Bringing cloud services to the edge - Vapor is building a distributed AWS

IT is a discipline full of recurrent cycles, themes and trends that ebb and flow over the decades. Due to the implications on investment and system design, one of the most significant is the oscillation between centralized, consolidated infrastructure and distributed systems.

We’ve been in a period of consolidation as the de-centralization of the PC client server and departmental Unix server era has succumbed to the technology and economics of large, central systems running dozens of virtual machines (VMs) attached to a SAN hosting massive shared storage arrays. Indeed, the consolidation that started at an organizational level has been extended by cloud operators that have built what amounts to global infrastructure utilities.

Cloud economies of scale and enormous capital and R&D budgets make operating a private data center less and less wise, however there are already signs that explosive increases in the scale and sources of data generation, the highly dispersed nature of application users and the increased need for real time results are exposing limitations to the cloud utility model.

Edge computing is the reactionary response to the inefficiencies of copying an increasing amount of data to a central location for processing only to send the results back from whence the data came. While Edge computing means that many workloads will move out from centralized enterprise and cloud facilities, done correctly, it doesn’t represent a repudiation of the tenets of centralized infrastructure and workload management that are fundamental to the cloud model.

A significant and open question is the form edge computing will take and whether the model is amenable to the same sort of industrial scale buildout, service management and automated administration as today’s cloud. Vapor IO is a startup with a healthy new cash infusion that thinks the answer is yes and it is building an edge cloud system that combines the best of cloud and distributed computing in a managed service built for the era of 5G networks, mobile users and IoT data sources.

Before diving into Vapor’s technology, it’s worthwhile exploring how is it that a 3-year old company with a few dozen employees might convince investors that there’s a business opportunity in building hundreds of autonomously managed small edge ‘data centers’ across the country.

The market for keeping things local

Edge computing is usually associated with small servers running a virtualization stack for local workloads and virtualized network services at a branch office or remote work site, often called a data center-in-a-box. However, the nexus of connected smart sensors and mobile devices that granularly record ever more data, along with machine learning models that can make sense of the information, mean that for many applications, centralizing data and analysis in a corporate or cloud data center is less and less viable.

As I wrote earlier this year, in discussing Swim.ai, there are already companies moving processing, particularly real-time machine learning applications, to the data source. Gartner estimates that 10 percent of an enterprise’s data is currently processed outside a data center or cloud service, however it expects the ratio to be fifty-fifty in four years. By another estimate, the market for Edge computing products and services will snowball at a 41 percent CAGR through 2025.

While a growth level mirroring that of today’s cloud services seems reasonable, I’m skeptical of the estimate’s absolute revenue figures since combining its ultimate projection of $3.24 billion in revenue with 41 percent growth over 7 years implies that the edge market is currently a minuscule $300 million, or about one-quarter the size of the that for Lamborghini super luxury sports cars. Nevertheless, the company’s conclusion that Edge computing for mobile networks is by far the largest segment, and will remain so with the growth of 5G networks, provides a clue to the strategy Vapor IO is pursuing.

Vapor IO - Building the Kinetic Edge network

With the proliferation of 5G, mobile networks will constitute a growing source of data and traffic, hence Vapor IO’s strategy of turning mobile base stations into miniature data centers is the logical starting point for an Edge cloud. As company founder and CEO Cole Crawford described to me, they plan to provide rack space, power, cooling, telemetry and a carrier-neutral, software-defined interconnect using modular data centers, i.e. shipping containers, positioned at carrier base stations. Vapor calls its system the Kinetic Edge network and it consists of the following components:

• Vapor Edge modular data centers positioned at select wireless base stations that are entirely automated, require no staff, include multiple layers of physical and logical security and house racks that are leased to multiple tenants in quarter-rack increments.
• Vapor Chamber equipment enclosures for leased equipment that include top-of-rack switches, a Vapor Edge Controller running Vapor’s proprietary SDN along with routing and system management software. The circular enclosure, which resembles an early Cray X-MP, is designed for maximum efficiency and cooling and is capable of handling 150 kWatt of equipment that Vapor claims can “deploy upwards of 15,000 CPU cores in a 2.88 m (9.45-foot) diameter footprint.” The chamber has nearly 100 sensors to monitor and tune airflow and other environmental conditions that are connected to its Edge Controller to allow both local (to leased systems in one of the racks) and remote (at Vapor’s operations center) management.
• Edge Interconnection to multiple backbone networks via an Internet exchange point along with a Vapor IO fiber ring network that connects Edge modules in the same vicinity into a low-latency metropolitan area network (MAN) for redundancy and scalability. Connections also include direct links to each of the major cloud providers via services like AWS Direct Connect and Azure ExpressRoute. Vapor claims that a six-site edge network can surpass the reliability of a tier 4 data center, delivering an astounding 12 nines of availability. Crawford contends that such bulletproof reliability is essential for emerging Edge applications such as autonomous vehicles and other medical and industrial applications.

Although Vapor IO will have just 13 sites deployed in the Chicago area by the end of the year, it’s plans are ambitious to the point of grandiosity. The goal of its Project Volutus is:

To build the world’s largest network of distributed edge data centers by placing thousands of Vapor Chambers at the base of cell towers and directly cross-connecting them to the wireless networks. This will make it possible to push true cloud capabilities to within yards of the end device or application, one hop from the wireless network.

The fact that Crown Castle, the largest provider of wireless infrastructure in the U.S., including 40,000 cell phone towers and a large metro fiber footprint, is a significant investor in Vapor provides the best evidence that the Volutus strategy is far more than wishful thinking.

Crawford says initial customers will be large cloud providers, carriers, and web-scale companies, think cloud providers, social networks and streaming media services, that need localized network access, particularly to mobile users and don’t have access to direct access to wireless infrastructure. The Volutus blog presents an intriguing usage scenario:

A public cloud provider would partner with Project Volutus to deploy its own cloud hardware in thousands of cell tower locations. The cloud provider would decide how it to present those capabilities to its end user customers—for example, as “edge regions,” premium priced “edge VMs” or perhaps bundled into an IoT service. Project Volutus operates the remote facilities and provides customers with real time telemetry and control services to remotely operate their equipment and network via APIs and browser-based tools.

My take

Edge computing will evolve into something far more extensive than today’s typical enterprise example of branch office hyperconverged systems running local virtualized workloads. As existing cloud providers, both IaaS and SaaS, other online services and large enterprises realize the wisdom of keeping data sources and running applications closer to users, particularly those accessing via mobile networks, the notion of an Edge cloud built from micro data centers becomes compelling.

Between a conversation with Crawford, the little technical detail Vapor IO has released and its financial endorsement by a company with massive wireless investments and expertise, I’m convinced that Vapor IO has the foundations of a viable edge cloud business. The proof will be in its 100-node buildout over the next couple years. Should these win customers from technically sophisticated cloud and online operators like Microsoft, Facebook and Netflix, expect to see others emulate the strategy.

Indeed, these might include the mega-clouds themselves since Microsoft already has extensive experience with modular data centers and the networking expertise to build a software-defined network backplane that could span both its private fiber and multiple carrier networks. Likewise, wireless carriers like Nokia that are building 5G platforms could partner with another tower operator to create an edge cloud system similar to Vapor’s.

The technology and market development for Edge computing will be exciting to watch over the coming years. I admire the audacious vision of Crawford and his Vapor IO colleagues, however I believe they will soon have competition as the technical and economic benefits of Edge cloud computing become obvious.