System on a Net
Locality of Reference Matters
In many ways we have spent the last twenty years building everything into a virtual world. The vast web ecosystems has transformed not only the way that we as consumers interact with services and our friends, it has changed the way business is done. Companies can stitch together experiences and services from building blocks provided by others. Small teams can reach hundreds of millions of subscribers almost overnight. The Internet became the Web became the Cloud.
Fast forward to now where the device side of the world is becoming more than just a window into the virtual world. We are connecting new kinds of things at the edge of the network. Devices that used to be solely physical are getting computing power and connectivity. The connectivity can be used for media (replacing wires) or control (replacing hard wired systems with software).
Unfortunately this device connectivity has historically developed in silos. The physical layers (the radios) have often been designed in a fashion where upper layer protocols are intertwined in an inextricable fashion. We describe the connected things by their form of connectivity: Bluetooth headset, Zigbee light switch. We never refer to our Cloud in this way. We don’t exclaim, “that’s a great fiber channel server”. The Internet always kept the physical connectivity out of the way of the developers so that each layer (physical connections and logical services) could innovate at their own pace. Those strong abstraction features need to be brought to the edges of the network.
So here we are. Working at the transition of what will be the next phase of the Internet and the connection of billions (or trillions) of new devices to the network. What impact will this have on the protocols and basic building blocks such as identity, security and privacy?
The answer is unclear. There is an enormous amount of interesting research and commercial effort in this area. At one end, there is the approach of extending the design metaphor of the cloud to new devices by creating proxies for each device in the cloud — in this architecture, the connections are now virtual so you can wire each and every thing to anything else. The “IFTT” service is a leading example of this approach that can quickly bring new experiences to consumers by connecting different services via the Cloud.
The best example I can point to at the other end of the design spectrum is the work around autonomous (usually swarming) robots. Companies like Fluid Robotics and others are making devices that crawl, swim, fly or walk and generally can coordinate with each other. In these intense real-time environments most designers have realized that you can’t go back to the Cloud for all the communication and instead these systems use local communication meshes to share sensor data, route planning, and otherwise drive coordinated behavior.
Get Rob Chandhok’s stories in your inbox
Join Medium for free to get updates from this writer.
I believe that somewhere in the middle of these two design patterns lies the hybrid approach that will come to consumer ecosystem faster than many might predict. By hybrid I mean a blend of local computing and communication that is thoughtfully connected to the Internet through well-defined and secure protocols resulting in a topology that is not unlike a system-on-a-chip (or “SoC”). In this topology, if data and software do not need to leave the local environment, they don’t have to. In a SoC you keep things local for many reasons with the chief ones being low latency access and power savings. In addition, the physical nature of the connection can often affect the security of the overall system. This is true in an SoC and it is true in a local context — something that we could call “system on a net” or “SoN” (even though I know that often means “self organizing network”).
What are the characteristics of a SoN?
- Locality of reference drives the topology.
Think about the history of computing processor design. When the first CPUs were built, the access to memory was quickly recognized as a limit to any increase in overall computing speed. Over time, this has turned into incredibly complex and hierarchical caching systems, from the pipelines in the CPU themselves to the level 1 (or 2 or 3 or 4) caches on the chip itself. Multi-core processors have to manage shared caches to drive parallel computing correctly. The same thing is necessary in local device communications. We are going to find ourselves wanting things to happen faster and faster as developers create new services and experience. We will want sensors and lights and media to move quickly and react to new kinds of software command and control. That simply cannot scale without managing the communication and data in a local-first way.
2. The topology can lead to better security by design.
A basic principle of security is to reduce the surface area by which an attacker can gain entry. If every device in a home, office, car, or train connects to each other via a connection to the cloud, the surface area of the system becomes so large as to affect any fundamental security design. In the home in particular this is troublesome as consumer devices from around the world are unlikely to maintain a high level of security quality in commodity devices. So a design that bounds the impact of a semi-trusted device would be an important design principle. I don’t want that cool connected toaster to be the infection path for my entire connected home. In addition, keeping communication local as much as possible provides a much better privacy by design baseline. In the same way we sandbox third party applications on our smartphones, we need to sandbox connected devices to reduce the opportunity for exploitation.
3. The Cloud remains the same useful interconnection point, but isn’t the only one.
Saying that there is only local connectivity is as ridiculous a design goal as saying there is only Cloud connectivity. Each SoN will have well-defined interfaces to the broader world. These include connections to the Cloud using, most likely, Cloud protocols like HTTP/JSON/REST and WebAPIs in general. This is analogous to the complex connections a system-on-a-chip has to the other components on the board near it. Many of these (if not most) are industry standard interconnections for memory, peripherals, and connectivity. But if you want something to run fast, you keep it local as much as you can. Games want to run in the GPU without touching external memory. The sensor “SoN” of the future will want to do the same as it takes that sensor input and applies deep learning to bring you better and better experiences. This deep learning can be computed locally and used locally.
Whether the SoN in question is in your office, your home, your car, or on your body the locality of reference is going to matter more and more. The companies that recognize this design pattern is coming and move towards the hybrid (and admittedly more complex) approach will be able to reach past their competitors in this new highly connected world.






