As AI continues to reshape how data centers operate, one of the biggest challenges has become keeping networks stable, responsive, and efficient — even as traffic grows more complex and workloads scale across thousands of GPUs. Nokia’s Event-Driven Automation (EDA) platform takes aim at that challenge, introducing a smarter, framework-driven way to manage the “nervous system” of AI backend networks.
What Is Event-Driven Automation?
At its core, Event-Driven Automation is about letting the network respond to change automatically.
Instead of relying on manual scripts or human intervention, EDA detects events — like configuration changes, performance shifts, or traffic congestion — and then reacts in real time with predefined, safe actions.
Think of it as a virtual engineer that constantly watches your network, understands context, and makes decisions based on intent. If something goes wrong, it can troubleshoot, push out a fix, or even reconfigure parts of the fabric autonomously
SaaS or On-Prem: Flexibility by Design
Nokia offers EDA in two deployment models:
- EDA SaaS – A cloud-hosted service that provides elastic scale and built-in analytics.
- EDA On-Prem – For organizations needing tighter control, it can run locally within the data center.
Both versions share the same automation engine, ensuring consistent behavior across environments whether managed by your team or hosted by Nokia
Built on Kubernetes: Automation Meets Abstraction
EDA borrows a page from Kubernetes — the technology that revolutionized cloud computing.
Just as Kubernetes automates workloads and maintains a desired state through its reconciliation loop, Nokia EDA applies the same declarative, abstract model to infrastructure.
You describe what your network should look like, and EDA figures out how to get there.
It supports YAML and JSON, keeps track of device schemas, and even uses Git for version control — ensuring every change is traceable, reviewable, and reversible.
Driving Network Downtime Toward Zero
Modern networks don’t fail often — but when they do, it’s usually during configuration or provisioning.
EDA addresses this head-on with various innovations:
- Source of Truth – EDA maintains a live, model-driven representation of your network. Every configuration and operational state is validated against it.
- Network-Wide Transactions – Configuration updates are pushed atomically across all devices, reducing risk and ensuring consistent changes.
- Revision Control – Using Git repositories, teams can roll back, audit, or compare configurations like software code.
Together, these capabilities aim for “five-nines” (99.999%) fabric availability — a goal validated by Nokia Bell Labs’ reliability model
Observability at Fabric Scale
A “fabric” is the underlying mesh of interconnected switches that link servers and GPUs inside a data center.
EDA provides deep fabric observability, offering dashboards for every level — from individual interfaces to overall topology views.
Operators can visualize:
- Queue depth
- Congestion metrics (PFC, ECN)
- Performance counters across GPU rails
EDA’s visualizations allow engineers to see congestion and imbalance in real time, replacing hours of manual troubleshooting with a few clicks
Simplifying Operations with Intent-Based Automation
Nokia’s EDA framework brings “intent-based” design to network operations — you define the desired outcome, not the low-level commands.
Highlights include:
- Zero-Touch Provisioning (ZTP): Automatically discovers and configures new devices.
- Multi-Vendor Support: Works with Arista, Cisco, and Nokia devices, ensuring interoperability in mixed environments.
- Live Query & Subscriptions: EDA can query active devices and subscribe to real-time telemetry updates, keeping data fresh and actionable.
This combination allows teams to focus on outcomes — not syntax.
Managing GPU Backend Fabric Congestion (Made Simple)
AI workloads are hungry — moving massive amounts of data between GPUs. That traffic can easily overwhelm a network if not managed carefully.
EDA uses advanced congestion control techniques to keep data flowing smoothly:
- PFC (Priority Flow Control): Acts like a “pause button” when buffers fill up, preventing packet loss.
- ECN (Explicit Congestion Notification): Adds early warning signals to traffic, letting devices slow down before congestion becomes critical.
- DCQCN (Data Center Quantized Congestion Notification): Combines both methods for fine-grained control — balancing performance and reliability.
In plain terms: instead of flooding the network and hoping for the best, EDA keeps communication between GPUs orderly and lossless
Why does this all matter?
AI clusters depend on predictable, high-speed connectivity — even a small hiccup can cause massive slowdowns in training or inference.
Nokia’s Event-Driven Automation helps prevent these issues before they happen, automating everything from configuration to fault recovery, and providing the visibility needed to keep large-scale GPU networks healthy.
The Bigger Picture
EDA isn’t just another management tool — it’s a framework for autonomous operations.
By combining Kubernetes-style abstraction, Git-based version control, and AI-driven reasoning, Nokia has built an automation layer ready for the new era of data-driven infrastructure.
For more details, you can explore:
For more details on NFD39 and Nokia’s presentation, you can explore:
- Introduction to Nokia with Andy Lapteff
- Exploring the Power and Potential of Enhanced AIOps with Nokia
- Nokia Event-Driven Automation (EDA) Multi Vendor – Deliver on the Promise
- Delivering Nokia Enhanced AIOps with the Right Foundations
- Root Cause Analysis with Nokia AI Operations Automation
- The Age of Operations: Third Party Views of DCF Innovation
- Networking Field Day 39
Disclosure: I occasionally attend events like Tech Field Day. While that might include some small perks such as travel assistance or swag from vendors, what I write and think is always 100% my own.
Ctrl+Alt+Route 
Leave a comment