A Cost Model for Edge–Fog–Cloud Inference Allocation in Health Monitoring

A Cost Model for Edge–Fog–Cloud Inference Allocation in Health Monitoring

Negar Shahiddokht1

1) Ph.D. in Artificial Intelligence, Amirkabir University of Technology, Tehran, Iran

IOI Code : AAHD-ABKBBA
Publication : National Conference on Information Technology, Nanotechnology, Artificial Intelligence and Technological Futures Studies(itnaf.ir)
Abstract :
    In IoT-based health monitoring systems, a key design question is which computational tier should execute the machine-learning inference task. The existing literature focuses mainly on improving model accuracy and selects the deployment tier empirically and case by case; consequently, the reported results depend on the particular testbed and do not generalise. This paper adopts a different approach: rather than reporting one specific implementation, it presents a closed-form analytical cost model that expresses end-to-end latency, communication load and device energy consumption as functions of system parameters. Building on this model, a three-tier edge–fog–cloud framework is designed together with a task-allocation policy based on the referral rate , defined as the fraction of events forwarded to the upper tier after lightweight screening at the edge. Break-even analysis shows that the conditions under which the proposed framework outperforms the cloud-only and edge-only architectures can be expressed as closed-form inequalities in terms of this same referral rate. A numerical evaluation of the model with a set of representative parameter values indicates that, at a referral rate of 15 per cent, alert latency, communication load and device energy consumption are reduced by approximately 82, 87 and 56 per cent respectively relative to the cloud-only architecture. It is emphasised that the reported figures are direct outputs of the analytical relations under the explicit assumptions stated in the paper, and not reports of experimental measurement; their value lies in exposing the behaviour and boundaries of the model rather than in claiming the performance of any specific implementation.

Keywords : internet of things smart city edge computing fog computing cost model task allocation