1. EE122: Multi-Access Aloha, Ethernet, 802.11
November 10, 2003

2. EECS 122: Introduction to Computer Networks Multiaccess Protocols
Computer Science Division
Department of Electrical Engineering and Computer Sciences
University of California, Berkeley
Berkeley, CA

3. Today’s Lecture: 21 Application Transport Network (IP) Link Physical 2
17, 18, 19, 20
Application 6
10,11
Transport
14, 15, 16
7, 8, 9
Network (IP)
21, 22, 23
Link
Physical 24
2/27/ :07 PM

4. Up Until Now..... Short-term contention is loss-less
main resource (link bandwidth) is controlled by router
router deals with short-term contention by queueing packets
switch algorithms and router buffers ensure no packets are dropped due to short-term contention
We have focused on long-term contention
queueing schemes (FQ, FIFO, RED, etc.)
end-to-end congestion control (TCP)
2/27/ :07 PM

5. What’s New in This Lecture?
Short-term contention leads to loss!
Lecture deals with networking over shared media
long-range radio
ethernet
short-range radio
Also known as “multiple-access”
don’t go through central router to get access to link
instead, multiple users can access shared medium
2/27/ :07 PM

6. Medium Access Protocols
Channel partitioning
Divide channel into smaller “pieces” (e.g., time slots, frequency)
Allocate a piece to node for exclusive use
Random access
Allow collisions
“recover” from collisions
“Taking-turns”
Tightly coordinate shared access to avoid collisions
2/27/ :07 PM

7. Problem in a Nutshell Shared medium Want high network utilization
If two users send at the same time, collision results in no packet being received (interference)
If no users send, channel goes idle
Thus, want to have only one user send at a time
Want high network utilization
TDMA doesn’t give high utilization
Want simple distributed algorithm
no fancy token-passing schemes that avoid collisions
2/27/ :07 PM

8. What Layer? Where should short-term contention be handled?
Network layer?
Application layer?
Link layer?
2/27/ :07 PM

9. Focus of Lecture Understanding basic algorithmic choices
Simple performance analysis
Will not stress the practical details
framing, packet formats, etc.
2/27/ :07 PM

10. Aloha Ralph Abramson left Stanford in search of surfing
Set up first radio-based data communication system connecting the Hawaiian islands
hub at alohanet HQ
many other sites on islands
Had two radio channels:
random access: sites sent data on this channel
broadcast: only used by hub to rebroadcast incoming data
2/27/ :07 PM

11. Aloha Transmission Strategy
When new data arrived at site, it was sent to hub
Site then listened to broadcast channel
if it heard data repeated, knew transmission was rec’d
if it didn’t hear data correctly, it assumed a collision
If collision, site then waited random delay before retransmitting
2/27/ :07 PM

12. Simple, but Radical
Aloha is to multiple access what Internet is to telephony
Previous attempts all partitioned channel
TDMA, FDMA, etc.
Aloha optimized the common case (few senders) and dealt with collisions through retries
sound familiar?
2/27/ :07 PM

13. Why is this better than TDMA?
In TDMA, you always have to wait your turn
delay proportional to number of sites
In Aloha, can send immediately
Aloha gives much lower delays, at the price of lower utilization (as we will see)
2/27/ :07 PM

14. Slotted Aloha Divide time into slots
Only start transmission at beginning of slots
Decreases chance of “partial collisions”
2/27/ :07 PM

15. Performance of Slotted Aloha
Time is divided into equal size slots (= packet transmission time)
Node with new arriving pkt: transmit at beginning of next slot
If collision: retransmit pkt in future slots with probability p, until successful.
Success (S), Collision (C), Empty (E) slots
2/27/ :07 PM

16. Slotted Aloha Efficiency
What is the maximum fraction of successful transmissions?
Suppose N stations have packets to send
Each transmits in slot with probability p
Prob. successful transmission S is (very approximated analysis!):
by a particular node i: Si = p (1-p)(N-1)
by exactly one of N nodes S = Prob (only one transmits) = N p (1-p)(N-1) <= 1/e = 0.37
but must have p proportional to 1/N
2/27/ :07 PM

17. Ethernet Shared medium (coax cable)
But, can “sense” carrier to see if other nodes are broadcasting at the same time
this sensing is subject to time-lag
can only detect those who were sending a short while before
Can monitor channel to detect collisions
once sending, can tell if anyone else is sending too
2/27/ :07 PM

18. CSMA and CSMA/CD Carrier sense multiple access: CSMA
listen before you start talking
CSMA with collision detect: CSMA/CD
stop talking when you hear someone else talking
2/27/ :07 PM

19. CSMA: Carrier Sense Multiple Access
CS (Carrier Sense) means that each node can distinguish between an idle and a busy link
Sender operations:
If channel sensed idle: transmit entire packet
If channel sensed busy, defer transmission
various retry algorithms
2/27/ :07 PM

20. CSMA collisions Collisions can occur: Collision: Note:
spatial layout of nodes along ethernet
Collisions can occur:
propagation delay means
two nodes may not
hear each other’s transmission
Collision:
entire packet transmission
time wasted
Note:
role of distance and propagation delay in determining collision prob.
2/27/ :07 PM

21. CSMA/CD (Collision Detection)
Collisions detected within short time
Colliding transmissions aborted, reducing channel wastage
Easy in wired LANs:
measure signal strengths,
compare transmitted, received signals
Difficult in wireless LANs
2/27/ :07 PM

22. CSMA/CD collision detection
2/27/ :07 PM

23. Ethernet Frame Structure
Sending adapter encapsulates IP datagram
Preamble:
7 bytes with pattern followed by one byte with pattern
Used to synchronize receiver, sender clock rates
2/27/ :07 PM

24. Ethernet Frame Structure (more)
Addresses: 6 bytes, frame is received by all adapters on a LAN and dropped if address does not match
Type: 2 bytes, indicates the higher layer protocol
E.g., IP, Novell IPX, AppleTalk
CRC: 4 bytes, checked at receiver, if error is detected, the frame is simply dropped
Data payload: maximum 1500 bytes, minimum 46 bytes
2/27/ :07 PM

25. Ethernet’s CSMA/CD Sense channel, if idle Receiver accepts:
If detect another transmission
Abort, send jam signal
Delay, and try again
Else
Send frame
Receiver accepts:
Frames addressed to its own address
Frames addressed to the broadcast address (broadcast)
Frames addressed to a multicast address, if it was instructed to listen to that address
All frames (promiscuous mode)
2/27/ :07 PM

26. Ethernet’s CSMA/CD (more)
Jam signal: make sure all other transmitters are aware of collision; 48 bits;
Exponential back-off
Goal: adapt retransmission attempts to estimated current load
Heavy load: random wait will be longer
First collision: choose K from {0,1}; delay is K x 512 bit transmission times
After second collision: choose K from {0,1,2,3}…
After ten or more collisions, choose K from {0,1,2,3,4,…,1023}
2/27/ :07 PM

27. Next Few Slides Practical aspects of ethernet
Then on to theoretical aspects
2/27/ :07 PM

28. Minimum Packet Size Why put a minimum packet size?
Give a host enough time to detect collisions
In Ethernet, minimum packet size = 64 bytes (two 6-byte addresses, 2-byte type, 4-byte CRC, and 46 bytes of data)
If host has less than 46 bytes to send, the adaptor pads (adds) bytes to make it 46 bytes
What is the relationship between minimum packet size and the length of the LAN?
2/27/ :07 PM

29. Minimum Packet Size (more)
Host 1
Host 2
a) Time = t; Host 1
starts to send frame
propagation delay (d)
propagation delay (d)
Host 1
Host 2
b) Time = t + d; Host 2
starts to send a frame
just before it hears from
host 1’s frame
propagation delay (d)
Host 1
Host 2
c) Time = t + 2*d; Host 1
hears Host 2’s frame 
detects collision
LAN length = (min_frame_size)*(light_speed)/(2*bandwidth) =
= (8*64b)*(2.5*108mps)/(2*107 bps) = 6400m approx
2/27/ :07 PM

30. Ethernet Technologies: 10Base2
10: 10Mbps; 2: under 200 meters max cable length
Thin coaxial cable in a bus topology
Repeaters used to connect up to multiple segments
Repeater repeats bits it hears on one interface to its other interfaces: physical layer device only!
2/27/ :07 PM

31. 10BaseT and 100BaseT 10/100 Mbps rate; later called “fast ethernet”
T stands for Twisted Pair
Hub to which nodes are connected by twisted pair, thus “star topology”
2/27/ :07 PM

32. 10BaseT and 100BaseT (more)
Max distance from node to Hub is 100 meters
Hub can disconnect “jabbering” adapter
Hub can gather monitoring information, statistics for display to LAN administrators
Hubs still preserve one collision domain
Every packet is forwarded to all hosts
Use bridges to address this problem
Bridges forward a packet only to the destination leading to the destination
2/27/ :07 PM

33. Gbit Ethernet Use standard Ethernet frame format
Allows for point-to-point links and shared broadcast channels
In shared mode, CSMA/CD is used; short distances between nodes to be efficient
Uses hubs, called here “Buffered Distributors”
Full-Duplex at 1 Gbps for point-to-point links
2/27/ :07 PM

34. Interconnecting LANs Why not just one big LAN?
Limited amount of supportable traffic: on single LAN, all stations must share bandwidth
Limited length
Large “collision domain” (can collide with many stations)
2/27/ :07 PM

35. Family of Backoff Algorithms
Use slotted transmission scheme
carrier sense is irrelevant in slotted model
When experience k’th collision for a particular packet, send packet with probability 1/f(k) in each successive slot (until transmitted)
Ethernet uses f(k)=2k (with a bound on k)
2/27/ :07 PM

36. Dynamics of Backoff Algorithms
Is ethernet asymptotically stable?
if we let number of hosts to increase without bound, will the channel have a nonzero throughput?
Answer: no (no matter what f(k) is)
Insight: jam medium for time taken to achieve sizable backlog; once this happens, very little chance channel will become unclogged
2/27/ :07 PM

37. Is Ethernet “Fair”? No, it has “capture effect”
Consider two nodes competing for the ethernet, both with an infinite set of packets to send
There is a finite chance that one will never send another packet, ever
Moreover, with probability one only one node will achieve an infinite number of transmissions
2/27/ :07 PM

38. Capture Effect
Assume there is a collision between a node with backoff counter 1 and backoff counter k
First node will win unless
it chooses the second slot (probability 1/2) and
other node chooses first slot (probability 1/f(k))
Probability that second node loses all subsequent collisions:
L=k (1-1/f(k)) ≈ exp(-k1/f(k))
if f(k) grows faster than linearly, then L is nonzero
2/27/ :07 PM

39. Wasted Time
If nodes don’t have infinite backlog of packets, then the nodes will “take turns” on long time scales; first one dumps its queue, then the other
For backoff functions faster than linear, but slower than exponential, the “wasted time” when switching over is small compared to the dumping time
For exponential backoffs, the wasted time is proportional to dumping times
2/27/ :07 PM

40. Summary of Backoff Slower than linear: Between linear and exponential
no capture
bounded utilization (due to ongoing collisions)
Between linear and exponential
capture
full utilization
Exponential and faster
bounded utilization (idle time between turns)
2/27/ :07 PM

41. Ethernet vs 802.11 Ethernet: one shared “collision” domain
802.11: radios have small range compared to overall system: collisions are local
collisions are at receiver, not sender
carrier-sense plays different role
CSMA/CA not CSMA/CD
collision avoidance, not collision detection
2/27/ :07 PM

42. 802.11
Designed for use in limited geographical area (i.e., couple of hundreds of meters)
Designed for three physical media (run at either 1Mbps or 2 Mbps)
Two based on spread spectrum radio
One based on diffused infrared
2/27/ :07 PM

43. Physical Link Frequency hoping Direct sequence Infrared signal
Transmit the signal over multiple frequencies
The sequence of frequencies is pseudo-random, i.e., both sender and receiver use the same algorithm to generate their sequences
Direct sequence
Represent each bit by multiple (e.g., n) bits in a frame; XOR signal with a pseudo-random generated sequence with a frequency n times higher
Infrared signal
Sender and receiver do not need a clear line of sight
Limited range; order of meters
2/27/ :07 PM

44. Collision Avoidance: The Problems
Reachability is not transitive: if A can reach B, and B can reach C, it doesn’t necessary mean that A can reach C
Hidden nodes: A and C send a packet to B; neither A nor C will detect the collision!
Exposed node: B sends a packet to A; C hears this and decides not to send a packet to D (despite the fact that this will not cause interference)! A B C D
2/27/ :07 PM

45. Multiple Access with Collision Avoidance (MACA)
other node in
sender’s range
sender
receiver
RTS
CTS
data
ACK
Before every data transmission
Sender sends a Request to Send (RTS) frame containing the length of the transmission
Receiver respond with a Clear to Send (CTS) frame
Sender sends data
Receiver sends an ACK; now another sender can send data
When sender doesn’t get a CTS back, it assumes collision
2/27/ :07 PM

46. Other Nodes
When you hear a CTS, you keep quiet until scheduled transmission is over (hear ACK)
If you hear RTS, but not CTS, you can send
interfering at source but not at receiver is ok
can cause problems when a CTS is interfered with
2/27/ :07 PM

47. Summary
Problem: arbitrate between multiple hosts sharing a common communication media
Wired solution: Ethernet (use CSMA/CD)
Detect collisions
Backoff exponentially on collision
Wireless solution: (CSMA/CA)
Use MACA protocol
Cannot detect collisions; try to avoid them
Distribution system & frame format in discussion sections
2/27/ :07 PM

48. What You Need to Know
Basics of Aloha and Ethernet contention algorithms
Basics of contention algorithm
2/27/ :07 PM