Swiss FederalInstitute of Technology

Computer Engineeringand Networks Laboratory

Power Management for Solar-Driven Sensor Nodes

Clemens Moser

( joint work with D. Brunelli, L. Thiele and L. Benini )

2Computer Engineering and Networks Laboratory TIK Clemens Moser

Outline

System Model Problem Statement Lazy Scheduling Admittance Test Simulation Conclusion

[1] C. Moser, D. Brunelli, L. Thiele, and L. Benini, “Real-time scheduling with regenerative energy.”, In The 18th Euromicro Conference on Real-Time Systems (ECRTS 2006), July 2006.[2] C. Moser, D. Brunelli, L. Thiele, and L. Benini, “Lazy scheduling for energy harvesting sensor nodes.”, In The Fifth IFIP Conference on Distributed and Parallel Embedded Systems (DIPES 2006), October 2006.

3Computer Engineering and Networks Laboratory TIK Clemens Moser

System Model

energy source

energy storage

computing device

tasks

PS(t)

PD(t)

a1, e1, d1 a2, e2, d2

DD

J1J1 J2

J2

SS

EC(t) ≤ CEC(t) ≤ C

4Computer Engineering and Networks Laboratory TIK Clemens Moser

System Model

Task Ji

can be preempted arrives at time ai

has deadline di

needs total energy ei to complete

can consume power

therefore, needs time

energy source

energy storage

computing device

tasks

PS(t)

PD(t)

a1, e1, d1 a2, e2, d2

DD

J1J1 J2

J2

SS

EC(t) ≤ CEC(t) ≤ C

5Computer Engineering and Networks Laboratory TIK Clemens Moser

Problem Statement Determine an optimal on-line scheduling

algorithm:If the task set is schedulable, it determines a feasible schedule.

Construct an admittance test:Determine, whether a set of event streams with a given characteristic is schedulable.

Nothing known so far … Nothing known so far …

6Computer Engineering and Networks Laboratory TIK Clemens Moser

Problem Statement - EDF

21

Greedy scheduling is not suited.

7Computer Engineering and Networks Laboratory TIK Clemens Moser

Problem Statement - ALAP

1 2

ALAP does not work either. And what happens if the energy storage is full?

8Computer Engineering and Networks Laboratory TIK Clemens Moser

Lazy Scheduling Algorithm

Rule 1:

Rule 2:

9Computer Engineering and Networks Laboratory TIK Clemens Moser

Optimality of Lazy Scheduling Algorithm

Theorem:If the Lazy Scheduling Algorithm LSA cannot schedule a given set of tasks, then no other scheduling algorithm can schedule it.

10Computer Engineering and Networks Laboratory TIK Clemens Moser

Sketch of Proof

Energy-Constrained

Time-Constrained

E

t

C

E

t

C

11Computer Engineering and Networks Laboratory TIK Clemens Moser

Admittance Test

Is the schedulingof the event streams feasiblewith LSA ?

Abstraction

Event stream: delay requirement energy request per eventarrival curve

Energy source: energy variability

12Computer Engineering and Networks Laboratory TIK Clemens Moser

Proof: The proof uses concepts of network calculus[Le Boudec, Thiran] and real-time calculus.

Admittance Test

Pmax

13Computer Engineering and Networks Laboratory TIK Clemens Moser

Simulation Results

Capacity savings of ~40% measured for random task sets for LSA with εl(Δ) compared to EDF

LSA with ES

LSA with εl

EDF

LSA with εu

14Computer Engineering and Networks Laboratory TIK Clemens Moser

1 2

Conclusions

SS

CC

DD

J1J1 J2

J2…

PS(t)

PD(t)

a1, e1, d1 a2, e2, d2

Scheduling Scenario

Optimal Lazy Scheduling

Admittance Test

Simulation Results

EDF LSA with εl

15Computer Engineering and Networks Laboratory TIK Clemens Moser

The Big Picture Modular Real-Time Analysis

16Computer Engineering and Networks Laboratory TIK Clemens Moser

The Big Picture Modular Real-Time Analysis