Energy-Aware Scheduling of Flow Applications on Master-Worker Platforms

Abstract

In this report, we consider the problem of scheduling an application composed of independent tasks on a fully heterogeneous master-worker platform with communication costs. We introduce a bi-criteria approach aiming at maximizing the throughput of the application while minimizing the energy consumed by participating resources. Assuming arbitrary super-linear power consumption laws, we investigate different models for energy consumption, with and without start-up overheads. Building upon closed-form expressions for the uniprocessor case, we are able to derive optimal or asymptotically optimal solutions for both models. Key-words: Scheduling, energy, master-worker platforms, communicationThis text is also available as a research report of the Laboratoire de l'Informatique du Parallélisme http://www.ens-lyon.fr/LIP. inria-00335255, version 1 -29 Oct 2008Ordonnancementénergétiquement efficace d'un flot de tâches sur une plate-forme maître-esclaves Résumé : Dans ce rapport, nousétudions l'ordonnancement d'une application composée de tâches indépendantes qui doiventêtre exécutées sur une plate-forme maître-esclaves hétérogène où le coût des communications ne peutêtre négligé. Nous proposons une approche bi-critère visantà maximiser le débit de l'application tout en minimisant l'énergie dissipée par les ressources de calcul utilisées. En supposant que les lois de puissanceélectrique consommée sont super-liméaires, nous considérons différents modèles de consommationénergétique, avec ou sans coût de démarrage. A partir de formes clauses pour le cas avec un seul processeur nous construisons une solution asymptotiquement optimale pour les deux modèles.

Energy‐aware scheduling of bag‐of‐tasks applications on master–worker platforms

“…Note that a similar result is given in [15,5] for the converse problem, namely maximizing the throughout subject to a prescribed bound on power consumption. Thanks to Proposition 1, there is no need to specify the throughput for each frequency on any given processor.…”

“…This average throughput is achieved under S by only using P u,i0 continuously, whereas S uses modes of P (1) and P (2) . According to our previous work on the ideal model [15], we know that the power consumption of S is then smaller than one of S. Furthermore, S pays less overheads than S (only two, instead of k j2 − k j1 + 1 in S), and smaller overheads (thanks to the non-decreasing behavior of the overhead function). We repeat this pattern for all sets of modes of P (1) in S.…”

“…We provided an optimal solution to these problems in [15,5], using the sole assumption that the power consumption is super-linear. A key step is to establish closed-form formulas linking power consumption and throughput on a single processor: Proposition 1.…”

“…We would like to use both modes, according to the results of [15,5] for the ideal energy consumption model: running mode P u,i0 during…”

Multi-objective list scheduling of workflow applications in distributed computing infrastructures

“…We calculate the energy consumption of an activity A i on a resource R j by multiplying its makespan by the power consumption v 3j , modeled as in [17,18]:…”