Parallel evolutionary algorithms for scheduling on heterogeneous computing and grid environments
Supervisor(es): Cancela, Héctor - Alba, Enrique
Resumen:
This thesis studies the application of sequential and parallel evolutionary algorithms to the scheduling problem in heterogeneous computing and grid environments, a key problem when executing tasks in distributed computing systems. Since the 1990's, this class of systems has been increasingly employed to provide support for solving complex problems using high-performance computing techniques. The scheduling problem in heterogeneous computing systems is an NP-hard optimization problem, which has been tackled using several optimization methods in the past. Among many new techniques for optimization, evolutionary computing methods have been successfully applied to this class of problems. In this work, several evolutionary algorithms in their sequential and parallel variants are specically designed to provide accurate solutions for the problem, allowing to compute an eficient planning for heterogeneous computing and grid environments. New problem instances, far more complex than those existing in the related literature, are introduced in this thesis in order to study the scalability of the presented parallel evolutionary algorithms. In addition, a new parallel micro-CHC algorithm is developed, inspired by useful ideas from the multiobjective optimization field. Eficient numerical results of this algorithm are reported in the experimental analysis performed on both well-known problem instances and the large instances specially designed in this work. The comparative study including traditional methods and evolutionary algorithms shows that the new parallel micro-CHC is able to achieve a high problem solving eficacy, outperforming previous results already reported for the problem and also having a good scalability behavior when solving high dimension problem instances.
In addition, two variants of the scheduling problem in heterogeneous environments are also tackled, showing the versatility of the proposed approach using parallel evolutionary algorithms to deal with both dynamic and multi-objective scenarios.
Esta tesis estudia la aplicación de algoritmos evolutivos secuenciales y paralelos para el problema de planicación de tareas en entornos de cómputo heterogéneos y de computación grid. Desde la década de 1990, estos sistemas computacionales han sido utilizados con éxito para resolver problemas complejos utilizando técnicas de computación de alto desempeo. El problema de planificación de tareas en entornos heterogéneos es un problema de optimización NP-difícil que ha sido abordado utilizando diversas técnicas. Entre las técnicas emergentes para optimización combinatoria, los algoritmos evolutivos han sido aplicados con éxito a esta clase de problemas. En este trabajo, varios algoritmos evolutivos en sus versiones secuenciales y paralelas han sido especificamente diseados para alcanzar soluciones precisas para el problema de planicación de tareas en entornos de heterogéneos, permitiendo calcular planificaciones eficientes para entornos que modelan clusters de computadores y plataformas de computación grid. Nuevas instancias del problema, con una complejidad mucho mayor que las previamente existentes en la literatura relacionada, son presentadas en esta tesis con el objetivo de analizar la escalabilidad de los algoritmos evolutivos propuestos. Complementariamente, un nuevo método, el micro-CHC paralelo es desarrollado, inspirado en ideas ítiles provenientes del área de optimización multiobjetivo. Resultados numéricos precisos y eficientes se reportan en el análisis experimental realizado sobre instancias estándar del problema y sobre las nuevas instancias especificamente diseñadas en este trabajo.
El estudio comparativo que incluye a métodos tradicionales para planificación de tareas, los nuevos métodos propuestos y algoritmos evolutivos previamente aplicados al problema, demuestra que el nuevo micro-CHC paralelo es capaz de alcanzar altos valores de eficacia, superando a los mejores resultados previamente reportados en la literatura del área y mostrando un buen comportamiento de escalabilidad para resolver las instancias de gran dimensión. Además, dos variantes del problema de planificación de tareas en entornos heterogéneos han sido inicialmente estudiadas, comprobándose la versatilidad del enfoque propuesto para resolver las variantes dinámica y multiobjetivo del problema.
2010 | |
Parallel evolutionary algorithms Scheduling Heterogeneous computing Grid computing |
|
Inglés | |
Universidad de la República | |
COLIBRI | |
http://hdl.handle.net/20.500.12008/2958 | |
Acceso abierto | |
Licencia Creative Commons Atribución – No Comercial – Sin Derivadas (CC BY-NC-ND 4.0) |
Sumario: | This thesis studies the application of sequential and parallel evolutionary algorithms to the scheduling problem in heterogeneous computing and grid environments, a key problem when executing tasks in distributed computing systems. Since the 1990's, this class of systems has been increasingly employed to provide support for solving complex problems using high-performance computing techniques. The scheduling problem in heterogeneous computing systems is an NP-hard optimization problem, which has been tackled using several optimization methods in the past. Among many new techniques for optimization, evolutionary computing methods have been successfully applied to this class of problems. In this work, several evolutionary algorithms in their sequential and parallel variants are specically designed to provide accurate solutions for the problem, allowing to compute an eficient planning for heterogeneous computing and grid environments. New problem instances, far more complex than those existing in the related literature, are introduced in this thesis in order to study the scalability of the presented parallel evolutionary algorithms. In addition, a new parallel micro-CHC algorithm is developed, inspired by useful ideas from the multiobjective optimization field. Eficient numerical results of this algorithm are reported in the experimental analysis performed on both well-known problem instances and the large instances specially designed in this work. The comparative study including traditional methods and evolutionary algorithms shows that the new parallel micro-CHC is able to achieve a high problem solving eficacy, outperforming previous results already reported for the problem and also having a good scalability behavior when solving high dimension problem instances. |
---|