Amalgamated Intermittent Computing Systems

Abstract

source type, it sends a notification. A battery-powered system is always on and thus can capture data and identify when the bell rang at time unit 2 (Figure 1a). However, an intermittent system that activates once every three seconds misses capturing the event (Figure 1b). Even if a battery-free node captures the sample, it may fail to finish processing it in time, resulting in an undetected event. Multiple identical nodes are also redundant because they will all be active simultaneously (Figure 1c).The opportunity lies in that only one of the many intermittent systems must capture and process the event successfully to identify if the bell rang. Thus, scheduling the wake-up and sleep cycle can amalgamate three identical intermittent systems with only one active node at any time. As a result, an amalgamated intermittent system captures and processes all the samples and can identify the bell ring at time unit 2 in Figure 1d. In summary, with sufficient intermittent systems working as an amalgamated system, we can emulate a battery-powered node when the power-off period does not exceed the energy buffer self-discharging time.Though scheduling a swarm of nodes to work collaboratively is not a new problem [54,55], these algorithms rely on active communication among the nodes. Frequent communication among multiple nodes is infeasible in intermittent computing systems, as communication is often the most power-hungry operation [23]. Though zero-power passive communication [20,50] is promising, these methods can communicate with only one node at a time, increasing communication time among all the nodes in a swarm. Moreover, while passive backscatter communication depends on an active receiver [50] or transceivers [20], they require precise (milliseconds or even nanoseconds) time synchronization to allow the sender and receiver to turn on the radio simultaneously. Bonito [21] proposes a learned wake-up schedule for effective communication, and Flync [20] provides an effective solution for neighbor discovery despite intermittency.However, these works only support one device and often require prior knowledge about the energy pattern. Moreover, required packet conflict recovery makes them insufficient for frequent communication across a swarm of systems.We propose AICS, a framework that amalgamates a swarm of intermittent nodes for prolonging the collective power-on period without inter-node communication. AICS allows each intermittent system to decide whether to wake up or go to sleep by predicting other nodes' behavior without communication. To increase successful capture and computing and maximize energy utilization, AICS reduces the number of simultaneously active nodes when the power-off period is less than the self-discharging time.Coalesced Intermittent Sensor (CIS) [41] is our closest related work which randomly wakes up each intermittent system to avoid concurrent active systems. However, it only considers the scenario where the same energy is available to every node of the swarm. Despite being an important scenario...

Energy-Aware Adaptive Sampling for Self-Sustainability in Resource-Constrained IoT Devices

“…Estimates suggest that there will be a deployment of approximately 30 billion Internet of Things (IoT) devices by the year 2025 [15,27]. Such a number of connected devices poses a significant problem regarding energy sourcing: if all these devices were battery-powered, it would require a tremendous effort to recharge/change batteries, and it would impose a serious toll on emissions, cost, and overall sustainability [19,30].…”