BiTEM: A Two-Tier Energy Efficient Resource Management Framework for Real-Time Tasks in Clusters

• Published in: Algorithms and Architectures for Parallel Processing pp. 494 - 508
• Main Authors: Huang, Wei, Shi, Jin, Wang, Zhen, Qian, Zhuzhong
• Format: Book Chapter
• Language: English
• Published: Cham Springer International Publishing 2015
• Series: Lecture Notes in Computer Science
• Subjects: Datacenter; DVFS; Energy aware; Energy prediction; Task scheduling
• ISBN: 9783319271217; 3319271210
• ISSN: 0302-9743; 1611-3349
• DOI: 10.1007/978-3-319-27122-4_34

Energy saving is a fundamental issue for clusters because huge energy consumption has profound impact on operating cost, system reliability and environment. Therefore, many techniques have been proposed to reduce energy consumption, among which, Dynamic Frequency and Voltage Scaling (DVFS) is recognized as an efficient technique. To save energy for DVFS-enabled clusters where independent real-time tasks are executed, we propose BiTEM which is a cooperative two-tier energy efficient management method including local DVFS control and global task scheduling. By using this method, the DVFS controller adjusts the frequencies of homogenous processors in each server at runtime based on the practical energy prediction. On the other hand, global scheduler assigns incoming tasks onto suitable processors on the designate servers based on the cooperation with the local DVFS controller. Each local DVFS controller responses minimum energy changes to the global scheduler to assist it in assigning tasks. The final evaluation results demonstrate the effectiveness of BiTEM on energy saving.