Performance for the masses

Experiments with A Web Based Architecture to Harness Volunteer Resources for Low Cost Distributed Evolutionary Computation

Merelo, Castillo, García-Sánchez, de las Cuevas, Rico, García-Valdez

Looking for a free high-performance infraestructure for evolutionary algorithms

It's a techno-social system

Security achieved through trust

Open source + open data

Open Science

But there are some challenges

Can we model the user behavior?

What's the relationship to the social network

Can we obtain better values than a local computer?

Initial experiments. Regular Ajax requests

Zipf at work here too

time vs rank
PUTs vs rank

Fits to a Generalized Extreme Values distribution

Working harder using web workers

PUTs vs rank

It's GEV all over again

PUTs vs rank
PUTs vs rank

Make it really tough

Keep them coming

Rastrigin, unique IP per hour vs rank

Fit to Weibull

Weibull

Volunteer EC is a thing

Leverage volunteer power

Keep volunteers engaged

Thanks

Follow us at @jjmerelo+ @geneura

Credits