GOssiping Optimization Framework (GOOF): A decentralized P2P architecture for function optimization

This thesis discusses the implementation of function optimization algorithms through distributed and decentralized processing in a peer-to-peer fashion. Our research is focused on a fully decentralized, general purpose P2P environment, with no special or ad-hoc facility for executing optimization tasks. Relevant information is exchanged among nodes by means of epidemic protocols, exploiting the overlay network topology formed by peers. A key issue in such a context is the relationship between the solution quality and the amount/kind of exchanged information among the various running instances. We propose and detail novel heuristics and hyper-heuristics. Experimental results obtained both in simulated and real P2P environments are presented and discussed as well. Distributed optimization has a long and rich history, but little has been done to make it exploit the (potentially) large computing facilities a reliable P2P network can provide. We propose a novel framework that aims at easing the burden of performing function optimization tasks in a decentralized P2P network of solvers. Our †̃GOssiping Optimization Frameworkâ€TM (GOOF) bridges the gap between P2P services that can provide large reliable networks of interconnected nodes and the needs of optimization practitioners who are often not able to find a reasonably simple way to run their algorithms in such a distributed environment.

Grantor dc:publisher

Università degli studi di Trento

Year dc:date

2010

Author dc:creator

• Biazzini, Marco

Contributors dc:contributor

• Montresor, Alberto

Statement dc:rights

• info:eu-repo/semantics/openAccess
• license:Tutti i diritti riservati (All rights reserved)

Language dc:language

eng

Identifier

http://dx.doi.org/10.15168/11572_367825
10.15168/11572_367825

OAI identifier oai:identifier

oai:iris.unitn.it:11572/367825