A Genetic Algorithm For Solving The First Price Sealed Bid Auction in Communication Networks

You might also like

Download as ppt, pdf, or txt
Download as ppt, pdf, or txt
You are on page 1of 10

A Genetic Algorithm for Solving the

First Price Sealed Bid Auction in


Communication Networks

Portilla-Figueras, S. Salcedo-Sanz, P. García-Díaz,


K. Hackbarth
Teoría de la Señal y Comunicaciones, Universidad de Alcalá
Ingeniería de Comunicaciones, Universidad de Cantabria

1
Index

 Intra-domain and Inter-domain Routing Problem


 Genetic Algorithms
 First Implementation
 Experiments and Results
 Conclusions

2
Routing Problem

 Two points of view:

 User-Services
 Internet traffic is growing at a very high rate.
 New broadband user, not only business but also residential
 New multimedia services “real time” demanding guaranteed QoS.

 Network operators
 Core network with enough capacity, but not the same in access.
 Free capacities that can be sold to ISP, Virtual Operatos or even
other network operators
 Interconnection problem to provide QoS Services.

3
Interdomain vs Intradomain
 Intradomain:  Interdomain
 We consider a single  We have to consider network of
administrative entity which different operators.
owns the network.  We have to establish
 It has to guarantee the QoS to agreements to meet the QoS
their customers, but obtaining requirements
the maximum benefit  Centralized or distributed
 Link to Route Scale approach.

4
Why Auctions?

 There is a current trend which applies economical concepts to


network design and optimization.
 It is based in game theory including auctions.
 The most suitable auction procedure for telecommunication
networks is based on Vickrey auctions:
 Incentive compatibility.
 Individual rationality.
 Efficiency.
 Even for very simple scenarios the problems gets NP-complex.

5
Genetic Algorithms

 Genetic algorithm are a family of computational models inspired by the


evolution.
 The “first” implementations was performed by J. Holland in 1975 which
is known as the Canonical Genetic Algorithm.
 Their encode a potential solution to a specific problem on a simple
“chromosome” data structure and applies some procedures (crossover,
mutation, elitism) to make it evolve.
 They can be seen as population based optimizers using a
corresponding fitness function which evaluates the goodness of the
solution.
 Why does it work?
 They consider the solution search space as hypercubes and the
chromosomes are hyperplane partitions of the hypercube
 Each time a “chromosome” is evaluated many hypercubes are evaluated in
an implicit parallel fashion.
 The cumulative effects of the hyperplane competitions (using the fitness
function) makes the algorithm converge towards the final solution.

6
Genetic Algorithms
 How it works?
 Generate a initial population, usually randomly.
 Evaluate the fitness of each individual of the population.
 Selection is applied to generate an intermediate population. Better
solutions have more possibilities or being selected to the next generation.
 Apply recombination (crossover) to generate the new population
 We also apply mutation to avoid fall in local minimums/maximums
 Evaluate the new fitnesses
 Repeat the process for all the generations.

7
Application of Genetic Algorithm
 Problem Codification  Utility Function
U i  a, c i    i  a   c i
If ci=0, we have the
Knapsack Problem

 Inelastic Demand Valuation Function

N i(ai)=iai<i.
 Fitness function, Maximize x p
i 1
i i

 Constraint, the capacity of the link, Q


N

x q
i 1
i i Q

8
Experiments & Results
 First Example (Only to see if it works)

 Second, a “real” case


Evolution of the Social Welfare

100
Benefit 91 80

Social Welfare
60
40
20
0
1 11 21 31 41 51 61 71 81 91
Generations

9
Conclusions

 Genetic algorithms have shown their capabilities to


solve auction based resource assignment problems
 Ok, we know that we are using too much algorithmic
power to such an easy problem.
 The work here is the starting point to tackle more
complex problems, specifically:
 Different types of user demands
 Different pricing schemes, (Vickrey)
 Defining a QoS route between two edge nodes.

10

You might also like