Welcome to ZDNet Scale, a new blog that looks at the infrastructure of the modern economy. If data is the new oil, the data centers behind the cloud, IoT and big data are akin to the bridges, canals, highways and power grids that move society. We'll use maps as metaphors for this tour of next-gen infrastructure. Let's get going.
Within an industrial park in a major metropolitan U.S. city, in the midst of an excavation project on one side and a bit of road construction on the other, rests a spotlessly clean, unmarked metal shed. If it were a garage, you could maybe park five cars in it. There's no activity here at the moment, but your first clue that it's not abandoned is the continual presence of security.
It's likely not the first image that springs to mind, or even the twelfth or fifteenth, when you consider what the touchpoint for the next revolution in information technology should look like. You might be more likely to conjure Apple's brilliant new, spaceship-like headquarters, one of Salesforce's new skyscrapers, one of Digital Realty's sprawling co-location complexes, or perhaps Microsoft's colossal, 700,000 square foot Azure complex outside of Chicago. Usually when one thinks of something in "hyperscale," one pictures very big, very solid, objects.
Data Center in Hand
Yet consider this: You hold in your hand a data center. The part that touches you directly may rest on your desk, fit in your pocket, or perhaps even sit on your wrist. What connects it to the part that makes it functional -- the servers and their processors, the network and its fabric, storage and its arrays -- is no less substantial than what connects those components to each other. The connections you can't see are the ones responsible for the applications you use.
So keeping in mind that scale can be an ethereal thing, consider once again this small, nondescript metal shed. Imagine it adjacent to a cell phone tower. If it's like many such towers in America, the local city council has probably ordered its site owner to disguise it like a pine tree. Place a small chunk of a data center inside this shed next to a hundred-foot fake tree. Then multiply that shed in your mind by hundreds of thousands.
Each of those micro data centers, as they're actually called, would instantly have the virtue of 4G, and soon 5G, connectivity. It may already be connected to some nearby enterprises by fiber optic cable. Electricity would already be both plentiful and available, with backup generators already on-site. And the classes of virtual infrastructure systems that already empower the applications running in private and hybrid clouds, could effectively weld all of these stations together into a global network mesh.
"How do we push those cloud capabilities to the edge of our network?" asked Egal Elbaz, AT&T's vice president for ecosystem and innovation, probably already having a good idea of what his answer would be. "Because we have the footprint, and because we have the evolution into 5G, and because ultra-low latency and high throughput are going to become characteristics of our network, and our ability to orchestrate in real-time, based on demand and consumption... then all of a sudden, you can see this vision."
Now we are talking scale.
Point of Embarkation
With this first edition of ZDNet Scale, we begin a kind of expedition beyond the surfaces of our devices, and onto the next frontier of the modern economy. Our forebears built roads, bridges, canals, aqueducts, highways, power grids -- the infrastructure of our society. Today, modern engineers, developers, architects, and operators are building the new support structures for an emerging society.
Special feature: Cloud Computing: Moving to IaaS
Our gathering point for this journey is on a kind of map. It's a way of visualizing the many aspects and facets of the world that we're taking part in building, to try to bring them into more familiar territory, if only conceptually. The places on this map are representative of the things we try to accomplish with technology, and the tools we use to do it. And the routes between those places are indicative of the ways we've tried to achieve these goals, yesterday and today.
Everything on this map is part of the modern data center. Some of it is "here" in your modern enterprise, some of it "out there" on the platforms to which your enterprise has outsourced some responsibility. No single part of this area is "the cloud," because the cloud can conceivably connect it all, and the cloud can also be "here" as well as "out there."
The river on this map is the gap between those IT assets you own and those you don't. Many of the resources and assets your enterprise needs to be functional, lie across that river. The best connections we've been able to make between the two sides, is represented on this map by a single bridge.
Waypoint #1 is at the base of a cell phone tower. And if we were to look at what's there right now, literally rather than figuratively, we'd find a toolshed.
The chamber of secrets
Inside this toolshed is a gunmetal grey cylinder encased in windowless door panels, resembling a rejected redesign for Doctor Who's TARDIS. Its name says as little about its purpose as the shed it's sitting in: a "Vapor Chamber." It's just small enough to be carried in an "Oversize Load" trailer along the highway. In fact, it was designed so that if a company wanted to purchase just one, it could be installed in a toolshed in the parking lot, or perhaps near the HVAC unit. There, from this discrete vantage point, it could replace the regional data center, or take the place of the public cloud.
The Vapor Chamber's producer is a startup company you might have seen before in ZDNet: Vapor IO. Its charismatic CEO is Cole Crawford.
"We tend to think about the edge as working backwards from the geography of data," Crawford told ZDNet. He described for us a world where latency -- the lag time in transmission -- is no longer noticeable from any location. In such a world, a social network user in rural India would perceive the same level of connectivity as another user in the middle of China, and another in San Francisco. In those localities where latency would normally be a noticeable factor, and geography would contribute to that problem, both data centers and bandwidth are brought closer to the user. Proximity compensates for physics.
"5G will be a forcing function to a much more localized Internet," said Crawford, referring to the suite of technologies that are expected to collectively replace 4G wireless communication. "If we just work backwards from the physics of the speed of light, from the amount of glass or fiber in the ground, in the form of MAE-East and MAE-West [the main Internet exchange connection points in North America] in the information highway, as it were, there isn't enough glass in the ground right now to transport the amount of traffic and data that will be here by 2025 -- 175 zettabytes of data, and 40 or 50 billion connected devices. The QoS on that, given our current architecture, would be atrocious."
"Latency is really a measurement of distance. There's natural laws of physics there that influence that latency, and there's nothing you can do on the network necessarily to change that significantly."— Scott Sneddon, Senior Director for SDN, Juniper Networks
The largest third-party owner of cell phone towers in North America is a company called Crown Castle. It leases its tower space and facilities to all the U.S.' leading carriers: Verizon, AT&T, T-Mobile, and Sprint. Earlier this year, Crown Castle entered into an astonishing deal, purchasing a major stake in Vapor IO. The two then launched what they call Project Volutus.
According to Vapor IO's announcement, "Project Volutus customers lease space, power, and networking in each of the desired cell tower locations and supply their own hardware to be rack-mounted in the Project Volutus facilities. Project Volutus provides the ability to directly cross-connect with the wireless network and also provides a fast fiber path back to the customer's centralized or regional data center, or an interconnect point."
The Vapor Chamber is, in this case, Volutus' preferred destination for these rack-mounts. Once populated with the servers of the customer's choice, it becomes a micro data center. You may be familiar with how Amazon, Microsoft, and Google built out their cloud service centers from servers assembled in shipping containers. Micro data centers (µDC) are the next generation of that concept: smaller still, more commercial grade, and hopefully at some future date, battle tested.
Vapor IO is not alone in introducing this concept; last September, colocation services provider DataBank partnered with cell tower site owner Vertical Bridge on a project to deploy µDCs in their own equipment sheds. Hewlett Packard Enterprise began seriously investing in the micro data center concept last year, along with new classes of servers such as Moonshot that may be better suited for edge deployments. In early October, HPE started phasing out its production of generalized, hyperscale-style servers, in an effort to devote more attention to what it calls the "intelligent edge."
Control of the river
How a µDC is designed and implemented will have a direct impact on where the "edge," in edge computing, ends up being located. Some vendors are saying there is, or will be, more than one edge: for instance, a "mobile edge" and a "compute edge" to accompany the "radio edge," where wireless transmission takes place. But the era of technology adopting weird and contorted shapes to accommodate vendors' own plans, may have already ended.
Scott Sneddon is a senior director for software-defined networking at Juniper Networks. Speaking with ZDNet, Sneddon offered this explanation of the problem edge computing -- and with it, µDCs -- seek to solve: specifically, latency.
"Let's assume everybody's building big, high-capacity networks -- which is largely the case -- and let's assume that there's plenty of bandwidth everywhere -- which isn't always the case, but often is, especially when we talk about interconnects between cloud data centers and large enterprise data centers. Then latency is really a measurement of distance. There's natural laws of physics there that influence that latency, and there's nothing you can do on the network necessarily to change that significantly."
If we could oversee the theatre of operations, if you will, the way Shelby Foote would oversee Chattanooga or Gettysburg, we'd see Vapor IO moving communications assets south, away from end customers and back toward the enterprise. The company foresees the building of a direct route for functionality, between the tower at the edge and the hyperscale data center at the heart of the cloud.
Meanwhile, across the river, enterprise IT is deploying its own hardware assets once again, this time a new generation of servers designed to run certain classes of workloads -- for example, real-time monitoring and analytics -- closer to the things being analyzed: coolant temperatures, tensile stress, human heartbeats.
For the second waypoint in this four-point journey, we'll introduce you to a major service provider that believes it may have already solved that problem. It has already deployed an edge network -- one which you're using right now, to read this very article.
Waypoint #3 will explore how the applications that serve the world's computers and devices, via data centers and the cloud, are catalyzing a kind of subdivision of the cloud into quality-of-service tiers. And at Waypoint #4, we'll end our journey with an examination of whether the servers we're building today are up to this task, or if we'll need new designs there as well.
Until next time, hold tight.
Journey Further -- From the CBS Interactive Network
- "The future of serverless cloud looks a lot like physical servers" by Matt Asay, TechRepublic
- "Vapor IO introduces the Open Data Center Runtime Environment" by David Chernicoff, ZDNet
- "The enterprise technologies to watch in 2017" by Dion Hinchcliffe, ZDNet
- "Vapor IO and the Emerging Market of Edge Computing" by Maxwell Cooter, The New Stack
- "HPE Computing for the Future: The Cloud, the Edge, and the Machine," by Scott M. Fulton, III, The New Stack
- "Vapor IO to Sell Data Center Colocation Services at Cell Towers" by Yevgeniy Sverdlik, Data Center Knowledge