1. Energy-Optimal Software Partitioning in Heterogeneous
Multiprocessor Embedded Systems
Michel Goraczko, Jie Liu (Microsoft Research, Redmond)
Dimitrios Lymberopoulos (Yale University)
Slobodan Matic (UC Berkeley)
Bodhi Priyantha Feng Zhao (Microsoft Research, Redmond)
Presentation at DAC 2008, Anaheim, CA
June 10th, 2008

2. Energy Usage in Embedded Applications
Patient monitoring
Smart environments
Mobile devices
Low duty cycle monitoring for long battery life
High throughput for realtime critical events processing.

3. Energy Performance Diversity
A single processor with DVFS may not be flexible enough.
Energy efficiency in embedded processors
Non-trivial wake-up latency and energy costs

4. Heterogeneous Multi-Processor Platforms
UCLA LEAP Platform
MSR mPlatform

5. Outline Introduction Design Flow Power State Machine
ILP Formulation and Optimization
A Sound Source Localization Case Study

6. Software Partitioning Problem
Given a time sensitive application, allocate software components to different processors to minimize energy consumption without violating timing constraints.
Tasks
Processor
modes
Timing Analysis
Task timing
Partitioning
Application structure/ requirements
Power model
Task-Processor-Mode assignments

7. Power State Machines STBY Power: ~0 mW IDLE Power: 0.25mW
60MHz Power: 141 mW
30MHz Power: 72 mW
7.5MHz Power: 20 mW
negligible
1.53 mJ
24.5 ms
0.1 mJ
1.4 ms
1.47 mJ
23.8 ms

8. Software Model Directed acyclic graph of tasks
Single-rate periodic execution
Known release time
Known end-to-end deadline
Worst case execution time:
Pre-assignments

9. ILP: Variables and Objective
Core binary variables
task-to-processor assignment;
task-to-mode assignment;
task transition assignment;
Core integer variables
task start time instances;
Derived variables:
In order to convert the problem into ILP formulations, need to further introduce auxiliary variables.
Objective: minimize total energy per iteration

10. ILP: Constraints
A task can only be allocated to one processor and one mode;
A processor can only execute one task at any time;
Waking up from sleep modes takes time;
Processor total utilization should be less than 1;
Tasks have dependencies with in an iteration;
Tasks have dependencies across iteration boundaries;
No task can start before its release time;
All tasks should finish by the deadline;

11. Case Study Sound Source Localization FFT SC FFT SC HT FFT SC FFT SC
VOTE HT
FFT SC
FFT SC
S – Audio Sampling
FFT – Fast Fourier Transform
SC – Noise Estimation &
Signal classification
HT – Hypothesis Testing
VOTE – Sound detection voting

12. Hardware Model Power Mode ARM7 @ 2.5V 60MHz full speed MSP430 @ 3V
141
10.8
1/2 speed 72
2.7
1/8 speed 20
1.4
Idle
0.25 ~0
Standby
Wake up
Energy (mJ)
Time (ms)
To full speed
1.5
24.5 ~0
0.006
To 1/8 speed
0.1
1.4

13. Task Profiling Proc Mode FFT (ms) SC(ms) HT (ms) ARM7 60MHz 7.8 4.4
111
30MHz
15.6
9.0
222
7.5MHz
39.6
23.3
567
MSP430
6MHz
99.2
37.2
3MHz
196 76
1.5MHz
394
152
0.75MHz
792
300

14. Partitioning Results (1)
ARM7 60MHz HT 50
100
150
Deadline: 128ms
Need 4 MSP430
60MHz
Total energy/iteration: 21.7mJ
Average power: 166.7mW
MSP-4 6MHz
FFT SC 50
100
150
MSP-3
6MHz
FFT SC 50
100
150
MSP-2
6MHz
FFT SC 50
100
150
MSP-1
6MHz
FFT SC 50
100
150

15. Scheduling Results (2) Deadline: 256ms Need 2 MSP430 ARM7 @ 30MHz
Total energy/iteration: 22.1mJ
Average power: 86.4mW HT 50
100
150
200
256
MSP4
6MHz
FFT SC
FFT SC 50
100
150
200
256
MSP3
6MHz
FFT SC
FFT SC 50
100
150
200
256
MSP2
6MHz 50
100
150
200
256
MSP1
6MHz 50
100
150
200
256

16. Scheduling Results (3) Deadline: 1000ms Need 2 MSP430 ARM7 @ 7.5MHz
4xFFT HT
Deadline: 1000ms
Need 2 MSP430
7.5MHz
Total energy/iteration: 16.2mJ
Average power: 16.2mW
200
400
600
800
1000
MSP4
6MHz SC SC
200
400
600
800
1000
MSP3
6MHz SC SC
200
400
600
800
1000
MSP2
6MHz
200
400
600
800
1000
MSP1
6MHz
200
400
600
800
1000

17. Conclusion Processor diversities can help energy saving.
Wakeup time and energy must be considered in software partitioning.
Optimal software partitioning is NP–hard, but can be formulated as an ILP problem.

18. Limitations & Future Work
Execution time variations
Aperiodic tasks
Lightweight heuristics for online scheduling