Sometimes, the terms we use create an impression that really isn’t justified. For example, we talk about “enterprise networks” as though they were a static model of connectivity when in fact they’ve undergone major changes over the last couple decades. And yes, there are more changes underway now, to the point where the concept of the enterprise network may become totally fuzzy.
At the start of this century, enterprise networks used to have two components, a data center piece that was really just emerging, and a WAN piece. The enterprise WAN, over time, has been subducted into MPLS VPN, and more recently SD-WAN, services. The data center network now seems threatened by the cloud. Enterprises themselves see these changes and have their own explanations for why they’re happening. In most cases, they’re right. In a few, not so much.
In the old days, the enterprise network was so WAN-focused that I can recall writing articles that criticized vendors for dawdling in the data center switch space. IBM at the time was cited by users as the company with the greatest strategic influence on their network decisions, and IBM at the time had just sold off their network equipment business to Cisco. There aren’t a lot of enterprise IT managers and planners of that period that are still available to talk with, but those who could chat with me about the period say that IBM’s strategic dominance was never due to its actual network products, but to its impact on application planning. “Changes in network requirements come from changes in applications and application usage,” one long-time CIO report told me.
It turned out that was an oversimplification, at least with regard to how drivers of network change evolved since the turn of the decade. Applications, it’s true, are fundamental to networking because they drive the benefits that justify the investment. But applications have to run on something, the IT platform, and that platform has been evolving under the pressure of technology change and an increased need to control costs. The biggest current drivers of that infrastructure-level change in the enterprise network have been, are, and will likely remain virtualization and componentization.
Virtualization was actually popularized by IBM with its VM operating system for its mainframes in the 1970s because the old model of a 1:1 relationship between applications and hosts proved inefficient as computers got more powerful. As hosting on servers whose compute power was based on microprocessors became more common, the architecture lent itself to virtual subdivisions within physical resources too. The concept of a “virtual machine” itself dates way back to the 1960s but the kind of virtualization we see today really dates from the early 1980s, and it exploded with VMware in the late ‘90s. Today, it’s used by every enterprise.
Componentization, meaning the creation of applications from discrete and at least semi-autonomous elements that might be shared across applications, is likewise a very early development, though in its earliest form it was more likely to be what we call “middleware” today. Today’s model evolved in no small way from the “shared repository” model, where corporate databases were used by multiple applications, and processing (access, update) were then also shared. Like virtualization, componentization has been adopted by every single enterprise.
Virtualization and componentization make a data center into a resource pool, a shared set of servers and databases that are exploited by applications in whatever way is convenient. It won’t surprise you that every enterprise says they have a “resource pool”. Obviously this pool of resources had to be connected, and that’s what gave rise to the modern notion of the data center network. Since this was evolving at the very time when we were seeing the enterprise WAN evolve into an MPLS VPN service, which meant only access routers were required, the result was what we see today, which is a data-center-centricity in enterprise networking.
The cloud is simply an extension of the virtualization and componentization trend, where hosting services that conform to that model are provided by a third party from a larger (more efficient, more elastic, more resilient) resource pool. About a third of enterprises see this, and most use the same software suite (VMware or Red Hat tools) to build applications in both data center and cloud. However, the cloud’s resource pool is a virtual network as much as a virtual host, and since most cloud users enter via the Internet, the adoption of the cloud as an application front-end (a set of user-facing components) has meant that part of the data center network and WAN have “separated” and virtualized as a single unit. We see things like SASE, which is a cloud-resident SD-WAN-like VPN extension, as a proof of this.
SASE is a critical step toward the future of networking because it’s the bridge between two truths. The first is that the data center is not going away, and in fact all evidence is that it’s growing. Over 90% of enterprises accept this as true. The second is that in order to unify the cloud’s virtual network and the VPN services companies use to connect fixed offices, you need the cloud to be a VPN element. Almost 100% of enterprises accept this as true.
The final factor in our evolution of the enterprise network is what’s effectively the virtualization of the user. The lockdowns of the pandemic and work-from-home combined to accentuate a truth that was already visible a decade ago, which was that a worker was the the target of empowerment and not the workplace. The Internet is ubiquitous. The cloud is likewise, and so the worker-facing piece of the application outward is now its own little virtual domain, with its own independent network. We’re not going to eliminate the data center, but what we are going to do, and what a quarter of enterprises already see, is that the data center is going to be an extension of the cloud and not the other way around. As workers move to an Internet-and-SASE model of connection to applications, they pull application drivers more and more into the cloud, and if applications are the dominant factor in networking, then the network goes there too. Site networking via VPNs gives way to virtual networking that connects users to their application components. The data center becomes the “repository host”.
This doesn’t mean that data center networks disappear, only that the focus becomes one of creating a common network-virtualization model that crosses the cloud-to-data-center boundary. The fact that IBM is still in the mainframe business proves that the core of business-critical applications, the processing of transactions against company databases, has a very high inertia. The question is really the extent to which the Internet-cloud combination and the data center network are really unified.
That almost surely depends on pricing issues more than on technology selection or evolution. As long as cloud providers levy traffic charges, it’s difficult to make the data center network look like a true extension of the cloud. The fact that one involves a hosting charge and the other is a capitalized resource isn’t an insurmountable problem, because decisions on where to put components within the data center are already made (by Kubernetes, for example) based on resource cost. The problem is that traffic volume is a difficult factor to predict, which means that if hosting choices have to consider it they’d likely be wrong. This means that there’s a persistent barrier to elastic movement across the combined “virtual data center”.
Economic factors enter into this in other ways too. Well over three-quarters of enterprises believe that economies of scale in hosting improve continuously as scale increases, which is not true. Almost three-quarters of enterprises think they pay too much for the cloud, and cloud providers are under pressure from Wall Street to sustain past levels of revenue growth. It’s hard to see all of these factors uniting in a way that encourages acceleration in cloud adoption, and it’s at least possible that the well-publicized examples of companies doing a reverse cloud migration are the tip of the iceberg. If that’s the case, then does the case for the subduction of networking into the cloud dissolve? Or are we perhaps simply seeing a transformation of the network model more than of the network host?
Enterprises have no real input on those questions today; they involves too many imponderables. They’re also questions that the dominant enterprise vendors aren’t particularly trying to answer. We’ll probably have to wait until vendor articulation and enterprise input matures before we can answer the questions with any certainty, but I think the cloud and the data center will logically merge and that “the enterprise network” will become a service of what comes out of that merger.
