Amol Bakshi, Jingzhao Ou, and Viktor K. Prasanna

1
An Integrated Design Environment to Evaluate
Power/Performance Tradeoffs for Sensor Network
Applications

• Amol Bakshi, Jingzhao Ou, and Viktor K. Prasanna

Dept. of Electrical Engineering -
Systems University of Southern California
Los Angeles, CA
Project URL http//milan.usc.edu/
funded by the DARPA Power-aware Computing and
Communications program
2
MILAN A Model-based Integrated Simulation
Framework

• A unified environment capable of
• modeling a large class of embedded systems and
  applications
• driving design space exploration tools for rapid
  evaluation of a large design space
• seamlessly integrating different widely-used
  simulators into a single framework for
  hierarchical simulation
• enabling rapid evaluation of different
  performance metrics such as energy, latency, and
  throughput

• Use coarse system models based on key parameters
• Reduce initial design choices
• Use low-level simulators to analyze the reduced
  design options
• Choose one (or more) designs for implementation

3
Design Flow in MILAN
Application (Task Graph)
Hardware Resources
4
I. Energy-Efficient Design of Sensor Network
Applications

• A modeling and simulation environment for
  power-aware design of a multi-node sensor network
• Multi-granularity simulation
• Simulator integration
• Results from Wattch simulation are used to
  automatically configure ns-2 parameters
• Results from Wattch/ns-2 are used to
  automatically refine parameters for high-level
  estimator

5
II. Energy-Efficient Design of Kernel
Applications for FPGAs
EAT (million nJ?slices?cycles) EAT (million nJ?slices?cycles) EAT (million nJ?slices?cycles) EAT (million nJ?slices?cycles)
Matrix Size Xilinx Uni-proc. Linear Array
3 ? 3 0.2 0.1 0.1
6 ? 6 41.2 3.9 3.5
9 ? 9 469.5 39.2 32.5
15 ? 15 10063.3 839.77 580.0