Evopar12 Validating a P2P EA

1,012 views
926 views

Published on

Published in: Technology
0 Comments
0 Likes
Statistics
Notes
  • Be the first to comment

  • Be the first to like this

No Downloads
Views
Total views
1,012
On SlideShare
0
From Embeds
0
Number of Embeds
572
Actions
Shares
0
Downloads
4
Comments
0
Likes
0
Embeds 0
No embeds

No notes for slide

Evopar12 Validating a P2P EA

  1. 1. Validating a Peer-to-Peer Evolutionary AlgorithmJ.L.J. Laredo*, P. Bouvry, S. Mostaghim, J.J. Merelo EvoPAR 2012, Wednesday 11th April, 2012 Malaga, Spain
  2. 2. P2P Optimization What? Why?Parallelization of time-consumingmeta-heuristics in P2P systems: Branch&Bound Massive Scalability Shorten convergence PSO EA … time BUT!! So far… simulation based experimentation
  3. 3. Outline Introduction  The Curse of Dimensionality  Parallel vs. Sequential P2P EC Model  Viability Validation in a real-environment  Simulations  Algorithmic results  Massive scalability Conclusions
  4. 4. The Curse of DimensionalityIntroduction - P2P EC Model - Validation - Conclusions
  5. 5. The Curse of DimensionalityIntroduction - P2P EC Model - Validation - Conclusions
  6. 6. The Curse of DimensionalityIntroduction - P2P EC Model - Validation - Conclusions
  7. 7. Parallel vs. SequentialIntroduction - P2P EC Model - Validation - Conclusions
  8. 8. Parallel vs. SequentialIntroduction - P2P EC Model - Validation - Conclusions
  9. 9. Evolvable Agent Overlay Network  Agent-based approach  Fine-grain parallelization  Spatially structured EA  Local Selection Physical NetworkIntroduction - P2P EC Model - Validation - Conclusions
  10. 10. Viability Simulator J.L.J. Laredo, A.E. Eiben, M. van Steen, J. J. Merelo. EvAg: a scalable peer-to-peer evolutionary algorithm. Genetic Programming and Evolvable Machines , 11(2):227-246. 2010. Simulator conducted experimentation!!!Introduction - P2P EC Model - Validation - Conclusions
  11. 11. Simulations Simulator Instance P Avg. N. Max. N. Gen Gen. L=48 390 85 140 Simulator conducted L=150 3000 173 250 experimentation!!!Introduction - P2P EC Model - Validation - Conclusions
  12. 12. Algorithmic Results https://forja.rediris.es/svn/geneura/drmWrapper Instance P Avg. N. Max. N. Gen Gen.http://www.hlrs.de/systems/platforms/nec-nehalem-cluster L=48 390 85 140Introduction - P2P EC Model - Validation - Conclusions
  13. 13. Massive Scalability https://forja.rediris.es/svn/geneura/drmWrapperhttp://www.hlrs.de/systems/platforms/nec-nehalem-clusterIntroduction - P2P EC Model - Validation - Conclusions
  14. 14. Conclusions  1st large parallel P2P EA experiment  188 computers x 8 cores x 2 threads = 3008 agents  Seamless scalability  Conducted experimentation meets simulations Generations Pop. Size S.R. Simulator 85 390 0.98 Parallel 40-80 400 0.8 T seq  Massive scalability: Speed up T par 1000Introduction - P2P EC Model - Validation - Conclusions
  15. 15. Thanks for your attention!!

×