Lecture notes

Switching Systems by Jorma Kekalainen

Switching Systems
Switch structures
Jorma Kekalainen

208

Structure of a switch
Switches can categorized as
circuit switches and
Packet switches

Circuit switching can use either of two

technologies:
the space-division or
the time-division switching or
combination of these.

209

Switch structures

Page 1

Lecture notes

Switching Systems by Jorma Kekalainen

Basic concepts

Accessibility
Blocking
Complexity
Scalability
Reliability
Throughput

210

Accessibility
A network has full accessibility ( connectivity)
when each inlet can be connected to each
outlet (in case there are no other I/O
connections in the network)
A network has a limited accessibility when the
above given property does not exist
Interconnection networks applied in todays
switch fabrics usually have full accessibility

211

Switch structures

Page 2

Lecture notes

Switching Systems by Jorma Kekalainen

Accessibility
Full accessibility

Limited accessibility

212

Blocking
Blocking is defined as failure to satisfy a
connection request and it depends strongly on
the combinatorial properties of the switching
networks

213

Switch structures

Page 3

Lecture notes

Switching Systems by Jorma Kekalainen

Blocking
Non-blocking - a path between an arbitrary idle inlet and
arbitrary idle outlet can always be established independent of
network state at set-up time
Blocking - a path between an arbitrary idle inlet and arbitrary
idle outlet cannot be established owing to internal congestion
due to the already established connections
Strict-sense non-blocking - a path can always be set up between
any idle inlet and any idle outlet without disturbing paths already
set up
Wide-sense non-blocking - a path can be set up between any idle
inlet and any idle outlet without disturbing existing connections,
provided that certain rules are followed.
These rules prevent network from entering a state for which new
connections cannot be made

Rearrangeably non-blocking - when establishing a path between

an idle inlet and an idle outlet, paths of existing connections may
have to be changed (rearranged) to set up that connection
214

Different sorts of blocking networks

215

Switch structures

Page 4

Lecture notes

Switching Systems by Jorma Kekalainen

Complexity
Complexity of an interconnection network is
expressed by cost index
Traditional definition of cost index gives the number
of cross points in a network
used to be a reasonable measure of space division switching
systems

Nowadays cost index alone does not characterize cost

of an interconnection network for broadband
applications
VLSIs and their integration degree has changed the way how
cost of a switch fabric is formed (number of ICs, power
consumption)
management and control of a switching system has a
significant contribution to cost
216

Complexity

217

Switch structures

Page 5

Lecture notes

Switching Systems by Jorma Kekalainen

Scalability
Due to constant increase of transport links and data rates on
links, scalability of a switching system has become a key
parameter in choosing a switch fabric architecture
Scalability describes ability of a system to evolve with
increasing requirements
Issues that are usually matter of scalability

number of switching nodes

number of interconnection links between nodes
bandwidth of interconnection links and inlets/outlets
throughput of switch fabric
buffering requirements
number of inlets/outlets supported by switch fabric

218

Example of scalability
A switching equipment has room for 20 line-cards and the
original design supports 10 Mbit/s interfaces (one per line card)
Throughput of switch fabric is scalable from 500 Mbit/s to 2
Gbit/s
When new line cards that each implement two 10 Mbit/s
interfaces are introduced, the interface logic may have to be
upgraded
When new line cards that implement a 100 Mbit/s interface (one
per line card) are introduced, the switch fabric has to be
upgraded (scaled up) to 2 Gbit/s speed and the interface logic
has to be upgraded to 100 Mbit/s speed
Buffering memories need to be replaced by faster (and possible
larger) ones
Larger number (>20) of line cards implies at least new physical
design
Increase of line rates beyond 100 Mbit/s means redesign of
switch fabric
219

Switch structures

Page 6

Lecture notes

Switching Systems by Jorma Kekalainen

Reliability
Reliability and fault tolerance are system measures that have an
impact on all functions of a switching system
Reliability defines probability that a system does not fail within
a given time interval provided that it functions correctly at the
start of the interval
Availability defines probability that a system will function at a
given time instant
Fault tolerance is the capability of a system to continue its
intended function in spite of having a fault(s)
Reliability measures:
MTTF (Mean Time To Failure)
MTTR (Mean Time To Repair)
MTBF (Mean Time Between Failures)

220

Throughput
Throughput gives forwarding/switching
speed/efficiency of a switch fabric
It is measured in bits/s, octets/s, cells/s,
packet/s, etc.
Quite often throughput is given in the range
(0 ... 1), i.e. the obtained forwarding speed is
normalized to the theoretical maximum
throughput

221

Switch structures

Page 7

Lecture notes

Switching Systems by Jorma Kekalainen

Switching mechanisms
A switched connection requires a mechanism that
attaches the right information streams to each other
Switching takes place in the switch fabric, the
structure of which depends on networks mode of
operation, available technology and required capacity
Communicating terminals may use different physical
links and different time-slots, so there is an obvious
need to switch both in time and in space domain
Time and space switching are basic functions of a
switch fabric

222

Space-Division Switch
In space-division switching, the paths in the
circuit are separated from one another
spatially.
This technology was originally designed for
use in analog networks but is used currently in
both analog and digital networks.
It has evolved through a long history of many
designs.

223

Switch structures

Page 8

Lecture notes

Switching Systems by Jorma Kekalainen

Space division switching

A space switch directs traffic from input links to
output links
An input may set up one connection (1, 3, 6 and 7),
multiple connections (4) or no connection (2, 5 and 8)

224

Crossbar Switch
Crossbar matrix introduces the basic structure of a
space switch
A crossbar switch connects n inputs to m outputs in a
grid, using electronic microswitches at each
crosspoint.
The major limitation of this design is the number of
crosspoints required.
To connect n inputs to m outputs using a crossbar
switch requires n
m crosspoints.
For example, to connect 1000 inputs to 1000 outputs
requires a switch with 1,000.000 crosspoints.
Such a switch is inefficient because only small
minority of crosspoints are use at any given time.
225

Switch structures

Page 9

Lecture notes

Switching Systems by Jorma Kekalainen

Crossbar switch with three inputs and four outputs

226

Crossbar switch matrix

Information flows are controlled (switched) by
opening and closing cross-points
m inputs and n outputs => mn cross-points
(connection points)
Only one input can be connected to an output at a
time, but an input can be connected to multiple
outputs (multi-cast) at a time

227

Switch structures

Page 10

Lecture notes

Switching Systems by Jorma Kekalainen

An example space switch

m1 -multiplexer used to implement a space switch
Every input is fed to every output mux and mux
control signals are used to select which input signal is
connected through each mux

228

Time-Division Switch
Time-division switching uses time-division
multiplexing (TDM) inside a switch.
The most popular technology is called the time-slot
interchange (TSI).

229

Switch structures

Page 11

Lecture notes

Switching Systems by Jorma Kekalainen

Time-Division Switch
The previous switch combines a TDM multiplexer, a
TDM demultiplexer, and a TSI consisting of random
access memory (RAM) with several memory locations.
The size of each location is the same as the size of a
single time slot.
The number of locations is the same as the number of
inputs (in most cases, the numbers of inputs and
outputs are equal).
The RAM fills up with incoming data from time slots
in the order received.
Slots are then sent out in an order based on the
decisions of a control unit.
230

Time-slot interchange

Each input line wants to send data to an output line according to the
following pattern:
1 3 2 4 3 1 4 2
231

Switch structures

Page 12

Lecture notes

Switching Systems by Jorma Kekalainen

Time division multiplexing

Time-slot interchanger is a device, which buffers m
incoming time- slots, e.g. 30 time-slots of an E1
frame, arranges new transmit order and transmits n
time-slots
Time-slots are stored in buffer memory usually in the
order they arrive or in the order they leave the
switch - additional control logic is needed to decide
respective output order or the memory slot where an
input slot is stored

232

Time-slot interchange

233

Switch structures

Page 13

Lecture notes

Switching Systems by Jorma Kekalainen

Time switch implementation example 1

Incoming time-slots are written cyclically into switch memory
Output logic reads cyclically control memory, which contains a
pointer for each output time-slot
Pointer indicates which input time-slot to insert into each output
time-slot

234

Time switch implementation example 2

Incoming time-slots are written into switch memory by using
write-addresses read from control memory
A write address points to an output slot to which the input slot
is addressed
Output time-slots are read cyclically from switch memory

235

Switch structures

Page 14

Lecture notes

Switching Systems by Jorma Kekalainen

Properties of time switches

Input and output frame buffers are read and written
at wire-speed, i.e. m R/Ws for input and n R/Ws for
output
Interchange buffer (switch memory) serves all inputs
and outputs and thus it is read and written at the
aggregate speed of all inputs and outputs
=> speed of an interchange buffer is a critical
parameter in time switches and limits performance
of a switch

236

Time-Space analogy
A time switch can be logically converted into a space
switch by setting time-slot buffers into vertical
position => time-slots can be considered to
correspond to input/output links of a space switch

237

Switch structures

Page 15

Lecture notes

Switching Systems by Jorma Kekalainen

Space-Space analogy
A space switch carrying time multiplexed input and
output signals can be logically converted into a pure
space switch (without cyclic control) by distributing
each time-slot into its own space switch

238

An example conversion

239

Switch structures

Page 16

Lecture notes

Switching Systems by Jorma Kekalainen

Time- and Space-Division Switch

Combinations
When we compare space-division and time-division
switching, some interesting facts emerge.
The advantage of space-division switching is that it is
instantaneous.
Its disadvantage is the number of crosspoints required to
make space-division switching acceptable in terms of
blocking.
The advantage of time-division switching is that it needs no
crosspoints.
Its disadvantage, in the case of TSI, is that processing each
connection creates delays.
Each time slot must be stored by the RAM, then retrieved and
passed on.

240

Properties of space and time switches

Space switches
number of cross-points
m input n output = mn
when m=n n2

output bit rate

determines the speed
requirement for the
switch components
both input and output
lines deploy bus
structure
=> fault location
difficult

Time switches
size of switch memory (SM)
and control memory (CM)
grows linearly as long as
memory speed is sufficient,
i.e. SM + CM + input
buffering + output buffering
= 2 2 number of timeslots
a simple and cost effective
structure when memory
speed is sufficient
speed of available memory
determines the maximum
switching capacity
241

Switch structures

Page 17

Lecture notes

Switching Systems by Jorma Kekalainen

Multistage Switch
The solution to the limitations of the crossbar switch is the
multistage switch, which combines crossbar switches in several
stages.
In a single crossbar switch, only one row or column (one path) is
active for any connection.
So we need N x N crosspoints.
If we can allow multiple paths inside the switch, we can
decrease the number of crosspoints.
Each crosspoint in the middle stage can be accessed by multiple
crosspoints in the surrounding stages.

242

Multistage Switch

243

Switch structures

Page 18

Lecture notes

Switching Systems by Jorma Kekalainen

Design a three-stage switch

1.

To design a three-stage switch, we follow these steps:

We divide the N input lines into groups, each of n lines.
For each group, we use one crossbar of size (n x k), where k is the
number of crossbars in the middle stage.
In other words, the first stage has N/n crossbars of (n x k)
crosspoints.

2.
3.

We use k crossbars, each of size [(N/n) x (N/n)] in the middle

stage.
We use N/n crossbars, each of size (k x n) at the third stage.
The total number of crosspoints are:

N
(n k ) + k N N + N (k n ) = 2kN + k N
n
n n n
n

244

Note

In a three-stage switch, the total

number of crosspoints is
2kN + k(N/n)2
which is much smaller than the number of
crosspoints in a single-stage switch (N2).

245

Switch structures

Page 19

Lecture notes

Switching Systems by Jorma Kekalainen

Example
Design a three-stage, 200 200 switch (N = 200) with k = 4 and
n = 20.
Solution
In the first stage we have N/n=200/20 or 10 crossbars, each of
size (nk)=20 4.
In the second stage, we have k=4 crossbars, each of size (N/n
N/n )= 10 10.
In the third stage, we have N/n=10 crossbars, each of size (k n)=4
20.
The total number of crosspoints is 2kN + k(N/n)2, or 2000
crosspoints.

Note. This is 5 percent of the number of crosspoints in a singlestage switch (200 200 = 40,000).

246

Blocking (1)
The multistage switch in the previous Example has one drawback
blocking during periods of heavy traffic.
The whole idea of multistage switching is to share the
crosspoints in the middle-stage crossbars.
Sharing can cause a lack of availability when the resources are
limited and all users want a connection at the same time.
Blocking refers to times when one input cannot be connected to
an output because there is no path available between them all
the possible intermediate switches are occupied.

247

Switch structures

Page 20

Lecture notes

Switching Systems by Jorma Kekalainen

Blocking (2)
In a single-stage switch, blocking does not occur because every
combination of input and output has its own crosspoint; there is
always a path.
Cases in which two inputs are trying to contact the same output do
not count. That path is not blocked; the output is merely busy.

In the multistage switch described in Example, however, only 4

of the first 20 inputs can use the switch at a time, only 4 of the
second 20 inputs can use the switch at a time, and so on.
The small number (k) of crossbars at the middle stage creates
blocking.

248

Blocking (3)
In large systems, such as those having 10,000 inputs
and outputs, the number of stages can be increased
to cut down on the number of crosspoints required.
As the number of stages increases, however,
possible blocking increases as well.
Note. Many people have experienced blocking on public
telephone systems caused by overload of the system.

249

Switch structures

Page 21

Lecture notes

Switching Systems by Jorma Kekalainen

Clos criterion
Clos investigated the condition of nonblocking in multistage
switches and came up with the following result:
In a nonblocking switch, the number of middle-stage switches must
be at least (2n-1).

In other words, we need to have

k (2n-1).
In this case, the total number of crosspoints is greater than or
equal to 4N[(2N) 1].
Thus, the minimum number of crosspoints according to the Clos
criterion is proportional to N3/2.
Note that the number of crosspoints is still increasing slower
than in a single-stage switch~N2

A Clos network is a kind of multistage switching network, first formalized by

250
Charles Clos in 1953. Clos networks are required when the physical circuit switching
needs exceed the capacity of the largest feasible single crossbar switch.

Minimum number of crosspoints

Calculate the minimum number of crosspoints with a fixed N by
using the Clos criterion.
Solution
Now we need to minimize the number of crosspoints with a fixed
N by using the Clos criterion k (2n-1).
We can take the derivative of the equation 2kN + k(N/n)2,
k=2n-1, with respect to n (the only variable) and find the value
of n that makes the result zero.
This n must be equal to or greater than (N/2)
In this case, the total number of crosspoints is greater than or
equal to 4N[(2N) 1].
In other words, the minimum number of crosspoints according to
the Clos criterion is proportional to N3/2.

251

Switch structures

Page 22

Lecture notes

Switching Systems by Jorma Kekalainen

Note

According to the Clos criterion:

n = (N/2)1/2
k 2n 1
Crosspoints 4N [(2N)1/2 1]

252

Example
Redesign the previous three-stage, 200 200 switch, using the
Clos criteria with a minimum number of crosspoints.
Solution
We let n = (200/2)1/2, or n = 10. We calculate k = (2n 1)= 19. In
the first stage, we have N/n=200/10, or 20, crossbars, each
with (nk)= 10 19 crosspoints.
In the second stage, we have k=19 crossbars, each with (N/n
N/n )= 10 10 crosspoints.
In the third stage, we have N/n=20 crossbars each with (k n)=
19 10 crosspoints.
The total number of crosspoints is 20(10 19) + 19(10 10) +
20(19 10) = 9500.

253

Switch structures

Page 23

Lecture notes

Switching Systems by Jorma Kekalainen

Comment about the previous example

If we use a single-stage switch, we need 200 x 200
40,000 crosspoints.
The number of crosspoints in this three-stage switch
is 24 percent that of a single-stage switch.
A lot more points (9500) are needed than in the first
Example (2000 5 percent).
The extra crosspoints are needed to prevent
blocking.

254

Comment 2
A multistage switch that uses the Clos criteria and a
minimum number of crosspoints still requires a huge
number of crosspoints.
For example, to have a 100,000 input/output switch,
we need something close to 200 million crosspoints
(instead of 10 billion with the single stage).
This means that if a telephone company needs to
provide a switch to connect 100,000 telephones in a
city, it needs 200 million crosspoints.
The number can be reduced if we accept blocking.
Today, telephone companies use time-division
switching or a combination of space- and time-division
switches.
255

Switch structures

Page 24

Lecture notes

Switching Systems by Jorma Kekalainen

Two stage switches

Time-Time (TT) switch

Time-Space (TS) switch
Space-Time (ST) switch
Space-Space (SS) switch
TT-switch gives no advantage compared to a
single stage T-switch
SS-switch increases blocking probability

256

Two stage switches

ST-switch gives high blocking probability (S-switch can develop
blocking on an arbitrary bus, e.g. slots from two different buses
attempting to flow to a common output)
TS-switch has low blocking probability, because T-switch allows
rearrangement of time-slots so that S-switching can be done
blocking free

257

Switch structures

Page 25

Lecture notes

Switching Systems by Jorma Kekalainen

Time multiplexed space (TMS) switch

Space divided inputs and
each of them carry a frame
of three time-slots
Input frames on each link are
synchronized to the crossbar
A switching plane for each
time-slot to direct incoming
slots to destined output links
of the corresponding timeslot

258

Connection conflicts in a TMS switch

Space divided inputs and each
of them carry a frame of three
time-slots
Input frames on each link are
synchronized to the crossbar
A switching plane for each
time-slot to direct incoming
slots to destined output links
of the corresponding time-slot

259

Switch structures

Page 26

Lecture notes

Switching Systems by Jorma Kekalainen

Connections through SS-switch

260

Three stage switch combinations

Possible three stage switch combinations:
Time-Time-Time (TTT) (not significant, no connection from PCM to
PCM)
Time-Time-Space (TTS) (=TS)
Time-Space-Time (TST)
Time-Space-Space (TSS)
Space-Time-Time (STT) (=ST)
Space-Time-Space (STS)
Space-Space-Time (SST) (=ST)
Space-Space-Space (SSS) (not significant, high probability of
blocking)

Three interesting combinations TST, TSS and STS

261

Switch structures

Page 27

Lecture notes

Switching Systems by Jorma Kekalainen

Three stage switches

Basic TS-switch sufficient for switching time-slots onto
addressed outputs, but slots can appear in any order in the
output frame
If a specific input slot is to carry data of a specific output slot
then a time-slot interchanger is needed at each output
=> any time-slot on any input can be connected to any time-slot
on any output => blocking probability minimized
Such a 3-stage configuration is named TST-switching

262

TST switch (1)

In a third option, we combine space-division and timedivision technologies to take advantage of the best of
both.
Combining the two results in switches that are
optimized both physically (the number of
crosspoints) and temporally (the amount of delay).
Multistage switches of this sort can be designed as
time-space-time (TST) switch.

263

Switch structures

Page 28

Lecture notes

Switching Systems by Jorma Kekalainen

TST switch (2)

Figure shows a simple TST switch that consists of two time
stages and one space stage and has 12 inputs and 12 outputs.
Instead of one time-division switch, it divides the inputs into
three groups (of four inputs each) and directs them to three
timeslot interchanges.
The result is that the average delay is one-third of what would
result from using one time-slot interchange to handle all 12
inputs.
The last stage is a mirror image of the first stage.
The middle stage is a space-division switch (crossbar) that
connects the TSI groups to allow connectivity between all
possible input and output pairs

264

Time-space-time switch

265

Switch structures

Page 29

Lecture notes

Switching Systems by Jorma Kekalainen

Time-Space-Space switch
Time-Space-Space switch can be applied to
increase switching capacity

266

Space-Time-Space switch
Space-Time-Space switch has a high blocking
probability (like ST-switch) - not a desired
feature in public networks

267

Switch structures

Page 30

Lecture notes

Switching Systems by Jorma Kekalainen

Graph presentation of space switch

268

Graph presentation of space switch

Switch structures

269

Page 31

Lecture notes

Switching Systems by Jorma Kekalainen

Graph presentation example

270

SSS-switch and its graph

presentation

271

Switch structures

Page 32

Lecture notes

Switching Systems by Jorma Kekalainen

Graph presentation of connections

272

Graph presentation of connections

Note:

Switch structures

273

Page 33

Lecture notes

Switching Systems by Jorma Kekalainen

Graph presentation of connections

274

Cost criteria for switch fabrics

Number of cross-points
Fan-out
Logical depth
Blocking probability
Complexity of switch control

275

Switch structures

Page 34

Lecture notes

Switching Systems by Jorma Kekalainen

Cross-points
Number of cross-points gives the number of on-off
gates in space switching equivalent of a fabric
Minimization of cross-point count is essential when crosspoint technology is expensive (e.g. electro-mechanical and
optical cross-points)
Very Large Scale Integration (VLSI) technology implements
cross-point complexity in Integrated Circuits (ICs) => more
relevant to minimize number of ICs than number of crosspoints
Due to increasing switching speeds, large fabric
constructions and increased integration density of ICs,
power consumption has become a crucial design criteria
higher speed => more power
large fabrics => long buses, fan-out problem and more driving
power
increased integration degree of ICs => heating problem
276

Fan-out and logical depth

VLSI chips can hide cross-point complexity, but

introduce pin count and fan-out problem
length of interconnections between ICs can be long lowering
switching speed and increasing power consumption
parallel processing of switched signals may be limited by the
number of available pins of ICs
fan-out gives the driving capacity of a switching gate, i.e.
number of inputs (gates/cross-points) that can be connected
to an output

Logical depth gives the number of cross-points a

signal traverses on its way through a switch
large logical depth causes excessive delay and signal
deterioration
Fan-out is a measure of the ability of a logic gate output, to drive a number of inputs of other
logic gates. A perfect logic gate would have infinite input impedance and zero output
impedance, allowing a gate output to drive any number of gate inputs. However, inputs of real
gates have capacitance as well as resistance. This capacitance will slow the output transition
277
of the previous gate and hence increase its propagation delay, which affects the maximum
speed of the overall system.

Switch structures

Page 35

Lecture notes

Switching Systems by Jorma Kekalainen

Illustration of cross-points, fan-out and

logical depth

278

Blocking probability
Blocking probability of a multi-stage switching network is
difficult to determine
Lees approximation gives a coarse measure of blocking
Assume uniformly distributed load
equal load in each input
load distributed uniformly among intermediate stages (and their
outputs) and among outputs of the switch

probability that a path between any two switching blocks is

engaged is p
Probability that all k paths between an input switching block and
an output switching block are engaged is
B= [1-(1-p)2]k ,
which is known as Lees approximation

279

Switch structures

Page 36

Lecture notes

Switching Systems by Jorma Kekalainen

Control complexity
Given a graph G, a control algorithm is needed to find and set up
paths in G to fulfill connection requirements
Control complexity is defined by the hardware (computation and
memory) requirements and the run time of the algorithm
Amount of computation depends on interconnection structure
and degree of blocking tolerated
In general, computation complexity grows exponentially as a
function of the number of terminals
There are interconnection networks that have a regular
structure for which control complexity is substantially reduced
There are also structures that can be distributed over a large
number of control units

280

Control complexity (cont.)

281

Switch structures

Page 37

Lecture notes

Switching Systems by Jorma Kekalainen

Example
Suppose that driving current of a switching gate (cross-point) is
100 mA and its maximum input current is 8 mA
How many output gates can be connected to a bus, driven by one
input gate, if the capacitive load of the bus is negligibly small ?
Fan-out= floor[100/8]= 12

How many output gates can be connected to a bus driven by one

input gate if load of the bus corresponds to 15 % of the load of
a gate input)?
Fan-out = floor[100/(1.15x8)] = 10
282

floor(x) is the greatest integer less than or equal to x.

Management complexity
Network management involves adaptation and
maintenance of a switching network after the
switching system has been put in place
Network management deals with

failure events and growth in connectivity demand

changes of traffic patterns from day to day
overload situations
diagnosis of hardware failures in switching system, control
system as well as in access and trunk network
in case of failure, traffic is rerouted through redundant built-in
hardware or via other switching facilities
diagnosis and failure maintenance constitute a significant part of
software of a switching system

In order for switching cost to grow linearly in respect

to total traffic, switching functions (such as control,
maintenance, call processing and interconnection
network) should be as modular as possible
283

Switch structures

Page 38

Lecture notes

Switching Systems by Jorma Kekalainen

Example

284

Erlangs blocking formula

Erlang-B, also known as the Erlang loss formula, is a formula for the blocking
probability derived from the Erlang distribution to describe the probability of
call loss on a group of circuits (in a circuit switched network, or equivalent).
It is, for example, used in planning telephone networks.
The formula is not limited to telephone networks, since it describes a probability
in a queuing system (albeit a special case with a number of servers but no buffer
spaces for incoming calls to wait for a free server).
The formula applies under the condition that an unsuccessful call, because the
line is busy, is not queued or retried, but instead really lost.
It is assumed that call attempts arrive following a Poisson process

call arrivals are independent,

the exponentially-distributed inter-arrival times with parameter (mean 1/)

Message length (holding times) are exponentially distributed although the B

formula turns out to apply under general holding time distributions.

285

Switch structures

Page 39

Lecture notes

Switching Systems by Jorma Kekalainen

Erlang B

The erlang (Erl) as a dimensionless unit is used in telephony as a

statistical measure of offered load.
calculated as the average arrival rate () multiplied by the average call
length (1/).

The Erlang B formula assumes an infinite population of sources (such as

telephone subscribers), which jointly offer traffic to N servers (such
as links in a trunk group).
The rate of arrival of new calls (birth rate) is not depending on the
number of active sources, because the total number of sources is
assumed to be infinite.
The rate of call departure (death rate) is equal to the number of calls
in progress divided by the mean call holding time.
The B formula calculates blocking probability in a loss system, where if
a request is not served immediately when it tries to use a resource, it is
aborted.
Requests are therefore not queued.
Blocking occurs when there is a new request from a source, but all the
servers are already busy.
The formula assumes that blocked traffic is immediately cleared.
The formula provides the GoS (grade of service) which is the
probability PN that a new call arriving at the circuit group is rejected
because all servers (circuits) are busy:
PN = EB(N,A) when A Erl of traffic are offered to N trunks (communication
286
channels).

Amount of traffic in Erlangs

Erlang defines the amount of traffic flowing through a
communication system - it is given as the aggregate holding time
of all channels of a system divided by the observation time
period
Example 1:
During an hour period, three calls are made (3 min, 12 min and 15
min) using a single telephone channel (per transfer direction) =>
the amount of traffic carried by this channel is (30 min/60 min)
= 0.5 Erlangs.
Since each call occupies two one-way channels, the total
generated traffic is 20.5 Erlangs.
Example 2:
A telephone exchange supports 1000 channels and during a busy
hour each channel is occupied 30 minutes on the average
=> the amount of traffic carried through the switch during the
busy hour is (1000 30 min / 60 min) = 500 Erlangs.
287

Switch structures

Page 40

Lecture notes

Switching Systems by Jorma Kekalainen

Erlangs blocking formula

Erlangs blocking formula
(Erlangs first formula)

AN
P( K = N ) PN = N N ! n = E B ( N , A)
A

n = 0 n!
Erlangs blocking formula applies to systems fulfilling conditions

a failed call is disconnected (loss system)

full accessibility
time between subsequent calls vary randomly
large number of sources
288

Example
E(5, 3) implies that we have a system of 5 inlets and
offered load is 3

Erlang-blocking calculated using the formula is 11 %

Tables and diagrams (based on Erlangs formula) have
been produced to simplify blocking calculations

289

Switch structures

Page 41

Lecture notes

Switching Systems by Jorma Kekalainen

Example
An exchange for 1000 subscribers is to be installed and it is
required that the blocking probability should be below 10 %. If
E1 links are used to carry the subscriber traffic to telephone
network, how many E1 links are needed?
- average call lasts 6 min
- a subscriber places one call during a 2-hour busy period (on the
average)
Amount of offered traffic is (2x1000x6 min /2x60 min) = 100
Erl.
Erlangs B formula gives N=97 for 10 % blocking and load of 100
Erl
=> required number of E1 links is ceil(97/30) = 4
(two incoming and two outgoing E1 links)

290

ceil(x) is the smallest integer greater than or equal to x.

Calculation

291

Switch structures

Page 42

Lecture notes

Switching Systems by Jorma Kekalainen

Traffic capability

292

Average utilization or efficiency (A/N) at

2% blocking probability

293

Switch structures

Page 43

Lecture notes

Switching Systems by Jorma Kekalainen

Blocking probabilities (%) calculated by

Erlangs B formula
CH ANN ELS
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25

A=0. 1

A=0.2

A=0.5

A=1

A=2

A=3

A=4

A=5

A=10

A=15

A=20

A=30

A=40

A=50

9,0909091
0,4524887
0,0150807
0,000377
7,54E-06
1,257E-07
1,795E-09
2,244E-11
2,493E-13
2,493E-15
2,267E-17
1,889E-19
1,453E-21
1,038E-23
6,919E-26
4,325E-28
2,544E-30
1,413E-32
7,438E-35
3,719E-37
1,771E-39
8,05E-42
3,5E-44
1,458E-46
5,833E-49

16,666667
1,6393443
0,1091703
0,0054582
0,0002183
7,278E-06
2,079E-07
5,198E-09
1,155E-10
2,31E-12
4,201E-14
7,001E-16
1,077E-17
1,539E-19
2,052E-21
2,564E-23
3,017E-25
3,352E-27
3,529E-29
3,529E-31
3,361E-33
3,055E-35
2,657E-37
2,214E-39
1,771E-41

33,333333
7,6923077
1,2658228
0,1579779
0,0157953
0,0013163
9,402E-05
5,876E-06
3,265E-07
1,632E-08
7,419E-10
3,091E-11
1,189E-12
4,246E-14
1,415E-15
4,423E-17
1,301E-18
3,614E-20
9,51E-22
2,378E-23
5,661E-25
1,287E-26
2,797E-28
5,827E-30
1,165E-31

50
20
6,25
1,5384615
0,3067485
0,0510986
0,0072993
0,0009124
0,0001014
1,014E-05
9,216E-07
7,68E-08
5,908E-09
4,22E-10
2,813E-11
1,758E-12
1,034E-13
5,746E-15
3,024E-16
1,512E-17
7,2E-19
3,273E-20
1,423E-21
5,929E-23
2,372E-24

66,666667
40
21,052632
9,5238095
3,6697248
1,2084592
0,344086
0,0859476
0,0190958
0,003819
0,0006944
0,0001157
1,78E-05
2,543E-06
3,391E-07
4,239E-08
4,987E-09
5,541E-10
5,833E-11
5,833E-12
5,555E-13
5,05E-14
4,391E-15
3,66E-16
2,928E-17

75
52,941176
34,615385
20,610687
11,005435
5,2157115
2,1864315
0,8132439
0,2703484
0,0810388
0,0220966
0,0055238
0,0012747
0,0002732
5,463E-05
1,024E-05
1,808E-06
3,013E-07
4,757E-08
7,135E-09
1,019E-09
1,39E-10
1,813E-11
2,266E-12
2,72E-13

80
61,538462
45,070423
31,067961
19,906687
11,716247
6,2748943
3,0420058
1,3339673
0,5307549
0,19263
0,0641688
0,0197403
0,0056398
0,0015039
0,000376
8,847E-05
1,966E-05
4,139E-06
8,277E-07
1,577E-07
2,867E-08
4,985E-09
8,309E-10
1,329E-10

83,333333
67,567568
52,966102
39,834289
28,486782
19,184726
12,051864
7,0047852
3,7457786
1,838457
0,8287368
0,3441188
0,1321784
0,0471843
0,0157256
0,004914
0,0014453
0,0004015
0,0001056
2,641E-05
6,289E-06
1,429E-06
3,107E-07
6,473E-08
1,295E-08

90,909091
81,967213
73,206442
64,666322
56,395218
48,45149
40,904078
33,831843
27,320794
21,458234
16,323233
11,973919
8,4338863
5,6819143
3,6496945
2,2301872
1,2948875
0,7142438
0,3745099
0,186905
0,0889232
0,0404033
0,0175636
0,0073176
0,002927

93,75
87,548638
81,403763
75,324733
69,322726
63,411084
57,605723
51,925557
46,39294
41,034054
35,879157
30,962563
26,322168
21,998293
18,03164
14,460212
11,315292
8,616888
6,3695015
4,5593216
3,153945
2,1051476
1,3543285
0,8393506
0,5010868

95,238095
90,497738
85,781686
81,093135
76,435799
71,813995
67,232758
62,697966
58,216481
53,796317
49,446818
45,178854
41,005014
36,939785
32,999693
29,203347
25,571358
22,126034
18,890791
15,889196
13,143603
10,673395
8,492963
6,6096717
5,0221778

96,774194
93,555094
90,343305
87,139501
83,944435
80,758951
77,583993
74,420626
71,270044
68,133597
65,012811
61,909409
58,825348
55,762844
52,724418
49,712932
46,731641
43,784243
40,874938
38,008488
35,190279
32,42638
29,723599
27,089523
24,532532

97,560976
95,124851
92,691829
90,262127
87,835986
85,413666
82,995456
80,581668
78,172646
75,768771
73,370459
70,978169
68,592409
66,213741
63,842785
61,480232
59,126846
56,783479
54,451079
52,1307
49,823522
47,530861
45,254186
42,995141
40,755565

98,039216
96,079939
94,122254
92,166252
90,212032
88,259699
86,309367
84,361158
82,415205
80,47165
78,530647
76,592365
74,656983
72,7247
70,795729
68,870302
66,948673
65,031118
63,117939
61,209465
59,306058
57,408112
55,516063
53,630385
51,751602

294

Blocking probabilities (%) calculated by

Erlangs B formula

295

Switch structures

Page 44

Lecture notes

Switching Systems by Jorma Kekalainen

Erlang B

296

297

Switch structures

Page 45

Lecture notes

Switching Systems by Jorma Kekalainen

298

Graphs for Erlangs blocking function

Horizontal axis: the offered traffic intensity A

The parameter of the family of curves: the size of
the system N
Vertical axis: blocking probability EB(N, A)
299

Switch structures

Page 46

Lecture notes

Switching Systems by Jorma Kekalainen

Required capacity as a function of the load

The following table gives the required number of trunks n as a function

of the offered traffic intensity A when the allowed blocking is 1 %.
The last column gives the required relative oversizing N/A, i.e. the ratio
of the number of trunks to the load.

When the traffic intensity a is large the Poisson fluctuations in the

occupancy are small in relative terms, and the required oversizing is
small
for Poisson distribution the standard deviation to mean ratio is
A/A = 1/A.
From the point of view of dimensioning the system it is very important
that the value of A on which the dimensioning is based has been
correctly estimated and the uncertainties in it have been properly
accounted for.
300

The calculated figures are shown in red

(B.H.T = Busy Hour Traffic = A
N=f(A,EB)

A=g(N,EB)

B.H.T.

Blocking

Lines

1.000

0.010

2.000

0.010

3.000

0.010

4.000

0.010

10

5.000

0.010

11

10.000

0.010

18

20.000

0.010

30

30.000

0.010

42

40.000

0.010

53

50.000

0.010

64

100.000

0.010

117

150.000

0.010

170

301

Switch structures

Page 47

Lecture notes

Switching Systems by Jorma Kekalainen

Erlang-B and Erlang-C Models

Model designed to predict blocking probability (Erlang-B) and
average call delay (Erlang-C) for a given number of channels and
traffic intensity
Valid for voice and traffic models conforming to the basic
assumption

Assumptions, Terminology and Parameters:

Channels Servers
Users Calls
Calls arrive according to a Poisson process with rate =
Call inter-arrival time is an exponentially distributed random variable,
with mean 1/.

Call duration is exponentially distributed random variable with mean

=1/
Traffic intensity or offered traffic, A = /

Note.
Note.

A[1-EB(N,A)] is referred to as the carried load

In this blocking model calls arriving while there are N
calls are blocked no buffering is employed.
302

Efficiency
Efficiency = A/N for a given blocking probability,
PN=EB(N,A)
Typical planning values for cellular and conventional
fixed networks are PN=EB(N,A) = 1% or 2%
Example: for PN=EB(N,A) = 1%

303

Switch structures

Page 48

Lecture notes

Switching Systems by Jorma Kekalainen

Example Erlang B
A cellular provider owns 100 cell sites each having 60 channels.
Assuming each user makes three calls per hour and the average
holding time per call is 1 minutes. Determine the total number of
subscribers that the service provider can support with a blocking
probability less than 2%.

Solution:

N=60/cell
EB(60,A)= 2%
A=49 Erl/cell
Asubs=/=(3/60)1=0.05 Erl/subs
Number of subs = total traffic/traffic per subs=49/0.05=980/cell
Total number of subs = 100 980 = 98000 subs
304

Erlang-C Model - Call Delay

The probability that an arriving call having to wait is
given by
AN
P(delay > 0) =
A N 1 Ak

N
A + N !1
N k = 0 k!
The average delay is given by
D = P(delay > 0 )

1
(N A)

The probability of the delay exceeding t time units is

given by
P (delay > t ) = P(delay > 0 ) exp[ ( N A)t ]
305

Switch structures

Page 49

Lecture notes

Switching Systems by Jorma Kekalainen

Example Erlang C
A company has a system with five private telephone lines
connecting two of its sites. The number of employees is 60
where on average each employee makes a three-minute
telephone call every hour.
a) What is the average delay for an employee to get access to
the telephone?
b) What is the probability of an employee waiting more than one
minute for the access to the telephone?
Solution:
a) N=5, A=60(1/60)*3=3 Erl, =1/3

306

Example
a) The average delay
D = P(delay > 0 )

1
( N A)

D0.35 min=21 s
b) The probability of the delay exceeding 1 min
is given by
P(delay > 1 min ) = P(delay > 0) exp[ ( N A) 1 min ]
P(delay > 1 min ) 0.12
Note. Blocking is

Switch structures

307

Page 50