DECICE Deploys Dedicated Development Infrastructure at E4 Computer Engineering
As part of the DECICE project, a dedicated and advanced development environment has been successfully deployed at partner site E4 Computer Engineering. This tailor-made infrastructure is designed to host the DECICE framework, providing project partners with the resources needed for efficient development, testing, and validation activities across the device-edge-cloud continuum.
The hardware selection was guided by three key criteria:
- Flexibility – to support a wide range of configurations and demands
- Heterogeneity – to enable testing across diverse hardware components
- Accessibility – to ensure seamless collaboration for all partners and developers
Infrastructure Overview
The deployed system features a high-performance, heterogeneous cluster that mirrors the architecture of modern cloud computing environments. It includes:
- 1 Master Node
- 3 Worker Nodes (without accelerators)
- 1 GPU-accelerated Worker Node
- Multiple IoT/Edge Devices
Where possible, all nodes utilize AMD x86_64 architecture, selected for its optimal balance of performance and cost-effectiveness.
Network Configuration
The system is connected via a high-speed, 100 Gbit/s Mellanox Ethernet network linking the master, worker, and GPU nodes. A 10 Gbit/s Ethernet switch supports internal communication, while IoT/Edge devices operate on a separate subnet to simulate real-world external connectivity scenarios.
This setup results in a robust, scalable platform suited for a broad spectrum of workloads, including cloud computing, edge computing, High Performance Computing (HPC), High Throughput Computing (HTC), Machine Learning (ML), and Artificial Intelligence (AI).
Key Hardware Components
Master Node
The central hub of the infrastructure, the master node manages login services, virtual machines, containers, monitoring, and shared storage via NFS. It features:
- AMD Epyc 7643 (Zen 3, Milan), 48 cores
- 256 GB DDR4 RAM
- SSD storage including 3 x 7.68 TB drives for shared access
Worker Nodes
These nodes provide the primary computing power for virtualization and development workloads. Each node includes:
- AMD Epyc 7543P (Zen 3, Milan), 32 cores
- 256 GB DDR4 RAM
- NVMe SSD storage
GPU Node
Dedicated to ML and AI workloads, especially for the project scheduling system, this node offers significant compute acceleration through:
- AMD Epyc 9334 (Zen 4, Genoa), 32 cores
- 384 GB DDR5 RAM
- NVIDIA RTX A30 GPU with 24 GB VRAM, Ampere architecture, and Multi-Instance GPU (MIG) support
IoT/Edge Devices
To simulate edge computing scenarios, the system includes:
- 1 Nvidia Jetson AGX Orin (GPU-enabled)
- 2 Raspberry Pi 5 units
These devices run KubeEdge, enabling integration with the core cluster. The Jetson Orin allows experimentation with AI workloads at the edge, reinforcing system heterogeneity.
Ethernet Switch
A 48-port 10 GbE SFP+ switch with six 100 GbE uplinks enables efficient network management and supports high-speed data transfer across the development environment. While peak network performance is not essential for current use cases, the switch ensures scalability and ease of operation.
This infrastructure milestone marks a critical step forward for DECICE, enabling a dynamic and scalable development platform. It reflects the project commitment to supporting cutting-edge research and development across cloud, edge, and AI domains.
Author(s): Elisabetta Boella, E4 COMPUTER ENGINEERING Spa
