Skip to content (access key 's')
Logo of Technion
Logo of CS Department
Logo of CS4People

The Taub Faculty of Computer Science Events and Talks

CSL Lucnheon: Real-Time Systems Scheduling for Energy Optimization
event speaker icon
Gil Kedar
event date icon
Thursday, 20.12.2012, 12:30
event location icon
Classroom 9 Taub Bld.
Advisor-Avi Mendelson.
Embedded real-time systems are continuously required to increase performance and to reduce power and energy consumption. Higher performance is needed in order to meet new demand of emerging applications. In order to meet such performance demand, modern processors implement complex enhancement mechanisms such as deeper pipeline, vector units, multi-cores and more. In real-time systems, energy consumption can be saved as long as the system guarantees meeting the tasks' real-time constraints. This energy concern becomes a major key factor in designing modern real-time systems. In order to address this issue, advanced modern processors commonly support low-power mechanism, such as dynamic voltage and frequency scaling (DVFS), to reduce the core dynamic and static power. However, as this research will show, using only the hardware power saving features without considering the cooperation between the hardware and the software is sub-optimal. The research we conduct so far was based on a main new theme that optimizing the energy consumption for real-time systems should be based on software / hardware co-designs. Common energy minimization algorithms consider only the thread's worst-case execution time[1] (WCET). We claim that by estimating thread's WCET for various core architectures and considering the threads interactions achieves a lower energy consumption. The research proposes a software / hardware co-design approach for improving the power management of real-time systems. The new innovative energy saving mechanisms, will consider both software related parameters, such as the tasks’ execution characterizations; e.g., miss ratio, IPC, as well as hardware related characterizations, such as the structure of the memory hierarchy and the relative energy consumption of different subsystems, etc. We show that system’s scheduler can efficiently use these parameters to achieve lower system energy consum