Issuu on Google+

Using pool-based evolutionary algorithms for scalable and asynchronous distributed computing

J. J. Merelo, A. M. Mora, C. M. Fernandes, M. G. Arenas, Anna I. Esparcia-Alcรกzar U. Granada + S2 Grupo http://geneura.wordpress.com


What are the ingredients for a massively parallel evolutionary algorithm?

#sofea via @jjmerelo @aiesparcia

2


How can you use a server/backoffice that does (almost) all the work?

#sofea via @jjmerelo @aiesparcia

3


Can you achieve fault-tolerance and asynchrony?

#sofea via @jjmerelo @aiesparcia

4


Pool CRUD Pools allow create, read, update and delete operations.

Fault-tolerant

Pool == conveyor belt

Pool == population

#sofea via @jjmerelo @aiesparcia

Scalable

5


Implementation matters Object stores: RDMBS or

Which frameworks can be used to implement pools?

NoSQL

- Latency + Concurrency

- Lack of control + Locality of writes

File synchronization systems #sofea via @jjmerelo @aiesparcia

6


Conclusions ●Tradeoff fault-tolerance/scalability. ●Difficulty of non-centralized non-asynchronous operation. ●Advantages: ● Availability of frameworks. ● Optimal CRUD operations. ● Availability of clients. ● Choice of languages.

●Paradigm mix! #sofea via @jjmerelo @aiesparcia

7


Thanks!

Any question? See you at EvoPar 2013 http://goo.gl/LtTCL @geneura

#sofea via @jjmerelo @aiesparcia

8


Using pool-based evolutionary algorithms forscalable and asynchronous distributed computing