• Computer Science > Distributed, Parallel, and Cluster Computing [Submitted on 18 Feb 2026] Title:How Reliable is Your Service at the Extreme Edge? • Analytical Modeling of Computational Reliability View PDF HTML (experimental)Abstract:Extreme Edge Computing (XEC) distributes streaming workloads across consumer-owned devices, exploiting their proximity to users and ubiquitous availability. • Many such workloads are AI-driven, requiring continuous neural network inference for tasks like object detection and video analytics. • Distributed Inference (DI), which partitions model execution across multiple edge devices, enables these streaming services to meet strict throughput and latency requirements. • Yet consumer devices exhibit volatile computational availability due to competing applications and unpredictable usage patterns. • This volatility poses a fundamental challenge: how can we quantify the probability that a device, or ensemble of devices, will maintain the processing rate required by a streaming service?

Article Summaries:

  • Researchers have developed an analytical framework to quantify computational reliability in Extreme Edge Computing (XEC), where streaming AI workloads run on consumer devices. The model defines reliability as the probability that a device’s instantaneous processing capacity meets a specified Quality‑of‑Service threshold. Closed‑form expressions are derived under two information regimes: Minimal Information (only declared operational bounds) and a historical‑data approach using maximum‑likelihood estimation. The framework extends to multi‑device deployments-series, parallel, and partitioned configurations-providing optimal workload allocation rules and device‑selection bounds. Validation with YOLO11m object‑detection workloads in emulated XEC environments shows close agreement between analytical predictions, Monte‑Carlo sampling, and empirical measurements.
  • A new analytical framework for Extreme Edge Computing (XEC) quantifies the probability that consumer‑owned devices can sustain the processing rate required by continuous AI inference workloads. The authors derive closed‑form reliability expressions under two information regimes: Minimal Information (only declared operational bounds) and a data‑driven regime using Maximum Likelihood Estimation from historical observations. The model extends to multi‑device deployments, providing reliability formulas for series, parallel, and partitioned configurations, and yields optimal workload‑allocation rules. Validation with YOLO11m object‑detection workloads in emulated XEC environments shows close agreement between analytical predictions, Monte Carlo sampling, and empirical measurements.

Sources: