The Race to Zero-Downtime: EDA Gives DC Networking an Edge

The most stable data center is the one nobody touches. That used to be the conventional wisdom among network engineers, at least until the explosion of AI workloads and cloud computing shattered that paradigm.

Today’s reality is different. Your data center network faces constant change, mostly due to the unpredictable demands of AI. And either you engineer it for stability under those conditions, or you’re headed for trouble.

“Historically, data center networks have been stable as long as no one touches them,” explains Sang Xulei, senior vice president and head of network infrastructure for Asia Pacific at Nokia. “However, with the demands of cloud IT and the explosion in AI workloads, network changes in the data center are a constant thing.”

This fundamental shift in operational requirements has pushed traditional network management approaches past their breaking point across the APAC region, where digital transformation is accelerating at a blistering pace. The answer? Event-driven automation (EDA) — a radical rethinking of how networks respond to change.

The polling problem

Before diving into how EDA works, it’s worth understanding the critical flaw in conventional network management: polling latency.

Traditional data center management relies on regular polling cycles — essentially rotating through network devices and asking for status updates. This approach creates a fundamental visibility gap where engineers are perpetually several minutes behind reality.

“In traditional polling-type management systems, you could be 5, 10, or even 15 minutes behind in what is happening on the network,” Sang notes. For workloads that demand near-instantaneous response times, this delay can be potentially catastrophic.

Event-driven systems eliminate this blind spot. When a network event occurs — whether that’s an alert, configuration change, or telemetry update — the platform automatically executes operational logic without waiting for the next polling cycle.

“The result is that the operations team sees live network data on their screens, instantaneous telemetry, and firsthand visibility of what is happening, giving them the ability to proactively resolve any issues,” says Sang. “This boosts network reliability and reduces operational costs.”

Digital twins: Network time machines

Among EDA’s most powerful innovations is the implementation of digital twin technology specifically engineered for network operations.

Nokia’s implementation creates what Sang describes as a “‘live’ digital twin of the network where you can apply changes.” This virtual doppelgänger allows engineers to validate modifications against an emulated environment before deploying to production infrastructure.

The system reports back on potential service-level impacts or topology changes, effectively providing a “dry run” before changes hit the live environment. For operations teams managing critical infrastructure, this capability fundamentally changes the risk profile of network modifications.

Perhaps even more valuable is EDA’s change-based revision control. “All changes on the network are stored as intent files,” explains Sang. “This means that operators can roll back the network to any time in the past. It’s like having your own DeLorean time machine — the operator can ‘go back in time’ to yesterday, last week, or even last month if they need to.”

Kubernetes: The invisible engine

Underpinning Nokia’s EDA platform is Kubernetes. Originally designed for hyperscale environments at Google, it enables the orchestration necessary for automated scaling and platform management.

“For Nokia, it allows the EDA platform to grow and scale effortlessly as customer networks evolve,” Sang explains. “EDA can spin up multiple virtual containers to horizontally scale functions as needed.”

For network engineers, the implementation details are abstracted away. “For day-to-day operations, Kubernetes is hidden inside the EDA system. It is purely the powerful engine that EDA uses to flex and grow as network workloads demand,” says Sang. “For the most part it is hidden away from the data center operators, so they do not need to be proficient in Kubernetes to get the best out of EDA.”

This abstraction represents a critical shift in how engineers interact with network infrastructure, moving away from device-level configuration toward intent-based management.

Embracing the multi-vendor reality

Modern data centers rarely run on a single vendor’s equipment. It is a reality that Nokia has embraced in its EDA development.

At launch, the platform supported Nokia’s own network operating systems (SR Linux and SROS), and recently added SONiC support, the open-source NOS gaining significant traction after Microsoft transferred oversight to the Linux Foundation.

“Customers in the data center have a strong preference to pick and choose the best offer for each role in their network, rather than compromising on a single vendor for all functions,” explains Sang. “For example, a data center may have one or two vendors for top-of-rack and end-of-row switching and then a third for their large-scale routing needs.”

This heterogeneous approach demands management solutions that work across platforms rather than creating additional silos. Nokia is expanding EDA to support third-party equipment, with efforts underway to officially support additional platforms.

AI Operations for the next generation

The network engineering talent pool is transforming. As veterans retire, they’re being replaced by engineers who grew up with touch interfaces and conversational UI, not command lines.

“The younger generations have grown up on more human-based interactions as the computer interfaces,” Sang notes. “Desktop operating systems, smartphones and tablets all function without requiring the user to be a UNIX guru or have a deep understanding of any command-line interfaces.”

EDA’s approach to this generational shift centers on AI-driven operations interfaces specifically designed for engineers who may lack deep vendor-specific knowledge.

“Operational interactions can be made in the native language, English or your own dialect,” explains Sang. “For instance, I can simply ask EDA ‘give me a list of all active ports with VLAN 1 on them’. EDA will respond with a listing and will also reply with the exact command it used to retrieve the information.”

This approach creates a natural learning curve where engineers can progressively master the platform’s more advanced capabilities while remaining productive from day one.

Solving AI’s unique networking challenges

AI workloads impose radically different requirements on network infrastructure compared to traditional computing workloads. Where traditional data centers were designed with some degree of multi-tiered contention, AI architectures demand non-blocking fabrics with vastly different performance characteristics.

“The other interesting aspect of AI networking is the cohesiveness that the operations tools need to understand the interactions between the end-to-end components; be that from the network to the DPU, to the GPU and the host,” Sang explains. “Tuning all these parameters and their respective automation silos is a significant part of the value that EDA brings to the operations team.”

As AI workloads transition from proprietary interconnects to Ethernet (with backing from the Ultra Ethernet Consortium), management platforms must adapt to these new requirements. EDA provides the end-to-end flow-based control and visibility needed for current and future AI workloads.

Intent-based networking: Business-first automation

Network changes have traditionally been implemented through device snippets—fragments of CLI code deployed device-by-device, creating a significant risk of inconsistency or incomplete implementation.

“The problem with this is that there is no end-to-end view of the change,” Sang explains. “Questions arise such as: did I update all my switches, could there be some that I missed?”

Intent-based networking (IBN) flips this model upside down, creating a business-oriented approach to network configuration rather than a device-by-device approach.

“EDA is natively intent-based driven so the operator can push network changes in a business-oriented fashion and get instant feedback that it’s been successfully deployed,” says Sang. “The operator could simply issue the native language ‘update all the top-of-rack switches to block TCP on port 1111 on their management interfaces’. EDA would make the change and report back with a list of all the switches that the change was applied.”

Automation becomes vital for quantum, 6G

The future of network automation extends well beyond data centers. As network speeds increase across domains — from 25G and 100G GPON in wired networks to 400G and 800G Ethernet in transport networks and the expanding footprint of 5G — the operational demands will only intensify.

“The one constant is that networks are reaching further and at faster speeds, which brings increasing demands on those who operate the networks to maintain reliability and performance,” Sang observes.

Looking toward technologies like quantum computing and 6G wireless, the need for comprehensive automation becomes even more acute.

“For Nokia, we will continue to evolve the management concepts around event-driven network automation and apply them to the various network domains that drive our cloud, internet, AI, and communication services,” says Sang. “It is an exciting time to be at the forefront of IP networking and to create technology that helps the world act together.”

The zero-downtime revolution has begun, and event-driven automation is leading the charge. For today’s data center engineers, the question isn’t whether to automate — it’s whether your automation strategy can keep pace with the future that’s already arriving.

Image credit: iStockphoto/sdecoret