Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

TriCloudEdge: A multi-layer Cloud Continuum

George Violettas, Lefteris Mamatas

TL;DR

This work presents TriCloudEdge, a practical three-tier Edge-Cloud AI continuum implemented on real hardware (Far Edge: ESP32-CAM, Edge: ESP32-S3, Cloud: AWS) and evaluated under two architectures: a Multi-Protocol stack (WebSocket/HTTP/MQTT) and a Zenoh-Unified approach. It systematically compares latency, throughput, and pipeline parallelism, revealing that Zenoh can reduce cross-layer overhead and improve throughput in asynchronous, multi-hop scenarios, while WebSocket delivers robust per-file transmission performance. The study demonstrates end-to-end AI workloads with privacy-preserving locality and provides detailed implementation challenges, statistical analyses, and reproducibility resources. The results inform practical trade-offs between integration complexity and performance, guiding future energy-aware offloading, security, and governance in edge-cloud deployments.

Abstract

TriCloudEdge is a scalable three-tier cloud continuum that integrates far-edge devices, intermediate edge nodes, and central cloud services, working in parallel as a unified solution. At the far edge, ultra-low-cost microcontrollers can handle lightweight AI tasks, while intermediate edge devices provide local intelligence, and the cloud tier offers large-scale analytics, federated learning, model adaptation, and global identity management. The proposed architecture enables multi-protocols and technologies (WebSocket, MQTT, HTTP) compared to a versatile protocol (Zenoh) to transfer diverse bidirectional data across the tiers, offering a balance between computational challenges and latency requirements. Comparative implementations between these two architectures demonstrate the trade-offs between resource utilization and communication efficiency. The results show that TriCloudEdge can distribute computational challenges to address latency and privacy concerns. The work also presents tests of AI model adaptation on the far edge and the computational effort challenges under the prism of parallelism. This work offers a perspective on the practical continuum challenges of implementation aligned with recent research advances addressing challenges across the different cloud levels.

TriCloudEdge: A multi-layer Cloud Continuum

TL;DR

This work presents TriCloudEdge, a practical three-tier Edge-Cloud AI continuum implemented on real hardware (Far Edge: ESP32-CAM, Edge: ESP32-S3, Cloud: AWS) and evaluated under two architectures: a Multi-Protocol stack (WebSocket/HTTP/MQTT) and a Zenoh-Unified approach. It systematically compares latency, throughput, and pipeline parallelism, revealing that Zenoh can reduce cross-layer overhead and improve throughput in asynchronous, multi-hop scenarios, while WebSocket delivers robust per-file transmission performance. The study demonstrates end-to-end AI workloads with privacy-preserving locality and provides detailed implementation challenges, statistical analyses, and reproducibility resources. The results inform practical trade-offs between integration complexity and performance, guiding future energy-aware offloading, security, and governance in edge-cloud deployments.

Abstract

TriCloudEdge is a scalable three-tier cloud continuum that integrates far-edge devices, intermediate edge nodes, and central cloud services, working in parallel as a unified solution. At the far edge, ultra-low-cost microcontrollers can handle lightweight AI tasks, while intermediate edge devices provide local intelligence, and the cloud tier offers large-scale analytics, federated learning, model adaptation, and global identity management. The proposed architecture enables multi-protocols and technologies (WebSocket, MQTT, HTTP) compared to a versatile protocol (Zenoh) to transfer diverse bidirectional data across the tiers, offering a balance between computational challenges and latency requirements. Comparative implementations between these two architectures demonstrate the trade-offs between resource utilization and communication efficiency. The results show that TriCloudEdge can distribute computational challenges to address latency and privacy concerns. The work also presents tests of AI model adaptation on the far edge and the computational effort challenges under the prism of parallelism. This work offers a perspective on the practical continuum challenges of implementation aligned with recent research advances addressing challenges across the different cloud levels.
Paper Structure (31 sections, 6 figures, 3 tables)

This paper contains 31 sections, 6 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Edge-AI Global Initiatives
  3. Related Work
  4. Edge Cloud Terms and Status
  5. Advantages of Edge-AI
  6. Edge-AI Issues and Challenges
  7. System Architecture & Reference Implementation
  8. Far Edge: Constrained but Real-Time
  9. Edge: Local Intelligence and Enhanced Capabilities
  10. Cloud: Centralized Analytics on a Global Scale
  11. Computational Power vs Low Latency
  12. Data and Model Flow
  13. Communication Protocols
  14. Comparative Analysis
  15. Methodology
  16. ...and 16 more sections

Figures (6)

  • Figure 1: Edge-Cloud AI Continuum
  • Figure 2: Far Edge - Edge - Cloud Continuum in action.
  • Figure 3: ESP32-CAM pipeline parallelism and per-protocol performance over a 3,000 images dataset.
  • Figure 4: Comparison of RTT to Edge vs Cloud (AWS).
  • Figure 5: Comparison of protocol throughput over multiple datasets.
  • ...and 1 more figures