Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

eScope: A Fine-Grained Power Prediction Mechanism for Mobile Applications

Dipayan Mukherjee, Atul Sandur, Kirill Mechitov, Pratik Lahiri, Gul Agha

TL;DR

eScope tackles the challenge of estimating energy use for mobile video analytics without hardware-specific power models or instrumented code. It collects device-level energy data and concurrent operator traces to train a run-time power predictor that maps operator execution times to power costs, yielding device-specific models deployed on the device. Across synthetic and real workloads, eScope achieves about 97% drain-interval prediction accuracy with under 3% overhead, rivaling or exceeding prior approaches while avoiding code instrumentation. The approach enables dynamic, input-aware energy management and can integrate with energy-aware placement and remote-offloading strategies, offering a practical path to extending mobile battery life in resource-intensive applications.

Abstract

Managing the limited energy on mobile platforms executing long-running, resource intensive streaming applications requires adapting an application's operators in response to their power consumption. For example, the frame refresh rate may be reduced if the rendering operation is consuming too much power. Currently, predicting an application's power consumption requires (1) building a device-specific power model for each hardware component, and (2) analyzing the application's code. This approach can be complicated and error-prone given the complexity of an application's logic and the hardware platforms with heterogeneous components that it may execute on. We propose eScope, an alternative method to directly estimate power consumption by each operator in an application. Specifically, eScope correlates an application's execution traces with its device-level energy draw. We implement eScope as a tool for Android platforms and evaluate it using workloads on several synthetic applications as well as two video stream analytics applications. Our evaluation suggests that eScope predicts an application's power use with 97% or better accuracy while incurring a compute time overhead of less than 3%.

eScope: A Fine-Grained Power Prediction Mechanism for Mobile Applications

TL;DR

eScope tackles the challenge of estimating energy use for mobile video analytics without hardware-specific power models or instrumented code. It collects device-level energy data and concurrent operator traces to train a run-time power predictor that maps operator execution times to power costs, yielding device-specific models deployed on the device. Across synthetic and real workloads, eScope achieves about 97% drain-interval prediction accuracy with under 3% overhead, rivaling or exceeding prior approaches while avoiding code instrumentation. The approach enables dynamic, input-aware energy management and can integrate with energy-aware placement and remote-offloading strategies, offering a practical path to extending mobile battery life in resource-intensive applications.

Abstract

Managing the limited energy on mobile platforms executing long-running, resource intensive streaming applications requires adapting an application's operators in response to their power consumption. For example, the frame refresh rate may be reduced if the rendering operation is consuming too much power. Currently, predicting an application's power consumption requires (1) building a device-specific power model for each hardware component, and (2) analyzing the application's code. This approach can be complicated and error-prone given the complexity of an application's logic and the hardware platforms with heterogeneous components that it may execute on. We propose eScope, an alternative method to directly estimate power consumption by each operator in an application. Specifically, eScope correlates an application's execution traces with its device-level energy draw. We implement eScope as a tool for Android platforms and evaluate it using workloads on several synthetic applications as well as two video stream analytics applications. Our evaluation suggests that eScope predicts an application's power use with 97% or better accuracy while incurring a compute time overhead of less than 3%.
Paper Structure (26 sections, 4 equations, 17 figures, 1 table)

This paper contains 26 sections, 4 equations, 17 figures, 1 table.

Table of Contents

  1. Introduction
  2. Related Work
  3. Motivating scenarios
  4. Challenges
  5. System Model & Problem Definition
  6. Proposed Approach
  7. Execution segment
  8. Power attribution
  9. Power prediction model
  10. Implementation
  11. Evaluation
  12. Experimental setup
  13. Feasibility of eScope for mobile devices
  14. Accuracy of power measurements
  15. Accuracy of eScope
  16. ...and 11 more sections

Figures (17)

  • Figure 1: An application can be divided into operators that are called in some sequence depending on the input configuration. Operators in a face recognition application.
  • Figure 2: Variation in power consumption of a fixed workload on two different phones
  • Figure 3: Input sensitivity to the frame rate on the power draw of a video analytics application
  • Figure 4: A snapshot of execution segments and discharge intervals for operators executing on the mobile device.
  • Figure 5: Sample training data for eScope from a video conferencing application, with the operator execution time and battery discharge interval being the independent and dependent variables respectively, from Eq \ref{['eq:final-energy-eq-escope']}.
  • ...and 12 more figures