Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Optimal energy-aware task scheduling for batteryless IoT devices

Carmen Delgado, Jeroen Famaey

TL;DR

The paper addresses the challenge of intermittent operation in batteryless IoT devices by formulating an optimal energy-aware task scheduler as a MILP that accounts for energy harvested, capacitor state, and per-task energy consumption and priorities. It decomposes applications into atomic subtasks, uses duty cycling to harvest energy during idle periods, and models the capacitor with an exponential charging/discharging behavior under an energy-harvesting source. The key contributions include (i) a globally optimal scheduling framework that avoids power failures and improves task success relative to energy-unaware baselines, and (ii) a study of the look-ahead horizon, showing that predicting energy for about eight future task executions suffices to achieve near-optimal performance in many cases. This work provides a principled basis for deriving practical heuristics and energy-prediction components to enable reliable, energy-aware operation of batteryless IoT nodes in real deployments.

Abstract

Today's IoT devices rely on batteries, which offer stable energy storage but contain harmful chemicals. Having billions of IoT devices powered by batteries is not sustainable for the future. As an alternative, batteryless devices run on long-lived capacitors charged using energy harvesters. The small energy storage capacity of capacitors results in intermittent on-off behaviour. Traditional computing schedulers can not handle this intermittency, and in this paper we propose a first step towards an energy-aware task scheduler for constrained batteryless devices. We present a new energy-aware task scheduling algorithm that is able to optimally schedule application tasks to avoid power failures, and that will allow us to provide insights on the optimal look-ahead time for energy prediction. Our insights can be used as a basis for practical energy-aware scheduling and energy availability prediction algorithms. We formulate the scheduling problem as a Mixed Integer Linear Program. We evaluate its performance improvement when comparing it with state-of-the-art schedulers for batteryless IoT devices. Our results show that making the task scheduler energy aware avoids power failures and allows more tasks to successfully execute. Moreover, we conclude that a relatively short look-ahead energy prediction time of 8 future task executions is enough to achieve optimality.

Optimal energy-aware task scheduling for batteryless IoT devices

TL;DR

The paper addresses the challenge of intermittent operation in batteryless IoT devices by formulating an optimal energy-aware task scheduler as a MILP that accounts for energy harvested, capacitor state, and per-task energy consumption and priorities. It decomposes applications into atomic subtasks, uses duty cycling to harvest energy during idle periods, and models the capacitor with an exponential charging/discharging behavior under an energy-harvesting source. The key contributions include (i) a globally optimal scheduling framework that avoids power failures and improves task success relative to energy-unaware baselines, and (ii) a study of the look-ahead horizon, showing that predicting energy for about eight future task executions suffices to achieve near-optimal performance in many cases. This work provides a principled basis for deriving practical heuristics and energy-prediction components to enable reliable, energy-aware operation of batteryless IoT nodes in real deployments.

Abstract

Today's IoT devices rely on batteries, which offer stable energy storage but contain harmful chemicals. Having billions of IoT devices powered by batteries is not sustainable for the future. As an alternative, batteryless devices run on long-lived capacitors charged using energy harvesters. The small energy storage capacity of capacitors results in intermittent on-off behaviour. Traditional computing schedulers can not handle this intermittency, and in this paper we propose a first step towards an energy-aware task scheduler for constrained batteryless devices. We present a new energy-aware task scheduling algorithm that is able to optimally schedule application tasks to avoid power failures, and that will allow us to provide insights on the optimal look-ahead time for energy prediction. Our insights can be used as a basis for practical energy-aware scheduling and energy availability prediction algorithms. We formulate the scheduling problem as a Mixed Integer Linear Program. We evaluate its performance improvement when comparing it with state-of-the-art schedulers for batteryless IoT devices. Our results show that making the task scheduler energy aware avoids power failures and allows more tasks to successfully execute. Moreover, we conclude that a relatively short look-ahead energy prediction time of 8 future task executions is enough to achieve optimality.
Paper Structure (14 sections, 25 equations, 12 figures, 3 tables)

This paper contains 14 sections, 25 equations, 12 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Related work
  3. System model
  4. Batteryless model
  5. Energy-aware task scheduling concept
  6. Optimization framework
  7. Input variables
  8. Decision variables
  9. Objective function and constraints
  10. Evaluation
  11. Simulation setup and methodology
  12. Energy-aware scheduler validation
  13. Influence of look-ahead window on performance
  14. Conclusions and Future Work

Figures (12)

  • Figure 1: Batteryless device intermittent behaviour
  • Figure 2: Electrical circuit model of a batteryless IoT device using a current source energy harvester
  • Figure 3: Energy-unaware vs energy-aware task schedulers in batteryless devices
  • Figure 4: General overview of the atomic tasks model
  • Figure 5: Detailed tasks of the Smart Building Application
  • ...and 7 more figures