Remember the scene in Jurassic Park where T-Rex is chasing the jeep with Laura Dern and Jeff Goldblum? A quick glance reminds us that “OBJECTS IN THE MIRROR ARE CLOSER THAN THEY APPEAR”, and that T-Rex is breathing down our necks.
That’s kind of the way it is with edge clouds. They’re closer than we think. The good news though is that they’re not trying to eat us for lunch.
The edge is actually already here
Edge clouds have started popping up in the form of virtual customer premise equipment (vCPE) solutions, providing traditional services like edge routing, SD-WAN, firewalls, intrusion detection/protection, load balancing, and such. And in the Telco space we’re seeing more specific solutions like virtual evolved packet core (vEPC) and virtual IP multimedia systems (vIMS), and network slicing with 5G.
For the next several years we will see an explosion of these services as virtualization and orchestration solutions continue to mature, coupled with the continued acceptance of whitebox devices as a general-purpose platform for both networking and compute.
The next edge will be even more pervasive
With the proliferation of the internet of things (IoT), edge clouds will move even closer, well, to the edge.
Current cloud models (public, private, hybrid) have proven themselves for traditional IT services where scale up is a key driver. But IoT flips that model on its edge, and scale out becomes the driver. Instead of thousands of cloud centers with millions of servers, we will need millions, maybe even billions, of edge clouds with tens (or even single digit) of servers.
Data and decisions drive the edge
Traditionally, IoT has been thought of as little single-purpose devices (e.g., sensors) that typically sit in bigger, yet highly distributed devices. These little IoT devices collect some data and either report it based on an event or condition, or periodically send it somewhere to track usage or capture empirical data.
The next generation of IoT dramatically changes all that. IoT becomes very data intensive. Hyper-intensive in fact for things like intelligent cars, immersive gaming, real-time recognition, security monitoring, and next-gen industrial control. Along with that data comes the need to make decisions in real-time (or very near). Many, if not most, IoT devices and applications will incorporate machine learning, which requires huge quantities of data to make decisions.
Moving all that data to a traditional cloud for analysis and processing simply won’t cut it. Even with the advent of 5G and the dramatic reduction in latency, the cycle for “sensing, analyzing and acting” will be too long if the data, and possibly additional decision making occurs in a remote cloud. The cycle must occur at the edge.
Why not just keep everything at the edge device?
With the progression of computing density and power, one might ask why not just keep everything at the IoT device. After all, the sense/analyze/act loop needs to be ultra-tight to facilitate real-time services – especially those for life-impacting applications.
Simply put, there’s too much data. The amount of data generated by most edge applications, and specifically those such as intelligent cars and imaging/recognition applications far exceeds current small-footprint storage capabilities (hard to believe, huh?). To make things work, the most critical data needs to be onboard. Secondary data can be moved to the edge. And least critical data for backend and deep analytics can be moved to traditional clouds.
Another aspect is manageability and maintainability. Updating applications, configuration data, and critical data stores in thousands of edge clouds is daunting, but still a lot easier and more reliable than doing so for millions, if not billions of end points.
Will we bury traditional clouds next to the mainframe computer?
Remember what you were doing the day client-server computing killed off the mainframe? Oh, wait… that didn’t happen, did it? Like mainframes, traditional scale-up clouds will continue to thrive long into the future.
Traditional clouds will play a pivotal role in the edge world. They will store or archive the massive amounts of data generated at the edge. They will provide backend services such as billing and mediation. But most importantly, they will provide deep analytics that will result in learning. And this learning will be returned to the edge where it will be used in the sense/analyze/act loop.
Edge clouds everywhere
Edge clouds will show up everywhere you can possibly imagine. As long as there is power and connectivity, the footprint of edge clouds will allow them to be placed wherever necessary to most appropriately serve their associated IoT end points.
Obvious locations such as cell towers, lamp posts, CORD centers (Central Office Rearchitected as a Datacenter), factory floors, sports arena, malls, etc., already have the power and infrastructure to support micro-cloud infrastructure. But again, with power and connectivity the sky’s the limit – think airplanes.
Edge clouds in the service life cycle
Edge clouds are a critical link in the life cycle of edge application. They enable IoT devices to focus on the critical functions of sensing, analyzing and acting by providing necessary compute, storage and networking resources at the right place and time. At the same time, they provide the gateway to backend services provided by traditional scale-up cloud and legacy services, enabling the deep analytics and learning that then feed back to the edge for improved application performance and enhanced user experiences.
Fueling the edge cloud engine
Building out the edge cloud will not be a simple task. While the IT industry has dealt with mega-scale infrastructure models in a scale-up world, it has not yet addressed the requirements of the hyper-distributed cloud landscape. While there are key industry initiatives underway such as ONAP (Open Network Automation Platform) and OSM (Open Source MANO), they are in early stages and currently fall short of delivering solutions that can be consumed by most service providers, operators and enterprises.
Regardless of the architectural approach, any solution for edge clouds will have to include at least three critical capabilities:
Hyper-scale infrastructure orchestration – reliable orchestration and management of millions of end points to support dynamic cloud services.
Service chain management – sophisticated policy management and workflow to enable deployment of complex service chains that span multiple network and cloud service points.
Integrated connectivity – seamless configuration and management of a wide variety of networking services to enable optimal connectivity between service points.
The road ahead
Back in 2013 I delivered a lecture series on the impact of digital technology for the Fromm Institute at the University of San Francisco. In my introduction, I shared that each year we are creating about 1.2 sextillion bytes of data – or the equivalent of about 320 times the contents of the Library of Alexandria for every person on earth. By all indicators, that number was conservative. Self-driving cars and LIDAR cameras are quickly proving my prediction wrong…
While we began by looking in the mirror to see what is closing in on us, the road ahead is what really matters. IoT is inevitable. Immersive computing will be everywhere and it will be hungry for data. It’s time to feed the beast – with edge clouds!