Keep up to date with our innovative initiatives.
Sign up here
As the digital landscape evolves, integrating intelligent technologies into edge computing is becoming essential for industries seeking to enhance operational efficiency and reduce latency. Effectively orchestrating resources in edge environments is crucial for deploying industrial applications, particularly in scenarios with diverse, distributed nodes, such as multi-AGV setups. The multi-AGV orchestration demonstration carried out under 6GSMART-EXP highlights how integrating 6GSMART-ICC’s on-premise edge orchestrator, i2Edge, with an intelligent placement algorithm addresses these challenges.
i2Edge is a modular, extensible orchestration framework that manages the lifecycle of industrial applications. It simplifies resource management, enabling applications to be deployed closer to data sources or end-users, significantly improving latency and overall efficiency.
The system’s orchestration capabilities are enhanced by an intelligent placement algorithm that dynamically selects the optimal nodes for application deployment based on criteria like CPU, storage, memory, and GPU availability. This ensures applications are placed on the most suitable nodes based on real-time network conditions.
In the demonstration, i2Edge efficiently handles multiple Automated Guided Vehicle (AGV) application placement requests, showcasing its ability to support complex industrial operations autonomously. The intelligent placement algorithm uses a Deep Q-network (DQN) model, normalizing system metrics to ensure data-driven decisions. These normalized values represent each AGV’s ability to meet specific application requirements, reflecting real-time resource status.
The integration between i2Edge and the placement algorithm allows seamless application deployment. When a request is initiated, the orchestrator collects data on resource availability and application requirements, which is sent to the placement algorithm. The algorithm normalizes the data, analyzes it, and computes the optimal node for deployment. The orchestrator completes the deployment, ensuring efficient resource use and performance optimization.
The experimental setup simulates a multi-AGV environment with four AGVs acting as edge computing nodes. Each AGV node has its own resource profile (CPU, memory, storage, GPU), allowing evaluation of the placement algorithm’s ability to balance workloads and match applications with the most suitable resources. The DQN model’s ability to learn and optimize GPU-specific scheduling allows it to outperform Kubernetes significantly, particularly in environments with increasing GPU requirements.
Authors: Javier Palomares, Professional Researcher and Estela Carmona Cejudo – Head of Innovation & Senior Researcher at the Software Networks Research Group at i2CAT
