The Future of...

The Future of an Intelligent Interconnection Fabric

How multi-domain orchestration is powering automated software-defined interconnection for digital infrastructure and application performance

Oleg Berzin
Rodney Elder

Today’s digital leaders value openness for greater application integration between legacy, on-premises infrastructure and hybrid multicloud architectures. To achieve this “open door” between physical systems, application processes and virtual services, companies must find a common language in which to communicate. This requires an intelligent interconnection fabric that offers the application programing interfaces (APIs) necessary to quickly and easily create integrate and orchestrate digital infrastructure anywhere an organization needs it.

Equinix, along with its interconnected ecosystem of more than 6,600 service provider partners, is continuously innovating to deliver multi-domain orchestration using APIs for Equinix Fabric™, Equinix’s vendor-neutral, global software-defined interconnection. In addition, APIs for Network Edge services from Equinix provides access to leading network functions virtualization (NFV) devices (SD-WAN, firewalls, VPNs, cloud routing) from leading vendors that can be deployed on anywhere on Platform Equinix® in minutes.

The following use case with Equinix, Ciena and Unitas Global demonstrates what is possible when you leverage an intelligent interconnection fabric using Equinix Fabric and Network Edge APIs to develop multi-domain infrastructures.

Equinix Developer Platform & Equinix Fabric™

Learn how Equinix Fabric securely and dynamically connects you to the globally interconnected digital ecosystem on Equinix platform.

Learn More
CTA Image_1-27-21_API
Video/media CDN takes advantage of multi-domain orchestration using automated software-defined interconnection without affecting underlying network routing in a service provider’s domain."

Equinix, Ciena and Unitas Global create award-winning orchestrated virtualized
network service

Interconnection is critical to enabling end-to-end services that rely on an infrastructure comprised of multiple providers. The com­­plexity of a multi-domain network is often further compounded by the fact that end-to-end service quality and reliability must be enforced in different administrative domains. At MEF19[i], Equinix, Ciena and Unitas Global demonstrated that this complexity and the associated increased probability of poor service quality or outages, can be effectively mitigated. As a result, the companies’ PoC won the MEF19 award for an Orchestrated Virtualized Network Service Implementation. [ii]

In the demonstration titled “AI-Driven Federated Network Operations for Ultra Resilient Services,” we demonstrated the ability of a video service provider to deliver orchestrated ultra-resilient services spanning multiple-operator networks with an aggregated fabric across the globe. Latency-sensitive gaming applications often rely on cloud and network services running over global interconnection fabric architectures, with distributed virtual and physical network functions in multiple operator domains. In addition to network and cloud resiliency, AI-driven network management is essential for these new ecosystems because it enables the prediction of SLA exceptions, re-routing, optimization, and other remediation actions on a very large scale.

In the PoC demonstration, we constructed a real multi-domain environment that showed how artificial intelligence (AI) can help a multi-domain service orchestration engine make decisions based on insights derived from telemetry data obtained from different independent service planes. By applying a closed-loop AI-assisted algorithm that learned to predict performance deterioration and redistribute application traffic, our team successfully maintained an optimal end-user quality of experience.

For the demonstration, we built a U.S./Canada network with locations in Ottawa, Hanover, Dallas, San Jose and the MEF19 venue in Los Angeles. The video service provider, who had a packet/optical transport network and a content delivery network (CDN) in Canada, wanted to expand its service in the U.S. to enable end users in Dallas and Los Angeles to stream video content. For this purpose, Equinix Fabric™ was used to interconnect the video service provider’s network to the access aggregation network provided by Unitas Global. The end users were connected to the Unitas Global aggregation switches using last mile Ethernet access in Dallas and Los Angeles, with Ciena 3906 uCPE representing customer locations.

In addition, the video service provider used Network Edge services from Equinix to deploy virtual routers and build distributed hybrid multicloud CDN points of presence (PoPs) in Equinix International Business Exchange™ (IBX®) data centers. The CDN PoPs were implemented in an Openstack Private Cloud that was used to host Akamai CDN Hyper Caches. The Openstack Private Cloud was also connected to multiple public clouds using AWS Direct Connect and Azure ExpressRoute private cloud interconnects on Equinix Fabric.

The Dallas customer location was connected using an end-to-end Ethernet Virtual Circuit (EVC) between the Ciena 3906 uCPE and the virtual router on Network Edge, enabling the customer site to join the hybrid multicloud network with minimal connectivity requirements (in this case a single EVC to reach all public and private clouds). The diagram below shows the logical connectivity. The network spanned multiple administrative domains and the infrastructure was rather complex.

AI-Driven Federated Domain Operations for Ultra Resilient Services

A fair question is: “How can the video service provider ensure acceptable quality of experience to its end users if the video traffic must pass multiple provider domains?” The answer to this question has to do with the AI-assisted closed loop automation. The principle of this closed-loop processing is shown in the figure below:

The closed-loop consisted of multiple stages: Sensing, Discerning, Deciding and Acting. The key to the sensing stage is the telemetry data that came from different service planes and across different service domains. In our demonstration, we used the video application plane data collected at the CDN service level including video control and processing events such as the number of connected video clients, start/stop of playback, join and buffering times and the video URLs. In addition to the video data, the network plane data and insights/metadata were collected from physical and virtual network functions in all provider domains.

The Discerning stage is a Machine Learning stage, where input data/insights are matched to known patterns. For example, a congestion onset pattern can be discerned and labeled with a given confidence level. This happens if the sensed video plane data indicates a growing number of connecting video clients, resulting in increased join and buffering times as the video URLs resolve to a common HyperCache or Origin Server. It does all of this while the network plane data/insights from the provider domains are also indicating increased link utilization in the network topology serving the video traffic.

The output of the Discerning stage is the operational recommendation to the video service provider (a human) to execute a corrective action. This leads to the Acting stage, where the Akamai CDN is reconfigured so new user requests are routed to the newly activated HyperCache locations in an Equinix IBX data center or on the uCPE itself.

One interesting observation is that the corrective action did not directly affect network routing or configurations in the service provider domains. In fact, such a requirement would have been practically problematic as it would mean that the video application provider would need to be given network provider control.

Instead, the reconfiguration of the video CDN (i.e., the activation of HyperCaches in locations closer to the end user and the reconfiguration of the CDN DNS) resulted in traffic being redistributed, avoiding congestion and preserving the end-user quality of experience.

It was a great experience and an honor to work with our partners at Ciena and Unitas Global, as well as my colleagues at Equinix, to demonstrate how an intelligent interconnected fabric can power a multi-domain digital infrastructure.

Left to right: Marco Naveda, Ciena; Oleg Berzin, Equinix; Anthony Thakur, Unitas Global; Ragu Ranganathan, Ciena; and Rodney Elder, Equinix.See what’s possible with Equinix Fabric and Network Edge by visiting the Equinix Developer Platform.


View the video of the AI-Driven Federated Domain Operations for Ultra Resilient Services PoC:


[i] MEF is a non-profit industry forum of network, cloud & technology providers. Together, MEF members develop standards, certifications and APIs to empower enterprise digital transformation.

[ii] Equinix, Ciena, Unitas Global MEF19 PoC Awards.


Oleg Berzin
Oleg Berzin Sr. Director Technology Innovation, Office of the CTO
Rodney Elder Global Principal, GSE/SSI