180 likes | 296 Views
Resource Discovery Using NeuroSearch Presentation for the Agora Center InBCT-seminar 13.11.2003. Mikko Vapa, researcher InBCT 3.2 Cheese Factory / P2P Communication Agora Center http:// tisu .it.jyu.fi/ cheesefactory. Resource Discovery Problem.
E N D
Resource Discovery Using NeuroSearchPresentation for the Agora Center InBCT-seminar 13.11.2003 Mikko Vapa, researcherInBCT 3.2 Cheese Factory / P2P Communication Agora Center http://tisu.it.jyu.fi/cheesefactory
Resource Discovery Problem • In peer-to-peer (P2P) resource discovery problem any node in the network can possess resources and also query these resources from other nodes Node1: Where is ? Node 2 Node 1 Node 4 Node 3
Resource Discovery Problem • The problem is multi-dimensionalbecause: • Location of resources can be arbitrary, it is not known beforehand and often changes constantly • Topology can be arbitrary and it might also change between each query • The querying node’s location in the network is arbitrary • Sufficient number of replies might depend on the queried resource • It is enough to locate one instance of a file or a processor, but the performance usually speeds linearly up when multiple instances are located • Some of the nodes might not behave as expected • For example some of the queries might be dropped by adversarial nodes requiring multiple query paths to be used • When the number of dimensions increase problem becomes complex
A Simple Solution for the Problem • Gnutella P2P network for example uses Breadth-First Search (BFS) flooding algorithm which sends query to all neighbors • Problems: all resources in the network can be found, but network gets congested and there are lots of useless packets Node 2: I have it! Node 4: Node 4 has it too! Reply Node 1: Where is ? Query Node 2 Query Query Node 1 Query Query Reply Node 4: I have it! Query Node 4 Node 3
Our solution: NeuroSearch • NeuroSearch resource discovery algorithm uses neural networks and evolution to adapt its’ behavior to given environment • neural network for deciding whether to pass the query further down the link or not • evolution for breeding and finding out the best neural network in a large class of local search algorithms Neighbor Node Forward the query Query Neighbor Node Forward the query
NeuroSearch’s Inputs • The internal structure of NeuroSearch algorithm • Multiple layers enable the algorithm to express non-linear behavior • With enough neurons the algorithm can universally approximate any decision function
NeuroSearch’s Training Program • The neural network weights define how neural network behaves so they must be adjusted to right values • This is done using iterative optimization process based on evolution and Gaussian mutation Define thenetwork conditions Iteratethousandsofgenerations Create candidate algorithmsrandomly Select the bestones for nextgeneration Breed a newpopulation Define the fitness requirementsfor the algorithm Finally select thebest algorithm forthese conditions Compare the bestone against other local search algorithms
Benefits • Universal: Whatever the peer-to-peer network conditions are a feasible solution algorithm can be found • Zero-configurable: There is no parameters that a designer would need to tune by hand • Supports various requirements: Designer can define what kind of an algorithm he/she wants to have • Rapid development: Even designing a simple algorithm for P2P network might take many months by human designer while with NeuroSearch this time is only couple of hours • Efficient: Multiple search strategies may in some cases be the only viable choice for example in mobile peer-to-peer networks where the environment is changing all the time
Well How Good Is The Algorithm? • We defined a peer-to-peer network scenario where: • 100 nodes form a power-law distributed P2Pnetwork having few hubs and lots of low-connectivity nodes • Resources are distributed based on the number of connections the node has meaning that high-connectivity nodes are more likely to answer to the queries • Topology is static meaning that nodes are not moving • Then we made a wish list for the algorithm and hoped that: • The algorithm should locate half of the available resources for every query • The algorithm should use as minimal number of packets as possible • The algorithm should always stop
Well How Good Is The Algorithm? • After a week we were ready to compare NeuroSearch’s invention against Breadth-First Search in 50-query test scenario • The measurements indicate that the optimization process can find an algorithm that: • finds half of the resources in the network with high probability • locates more resources than BFS with maximum number of 2 hops (BFS-2), but sill fewer than BFS-3 • consumes a bit more packets than BFS-2, but significantly less than BFS-3 • adapts to the peer-to-peer environment taking advantage of the environments resource distributions and topological features • Conclusion is that the approach is feasible, but not yet optimal
Future Work • Now the first version of NeuroSearch is ready and analyzed • The short-range future work of NeuroSearch includes (master’s thesis topics): • Comparison to other existing resource discovery algorithms • Reduction of resource discovery problem to route discovery problem and analysis of behavior in dynamic conditions • Analysis of the effects of increasing NeuroSearch’s brain power and peer-to-peer network size • New input types to feed NeuroSearch with more information • More neurons to allow NeuroSearch to make wiser decisions • Studying the scalability factors affecting NeuroSearch when the network size grows • Hybridization of evolutionary optimization method with local optimization method
Future Work • The long-range future work of NeuroSearch includes: • Parallelization of evolutionary optimization method for speeding up the convergence time (free master’s thesis topic) • Multicriteria objective functions for NeuroSearch for maximizing the lifetime of battery powered mobile P2P networks (free master’s thesis topic) • Development of new ad hoc protocol based on NeuroSearch (free dissertation topic) • Combination of topology management algorithm with NeuroSearch for finding optimal P2P network (dissertation topic) • Extension of NeuroSearch to ontology-based queries and reduction of query traffic using varying ontologies (dissertation topic) • Study of NeuroSearch’s performance under attack, random failure and deceptive scenarios (free dissertation topic) • Online adaptation of NeuroSearch in real P2P environment (free dissertation topic)
References • Vapa M., Kotilainen N., Kainulainen H., Vuori J., ”Resource Discovery in P2P Networks Using Evolutionary Neural Networks”, submitted to IEE Electronics Letters, June 2003 • Vapa M., Kotilainen N., Auvinen A., Töyrylä J., Hyytiälä H., Vuori J., ”NeuroSearch: evolutionary neural network resource discovery algorithm for peer-to-peer networks”, being submitted to Elsevier Science Ad Hoc Networks Journal, November 2003
Thank You! Any questions?