- Full title Arquitectura de agentes autónomos para el modelado de usuarios y la gestión óptima de un centro de supercomputación
- Funding entity Xunta de Galicia
- Researchers Francisco Javier Bellas Bouza, Richard J Duro Fernández, José Antonio Becerra Permuy, Fernando López Peña, Santiago Vázquez Rodríguez, Alejandro García del Valle, Maria Luisa Sánchez Simon, Pablo Fariñas Alvariño, Roberto José Perez Serantes, Jesús Angel Gomollón García, José Santos Reyes, Francisco Javier Perez Castelo, Adolfo Lamas Rodríguez, Francisco Carlos Oliver Charlón, Juan Luis Crespo Mariño, Arturo Nieto de Almeida
- Start date 20/08/2003
- End date 30/09/2006
- Centro de Supercomputación de Galicia
- Universidade da Coruña
Description
The way a scheduler for a process queue manager works is determined by numerous parameters. To work properly and meet expectations, it has to be adjusted by a high-performance computing centre. That is a complex process that does not usually take into account crucial issues. Many of these parameters are based on subjective criteria about the way the server is perceived and vary from one user to another. Thus, the objectives achieved in this project are to:
- Model users through intelligent autonomous agents that adapt to changing user behaviour
- Characterise user satisfaction with the service
- Schedule processes with evolutionary techniques that consider various aspects and optimise parameters according to the centre´s decision, including average satisfaction, waiting times and makespan.
Thus, we have obtained an integrated JMS Sun Grid Engine scheduler that dynamically adapts in real time to unstable user behaviour. What is more, it is capable of considering trivial and subjective aspects, like user satisfaction. More importantly, the centre can change its behaviour in a simpler and more natural way than ever before.