Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Lightweight Latency Prediction Scheme for Edge Applications: A Rational Modelling Approach

Mohan Liyanage, Eldiyar Zhantileuov, Ali Kadhum Idrees, Rolf Schuster

TL;DR

The paper tackles the challenge of end-to-end latency prediction for real-time edge offloading by introducing a lightweight rational-exponential delay formulation that uses readily available features like frame size, arrival rate, and link utilization, avoiding intrusive probing. It demonstrates that the Rational-Exponential model achieves state-of-the-art predictive accuracy (MAE ≈ 0.0115, $R^2$ ≈ 0.9847) with minimal inference time, outperforming linear, polynomial, sigmoid, MLP, and other rational approaches. The methodology combines real-time network monitoring with a delay-aware decision engine to select QoS-friendly offloading targets in a campus 5G setup, validated through 5-fold cross-validation and residual analysis. Overall, the work offers a practical, scalable solution for robust, low-latency offloading in multi-tier edge environments.

Abstract

Accurately predicting end-to-end network latency is essential for enabling reliable task offloading in real-time edge computing applications. This paper introduces a lightweight latency prediction scheme based on rational modelling that uses features such as frame size, arrival rate, and link utilization, eliminating the need for intrusive active probing. The model achieves state-of-the-art prediction accuracy through extensive experiments and 5-fold cross-validation (MAE = 0.0115, R$^2$ = 0.9847) with competitive inference time, offering a substantial trade-off between precision and efficiency compared to traditional regressors and neural networks.

Lightweight Latency Prediction Scheme for Edge Applications: A Rational Modelling Approach

TL;DR

The paper tackles the challenge of end-to-end latency prediction for real-time edge offloading by introducing a lightweight rational-exponential delay formulation that uses readily available features like frame size, arrival rate, and link utilization, avoiding intrusive probing. It demonstrates that the Rational-Exponential model achieves state-of-the-art predictive accuracy (MAE ≈ 0.0115, ≈ 0.9847) with minimal inference time, outperforming linear, polynomial, sigmoid, MLP, and other rational approaches. The methodology combines real-time network monitoring with a delay-aware decision engine to select QoS-friendly offloading targets in a campus 5G setup, validated through 5-fold cross-validation and residual analysis. Overall, the work offers a practical, scalable solution for robust, low-latency offloading in multi-tier edge environments.

Abstract

Accurately predicting end-to-end network latency is essential for enabling reliable task offloading in real-time edge computing applications. This paper introduces a lightweight latency prediction scheme based on rational modelling that uses features such as frame size, arrival rate, and link utilization, eliminating the need for intrusive active probing. The model achieves state-of-the-art prediction accuracy through extensive experiments and 5-fold cross-validation (MAE = 0.0115, R = 0.9847) with competitive inference time, offering a substantial trade-off between precision and efficiency compared to traditional regressors and neural networks.

Paper Structure

This paper contains 17 sections, 2 equations, 2 figures, 3 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Related work
  3. Analytical Approaches
  4. Machine Learning-Based Models
  5. Hybrid and Adaptive Models
  6. Scenario: QoS-Aware Server Selection in Edge Video Analytics
  7. Methodology
  8. Network Monitoring and Delay Estimation
  9. Delay-Aware Decision Logic
  10. Rational-Exponential Model for Delay Prediction
  11. Experimental Results and Discussion
  12. Experimental Setup
  13. Feature Engineering for Edge Delay Modeling
  14. Initial Feature Set
  15. Feature Selection via Correlation and Sensitivity Analysis
  16. ...and 2 more sections

Figures (2)

  • Figure 1: Network model for experimental setup
  • Figure 2: Residual analysis of the Linear Regression (left) and Rational-Exponential (right) models plotted against Utilization