 0, 1, 2, or 3 times
 Nth time (n<=10): wait 0, 1, …, 2n-1 times
 Max wait 1023 frames; give up after 16 attempts

19
Capture Effect
 Randomized access scheme is not fair

 Suppose stations A and B always have data to send

 They will collide at some time
 Both pick random number of “slots” (0, 1) to wait
 Suppose A wins and sends
 Next time the collide, B’s chance of winning is halved
» B will select from 0,1,2,3 due to exponential back-off

 A keeps winning: said to have captured the channel

20
Ethernet Performance
 Much better than Aloha or CSMA in practice

 Source of protocol inefficiency: collisions

 More efficient to send larger frames
» Acquire the medium and send lots of data
» Less time on arbitration (figuring out who gets to send)

 Less efficient if
» More hosts – more collisions needed to identify single sender
» Smaller packet sizes – more frequent arbitration
» Longer links – collisions take longer to observe, more wasted
bandwidth

21
Contention-free Protocols
 Problem with channel partitioning
 Inefficient at low load (idle subchannels)
 Problem with contention-based protocols
 Inefficient at high load (collisions)

 Contention-free protocols
 Try to do both by explicitly taking turns
 Can potentially also offer guaranteed bandwidth, latency, etc.
Two contention-free approaches
Polling Token Passing
 Master node “invites” slave  Control token passed from one
nodes to transmit in turn node to next sequentially.
 Request to Send (RTS), Clear  Point-to-point links can be fast
to Send (CTS) messages  Problems:
 Problems:  Token overhead
 Polling overhead  Latency
 Latency  Single point of failure (token)
 Single point of failure (master)
Token Ring (802.5)
C B
Direction of
D transmission
A nodes

 Token rotates permission to send around ring of nodes

 Sender injects packet into ring and removes later
 Maximum token holding time (THT) bounds access time
 Early or delayed token release
 Round robin service, acknowledgments and priorities
 Monitor nodes ensure health of ring (alerts on failures)
SD AC FC Destination Source Checksum ED FS
Data
(1) (1) (1) Address (6) Address (6) (4) (1) (1)
FDDI
(Fiber Distributed Data Interface)
 Roughly a large, fast token ring
 100 Mbps and 200km vs 4/16 Mbps and local
 Dual counter-rotating rings for redundancy
 Complex token holding policies for voice etc. traffic

 Token ring advantages Break!

 No contention, bounded access delay
 Support fair, reserved, priority access
 Disadvantages
 Complexity, reliability, scalability
Why Did Ethernet Win?
 Failure modes
 Token rings – network unusable
 Ethernet – single node detached

 Good performance in common case

 Completely distributed, easy to maintain/administer

 Easy incremental deployment

 Volume  lower cost  higher volume ….

26
Wireless Media Access
Wireless is more complicated than wired …

 Cannot detect collisions

 Transmitter swamps co-located receiver
 Collision avoidance

 Different transmitters have different coverage areas

 Asymmetries lead to hidden/exposed terminal problems
» Just because A is talking to B, doesn’t mean that C can’t talk to D
» A is aware of B and B is aware of C, but A and C are unaware of
each other (they are hidden nodes). Both will send to B and frames
will collide at B without A or C realizing
 Also can use contention-free protocols (RTS/CTS)
Hidden Terminals

A B C

transmit range
 A and C can both send to B but can’t hear each other
 A is a hidden terminal for C and vice versa
 Packets will collide at B and create interference
 CSMA will be ineffective – want to sense at receiver
Exposed Terminals

A B C D
transmit range
 B, C can hear each other but can safely send to A, D
 B and C won’t transmit, when they should
CSMA with Collision Avoidance
(CSMA/CA)
 Since we can’t detect collisions, we try to avoid them
 When medium busy, choose random interval
(contention window)
 Wait for that many idle timeslots to pass before sending
 Remember p-persistence … a refinement
 When a collision is inferred, retransmit with binary
exponential backoff (like Ethernet)
 Use ACK from receiver to infer “no collision”
 Again, exponential backoff helps us adapt “p” as needed
 CSMA/CA vs. CSMA/CD
 In CSMA/CA, backoff before collision
 In CSMA/CD, backoff after collision
RTS / CTS Protocols (MACA)

RTS
A B C D
CTS

Overcome exposed/hidden terminal problems with

contention-free protocol
1. B stimulates C with Request To Send (RTS) packet
2. A hears RTS and defers to allow the CTS
3. C replies to B with Clear To Send (CTS)
4. D hears CTS and defers to allow the data
5. B sends to C
IEEE 802.11 Wireless LAN
 802.11b  802.11a
 2.4-5 GHz unlicensed  5-6 GHz range
radio spectrum  up to 54 Mbps
 up to 11 Mbps  802.11g
 direct sequence spread  2.4-5 GHz range
spectrum (DSSS) in  up to 54 Mbps
physical layer  All use CSMA/CA for
» All hosts use same code multiple access
 Widely deployed, using  Optional RTS/CTS
base stations  All have base-station and
ad-hoc network versions
IEEE 802.11 Wireless LAN
 Wireless host communicates with a base station
 Base station = access point (AP)
 Basic Service Set (BSS) (a.k.a. “cell”) contains:
 Wireless hosts

 Access point (AP): base station

 BSS’s combined to form distribution system (DS)

802.11 Twists
 How to support different speeds on same channel?
 Physical layer header encoded at lowest bitrate and indicates
bitrate of rest of packet

 Network Allocation Vector (NAV)

 Each frame contains field that indicates the amount of time
that will be used for the communication (channel reservation)
 All receivers defer transmit for that time
 Allows for long or multi-frame exchange
Misc issue:
Addressing Alternatives
 On a broadcast channel all nodes receive all packets
 Addressing determines which packets are kept and which
packets are thrown away
 Packets can be sent to:
» Unicast – one destination
» Multicast – group of nodes (e.g. “everyone playing WoW”)
» Broadcast – everybody on wire

 Dynamic addresses (e.g. Appletalk)

 Pick an address at random
 Broadcast “is anyone using address XX?”
 If yes, repeat
 Static address (e.g. Ethernet, 802.11, TokenRing, etc)
IEEE addressing (Ethernet, etc)
 Addresses: 6 bytes (48 bits)
 Each adapter is given a globally unique address at
manufacturing time
» Address space is allocated to manufacturers
 24 bits identify manufacturer
 E.g., 0:0:15:*  3com adapter
» Frame is received by all adapters on a LAN and dropped if
address does not match
 Special addresses
» Broadcast – FF:FF:FF:FF:FF:FF is “everybody”
» Range of addresses allocated to multicast
 Adapter maintains list of multicast groups node is interested in
 Practical problems: non-unique addresses
Summary
 Ways to share a channel
 Subdivide the channel into subchannels
 Contention-based protocols
» Try and retry if it fails
 Contention-free protocols
» Explicit control over who gets to send at each time
 Particular issues for wireless
 Hidden/exposed terminal problems
 Addressing

 For next time: bridging/switching, read P&D 3-1-3.2,

3.1