Network Infrastructure for the Era of Cloud and AI

The network, at the very heart of the digital transformation, is the catalyst behind the rapid adoption of the cloud and the exponential growth of AI. It’s under immense pressure to meet the ever-increasing demand for scale, agility, and cost-effectiveness. Over the past two decades, we’ve witnessed significant shifts in networking approaches. It’s crucial to understand our past, avoid past mistakes, and build on our successes to navigate the future.

The physical networks are the traditional model of connectivity. In this model, an organization procures and installs physical network equipment, i.e., routers and switches, for all the desired locations, such as branches, campuses, and data centers. Locations are often interconnected using private circuits, like MPLS, delivered by the service provider. Because organizations cannot install physical equipment inside the public cloud, such networks rely on cloud cross-connects, e.g., AWS Direct Connect, Azure ExpressRoute, and Google Cloud Interconnect, commonly available at the colocation facilities. Organizations also provide physical network equipment at the colocation. Some colocation providers offer virtual network equipment (virtual routers) to reduce reliance on hardware. The cross-connects are augmented with public cloud service providers’ cloud-native capabilities, allowing them to extend network connectivity to the public cloud workloads.

Physical networks establish a strong link between the underlying network infrastructure and the connectivity services. Overlay networks, in the form of SD-WAN and SASE, disrupted this bond, allowing organizations to separate the concepts and distinguish between an underlay and an overlay. While innovative, this model presents its own challenges, requiring organizations to perform hardware or software refreshes. However, it also enhances innovation agility, as it occurs in software that powers the overlay rather than hardware that powers the underlay. Virtual network equipment (virtual routers), deployed in organizational cloud accounts, extends the network into the public cloud, providing direct-to-cloud connectivity using the Internet as an underlay transport rather than the cloud cross-connects at the colocation facility.

It would seem that overlay networks have effectively mitigated the shortcomings of the physical networks, allowing organizations the freedom of choice for transport with MPLS and the Internet, easy cloud expansion with direct connectivity into the public cloud, avoiding colocations, and IT agility powered by software innovation. Where do we go from here?

The main challenge behind physical and overlay networks is the deep networking (and security) intelligence embedded at the network edge. A physical edge in on-premises sites and a virtual edge in the public cloud. It locks organizations into a vicious cycle of hardware refreshes and software upgrades driven by the need to address the latest business demands for cloud and AI connectivity (and security) services. Suitable for the manufacturing vendors, but not so good for the end-customers. Furthermore, it creates a hyper-complex administrative environment requiring heavy investment in tooling and skillsets.

What if we could draw the network (and security) intelligence away from the network edge, placing it at the network backbone? Of course, we would still need some basic networking capabilities at the edge. Still, those should be insignificant, such as sending traffic to the backbone where all the intelligent functions would now reside. This approach would eliminate the need for hardware refresh unless mandated by the end-of-support from the equipment manufacturer. It would also eliminate the need for the overlay network because the backbone would have achieved the same goal of separating the underlying infrastructure from the advanced connectivity services. If such a network backbone were to run inside the public cloud, it would eliminate the need for virtual routers by replacing them with cloud-native capabilities. It almost sounds too good to be true.

A network backbone like that would require network infrastructure to be offered by the provider (rather than built by the end customers) and positioned worldwide for the closest proximity to where it’s needed, on demand. Organizations could leverage the instance of the network infrastructure (a tenant) to build the backbone and connect their branches, campuses, data centers, and cloud workloads effortlessly, providing advanced networking (and security) capabilities in a fraction of the time.

This infrastructure would bring extreme IT agility benefits, replace traditional network complexity layers with modern network-as-a-service, and lower the overall cost of ownership. All that without replacing hardware, deploying software, and understanding the intricacies of cloud and AI networking.

This is the next step in the network’s evolution in the cloud and AI era.


Author's Bio

David Klebanov

Partner Solutions Leader

David Klebanov is a 25 years networking industry veteran. Throughout his career David has engaged in architecting and building a multitude of multi-dimensional network solutions for a variety of enterprise and service provider customers. His expertise extends across the wide area network, data center and the cloud. David currently leads partner solutions for Alkira, the company reinventing networking for the era or cloud and AI. Prior to Alkira, David led technical marketing at Viptela, a company that started a multi-billion dollar SD-WAN market and was acquired by Cisco in 2017.