- Can a startup build global WAN facilities using nothing but the cloud? It may be impractical, but economics are trending toward possible.
- Despite serious roadblocks today, the idea of sourcing/scaling facilities on-demand, and using them on a pay-as-you-use basis, seems tempting.
As a thought exercise, imagine what network function virtualization (NFV) will look like when the technology reaches its theoretical end state. Instead of switch/routers running on purpose-built hardware, the routing function could run entirely on software in virtual machines. These virtual machines in turn run on generic high-performance computing platforms located in large regional data centers and strategically distributed worldwide. What’s more, those large regional data centers happen to have plenty of competitive fiber linking them. That means plenty of carrier competitors to offer commodity priced, flexible high-performance switched Ethernet connectivity between sites. The result is a completely virtual global WAN operator, one that can ramp up and tear down both router horsepower and corresponding capacity, based on customer need. There would be no networking sunk costs – the primary investment would be in orchestration software, and in operations support/billing systems.
There are plenty of reasons why this model doesn’t work – at least not yet. For starters, software routers lag in performance over purpose-built hardware at high speeds. Also, if a cloud provider charges for data ingress/egress, pushing constant, large volumes of traffic across a VM router is going to break the model. In fact, the pay-as-you-go model probably won’t be as cost-effective as a more static, longer-term contract. The theoretical virtual WAN services company can offer dynamic service between regional data centers – but to connect to other locations will mean going across more static services, whether it sources WAN services from other providers or supports connectivity over the public Internet.
But it’s already possible for someone to become a completely virtual provider of a global WAN, working only through Equinix and other major data center facilities owners. If the economics don’t yet line up yet, pricing from public cloud providers continues to erode steadily. Major public cloud providers such as AWS or Google likely have little interest in also becoming retail global WAN operators – too many other complexities involved in that business – but they could provide a foundation for a specialist with carrier ambitions. Whatever company were to decide to try this model could make hay, marketing (misleading conclusions or not) that other network providers connect to the cloud, but its routers are located inside the same cloud services that customers want to reach. Certainly there would be plenty of opportunities for such a provider to bundle other NFV functions for clients: centralized managed customer router instances, firewalls, WAN optimization, load balancing, etc.
Fortunately for network providers, this is all purely a thought exercise. There are plenty of reasons why this model couldn’t work, as it wouldn’t be cost-effective. For now.