CENG 460 ELEC 510

1
CENG 460 / ELEC 510

• Visual Aids
• Eric G. Manning, P.Eng., FIEEE

2
Section 1

• Origins
• Telephony concept of circuit switching,
  Strowgers Ericssons switches
• Text Sec. 1.1 telephone model
• Sec 1.3, 1.4 origins

3
Origins

• To understand
• The Internet
• Tcp/IP
• ATM ... etc etc
• We first need to understand
• Multiplexing
• Circuit switching
• Packet switching

4
Read in the text

• Chapter 1 Overview

5
Switching means

• Selectively connecting subscribers
• In pairs point-to-point or
• In groups multicasting
• or all of them - broadcasting
• Graphically . . .

6
Point-to point connections
subscribers
Multicast group
broadcast -everyone to everyone
7
Why connect selectively?

• Why not just connect everyone to everyone?
  broadcast

8
Circuit Switching and

• The Telephone Network
• Switched Voice Network,
• Public Switched Telephone Net

9
Manual circuit switchinglate 19th century
switchboard
Plug-cord
10
Note properties

• End-to-end physical path or connection
• Sequencing preserved
• ABCDE in implies ABCDE out
• Nothing is lost or inserted
• Pretty obvious for this kind of switch
• Less obvious for other kinds!

11
Automated Switching

• Erwin Strowger
• undertaker, Kansas City, around 1910
• Competition problem
• His solution
• Strowgers Automatic Switch

12
Strowgers Step By Step Switch
Rotating shaft
Sliding arm
10 X 10 Array of contacts
13
Functionality

• Any ONE of the 10x10 array of contacts can be
  connected to the rotating wiper arm
• A 1/100 multipexor or selector
• For telephone switching ?

14
Strowger Switch in telephony

• Can connect one of 100 phones to a given phone
• How to connect one of 100 phones to any other one
  of the 100 phones?
• How many concurrent calls?
• What to do to have n concurrent calls?

15
Building a circuit between source destination

Circuit with 50 VDC battery
16
Circuit . . .

Could be 10 or more links
17
Strowger Switch in telephony

• What if the town has 100 phones?
• How many switches for one call out of 104 phones?
• For k concurrent calls and 104 phones, how many
  switches?
• Why is the switch called Step By Step?

18
Step by Step SxS

• World telephone network was entirely SxS or
  manual, until 1948
• L M Ericssons CROSSBAR SWITCH

19
Ericssons CROSSBAR

• Motive
• Reduce the step by step circuit build time
• upwards of 10 seconds on trunk calls
• Save money
• The design

20
CROSSBAR

Crosspoint detail
horizontals
end -marked
verticals
21
CROSSBAR System

Trunks to other switches
marker
Crosspoint Switching fabric
Local or Subscriber loops
22
Next step software control

• Voice Circuit switch with stored program control
• Concept replace the wired-logic marker with a
  cpu

23
Next step software control

• Claimed benefits
• Flexibility for adding new customer features e.g.
  conference calling
• just add some code !
• Ability to exploit solid state electronics
  other computer economies to save
• trendiness

24
First commercial systemNumber 1 Electronic
Switching System ESS

• Developed by Bell Telephone Labs.
• Custom-designed
• cpu SSI
• Program store - ROM - Permanent Magnet twistor
• Call store - RAM - ferrite sheet - 5.5 msec cycle
  time
• switching fabric
• arrays of relays with wire interconnects
• no more rotating brass rods
• not end-marked!

25
ESS 1

• Reliability requirement
• to compete with crossbar systems
• Real time requirements
• Digit reception, dial tone supply, routing etc
  all to be done in real time

26
Gross architecture
CPU

I/O bus
Line scanner
Crosspoint driver
Crosspoint array
27
Fault tolerance

• Hardware fault detection diagnosis
• Duplicate match cpus
• Interpret state of cpu upon mismatch to diagnose
  fault
• Diagnostic dictionary developed by physical fault
  insertion
• Does this make sense in 2003?

28
Fault tolerance

• Software fault correction
• Call processing code organized as loop
• Timer timeout triggered by excessive looping time
• Audit programs executed in sequence

29
Audit programs

• Phase 1
• Re-initialize constants
• Phase 2
• Correct pointers of doubly-linked lists
• Phase 3
• Correct crosspoint map
• by tearing down all calls!

30
Audit programs

• Phase 4 Emergency action
• Try combinations of
• 2 cpus
• 2 callstores
• 2 program stores
• 2 busses
• Until one which allows program looping is found

31
Children of ESS-1

• SP-1 Northern Electric Co
• used a crossbar fabric! why?
• DMS-100 Nortel Networks Inc.
• time-domain switching of digital signals,
• but still a circuit switch
• ESS-5 Lucent Technologies
• No new circuit switches planned

32
Section 1.2

• PSTN topology

33
Hierarchical structure augmented tree -rational

Two in Canada
Class 1 switch Class 2 Class 3 Class 4 Class
5
Toll network
Extra links
Serving Central Office
34
Routing algorithm

• Go UP until the destination is in the subtree
  routed at this node, then
• Go DOWN to the destination
• Loop free
• Alternate routes via extra links in the tree
• 1 call in 106 goes all the way up
• the ultimate route
• Internet counterpart -
• no backbone net on the path-
• just NOW being implemented...peering agreements
• Jiddah, Saudi Arabia to Mecca, Saudi Arabia via
  Washington DC !

35
Canadian Voice Network Summary

• Formerly analog transmission
• Now digital
• Formerly space division multiplexing switching
• Now mostly time division
• Formerly electromechanical
• Now entirely electronic solid state crosspoints

36
Voice Network Summary

• Service provided
• Approx. 2 kHz bandwidth channel
• Quality of Service QoS
• Constant bandwidth
• 1-10 seconds to build a circuit
• Mean call holding time 3 minutes
• Residential phone utilization 10
• critical fact for deciding blocking factor
• big trouble at present

37
Voice Network used for data

• Used modems to convert digital data to analog
  tones
• OK except for
• Connect time too slow
• Circuit utilization too low
• Data rate
• Cost too high
• Bit Error Rate 10-2 way too high!!
• still used today AOL for Internet Access

38
And now for something completely different . . .

• The antithesis
• of circuit switching . . . .

39
Message Switching!

40
Section 1.3 Message Switching

• Data networks, manual torn-tape switch. Automated
  version, notion of message with address.
• Problems Speed, cost, reliability.

41
Telegraph network ca. 1890the Channel, with
coding
Paper tape reader
Paper tape punch
Baudot Code
Paper tape 3/4 inch wide, 5, 7 or 8 holes
0 0 0 0 0
0 0 0 0 0
O 0 0 0
42
Manual message switch

• Messages have tape format

Where- To
Where- from
Message text . . .
Canadian National Railways
TELEGRAM TO Mom FROM
Dilbert __________________________________ MESSAG
E Hi Mom! - Dilbert
From the Paper Telegram Form . . .
43
Manual Message switch3 lines in X 3 lines
out in this example
IN
OUT
punches
readers
44
The operator

• Tears each message off a reel of incoming
  punched tape
• Reads the TO field
• Carries the message across the room to the output
  side
• Feeds it into the punch
• connected to the telegraph wire
• for the next destination,where it is punched
  again

45
Hence the name

• Torn-Tape Message Switching Center

46
Message Switch Version 2

• Automated
• Ca. 1975
• Computerized, of course

47
Automated Message Switch
CPU

Large disc
Low-speed lines
I/O controller
48
Automated Message Switch

• Developed around 1975
• Used by banks airlines to switch terminal
  traffic to a host
• Message format

TO FROM VARIABLE LENGTH TEXT
49
SWITCH FUNCTIONALITY

• Receive a message
• Store it on disc
• Analyze the header to determine outgoing line
  routing
• enqueue on outgoing line
• Transmit when line is free

50
Properties

• Fine, except
• Slow
• unreliable

51
Why slow?

• Variable message length
• Memory fragmentation of primary memory Mp
• need to store message on slow disc
• disc arm movement time about 30 msec

52
Why unreliable?
terminals
switch
Star topology Broken line subscriber
failure Broken sw system failure!
53
BUT, some positive features

• No circuit setup time
• Errors can be detected corrected EDAC by CRC
  codes retransmission ARQ
• SO, used today for ATMs airline reservation
  terminals
• Next step

54
Notion of packet Barans work.

• Packet nets easing the reliability, speed and
  cost problems of circuit message switching.
• New problems rear their heads.
• Read Secs. 10.6ps principles
• , 10.7 readahead X.25 of text.

55
Barans Bright Idea

• Impose a maximum message length
• _at_ 256 bytes
• no memory fragmentation - buffers are either
  full or empty, never partly-full
• packets can usually be held in primary memory, so
• No disc needed
• fast store forward 100 msec - 1 msec

256
Buffer memory map
56
Barans Bright Idea
To from payload PACKET!
57
Barans Bright idea

• 2. Mesh network, not star or tree
• Many switches
• multiple routes
• Fault tolerance to link or switch failures
• nontrivial routing problem

sw
sw
sw
sw
58
Barans Bright idea

• 3. Hot potato routing adaptive
• record path lengths delays seen from S to D
  via each incoming link to D.
• Use smallest-delay link for outgoing packets
• from D to S
• Net result fast, efficient, fault tolerant
  network
• Algorithm does NOT require knowledge of the
  network topology!

59
Questions arise . . .

• Big messages?
• Fragment,
• Send
• reassemble

. . .
60
Questions arise . . .

• Sequencing?
• Packets can get out of sequence with multiple
  routes adaptive routing
• Sequence numbers!
• Sort at reassembly time - for data
• Discard out-of-sequence packets - for digitized
  voice

61
Questions arise . . .

• Flow control?
• Sender may send faster than receiver can receive
• Solution Rcvr issues credits to sender
• -send 5 packets

62
Questions arise . . .

• Duplicates?
• Filter them out
• How can they occur??

63
Summary

• Combined effect of
• fragmentation reassembly
• Sorting by sequence nr
• Retransmission of missing packets
• Dupe filtering
• Simulates the properties of a circuit
• Name Virtual circuit or virtual call

64
Questions arise . . .

• Who does all this?
• Transport station, e.g. Cyclades ts, Internet
  tcp, CLP . . .

host
ts
psn
65
Summary circuit vs packet switching
switch
switch
S
D
66
Circuit switching

• Switch physically joins incoming link to outgoing
  link
• Creates an end-to-end from S to D path
• Good for long conversations at constant data rate
• Telephone calls
• File transfer . . .

67
Packet switching

• No end-to-end path
• Switch
• stores packet
• selects outgoing link
• Enqueues packet on outgoing link
• Good for ???

68
Comparison

• Packet switching
• One stream follows many paths
• One link or path sees many streams
• Circuit switching
• One stream follows ONE path
• One link or path sees one stream

69
Packet switching
sw
2
3
1
6
5
sw
sw
4
One stream many paths One link many streams
sw
70
Packet network structure

• Over-connected
• For reliability
• To help relieve congestion
• Typical connectivity about 1.2 to 1.5
• number of independent paths from S to D,
  averaged over all pairs (S,D)
• Why not fully connected?

71
Packet network structure

• Is this the optimal topology??
• For a given traffic matrix
• For a given function to optimize cost and
• For given constraints connectivity, path length
• NP-hard problem
• to be made

72
Packet Switching Implications of small delays
video

• Packet must be stored retransmitted at each
  switch
• Line service time Length L
• must be small
• How??
• small numerator or big denominator . . .

Linespeed S
73
Small line service time

• 1 short packets
• eg 53 bytes in ATM, or
• 2 fast lines
• eg 1012 bits/sec with optical fibre
• Role of latency propagation delay? . . .

74
Role of latency propagation delay?

• 1 must be small
• Added on to line service time to calculate total
  link delay
• 2 acts as a shift register memory
• 109 bps line speed, 2 000 bit packets
• 2 10-6 sec line service time. If
• latency 2 msec 200 - 400 miles of wire or
  fibre
• There are 103 packets simultaneously in flight!
• Flow control sabotaged

75
For small total end-end delay

• We need
• small line service time
• fast lines
• Short packets
• Small latency
• cant control it, except by routing to minimize
  path length! and
• Short queues on links

76
queue of packets awaiting transmission on a link?

queue server
mean arrival rate lambda, exponentially
distributed
service rate 1/m Poisson distributed
If there are n packets in the queue and each
takes L/S seconds to be transmitted on the
link, the last packet waits for nL/S seconds
to get on the link...
77
Short queues. . .

• Queuing theory says
• Mean queue length N r / (1-r)
• Where r is link utilization 0
• Plot N versus r
• Clearly a low-delay net will have low line
  utilization. Expensive??
• __________________
• for exponentially-distributed arrivals, and
  Poisson-distributed service times

78
Packet nets expensive?

• Links are often optical fibre transmission
  systems
• Can cost 10 MILLION AND UP
• So low-link-utilization means expensive??
• Maybe not
• Line utilization of PSTN telephone network is
  even worse !

79
Summary

• CIRCUIT
• No address
• Setup required
• ONE path per stream
• No
• Out of sequence
• Lost
• Inserted payloads
• Muxes the medium in space, time or frequency

• PACKET
• Address
• No setup
• Many paths/stream
• Can have
• Out of sequence
• Lost
• Inserted payloads
• Muxes the medium in time

80
Sec. 1.5 Varieties of SwitchingMultiplexing and
concentration

• Space, time frequency resources. Space, time
  frequency-division styles of
• switching
• multiplexing
• concentration
• Read text Ch. 5 modulation, Sec 8.1 FDM
• Sec 8.2 Synchronous TDM , 8.3 Statistical TDM

81
Switching, Mux-ing, concentration

• A channel has 2 resources
• Time and
• Frequency
• A set of channels has 3 resources
• Space
• Time and
• frequency

82
Switching, Mux-ing, concentration

• We can share the channel with respect to
• Space
• Time or
• Frequency
• or code, see Chapter 9, spread-spectrum

83
Kinds of channel-sharing

• We can switch

Box has n1 inputs, m1 outputs Can connect input
i to output j
84
Kinds of channel-sharing

• We can mux or multiplex

Box has N1 inputs 1 output Capacity out S
capacities in S Cin Cout
85
Kinds of channel-sharing

• Or concentrate

S Cin Cout
86

• On any resource
• Space
• Time
• Frequency or code

87
And

• 2 varieties of TD mux
• Smart or statistical and
• Dumb or fixed

88
Next

• Some of

89
First. . .

• Time Division Multiplexing or TDM dumb

90
Dumb TDMwhy was it invented?

• The intracity trunking problem
• Need to carry many calls on one wirepair
• Digital encoding of speech one way
• Pulse Code Modulation or PCM
• one of many techniques including
• Pulse Amplitude Modulation
• Pulse Position Modulation , . . .

91
PCM
7
1.How often should we sample?? 2.How many
sampling buckets? implies bits/sample
6
5
4
3
2
1
0
0 0 0
0 1 0
1 0 0
1 0 0
0 1 0
0 0 1
92
How often should we sample?

• The Sampling Theorem
• A periodic function of time with maximum
  frequency component F need only be sampled 2F
  times per second.
• Think of the Fourier series expansion truncated
  at the frequency-F term.
• A sine of frequency F can be determined by its
  zero-crossing points - 2F per second of them.

93
How many buckets? Or,how many bits/sample

• No neat theory, but Bell Labs empirical result
• Quantization noise of 4-KHz band-limited speech
  is OK at 7 bits per sample or 128 buckets

94
Put it together

• For 4-KHz band-limited speech
• 8 000 samples/second Sampling Theorem
• Each of 7 bits 56 000 bits/sec suffices.
• Add 1 control bit /sample for
• 8 bits/sample 8 000 sample/sec 64 Kb/s
• the famous T0 or DS0 Transmission Rate
• Uncompressed PCM toll-quality speech.
• compression algorithms can reduce it to 10 Kb/s
  or better used in Voice Over IP

95
Next trick the FRAME concept

• How to put , say, 24 PCM-encoded speech channels
  on a wirepair?

8 bits per slot, holds 1 sample
F 1 2 3 4 5 . . . 24 F 1 2 3 . . .
time
a frame
Framing Symbol marks start/end of Frame
96
Frame arithmetic

• 8 000 samples/sec 8 bits/sample 64 Kb/s
• 24 time slots/frame 8 bits/slot 192
  speechbits / frame
• 192 speechbits 1 framing bit 193 bits/frame
• 8 000 frames/sec 193 bits/frame
• 1.544 Mb/sec -- the T-1 Carrier Rate

97
Bell System TDM Frame Hierarchy

98
Bell System Frame Hierarchy

• For wire, fibre and microwave, time-divided
• For optical fibre the Optical Carrier hierarchy

99
Why do we need these standard rates?

• Provide transmission channels at these speeds,
  then
• N slower channels will MUX neatly into one faster
  channel,with minimal waste
• E.g. 4 T-1s MUX into one T-3
• 4 OC-3s MUX into one OC-12
• Mux-makers build boxes to do this

100
Back to PCM

• One way to carry analog data voice on a digital
  channel
• Involves encoding and modulation
• More generally

How to carry data on An channel??
Analog digital
Analog digital
101
READ THIS!!

• For the answer, read . . .
• Textbook Chapter 5.
• For Dumb or synchronous TDM read
• Textbook Sec. 8.2

102
Section 1.6

• Pure TDM switching
• Hybrid TDM switching space-time-space
• Read
• Sec. 10.1 communication nets
• Sec. 10.2 circuit switching
• Sec. 10.3 single node

103
TDM switch

104
Pure TDM switch-no space or frequency division
Frames
Frames
switch
1 channel in 1 channel out It permutes
timeslots Time Slot Interchanger
105
Real TDM switch

• 1 input, 1 output not very practical
• So we need a hybrid
• Time and space division
• One possibility Space-Time-Space
• Space division followed by time division fby
  space division

O1 O2 O3
I1 I2 I3
TSi of Ij - TSk of Ol
Nortel DMS family, Lucent ESS-5
Crosspoint array
106
The two kinds of TDM

• Synchronous or dumb TDM
• reserves TSk of Ol for my call
• whether I use it or not
• Asynchronous or smart TDM
• Gives me the time slot iff I have data to send
• Cant identify my call by time slot position
• Must use some kind of address . . .

107
Smart TDM switch
Reminds us of ???
108
More thorough exploration of

• Interesting space of ways to share channels . . .

109
Space Time frequency
Mux concentrate switch
110
Time-Division Mux dumb smart
F 3 2 1 F 3 2 1
F
dumb smart
A
1 2 3
A
A
A
B
A
B
A
3
2
1
A
3
A
B
Dumb Frame-formatted output line Second-level
mux?
111
T-D switch, dumb circuit, frame-formatted
Ij TSi -- Ok TSl
Ij
Ok
Frame-formatted lines in out
112
Time-division switch,smart, packet-formatted
X
What is it?
113
Space-division mux
Vacuous!
114
Space-division switch
115
Frequency-division muxs
One phone call Occupying frequency spectrum 0-4
kHz
A
1 2 3
MUX
1 2 3
KHz
100 104 105 109 110 114
0 4 KHz
Ok
Whats inside the box? Whats a 2nd level mux
Look like? Nortel LD-series
First-level edge, in internet-speak MUX
116
Frequency-division switch
Whats inside the box?
Practical example Wireless repeater
One wire, or fiber, Or wireless
signal, Frequency-divided
One wire, or fiber, Or wireless
signal, Frequency-divided
117
Summary

• Sec. 1.5 Varieties of SwitchingMultiplexing and
  concentration

Space, time frequency resources. Space, time
frequency-division styles of switching multiplex
ing concentration Read text Ch. 5
modulation, Sec 8.1 FDM Sec 8.2 Sync.TDM,
8.3 Statistical TDM,
118
Summary
Section 1.6

• Pure TDM switching
• Hybrid TDM switching space-time-space
• Read
• Sec. 10.1 communication nets
• Sec. 10.2 circuit switching
• Sec. 10.3 single node

119
Section 1.7

• Preliminaries setting the stage for packet nets
  Arpanet Cyclades. Setting the stage for line
  disciplines Level Two protocols. Handy
  notations.

120
Packet net structure
PSN
Host computer
Host computer
Hosts subnet computer network, e.g. Cyclades
or Arpanet
PSN
Communications subnetwork, or subnet e.g. Cigale
121
Situation

• French telephone network PSTN in 1972
• 1-10 seconds to build a circuit
• 300 b/sec modems 1 000
• 56 Kb/s modems 20 000
• BER 10 - 2

122
Situation

• Hosts
• Mainframe computers from IBM, GE, CII, . .
• Vendors proprietary OSs
• Wanted
• Terminal-to-computer service telnet
• File transfer service ftp
• Unheard-of
• Electronic mail, webpages

123
Cyclades designers problem
CII
IBM
GE
M O D E M
M O D E M
PTT leased analog lines
FTP, telnet services
124
Setting the stage for

• Line disciplines or
• Level 2 protocols

125
Point-to-point line

• How about
• EDAC
• Flow control
• Synchronizing
• Data format packet? Frame? Or?

126
Multidrop line

• Addressing?
• Polling?
• Access protocol?

host
terminals
127
Handy notation linear graphs
Directed edge
Edge, Arc, link
136.2
Weighted edge
Cycle, circuit
128
Handy notation linear graphs
root
A tree
E1 E2 E3
A
B
E1,E2,E3 cutsetA,B
129
Handy Notations Queuing theory
server
Arrival process
To more queues
queue
Service process
Arrival process interarrival time distribution
exponential? Server process service time
distribution Poisson? Server utilization r Mean
queue length is r / (1- r ) under above
assumptions
130
Queuing models OpNET
I1 I2 I3
O1 O2
Queuing model of a packet switch