A technique to reduce preemption overhead in real-time multiprocessor task scheduling
SCIE
SCOPUS
- Title
- A technique to reduce preemption overhead in real-time multiprocessor task scheduling
- Authors
- Jung, KJ; Park, C
- Date Issued
- 2005-01
- Publisher
- SPRINGER-VERLAG BERLIN
- Abstract
- Partitioning and global scheduling are two approaches for scheduling real-time tasks in multiprocessor environments. Partitioning is the more favored approach, although it is sub-optimal. This is mainly due to the fact that popular uniprocessor real-time scheduling algorithms, such as EDF and RM, can be applied to the partitioning approach with low scheduling overhead. In recent years, much research has been done on global real-time multiprocessor scheduling algorithms based on the concept of "proportionate fairness". Proportionate fair (Pfair) scheduling [5][6] is the only known optimal algorithm for scheduling real-time tasks on multiprocessor. However, frequent preemptions caused by the small quantum length for providing optimal scheduling in the Pfair scheduling make it impractical. Deadline Fair Scheduling (DFS) [1] based on Pfair scheduling tried to reduce preempt ion-related overhead by means of extending quantum length and sharing a quantum among tasks. But extending quantum length causes a mis-estimation problem for eligibility of tasks and a non-work-conserving problem. In this paper, we propose the Enhanced Deadline Fair Scheduling (EDFS) algorithm to reduce preemption-related overhead. We show that E-DFS allows us to decrease quantum length by reducing overhead and save wasted CPU time that is caused by preemption-related overhead and miss-estimation of eligibility.
- URI
- https://oasis.postech.ac.kr/handle/2014.oak/24242
- DOI
- 10.1007/11572961_46
- ISSN
- 0302-9743
- Article Type
- Article
- Citation
- LECTURE NOTES IN COMPUTER SCIENCE, vol. 3740, page. 566 - 579, 2005-01
- Files in This Item:
- There are no files associated with this item.
Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.