Europar-08 Peer-to-Peer Evolutionary Algorithms

P2P EVOLUTIONARY ALGORITHMS: A SUITABLE APPROACH FOR

TACKLING LARGE INSTANCES IN HARD OPTIMIZATION PROBLEMS

J.L.J. Laredo1, A.E. Eiben2, M. van Steen2, P.A. Castillo1, A.M. Mora1, J.J. Merelo1

Dept. of Architecture and Computer Technology1

University of Granada

Dept. of Computer Science2

Vrije Universiteit Amsterdam

Scope

First steps to know if talking about P2P

technology applied to EA makes sense

MASSIVE SCALABILITY

Outline

Introduction

Context of the work: Our Model

Experimental Setup

Results

Conclusions

Introduction

Context of the work: Our Model

Experimental Setup

Results

Conclusions

EA

P2P Networks

P2P and dEA issues

Outline

Introduction

Context of the work: Our Model

Experimental Setup

Results

Conclusions

Overall Architecture

Evolvable Agent

Massive Parallelization

Outline

Outline

Introduction

Context of the work: Our Model

Experimental Setup

Results

Conclusions

Introduction

Context of the work: Our Model

Experimental Setup

Results

Conclusions

Outline

Introduction

Context of the work: Our Model

Experimental Setup

Results

Conclusions

Outline

Introduction

EA

P2P Networks

P2P and dEA issues

Bio-inspired optimization methods based on population

1 0 0 1 1 1 1 1 1 1

1 0 1 1 0 0 1 0 0 1

1 0 0 1 1 1 0 1 1 1

1 0 0 0 1 0 0 0 0 0

1 0 0 0 0 0 0 1 1 1

14)(xf

25)(xf

10)(xf

8)(xf

4)(xf

Introduction

EA

P2P Networks

P2P and dEA issues

Selection

Bio-inspired optimization methods based on population

1 0 0 1 1 1 1 1 1 1

1 0 1 1 0 0 1 0 0 1

1 0 0 1 1 1 0 1 1 1

1 0 0 0 1 0 0 0 0 0

1 0 0 0 0 0 0 1 1 1

14)(xf

25)(xf

10)(xf

8)(xf

4)(xf

1 0 0 1 1 1 1 1 1 1

Introduction

Selection

Recombination

Bio-inspired optimization methods based on population

1 0 1 1 0 0 1 0 0 1 14)(xf

25)(xf

1 0 1 1 0 1 1 1 1 1

EA

P2P Networks

P2P and dEA issues

30)(xf

1 0 1 1 0 1 0 1 1 1 29)(xf

Introduction

EA

P2P Networks

P2P and dEA issues

Overlay Network

Physical Network

Introduction

EA

P2P Networks

P2P and dEA issues

• Volunteer Computing

• Convergence P2P GRID

• Application Level Networks

•Dynamic neighborhood

(i.e. Small World)

EA

P2P Networks

P2P and dEA issues

Introduction

P2P = LARGE SCALE

EA Population Size Є (40, 400)

1000s peers are inhabited places

Introduction

EA Population Size Є (40, 400)

Wrong assumption!!!

• Population Size depends on the problem instance complexity

• Larger Instances require larger population sizes

• Practitioners tackle large instances with few individuals due to a

lack in resources

• We have many resources in P2P systems

EA

P2P Networks

P2P and dEA issues

Introduction

The Population Size of a P2P EA should be

determined following a scalability analysis

EA

P2P Networks

P2P and dEA issues

Overall Architecture

Evolvable Agent

Massive Parallelization

Context of the work: Our Model

Population of Evolvable Agents

Population Structure: Gossiping protocol newscast

Evolvable Agent

St Initialize

DO

Sols Selection

St +1 Recombination(Sols,Pc )

Evaluation (St +1)

If St +1 better than St

St St + 1

Overall Architecture

Evolvable Agent

Massive Parallelization

Context of the work: Our Model

Overall Architecture

Evolvable Agent

Massive Parallelization

• Cuadratic number of edges

• High clustering coefficient

• Lineal number of edges

• Low clustering coefficient

• Lineal number of edges

• High clustering coefficient

Context of the work: Our Model

Experimental Setup

Test-bed

PeerSim

Parameters

Population size Bisection, 51200

Crossover DPX, Pc = 1.0

Mutation No mutation

Selection Binary Tournament

PROBLEMS

Instances MMDP, wP-PEAKS, P-PEAKS

GA EvAg Newscast

0.98 of SR

MMDP

Results

WPPEAKS

Results

PPEAKS

Results

Conclusions

Agent-based approach for dEA on P2P

Algorithmically viable, 0.98 SR

Massively scalable

Extra peers don’t decrease the performance

Large instances, large number of resources

Future Works: DOP, Self-adaptive population,

Churn, real environment

THANK YOU!