EdgeNet: A Global Cloud That Spreads by Local Action Justin Cappos NYU Tandon School of Engineering NYU New York, NY jcappos@nyu.edu Matthew Hemmings US Ignite Victoria, Canada discount.yoyos@gmail.com Rick McGeer US Ignite Orinda, CA rick.mcgeer@us-ignite.org Albert Rafetseder NYU Tandon School of Engineering NYU Vienna, Austria albert.rafetseder@univie.ac.at Glenn Ricart US Ignite Salt Lake City, UT glenn.ricart@us-ignite.org I. INTRODUCTION EdgeNet is a distributed edge cloud, in the family of PlanetLab[1], GENI[2], Canada’s SAVI infrastructure[3], Japan’s JGN-X[6], Germany’s G-Lab[5], and PlanetLab Europe. EdgeNet is designed with the experiences of its predecessors in mind, and has chosen design features to avoid the pitfalls that befell earlier systems. Moreover, edgeNet takes advantages of recent technological advances unavailable to its predecessor systems. EdgeNet is a software-only infrastructure. Previous infrastructures all used dedicated hardware, which had three major negative effects. ● Scalability was limited. PlanetLab eventually achieved 700 sites, but maintenance of those sites was a full-time occupation for a team of six, largely troubleshooting node failures. ● Hardware refresh formed a major expense for these systems, since about ⅓ of capital costs are required annually for hardware refresh. ● Local system administrators found maintenance of special-purpose hardware a burden. A. Why an Edge Cloud? EdgeNet is the platform for a new class of applications and services: Cloud-in-the-loop systems. Lightweight devices (smartphones, sensors, actuators) interact with the real world (things and people) and can do some lightweight computation, but their computational power is soon exhausted. Conversely, Cloud systems are arbitrarily powerful, and so the natural pairing is lightweight devices with Cloud systems: devices offload storage and computing to Cloud nodes. However, Cloud systems are sparse; Google Cloud has 11 Points of Presence in North America in five regions (a sixth is planned). Distance is time; device-Cloud latency is on the order of tens of milliseconds or more. This means that a Cloud node and a device can have at most a few transactions per second, severely limiting the classes of applications which feature device-Cloud interaction. Reducing the latency between device and Cloud opens up a broad array of new services, including data-intensive visualizations, virtual reality on thin devices, prediction and analysis for the Internet of Things, including real-time video processing and analytics. B. Why Software-Only? EdgeNet nodes are VMs which run on general-purpose computing nodes; these may be local Clouds, servers, or even personal or embedded computers running the appropriate software. EdgeNet is software-only because edge Clouds must be orders of magnitude larger than existing commercial Clouds. This is due to a theorem from physics and plane geometry. Getting a Cloud node within k milliseconds of a point on a plane requires that the Cloud node must be within k/c meters of the point, where c is the speed of light. Thus, getting a Cloud node within k milliseconds of anywhere requires covering the plane with circles of radius (k/c). The area of a circle is proportional to the square of its radius, so this amounts to covering the plane with circles of area (k/c) 2 . Since the area of the plane is unchanged, reducing k by a factor of x requires increasing the number of circles (and thus the number of POPs) by a factor of x 2 . Cutting latency by a factor of 10 means 100 POPs for every Cloud POP today. This is an enormous and slow undertaking, if done as a Cloud with dedicated hardware and bespoke software. However, there is another model for building infrastructure. In the late eighties and early nineties, digital libraries appeared to he a massive undertaking, with enormous disk farms and city-scale computing. There was no centralized, massive buildout; rather, a simple piece of software was made available for download by the National Center for Supercomputing Applications. Within a decade, the World Wide Web had spread to millions of sites and created a digital library of massive, worldwide scale by the local, individual action of millions of individuals and organizations. EdgeNet’s vision is to replicate the experience of the World Wide Web: to build a world-girdling, always-available, next-to-you-wherever-you-are Cloud through the actions of millions of individuals and organizations. XXX-X-XXXX-XXXX-X/XX/$XX.00 ©20XX IEEE