Local Area Networks:

1
SECTION 13

• Local Area Networks
• Ethernet

2
Overview

• Abramsons Multiple Access idea leads to
• Aloha Net
• And then to Carrier Sense Multiple Access CSMA
• And finally to Ethernet CSMA-CD
• Fitting it into the ISO layers- where?
• IEEE Standard 802.3

3
textbook

• Sections 13.2,.3,.4,.6, 14.1 6th edition
• Chapter 15, and
• Sections 16.2, .3, .4, .6 7th Edition

4
Abramsons Idea CSMA

• Why did he think of it?
• Univ of Hawaii mainframe computer and remote job
  entry terminals
• Arpanet impractical
• Telephone network needed for links a disaster
• Mountainous terrain suggests
• VHF radio links with mountaintop repeaters
• How to handle contention??

5
Contention
f1 no problem f2 problem!
Sends on f2
Sends on f1
Rcves on f2
Sends on f2
mainframe
Sends on f2
Remote terminals
6
Contention what to do?

• HDLC solution poll
• Not attractive complicated time consuming
• Abramsons idea about contention
• Just let it happen.
• If 2 packets collide,
• Error Detection will detect it
• Correct by ARQ retransmission

7
Result Aloha Net

• The channel is called
• Aloha Channel, or
• Multiple Access without Polling, or MA for short.

8
Thruput vs. offered load-your normal channel
thruput
Channel capacity ignoring overheads
Offered load
9
Aloha Channel - flooding

• As offered load rises
• More and more collisions gt
• More and more retransmissions gt
• Still more collisions, until eventually
• Every packet suffers collision, I.e.
• Throughput -gt 0
• Channel is called flooded
• Only cure
• all transmitters be quiet.
• Wait until the channel is empty.
• start again.

10
Aloha Thruput vs. offered load --showing
flooding
thruput
channel capacity
wastage due to retransmits etc
Max Thruput ??
Flooding point!
Offered load
11
The Aloha Result

• Large question If the point-to-point channel
  capacity is 1, whats the capacity in Aloha
  mode?? less!
• Answer 1/2e
• Why?

12
proof

• Let time duration of packets t
• Assume
• All packets are full
• On collision, both colliding packets are
  clobbered gt both must be re-transmitted
• Let number of active stations k

copy
13
The Aloha Result

• Let
• uncontended capacity be 1
• message packets be packets sent for the first
  time
• Retransmissions be retransmitted packets
• l mean rate of sending message packets per
  terminal so
• r k l mean total rate of sending message
  packets summed over all terminals

copy
14
The Aloha Result

• r k l mean total rate.
• relationship of r and t
• Suppose t 0.1 sec , then rmax 10 pkt/s
• And rmax t 1 so
• r t measures channel utilization

15
The Aloha Resultcollisions retransmissions
u2
uk
S
16

• Let
• R nr of message packets retransmissions from
  all k users per unit time. SO
• R gt r iff there are any retransmissions and
• Rt is the channel utilization considering BOTH
  messages and retransmissions

17

• Lets calculate
• Rt f rt the utilization

P1
t1-t t1 t1t
Collisions with packet P1 occur anywhere in a
window of width 2t
18

• Assume intervals between packet start times are
  independent exponentially distributed. So
• Prno start in interval T e-RT
• Collisions
• P1 starting at t1 will collide with some P2 if
  there exists a start time in the interval
• ( t1-t, t1t ) of width 2t so
• Prcollision with P1 1- Prno collision in 2t
  
• 1 - e-2Rt

copy
19

• Hence mean number of collisions per unit time
  R1 - e-2Rt so that
• R nr packets nr retransmits
• r R1 - e-2Rt
• so that rt Rt e-2Rt of form
• y xe-2x

copy
20
the Aloha Result, finally

• y xe-2x Wheres the maximum??
• y e-2x xe-2x -2 e-2x 1-2x
• So y 0 gt x 1/2
• So Rt 1/2 maximizes channel utilization rt and
  rt 1/2e is the maximum value

21
The Aloha Result

• If channel capacity is C, then capacity operated
  in Aloha mode is
• C/2e

22

• Beyond offered load 1/2e retransmissions
  increase without bound -- flooding
• If packet start events are synchronized, capacity
  doubles see why? and capacity 1/e -- Slotted
  Aloha

23

• If carrier sense is added we get CSMA and channel
  capacity approx. 0.5
• CSMA need not be run on terrestrial radio works
  OK on satellites too
• Finally, CSMA on a wirepair or Coax
• plus Collision Detect listen while you speak is

24
EtherNet CSMA-CD on guided media

• Invented by Bob Metcalfe
• Originally ran at 1 Mb/s on co-ax
• Then 10Mb/s, 100Mb/s and 1Gb/s Gig-E
• Standardized as IEEE 802.3 in 1985

25
EtherNet

• Where to fit it into the ISO stack??
• Never imagined when stack defined
• L2 link makes sense for the frame transport
• Where to put CSMA-CD?
• New sublayer! -- Layer 2B

26
model of Ethernet Station showing Layers
onwards upwards
Logical Link Control LLC
HDLC-ish CSMA-CD
Medium Access Control MAC
Phys Layer Signalling PLS
ISO L1
Attachment Unit Ifce AUI
MAU
Medium Dependent Ifce MDI
Physical Medium Attachment
Medium
27
Primitives of ISO stack
N
N
REQ
CONFIRM
INDICATION
N-1
N-1
28

SERVICES PROVIDED TO SUPERIORS, ACROSS IFCES
SEND FRAMES RCVE FRAMES FRAMING SIGNALS 2
SERIAL BIT STREAMS
LLC MAC PHYS
29
ETHERNET

• Physical layer notation
• ltdatarate in Mb/sgtltmediumgtltsegment length in 100
  metersgt
• 10 BASE 5
• 10 BROAD 20
• Misnomer 10 Base T
• 10 Mb/s, baseband, on twisted pair
• marketing bafflegab

30
Standards stuff 802.3

• Functional spec of svces provided by MAC to LLC
• No implication about how to implement!

31
MAC services provided to LLC

• MA_data_request
• destination
• M_payload
• service_class
• To send a payload to the peer LLC
• Note that MAC makes up the frame
• downcall

32
MAC services provided to LLC

• MA_data.confirm
• status
• Ack back to requesting LLC from MAC
• LOCAL significance only
• Status failure gt
• Too many collisions, medium failed, . . .
• upcall

33
MAC services provided to LLC

• MA_data.indication
• DA / dest addr
• SA / src addr
• m_sdu / payload
• status
• Heres a frame for you
• upcall

34
802.3 ethernet LLC frame, orHDLC reworked
Octets 7 1 2 or 6 2 or 6 2 Variable 1
Preamble SOF SA DA Len Datapad FCS preambl
e
Creates periodic waveform at L1 Manchester
encoded for bit sync PLL. Format 101010 . . .
10 Format 10101011 for frame sync Source address
always 6 bytes nowadays Destination address
ditto Helps transparency Pad to achieve minimum
allowed framesize CRC 32 26 23 22 16 12 11 10 8
7 5 4 2 1 0
35
Ethernet

• Addresses can be
• Individual of form 0xxxxx
• Group multicast 1yyyyy
• Broadcast 111111
• IT IS A BROADCAST MEDIUM!

36
Chores for MAC

• Encapsulate data in frame
• Addressing
• Error detection
• Access the medium via CSMA-CD I.e.
• Collision detect
• Collision recovery rest done at L1
• Specifics

37
Chores for MAC

• Transmit, no-collision case
• Compose frame
• proc Trans Data Encap
• Sense carrier
• proc Trans Media Access Mgmt
• Signal provided by L1 Phys Layer Signalling PLS
• ON CLEAR
• Wait a short time for medium to recover
• Pass bit-serial stream to PLS
• PLS generates electrical signals senses
  collision

38
Chores for MAC

• Receive, no-collision case
• Dual of the above
• Exercise for the reader

39
Chores for MAC

• Transmit, collision encountered
• Define transmit window D1 Ta, Tb where
• Ta is time I start to send
• Tb is earliest time when ALL stations have heard
  my carrier
• If collision, additional time D2 elapses before I
  can hear the collision

40

• Transmit, collision encountered
• D1 and D2 determined by round trip propagation
  time on the medium
• Long D1, high channel data rate gt many useless
  bits sent before Collision Detected
• Geosynchronous Satellites bad news here!
• 250 msec roundtrip latency

41

• Transmit, collision encountered
• ON COLLISION
• PLS turns on signal CD Collision Detected
• MAC must
• enforce collision everywhere by sending JAM bits
• a bit sequence long enough to ensure CD ON at
  all stations
• Reschedule transmission
• How??

42
Reschedule transmission

• Immediate retransmit a bad idea
• Retransmit after constant K sec a bad idea

43
Reschedule transmission

• Exponential backoff -Concepts
• The more we have to re-try the longer we should
  wait congestion happening
• The colliding parties should not re-transmit at
  the same time !
• Exponential rule
• back off rn scheduling quanta for nth retry
  where
• 0 lt r lt 2k where
• k minn, 10
• R is random, uniformly distributed in that
  interval
• and

44

• Exponential backoff
• Boundary conditions
• Scheduling quantum gt collision window
• Scheduling quantum gt frame xmit time
• WHY backoff?
• Trying to avoid flooding
• Why random?

45
ethernet

• practicalities

46
ethernet

• Some design numbers
• The 10 BASE 5 implementation of IEEE 802.3
• Interframe gap 9.6 msec
• Backoff limit 10 attempts
• JAM 32 bits
• Max framesize 1518 octets
• Min framesize 64 octets

47
How to screw up an ethernet

• Provisioning close to the flooding region
• So-called Ethernet Limit
• Space repeaters too far apart
• Its a passive system
• Too many stations between repeaters
• Coax cable bends too sharp -
• causes change of characteristic impedance
• mismatch
• reflected power...
• Twisted-pair in noisy environments
• use fibre or Token Bus

48
Topological arrangement
The medium
49
Physical arrangement in office buildings
stations
Wiring closet
50
Modern post-1990 ethernet

• Flooding worries
• Security worries suggest

51
Put a frame switch in the wiring closet!
Frame switch L2 packet sw
So-called switched ethernet hub lt 100 at
Future Shop Yielding Switched Ethernet
52
Switched ethernet

• Can there be any collisions?
• Do we need CSMA-CD?
• Why bother with ethernet interface??

53
Onwards upwards

• 1 BASE 5 1981
• 10 BASE T 1988
• 100 BASE T 1993
• Switched hubs for 50 at FutureShop 1996
• Gigabit ethernet entirely switched 2000
• Optical ethernet ditto 2000
• GigE as a broadband loop to the home?
• --dates approximate