Asymmetry-Aware Scheduling in Heterogeneous Multi-core Architectures

Abstract : As threads of execution in a multi-programmed computing environment have different characteristics and hardware resource requirements, heterogeneous multi-core processors can achieve higher performance as well as power efficiency than homogeneous multi-core processors. To fully tap into that potential, OS schedulers need to be heterogeneity-aware, so they can match threads to cores according to characteristics of both. We propose two heterogeneity-aware thread schedulers, PBS and LCSS. PBS makes scheduling based on applications’ sensitivity on large cores, and assigns large cores to applications that can achieve better performance gains. LCSS balances the large core resource among all applications. We have implemented these two schedulers in Linux and evaluated their performance with the PARSEC benchmark on different heterogeneous architectures. Overall, PBS outperforms Linux scheduler by 13.3% on average and up to 18%. LCSS achieves a speedup of 5.3% on average and up to 6% over Linux scheduler. Besides, PBS brings good performance with both asymmetric and symmetric workloads, while LCSS is more suitable for scheduling symmetric workloads. In summary, PBS and LCSS provide repeatability of performance measurement and better performance than the Linux OS scheduler.
Type de document :
Communication dans un congrès
Ching-Hsien Hsu; Xiaoming Li; Xuanhua Shi; Ran Zheng. 10th International Conference on Network and Parallel Computing (NPC), Sep 2013, Guiyang, China. Springer, Lecture Notes in Computer Science, LNCS-8147, pp.257-268, 2013, Network and Parallel Computing. 〈10.1007/978-3-642-40820-5_22〉
Liste complète des métadonnées

Littérature citée [19 références]  Voir  Masquer  Télécharger

https://hal.inria.fr/hal-01513772
Contributeur : Hal Ifip <>
Soumis le : mardi 25 avril 2017 - 14:33:38
Dernière modification le : mardi 25 juillet 2017 - 14:33:44
Document(s) archivé(s) le : mercredi 26 juillet 2017 - 14:08:58

Fichier

978-3-642-40820-5_22_Chapter.p...
Fichiers produits par l'(les) auteur(s)

Licence


Distributed under a Creative Commons Paternité 4.0 International License

Identifiants

Citation

Tao Zhang, Xiaohui Pan, Wei Shu, Min-You Wu. Asymmetry-Aware Scheduling in Heterogeneous Multi-core Architectures. Ching-Hsien Hsu; Xiaoming Li; Xuanhua Shi; Ran Zheng. 10th International Conference on Network and Parallel Computing (NPC), Sep 2013, Guiyang, China. Springer, Lecture Notes in Computer Science, LNCS-8147, pp.257-268, 2013, Network and Parallel Computing. 〈10.1007/978-3-642-40820-5_22〉. 〈hal-01513772〉

Partager

Métriques

Consultations de la notice

43

Téléchargements de fichiers

19