Computer Simulation of Computer Networks
CE321
Network Engineering
Introduction and
“beyond Cisco CCNA” material
v 2.3
CE321 – beyond Cisco notes ‹#›
1
Part 1: Circuit Switched Systems
CE321 – beyond Cisco notes ‹#›
2
Evolution of Networking
Networking started with the telegraph and then the telephone
Much of what takes place in networking today has its roots in the telephony network
Telephony networks are inherently circuit switched, whereas modern networks carry packet switched data
Even though modern networks carry packets, the underlying network has to have some type of “circuit” (e.g. fibre, cable, wavelength in fibre, time-slot in time division multiplexing…)
To start the story we will explore circuit switched systems.
CE321 – beyond Cisco notes ‹#›
Converging telecommunications and data communications
Telecommunications (e.g. plain old telephone service – POTS – see diagram) and data communications (e.g. Internet) are converging
One network is cheaper to manage and maintain than two
In telecommunications, communication takes place via circuits:
Network resources reserved
Information transfer takes place (e.g. phone call)
Resources released afterwards
node
subscriber A
link
subscriber B
CE321 – beyond Cisco notes ‹#›
University of Essex
Department of Electronic Systems Engineering
Computer Networks
2005
Dr. David K. Hunter
Page 4
Digital transmission techniques
Data is inherently digital; speech is inherently analogue
So why convert speech into digital form?
Immunity to noise
Difficult to mistake a ‘0’ for a ‘1’ or vice-versa at the receiver
Received signal is compared with a threshold midway between 0 and 1
Integration with data
One network handles both voice and data (reduced cost)
sampler
quantiser
coder
A/D: analogue to digital converter
t
amplitude
analogue in
10011101
digital out
CE321 – beyond Cisco notes ‹#›
University of Essex
Department of Electronic Systems Engineering
Computer Networks
2005
Dr. David K. Hunter
Page 5
Sampling
Sampling: representation of a time-continuous signal in discrete instances
To sample, the signal is probed at regular intervals of T seconds
Equivalent to multiplying analogue signal by sampling pulses
T known as the sampling period
What value should T have?
t
t
t
analogue
signal
sampling
pulses
sampled
signal
T
CE321 – beyond Cisco notes ‹#›
University of Essex
Department of Electronic Systems Engineering
Computer Networks
2005
Dr. David K. Hunter
Page 6
Nyquist’s theorem
F = 1 / T is the sampling frequency
Nyquist’s theorem: In order to recover an analogue signal from digital samples, require F 2W
W is the signal bandwidth
For speech, W 4 kHz
Hence F = 8 kHz, which is a standard frequency
T = 125 ms
frequency
amplitude
of typical
speech spectrum
0.3 kHz
3.4 kHz
CE321 – beyond Cisco notes ‹#›
University of Essex
Department of Electronic Systems Engineering
Computer Networks
2005
Dr. David K. Hunter
Page 7
Quantisation and coding
Quantisation: approximation of continuous amplitudes by discrete values
Maximum quantisation distortion is half the quantisation interval i.e. d / 2
Mean quantisation noise power = d2 / 12
Coding: for linearly quantised speech, at least 212 = 4096 levels required
In linearly quantised coding, all quantisation intervals are equal
In Pulse Code Modulation (PCM), each quantized sample is coded into binary “0” and “1” pulses
Hence number of bits required is log2 4096 = 12 bits
0
1
2
3
4
quantisation scale
2
3
decision level,
half way
between 2 & 3
sample
A
quantised
Q(A)
sample
B
quantised
Q(B)
d
CE321 – beyond Cisco notes ‹#›
University of Essex
Department of Electronic Systems Engineering
Computer Networks
2005
Dr. David K. Hunter
Page 8
Companding
Companding: make quantisation intervals progressively larger as the amplitude reaches higher positive or negative values
Reduces coded sample size by 4 bits, yielding 8-bit samples instead of 12 bits
No perceptible degradation in speech quality
Hence for PCM telephony, bitrate is 8000 8 = 64000 bit/s = 64 kbit/s
Standard value which occurs throughout digital telephony and this is called the G.711 codec.
analogue in at Tx
analogue out at Rx
CE321 – beyond Cisco notes ‹#›
University of Essex
Department of Electronic Systems Engineering
Computer Networks
2005
Dr. David K. Hunter
Page 9
Circuits and packets
The basic objective in telecommunications is to establish circuits between users
Permanent circuit: communications channel is exclusively allocated to the call for its duration (e.g. plain old telephone service – POTS)
Released when call is terminated
These will be studied in this part of the course
Virtual circuit: communications channel is held by the virtual circuit if there is data for transmission, otherwise it’s available for other virtual circuits
The route for the virtual circuit remains set up throughout
Hybrid between circuit switching and packet switching
In data communications, information is generally split into small units called packets
Often based on store-and-forward
Each packet received by a node is stored in memory
A decision is made about whether to send (forward) it to another node, and if so, which one
If necessary, the packet is read out of memory and forwarded to the next node
CE321 – beyond Cisco notes ‹#›
University of Essex
Department of Electronic Systems Engineering
Computer Networks
2005
Dr. David K. Hunter
Page 10
Control signalling in telephone networks
Apart from voice signals in telephone networks, there is also “control signalling”, often just called “signalling”
Signalling is information exchange to manage the network, and to establish, maintain and terminate calls:
Supervisory – detects changes in status of a line (on/off)
The supervisory circuit then generates a predetermined response
Request for service – subscriber off-hook
Call termination – subscriber on-hook
Address – identifies a subscriber e.g. when dialling a phone number
Subscriber to exchange – pulse dialling, DTMF (dual tone multi frequency)
Between exchanges – e.g. using tones in the channel used for the call
Call information – audible tones providing information to subscriber about call status e.g. dial tone, ringing tone, busy signal etc
Network management – maintenance, billing, diagnostic information and overall operation of the network
Inchannel signalling – call and signalling on same channel
Common channel signalling – control signals for several voice channels carried on a separate (common) channel
CE321 – beyond Cisco notes ‹#›
University of Essex
Department of Electronic Systems Engineering
Computer Networks
2005
Dr. David K. Hunter
Page 11
Subscriber interconnection – complete mesh
N selectors, each with N – 1 outlets so N(N – 1) / 2 = (6 5) / 2 = 15 lines
Simple control: “selection” by subscriber
Simple signalling: “alerting” via ringing
Simple switching but costly transmission
Particularly for large N and large distances between subscribers
N = 6:
6 subscribers
CE321 – beyond Cisco notes ‹#›
University of Essex
Department of Electronic Systems Engineering
Computer Networks
2005
Dr. David K. Hunter
Page 12
Centralised switching system
N lines for signalling and N lines for voice
Alternatively, voice and signalling could share the same line
Fewer lines, but more complex switching
Private Automatic Branch
Exchange (PABX)
voice
signalling
CE321 – beyond Cisco notes ‹#›
University of Essex
Department of Electronic Systems Engineering
Computer Networks
2005
Dr. David K. Hunter
Page 13
Full mesh networks vs star networks
Rough estimate of full mesh network cost:
Add up the length of all the N(N – 1) / 2 transmission links between nodes (in metres), and multiply the total by the cost per metre
Transmission cost only
No switching cost
Rough estimate of star network cost:
Add up the length of all the N transmission links between each node and the switch (in metres), and multiply the total by the cost per metre
This is the transmission cost
Add on the cost of the central switch itself
This is the switching cost
The star network has lower transmission cost
The full mesh network has no switching cost
Selectors at each subscriber are ignored in this simple analysis
CE321 – beyond Cisco notes ‹#›
University of Essex
Department of Electronic Systems Engineering
Computer Networks
2005
Dr. David K. Hunter
Page 14
Switching and transmission
No subscriber can connect
back to itself
Corresponding crosspoints omitted
N(N – 1) crosspoints
1
6
1
6
The trunk links between switching centres are shared between all subscribers as needed
crossbar switch
transmission
switching
centre
switching
centre
trunk links,
forming a
trunk group
represents
2
3
4
5
2
3
4
5
CE321 – beyond Cisco notes ‹#›
University of Essex
Department of Electronic Systems Engineering
Computer Networks
2005
Dr. David K. Hunter
Page 15
Electromechanical crossbar switch
CE321 – beyond Cisco notes ‹#›
University of Essex
Department of Electronic Systems Engineering
Computer Networks
2005
Dr. David K. Hunter
Page 16
Crossbar switching equipment
CE321 – beyond Cisco notes ‹#›
University of Essex
Department of Electronic Systems Engineering
Computer Networks
2005
Dr. David K. Hunter
Page 17
Cost optimisation in large networks
A large geographical area can be divided up into many smaller areas
Each smaller area has its own star network, with a central switch
These are then connected up in one large star network, with its own
central switch
More switching centres but less transmission cabling
Area per exchange can be selected to minimise overall cost
C
L
L
L
L
local switching
centre
area per local switching centre
0.0
1.0
2.0
3.0
4.0
5.0
6.0
0.0
1.0
2.0
3.0
total cost
transmission cost
switching cost
cost
central switching
centre
CE321 – beyond Cisco notes ‹#›
University of Essex
Department of Electronic Systems Engineering
Computer Networks
2005
Dr. David K. Hunter
Page 18
Telecommunications traffic
Telecommunications networks, like roads, are said to carry “traffic”
Telephone calls or data messages, instead of vehicles
A road gets jammed if there are too many cars and lorries, and a network becomes congested if there is too much traffic
Results in busy tone, or packet loss
Short of providing an infinite amount of resources, it’s impossible to guarantee that congestion will never occur
However, it’s possible to “dimension” a system so that the probability of congestion is acceptably low
The first work on “teletraffic theory” was published by the Danish scientist Agner K. Erlang in 1917
This section is a simple introduction to the basic concepts
CE321 – beyond Cisco notes ‹#›
19
Modelling circuit-switched telephony
Mathematical models analyse networks under particular traffic conditions
Complex models exist for e.g. IP datagrams carrying video traffic
We will look at the very simplest model: a concentrating switch in circuit-switched telephony
Modelling random phenomena which cannot be predicted in advance
We provide a formula for blocking in the concentrating switch above
Certain statistical assumptions are made, e.g. sources are independent
concentrating
switch
trunk group
N, A
very large number of sources
number of links
overall traffic offered
CE321 – beyond Cisco notes ‹#›
20
Reminder: Probability Density Function
Dice throws recorded, experimental results are shown (above) as PDF above
Area under PDF adds to 1
Infinite number of throws needed to get to “true” curve
Each number on the dice is equally likely for a perfect die:
Uniform Distribution
10 throws
103 throws
107 throws
CE321 – beyond Cisco notes ‹#›
21
Reminder: PDF of non-uniform distrubutions
Exponential distribution very important
Many natural systems behave like a Poisson process
That is time between events have exponential distribution (and other properties such as independence….)
Network events tend to behave like a Poisson process
infinite sided dice “loaded” with exponential distribution, value is time to next event
Imagine a “loaded” die
1 most likely, 2 less etc….
Example left is a negative exponential PDF (often just “exponential”)
Mean is
e=2.71828
CE321 – beyond Cisco notes ‹#›
Negative exponential call holding time
Probability distribution function of 7358 local call holding times
Observed statistics are good fit to negative exponential distribution
Poisson arrival process reasonable since many independent sources
CE321 – beyond Cisco notes ‹#›
23
Blocking, and performance metrics
Non-blocking
All demands receive immediate service
Not possible really for whole network, but can be used for part of it
For example, can apply to a single switch
Blocking, with various responses:
Loss – call is rejected, resulting in busy tone (Erlang-B formula – see later)
Delay – call held in a queue
Alternate route tried through the network
Combinations of these
A more advanced study would provide upper bounds on the following performance metrics:
Loss probability
Delay probability – probability of any delay occurring
Mean delay – in units of seconds or mean call holding time
Packet loss probability
Other statistics, such as probability of very large values, not just mean
All these are measured and calculated under normal load
Must allow for system growth and overload
CE321 – beyond Cisco notes ‹#›
24
Call holding times and the Erlang unit
Mean holding time per call h is an important statistic
Traffic A = lh is measured in Erlangs
l is the arrival rate for all (very many) sources – the actual number of sources is not important, only the overall arrival rate
A is the average number of concurrent calls carried by the incoming circuits – a dimensionless quantity
Examples of the Erlang unit
1 Erlang refers to a single circuit being in continuous use, or two circuits being in use 50% of the time
If an office has two telephone operators who are both busy all the time, that represents 2 Erlangs of traffic
If a radio channel is occupied for 30 minutes during an hour, it is said to carry 0.5 Erlangs of traffic
CE321 – beyond Cisco notes ‹#›
25
Infinite source, finite server: the Erlang-B formula
A = lh, arrival rate constant = l, representing all potential users
Number of potential users very large
Mean call holding time h for each active user
Blocking occurs if all N trunks are already in use; blocked calls are lost
B(N, A) is a well-tabulated expression known as the Erlang blocking formula, Erlang loss formula, or Erlang-B formula
The Erlang-B formula holds regardless of the statistics of the call holding times
Usual blocking probability (grade of service) was 0.01 or 1%
CE321 – beyond Cisco notes ‹#›
26
Example – telephone trunks
A group of telephone trunk links is offered traffic at an intensity of A = 10 Erlangs
Blocking may be calculated from Erlang’s formula
B(N, A)
Some values of B(N, 10) are tabulated opposite
N is the number of outgoing trunks
N B(N, 10) N B(N, 10)
1 0.909 13 0.084
2 0.820 14 0.057
3 0.732 15 0.036
4 0.647 16 0.022
5 0.564 17 0.013
6 0.485 18 0.007
7 0.409 19 0.004
8 0.338 20 0.002
9 0.273 21 0.0009
10 0.215 22 0.0004
11 0.163 23 0.0002
12 0.120 24 0.0001
CE321 – beyond Cisco notes ‹#›
27
Graph of the Erlang-B formula
CE321 – beyond Cisco notes ‹#›
Graph of the Erlang-B formula with log y-scale
CE321 – beyond Cisco notes ‹#›
Time Division Multiplexing (TDM)
Time-division multiplexing (TDM)
Each user or message uses the transmission medium at a different time
Interleaving of samples of channels (SDH or PDH)
Each frame contains several timeslots
Four timeslots in above example
Packet switching
May be considered as a variant of TDM
Each packet has a header indicating its destination
Unlike pure TDM, where destination can be deduced from position in frame (as above)
No frame structure
1
2
3
4
#1
#2
#3
#4
4
3
2
1
4
3
2
1
multiplexed data
stream at higher
bitrate
1
2
3
4
1
2
3
4
1
2
3
4
data
3
data
11
data
26
timeslot
frame
CE321 – beyond Cisco notes ‹#›
University of Essex
Department of Electronic Systems Engineering
Computer Networks
2005
Dr. David K. Hunter
Page 30
Basic TDM group (frame)
A frame has 30 speech channels, each held in a timeslot
Timeslots are transmitted one after another to form the frame
One timeslot = 8 bits (64 kbit/s speech)
One frame = 32 timeslots (30 speech, 2 signalling)
Uses 32 8 = 256 bits
Frame duration = 125 ms = 1 / 8000 s
Implies data rate of 2048 kbit/s
Timeslot 16: either inchannel or common-channel signalling
North America has different standard: frames of 24 timeslots
Implies data rate of 1536 kbit/s
These groups of 2048 or 1536 kbit/s are multiplexed by SDH
0
1
2
15
16
17
31
15 speech channels (1-15)
15 speech channels (16-30)
frame
alignment
signalling
frame of
256 bits
CE321 – beyond Cisco notes ‹#›
University of Essex
Department of Electronic Systems Engineering
Computer Networks
2005
Dr. David K. Hunter
Page 31
TDM multiframe
Multiframe is said to consist of 16 TDM frames, one after another
Data rate is still 2048 kbit/s
Multiframe concept was invented to define how signalling information should be spread over all the frames in a multiframe
Timeslot 16 in each frame N within a multiframe refers to a particular pair of channels (N and N + 15) in each frame
N can be any number between 1 and 15 inclusive
0
1
2
15
16
17
31
15 speech channels (1-15)
15 speech channels (16-30)
frame
alignment
signalling
0
1
2
N
15
multiframe of
16 frames (2 ms)
– same bitrate
frame of
256 bits
frames
timeslot 16
multiframe alignment
timeslot 16
signalling, channel
N of frame
signalling, channel
N + 15 of frame
CE321 – beyond Cisco notes ‹#›
University of Essex
Department of Electronic Systems Engineering
Computer Networks
2005
Dr. David K. Hunter
Page 32
Generic Telecommunications Network Hierarchy
Core
(Tier 1)
Metro (Tier 2)
Access (Tier 3)
Metro (Tier 2)
Access (Tier 3)
Traditionally, telecommunications networks are constructed in tiers:
Core – High capacity inter-city links with few intermediate nodes – long reach
Metro – Transports traffic between urban areas within a city – intermediate reach
Access – Collects traffic from local access points – short reach
Each tier has different traffic distributions (uniform, adjacent or hubbed), various bit-rates and consequently, different architectures (linear, ring or mesh)
CE321 – beyond Cisco notes ‹#›
33
Customer and Provider Connectivity
Provider Core
Modem
Router
ISP
POP
Data (home)
DSLAM
xDSL
Dedicated
“fat pipe”
(e.g.WAN i/f)
SDH
IP over SDH
Voice PBX
IP
ATM over SDH
(Circuit-Switched)
ATM
Provider Edge
SDH Transport Network
N-ISDN
Customer
DSLAM = Digital Subscriber Line Access Multiplexer, SDH = Synchronous Digital Hierarchy
Terminal multiplexers (TM) or digital cross-connects (DXC)
Voice (home)
Downstream bandwidth
2 Mb/s (E1)
56 kb/s
2 Mb/s (E1)
> 512 kb/s
144kb/s
4 kHz
PSTN
CE321 – beyond Cisco notes ‹#›
34
Synchronous Digital Hierarchy (SDH)
Two related and interoperable standards:
SDH (Europe)
SONET (Synchronous Optical Network – North America & Japan)
Set up high-capacity paths for weeks, months or years
Paths defined by overall forecast of subscribed demand
Can extract and insert individual lower bitrate channels easily from a high bitrate channel
PDH (an earlier technique) required inefficient and expensive “multiplexer mountains” to do this
Based on building blocks of 51.84 Mbit/s (SONET) and 155 Mbit/s (SDH)
Can carry a wide variety of signal types (next slide)
Can be multiplexed up to higher bitrates (see table)
The framing period is 125 ms = 1 / (8 kHz)
CE321 – beyond Cisco notes ‹#›
University of Essex
Department of Electronic Systems Engineering
Computer Networks
2005
Dr. David K. Hunter
Page 35
SDH capabilities
Can carry multiple lower level signals in these building blocks
2 Mbit/s or 1.5 Mbit/s TDM groups carrying multiple 64 kbit/s voice circuits
Ethernet – an important LAN standard
IP (Internet Protocol) – discussed previously
PDH signals – for backwards compatibility
New telecoms technologies must be compatible with old technologies
ATM (Asynchronous Transfer Mode)
Fibre Channel – used for SANs (storage area networks), i.e. connecting servers to disk controllers
Can accommodate future applications and their signal types
Defines extensive network management capabilities as part of the standard
Facilitate overall running of the network
CE321 – beyond Cisco notes ‹#›
University of Essex
Department of Electronic Systems Engineering
Computer Networks
2005
Dr. David K. Hunter
Page 36
Mapping data onto SDH
Data (IP, IPX, MPLS etc)
Fibre or Wavelength Division Multiplexing (WDM)
SDH/SONET
ATM
Private Lines
HDLC
Ethernet
X.86
PPP
POS
Voice
Ethernet has emerged as the dominant technology for LANs and enterprise networking
Storage devices are appearing with Gigabit Ethernet (GigE) interfaces
Interconnects among ISP Points of Presence (POPs) have GigE interfaces
Ethernet LANs require to be extended across WANs
The question is not really now how to transport IP, but how to transport Ethernet
CE321 – beyond Cisco notes ‹#›
37
The SONET/SDH Hierarchy
Optical SONET SDH Data rate Overhead rate Payload rate
level level level (Mb/s) (Mb/s) (Mb/s) (electrical) (electrical)
OC-1 STS-1 —- 51.84 1.728 50.112
OC-3 STS-3 STM-1 155.52 5.184 150.336
OC-12 STS-12 STM-4 622.08 20.736 601.344
OC-48 STS-48 STM-16 2488.32 82.944 2405.376
OC-192 STS-192 STM-64 9953.28 331.776 9621.504
OC-768 STS-768 STM-256 39,814.32 1327.104 38486.016
Electrical SONET signal is a Synchronous Transport Signal (STS)
Electrical SDH signal is a Synchronous Transport Module (STM)
Optical SDH/SONET signal is an Optical Carrier (OC)
SDH basic rate (higher than SONET) – chosen to accommodate the commonly used PDH signals
CE321 – beyond Cisco notes ‹#›
38
9 rows
Synchronous Transport Module (STM-1)
9 columns
270 columns
125 µs
Repetitive frame structure with a periodicity of 125 µs, the same as that of primary rate (E1)
9 × 270 = 2430 bytes in 125 µs i.e. 2430 × 8 × 8000 = 155.52 Mbit/s aggregate bit-rate
9 × 261 = 2349 bytes in 125 µs i.e. 2349 × 8 × 8000 = 150.336 Mbit/s useful payload per frame
Transport overhead (3%)
Usual Representation
Byte-wise Representation
64 kbit/s
CE321 – beyond Cisco notes ‹#›
39
Each tributary to the multiplex has its own payload area, occupying a group of columns, known as a tributary unit (TU)
In North America, this is known as a virtual tributary (VT)
Several TUs can be combined into a tributary unit group (TUG) in predefined ways
Tributary units and groups
9 rows
9 × 64 = 576 kbit/s per column
2 Mbit/s PCM channel (E1)
occupies four columns
Tributary Unit (TU)
9 rows
9 columns
Transport overhead (3%)
CE321 – beyond Cisco notes ‹#›
40
SDH/SONET Network Elements
Add/drop multiplexer (ADM)
Drops and adds entire multiplexes – e.g. 2 Mb/s or 155 Mb/s – from higher bitrate stream
Terminal multiplexer (TM)
Multiplexes traffic to next highest multiplexing level
Demultiplexer carries out the reverse function
Digital cross-connect (DXC)
Interconnects virtual containers (VCs) between ports
Lower-order e.g. VC-12 (2 Mbit/s)
Higher-order e.g. VC-4 (155 Mbit/s)
Differs from a switch
Operates under network management control, not user control
Connections (paths) set up over weeks or months, not minutes
Regenerator
Deployed when distance > 40-60 km in order to “clean up” the signal
Exercise – search the web for data on SDH/SONET network element products
Manufacturers include Nortel, Alcatel, Huawei and Ericsson
NxN
CE321 – beyond Cisco notes ‹#›
41
Digital Cross-Connects (DXCs)
STM-1 (155 Mbit/s)
STM-4 (622 Mbit/s)
STM-4 (622 Mbit/s)
STM-4 (622 Mbit/s)
Lower-order DXC granularity (VC-12)
Lower-order SDH cross-connects switch at the VC-12 level (2 Mbit/s) or lower
All SDH cross-connect architectures are capable of timeslot interchange (unlike early SONET equipment in USA) thus allowing freedom of network capacity allocation
The desired cross-connect granularity depends on the networking requirements
Higher-order DXC granularity (VC-4)
STM-1 (155 Mbit/s)
STM-4 (622 Mbit/s)
STM-4 (622 Mbit/s)
STM-4 (622 Mbit/s)
STM-1 (155 Mbit/s)
STM-4 (622 Mbit/s)
STM-4 (622 Mbit/s)
STM-4 (622 Mbit/s)
Different granularity cross-connects are available:
Higher-order SDH cross-connects switch at the VC-3/4 level (44/155 Mbit/s)
CE321 – beyond Cisco notes ‹#›
42
Ring A
Typical SDH/SONET Network
Ring B
ATM or Frame Relay Switch
ATM or Frame Relay Switch
IP Router
IP Router
DXC interconnects rings
Point to point segment
Tributary Interfaces
Trunk Interfaces
ADM
ADM
ADM
ADM
ADM
ADM
TM
TM
Regenerator
CE321 – beyond Cisco notes ‹#›
43
The Evolution of a SDH Transport Networks
10 Gbit/s
Very large ISP
Distance
Capacity
STM 4
Rings
155 Mbit/s
2.5 Gbit/s
622 Mbit/s
2 Mbit/s
Large business/ISP
WDM
Ring/mesh
OADM
Access
Ring
STM-1
Small/medium business
STM 64
Rings
E1
Access spurs
E1
STM 16
Rings
Future customers
“Raw” wavelength (no specified protocol)
CE321 – beyond Cisco notes ‹#›
44
Wavelength Division Multiplexing (WDM)
Channel rates
140 Mbit/s
2.5 Gbit/s
10/40 Gbit/s
Year 1980:
1 channel
year 1990:
2 channels
year 2000:
> 50 channels
Fibre attenuation
Windows of low loss in fibre
WDM
0.5
0.5
1.0
dB/km
dB/km
1.0
1.3
1.5
1.5
lm
/m
lm
/m
1.3
1.5
0.5
dB/km
1.0
1.3
lm
/m
Region in which optical erbium-doped fibre amplifiers (EDFAs) operate (1550 nm window)
In the 1970s, suitable lasers
only existed operating at
850 nm. This wavelength is
still used for short distances
in 10 Gb/s Ethernet
(10GBASE-SR).
CE321 – beyond Cisco notes ‹#›
45
Parallel fibres versus WDM
10 Gb/s
receiver
10 Gb/s
transmitter
10 Gb/s
receiver
10 Gb/s
transmitter
10 Gb/s
receiver
10 Gb/s
transmitter
10 Gb/s
receiver
10 Gb/s
transmitter
SDM: Space Division Multiplexing
Electrical regenerators
(1 per fibre)
4 fibres
4 fibres
10 Gb/s
transmitter
10 Gb/s
transmitter
10 Gb/s
transmitter
10 Gb/s
transmitter
10 Gb/s
transmitter
10 Gb/s
transmitter
10 Gb/s
transmitter
10 Gb/s
receiver
WDM: Wavelength Division Multiplexing
Optical Amplifiers (EDFAs – Erbium Doped Fibre Amplifiers)
(1 per fibre – many wavelengths)
Mux
De-mux
l2
l3
l4
l1
l2
l3
l4
1 fibre
1 fibre
l1
CE321 – beyond Cisco notes ‹#›
46
l3
Enhancing SDH with WDM
Stacked SDH rings within a fibre reduces problem of “fibre exhaustion”
l2
Wavelengths
STM-x rings
STM-x ring
l1
ADM
“coloured” trunk interfaces (15xxnm)
Wavelength determined by channel
ln
CE321 – beyond Cisco notes ‹#›
47
Network Resilience
SDH/SONET networks are often based upon physical ring topologies
A ring always has a diverse path that can be used as an alternative path in case of failure
Network resilience can be implemented at different layers of the network
SDH/SONET is able to protect traffic in a guaranteed time of 50 milliseconds after detecting the failure
Protecting closer to the physical layer masks physical failures from the higher network layers, which typically take longer to find an alternative path
Types of protection
Unidirectional: Only the channel in the failed direction is switched to the protection channel
Bi-directional: Channels in both directions are switched as a result of a failure in one direction
Revertive: Traffic is switched back to the working line when the working line has recovered from the failure
Non-revertive: A switch to the protection line is maintained even after the working line has recovered from the failure
CE321 – beyond Cisco notes ‹#›
48
1+1 Protection in SDH/SONET
Sender transmits outgoing traffic onto two fibres
Receiver selects better of two signals
Signal Fail (SF) – LOS, LOF or AIS-L
Signal Degrade (SD) – excessive BER (Bit Error Rate)
Unidirectional by default: bi-directional may be provided
Non-revertive by default: revertive may be provided
1:1 (one-to-one) protection is similar, but there is also a switch at the transmitter
Transmission only takes place on one fibre at once
Below shows a point-to-point 1+1 protection scheme, this is often termed linear protection.
Tx
Rx
Tx
Rx
W
P
W
P
signalling can be used for bi-directional switching
1
2
3
LOS = Loss of signal, LOF = loss of frame, AIS-L = Alarm indication signal – line
CE321 – beyond Cisco notes ‹#›
49
1:N (1 to N) Linear Protection in SDH/SONET
1 fibre protects N working fibres
Revertive only
Bidirectional is default, unidirectional can be provisioned
Transmitter and receiver must coordinate use of the protection fibre (uses K1/K2 bytes in frame overhead)
Protection line can carry unprotected traffic
Example: 1:3 protection (showing one direction only)
W1
W2
W3
P
1
Tx 1
Tx 2
Tx 3
3
5
CE321 – beyond Cisco notes ‹#›
50
D
A
C
B
protection
path
working
path
path
selector
signal
entry
signal
exit
fault-free state
fibre
pair
D
A
C
B
protection
path
working
path
path
selector
signal
exit
fibre cut
signal
entry
1
2
SDH/SONET Protection
1+1 applied to a ring (2-fibre DPRing – Dedicated Protection Ring,
UPSR – Unidirectional Path Switched Ring)
CE321 – beyond Cisco notes ‹#›
51
D
A
C
B
fault-free state
working
fibre
pair
protection
fibre
pair
Ring Protection
four fibres in total
4-fibre MS-SPRing (Multiplex Section Shared Protection Ring) or
BLSR (Bidirectional Line Switched Ring)
CE321 – beyond Cisco notes ‹#›
52
D
A
C
B
fibre cut – working span only
4-fibre MS-SPRing/BLSR
D
A
C
B
fibre cut – working and protection spans
Four fibres connected between each pair of adjacent nodes
Two fibres used for working channels, two used for protection
Two protection mechanisms
Working span failure – similar to 1:1 (1 to 1) protection switching (left-hand diagram)
Working and protection span failures – ring re-route (right-hand diagram)
Can protect against some (but not all) simultaneous faults
CE321 – beyond Cisco notes ‹#›
53
D
A
C
B
fault-free state
first half of capacity
in fibre pair
(working)
second half of
capacity in fibre
pair (protection)
2-fibre MS-SPRing/BLSR
two fibres (go and return)
D
A
C
B
fibre cut
reserved bandwidth (timeslots) used for re-routed traffic
Similar to 4-fibre BLSR except:
Span protection mechanism is not provided
Instead of separate protection fibres, half of capacity (i.e. half of the timeslots) of each fibre in each direction is reserved for protection
CE321 – beyond Cisco notes ‹#›
54
Growth in data vs voice
Consequently, instead of a circuit switched network optimized for real-time services like voice that also carries data – now we need a packet based network optimized to carry data that also carries voice.
Thus the motivation to move to voice over IP (VoIP)
A single packet network can carry all data types in a single network (voice, data, video), see Cisco Chapter 6 for the QoS mechanisms to allow sharing the data types
Unlike a circuit switched network a packet based network can allow other traffic (e.g. best effort data) to use capacity not used by voice/video (and vice versa) – this is called statistical multiplexing gain and means a shared packet based network will be cheaper.
The networks studied so far grew from a network designed around voice circuits (TDM with multiple of 64 kb/s channels)
This was because the network was dominated by voice circuits before approximately year 2000, this is also true now for mobile networks (see right).
Now data is dominating the network
CE321 – beyond Cisco notes ‹#›
Traffic calculation for VoIP compared to Circuit-Switched
Imagine circuit switched scenario:
l = 6 calls per minute with mean holding time of h= 2 mins, requirement for 1% blocking probability.
Traffic in Erlang A = lh = 6 * 2 = 12
From graph of Erlang-B earlier this implies the number of voice circuits required is T=20
If using standard voice circuits with G.711 codec with 64 kb/s capacity required is: 64000 * 20 = 1.28 Mb/s
In practice this may mean using a full E1 circuit (capacity 30 voice calls) with a bit-rate of 2.048 Mb/s
ie the smallest circuit-switched bearer with the rest of the capacity unused
CE321 – beyond Cisco notes ‹#›
Calculation for G.711 codec in packet network
Consider VoIP packet
VoIP is transmitted in IP with UDP (why not TCP?)
Additionally real-time transport protocol (RTP) is used (see next slide) to synchronise audio and identify the source
The voice stream has to be broken down into packets, but what size?
Too small: highly inefficient transport due to packet headers
Too large: excessive delay due to packetization delay and risk of loosing large amount of data if packet is lost
Common size is 20 ms of audio per-packet
See Cisco Chapter 6 maximum voice packet delay of 150 ms is a common requirement, and packetization is just one part of the delay.
CE321 – beyond Cisco notes ‹#›
RTP Header
RTP header used for many media streams carried in UDP
First 12 bytes are compulsory (others optional)
Sequence number used to keep packet order (remember UDP does not have a sequence number)
Timestamp is used to align different media streams
SSRC – synchronization source identifier identifies different streams e.g. video or audio of a video conference
CE321 – beyond Cisco notes ‹#›
Calculation for G.711 codec in packet network (continued)
G.711 codec 64 kb/s
20 ms of audio/packet ⇒ 64000 × 0.02/8 = 160 bytes/packet
Assuming an Ethernet/IP/UDP/RTP frame:
Protocol Layer Size (bytes)
Data (media see above) 160
RTP header 12
UDP header 8
IP header (without options) 20
Ethernet MAC (addresses + ethertype) 14
Ethernet physical layer (preamble, start of frame, FCS) 12
Total 226
Bit rate is thus 226 × 8 / 0.02 = 90.4 kb/s
This would make VoIP less efficient, but as well as networks codecs have improved …
CE321 – beyond Cisco notes ‹#›
Example voice codecs
Codec Typical use Bit-rate (kb/s)
G.711 PSTN telephony 64
Adaptive multi-rate (AMR) Mobile phone (GSM) 12.2 (but other rates possible, minimum for “toll quality is 7.4 kb/s)
G.729 Low-bit rate VoIP applications 8
Performing the same calculation as previous for AMR at 7.4 kb/s:
20 ms of audio/packet ⇒ 7400 × 0.02/8 = 19 bytes/packet (rounded)
Same headers as previous gives total packet 85 bytes
Bit rate is thus 85 × 8 /0.02 = 34 kb/s
Additionally silence suppression (not sending when not talking) can reduce on average by about 30%
Compression of IP/UDP/RTP header to 2 bytes is possible (but not when mixing with other IP traffic)
CE321 – beyond Cisco notes ‹#›
How much to allow for multiple VoIP calls?
While Erlang-B is only strictly valid for circuit switched networks, it can be used as a guide for the number of consecutive calls
For example our previous scenario (6 calls per minute, 2 minute call duration, 1% blocking probability) needed 20 circuits. This is also a good estimate for the number of concurrent calls to achieve reasonable quality in VoIP.
So for the previous AMR wideband we have 34 kb/s with 20 concurrent calls ⇒ 34000 × 20 = 0.68 Mb/s
This presents a large saving compared to a 2.048 Mb/s E1 and if there happen to be no voice calls this capacity could be used by other traffic types.
CE321 – beyond Cisco notes ‹#›
Summary: Part 1 Circuit Switched Networks
At fundamental level all operator networks are circuit switched (a physical fibre or other point-to-point network)
However, increasingly the switched networks are used to interconnect packet-switched systems (mostly IP).
A converged IP network allows one network to carry voice, video and data – this is cheaper as it allows statistical multiplexing and reduces the number of networks that need to be provisioned and mangaged
Thus operators are changing to packet switched networks for all services.
Part 2 considers Multiprotocol label switching (MPLS), the predominant packet switching technology used by operators to replace SDH/SONET switching.
SDH/SONET may still be used for Point-to-point data between MPLS switches, but in new links Ethernet is now being used.
CE321 – beyond Cisco notes ‹#›
pdf(x) = λe−λx
pdf(x)=le
-lx
1/ λ
1/l
B N,A( ) =
AN N!
Ai i!
i=0
N
∑
=
AN N!
1+ A+ A
2
2
+
A3
6
++
AN
N!
n!=1×2×3××n
BN,A
()
=
A
N
N!
A
i
i!
i=0
N
å
=
A
N
N!
1+A+
A
2
2
+
A
3
6
++
A
N
N!
n!=1´2´3´´n
T=1
T=2
T=5
T=10 T=20
1e−04
1e−03
1e−02
1e−01
1e+00
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
Erlang E
B
lo
ck
in
g
Pr
ob
ab
ili
ty
P
(E
, T
)
T=1
T=2
T=5
T=10 T=20
1e−04
1e−03
1e−02
1e−01
1e+00
0123456789101112131415161718192021222324
Erlang E
B
l
o
c
k
i
n
g
P
r
o
b
a
b
i
l
i
t
y
P
(
E
,
T
)
/docProps/thumbnail.jpeg