On the Modeling of Reliability in Extreme Edge Computing Systems

TL;DR

This paper develops a probabilistic reliability model for extreme edge devices in edge computing, analyzing their ability to execute tasks on time amidst resource uncertainty, supported by experimental validation.

Contribution

It introduces a novel reliability model for EEDs considering resource availability randomness and evaluates task execution reliability in EEC systems.

Findings

Reliability of EEDs varies with resource availability patterns.

Task execution success depends on probabilistic resource models.

Experimental results validate the reliability analysis across scenarios.

Abstract

Extreme edge computing (EEC) refers to the endmost part of edge computing wherein computational tasks and edge services are deployed only on extreme edge devices (EEDs). EEDs are consumer or user-owned devices that offer computational resources, which may consist of wearable devices, personal mobile devices, drones, etc. Such devices are opportunistically or naturally present within the proximity of other user devices. Hence, utilizing EEDs to deploy edge services or perform computational tasks fulfills the promise of edge computing of bringing the services and computation as close as possible to the end-users. However, the lack of knowledge and control over the EEDs computational resources raises a red flag, since executing the computational tasks successfully becomes doubtful. To this end, we aim to study the EEDs randomness from the computational perspective, and how reliable is an EED in terms of executing the tasks on time. Specifically, we provide a reliability model for the EEDs that takes into account the probabilistic nature of the availability of the EEDs' computational resources. Moreover, we study the reliability of executing different types of computational tasks in EEC systems that are distributed across the EEDs. Lastly, we carry out experimental results to analyze the EEDs and the EEC systems' reliability behavior in different scenarios.