Download presentation
Presentation is loading. Please wait.
Published byRomeo Woosley Modified over 10 years ago
1
Vinay Devadas Hakan Aydin Professor : Chen, Ya-Shu Presenter : Ho, Shu-Wei
2
Outline Break-even time(B) Next Device Usage Time(NDUT(t)) Conservative Energy-Efficient Device Scheduling(CEEDS) Device Forbidden Region(DFR) Feasibility Dynamic Power Management(DPM) through Device Forbidden Regions Comparison Conclusion
3
Break-even time : The power consumption in active state : The power consumption in sleep state : The time overhead to perform a transition from active to sleep state : The time overhead to perform a transition from sleep to active state : The energy overhead to perform a transition from active to sleep state : The energy overhead to perform a transition from sleep to active state : The break-even time of the device
4
Next Device Usage Time(NDUT(t)) Prediction Current time(t) Tasks in ready queue Release time of jobs that request Current tNDUT(t) Task use
5
Conservative Energy-Efficient Device Scheduling Predicting NDUT(t) Example : = = 1000 = = 4000 = 990, = = 495 = 20, = = 10 T1 : (2000, 1000) ; T2 : (4000, 1000)
6
Conservative Energy-Efficient Device Scheduling
7
Device Forbidden Region(DFR) Rat-Monotonic scheduling(RMS). Duration Period Expected Energy Savings(EES) of
8
Feasibility Time demand function of Corollary : A set of periodic tasks can be feasibly scheduled by DFR-RMS if i t 0 t Ti,
9
DPM through Device Forbidden Regions
10
Comparison Always on (AON) --Where the devices remain in active state throughout the simulation (no dynamic power management). CEEDS --Adapted to RMS settings. This scheme performs device shutdown and activation based solely on next device usage time predictions. DER-RMS
11
Comparison
12
Conclusion Compare with original CEES, no matter with DVS or without DVS, RMS-DFR can save up to 27% energy consumption, under the worst condition, it can also save 20% consumption.
Similar presentations
© 2024 SlidePlayer.com Inc.
All rights reserved.