Description
Date depot: 1 janvier 1900
Titre: Scheduling on new computing platforms
Directrice de thèse:
Safia KEDAD-SIDHOUM (CEDRIC)
Domaine scientifique: Sciences et technologies de l'information et de la communication
Thématique CNRS : Non defini
Resumé:
The applicative framework is the efficient management of resources
(processing, memory, network) on new large-scale computing platforms
composed of a huge number of hierarchical computing units (multi-core machines with GPU accelerators). These new features created new hopes for the users, but they also induce new problems. The existing software tools are not well-suited in practice. In particular, they target mostly the maximum completion time of a set of applications. We aim in this Ph-D to study other (unconventional) objectives like lateness or fairness.
Moreover, the always growing computing power of the new parallel
platforms created new objectives. Thus, we need to study this problem
from a multi-objective perspective for taking into account
simultaneously these contradictory attempts.
The proposed approach will consist in determining adequate execution
models, able to take into account the characteristics of the new
platforms. We are interested in designing scheduling algorithms
determined by exact methods for some specific applications and
approximation algorithms with performance guarantee in more general
settings.
Doctorant.e: Monna Florence