OSM PoC 11 Deployment of AI-Agents in OSM
PoC Team Members
All Poc team members are OSM members or participants
- Main contacts:
- Ignacio Labrador (Ignacio.labrador _at_ atos.net)
- Luis Gómez (luis.gomez.external _at_ atos.net)
This PoC showcases how to deploy AI-Agents using HELM-Charts running in VNF Execution Environments in OSM. These AI-Agents have been designed to trigger VNFs scaling actions and enrich OSM by making it possible to address complex data classification or prediction problems (e.g., based on CPU, RAM or network usage metrics). In the particular case of this PoC a specific AI-Agent queries an AI Models Server hosting an AI Model consisting on a Convolutional Neural Network, that performs image classification on actual road traffic images. Based on the AI-Model responses about the traffic status, the deployed AI-Agent will trigger VNF scale-in/scale-out actions. Beyond this particular AI model case, the set-up of the AI-Agents to be used in one type of problem or another is done by just configuring the HELM-Chart descriptors.
PoC Key Takeaways
- Integration AI-Agents in OSM allows to act in a proactive way before a mitigation action is required.
- The new AI-Agents model is compatible with the current monitoring system in OSM.
- It gives the opportunity to integrate heterogeneous data sources (infrastructure metrics, NS data and external) to reach automated ambitious decisions.
- Open Solution:
- It enables to deploy different ML models defined by the NS designer.
- It is possible to use different AI/ML frameworks at the OSM user's discretion.
- The solution provides a scalable architecture (multiple AI-Agents and ML models can be deployed to satisfy different requirements).
OSM Version and Components
OSM version and main components leveraged by the PoC:
- OSM Release EIGHT over K8s (subsequent releases might also be used).
- AI-Agent running at Execution Environment.
- AI Models Server (TensorFlow Serving v2.3.0).
- Presentations and demos showed in OSM TECH meetings:
- OSM-Mid Release NINE - 10th Sept’20;
- OSM Tech meeting 22 Oct’20.
Video work in progress.