Network Dynamics in Bandwidth Provisioning

1
Network Dynamics in Bandwidth Provisioning

• A proof of concept
• Sébastien Rumley

2
Simple problem
A
B

• A has a parcel for B

The operator also delivers parcels to C,
DZ Questions - how many cars does the
operator need ? - how big should be the cars
? Problem gets complicated !!!
3
Simple problem
A
B

• A has a parcel for B

The operator also delivers parcels to C,
DZ Questions - how many cars does the
operator need ? - how big should be the cars
? Problem gets complicated !!!
4
Complicated problem
E, H
C, D
The operator has a network of customers (both
sending and receiving parcels) connected with
roads, but
4
2
A
B
5
3
1

• Many itineraries exist between 1 (where A is) and
  5. Which one choose?
• Shorter distance?
• Shorter time?
• Traffic jams ?

Should the driver pick up parcels from 2 to 3
and 2 to 5 when arriving in 2? Car has to leave
the highway Operator needs a an operation
centre in 2 Parcels have to be ordered in the
car, and labelled Car is almost empty between 1
and 2, as place should be available arriving in
2 Problem gets even more complicated
5
Complicated problem
E, H
C, D
The operator has a network of customers (both
sending and receiving parcels) connected with
roads, but
4
2
A
B
5
3
1

• Many itineraries exist between 1 (where A is) and
  5. Which one choose?
• Shorter distance?
• Shorter time?
• Traffic jams ?

Should the driver pick up parcels from 2 to 3
and 2 to 5 when arriving in 2? Car has to leave
the highway Operator needs a an operation
centre in 2 Parcels have to be ordered in the
car, and labelled Car is almost empty between 1
and 2 Problem gets even more complicated
6
Very complicated problem
E, H
C, D
4
Operator has so much parcels to transfer from 2
to 4 that cars are not big enough
2
A
B
5
3
1

• The operator can buy vans or trucks, or even
  planes
• High costs (both to buy and to maintain)
• And if the number of parcel diminishes?
• If a buy a plane, can I use it tomorrow?

The operator deals with a truck company, which
will transport all parcels between 2 and 4. The
operator has to pay for it The operator has to
trust the truck company The truck has to
transport parcel the same day
The truck company has other customers Questions
- how many trucks? - how big should be the
trucks ? Déjà vu !!
7
Very complicated problem
E, H
C, D
4
Operator has so much parcels to transfer from 2
to 4 that cars are not big enough
2
A
B
5
3
1

• The operator can buy vans or trucks, or even
  planes
• High costs (both to buy and to maintain)
• And if the number of parcel diminishes ?
• If a buy a plane, can I use it tomorrow?

The operator deals with a truck company, which
will transport all parcels between 2 and 4. The
operator has to pay for it The operator has to
trust the truck company The truck has to
transport parcel the same day
The truck company has other customers Questions
- how many trucks? - how big should be the
trucks ? Déjà vu !!
8
More complexity?
And if A has no more parcels to send B?

• What is the influence over traffic of parcels
  between 2 and 4?
• Should the truck be sold?
• Will the node 1 be further included inside the
  operator network?
• Should the contract with truck company be broken?
• And if truck company has a contract with a plane
  company?

And if operator changes its strategy? We have to
drive only filled car!, says the director, even
if this means denial of service in several cases.
9
More complexity?
And if A has no more parcels to send B?

• What is the influence over traffic of parcels
  between 2 and 4?
• Should the truck be sold?
• Will the node 1 be further included inside the
  operator network?
• Should the contract with truck company be broken?
• And if truck company has a contract with a plane
  company?

And if operator changes its strategy? We have to
drive only filled car!, says the director, even
if this means denial of service in several cases.
10
In communication networks

• If a car is an ADSL connection (1 Mbit/s)
• ? A single channel on an optical fibre is jumbo
  ferry carrying several thousands of cars (1
  to 40 Gbit/s)
• One fibre has several dozens of channels
  (Tbit/s)
• One cable may have hundreds of fibres (Pbit/s)
• ? One cable can transport billions of ADSL
  connections ( cars)
• With (upcoming) optical switching, an optical
  channel can be switched in a very short time
  (ms)
• ? As a plane in parcel analogy
• ? An optical channel operator can thus react
  quickly to changes in the demand.
• ? Allows dynamic bandwidth provisioning from 1
  Mbit/s to 40 Gbit/s

11
Realisation - the model

• The model is made of 3 layers

? User layer (customers) Models user to user
connection (A to B)
? Intermediate layer (operator) Models
operators centre to centre connections
? Physical layer (roads) Cities to cities
connections
12
Generation of demands
At each time step, two customers are picked at
random, and

• If the corresponding connection if OFF, it is
  switched ON
• ? Operator has to furnish the connection
• ? It has to add new cars and new route if needed
• This can change in turn the amount on traffic on
  roads
• and may be traffic jams

- or -
If ON, it is switched OFF ? Operator has to
recycle the vehicles
13
State annotation
After each time step

• The configuration of each layer is annotated and
  listed

With this list of configurations (states), we
analyse the influence of operator decisions
? One the roads occupation ? On the number
and types of connections the operator as to
setup ? On the quality of service offered to the
customers (blocking prob.) As the decisions are
taken online ? Two sequence of decisions can
lead to very different results and
situation ? How different are these
situations? ? Does situation exists where it is
important to take the right decision?
14
Simulations
Two topologies have been used
Centric topology 8 nodes10 linkscapacity of 24
units on links
Bi-polar topology 11 nodes13 linkscapacity of 6
units on linksexcepted on central link (24)
15
Results (centric)
Four different strategies experiment executed
twice (6000 iterations) Data extracted Number of
different configurations experienced by each
layerNumber of times the most experienced
configuration arisedAverage utilisation of the
links of the layer
16
Most visited state (centric)
Configuration on the physical layer mostly used
Degree repartition of theof the physical states
17
Results (bi-polar)
18
Most visited state (bi-polar)
Configuration on the physical layer mostly used
19
Comparisons between strategies
Centric Bi-polar
20
Conclusions
A dynamical networks approach for multilayer
communication networks analysis seems
interesting, particularly if layers are
configured dynamically. It offers a good
complement to classic teletraffic theories and
Erlang laws, which struggle to deal with multiple
layers. It offers also a good alternative to
classic traffic simulator (which consider only
one dynamic layer at the same time). Model
realised is absolutely not complex enough to
return not already known results. 6000 iterations
to obtain the proof that our routing algorithm
works