IoT architecture layers and topologies are in a state of flux; how IoT grows and what options emerge for new network design will have a tremendous effect on how IoT expands as the world moves forward.
No proliferating technology — not telephones, televisions, automobiles, computers, video games or even cellphones — has ever grown and spread across the landscape as rapidly as IoT. Of course, computers, video games and cellphones are all, by definition, now part of IoT. IoT even absorbed innocuous devices, such as wristwatches, thermostats, video cameras, door locks, and electrical sockets and switches in buildings.
By the end of next year, the number of devices connected to the internet will approach 40 billion. The problem is that much — if not most — of that data is bound for clouds. The number of IoT devices grows by billions every year, but the number of cloud platforms or additional servers within existing clouds only increases at less than 1% of that rate. That’s a serious traffic problem in the making.
IoT proliferation doesn’t just create unprecedented digital traffic and all the associated problems; it creates entirely new categories of security concerns because it expands the attack surface by an order of magnitude. The growth increases the need for cleaning and validating data before it even reaches the cloud. IoT expansion has opened up opportunities for new application categories that need machine learning and AI beyond the enterprise firewall, and IoT requires new standards for data usage and storage. Put simply, the need for new and better IoT architecture layers and topologies has never been greater.
Edge computing addresses IoT architecture layers’ needs
The existing IoT architecture is based on a tried-and-true model that is highly scalable and extensible, and it accommodates a broad range of topologies. Even this entrenched model, which consists of three IoT architecture layers, is currently undergoing big changes. The three layers are the following:
- Device layer. This is the client side where all devices, including sensors, switches, actuators and cameras, gather or respond to data live.
- Gateway layer. This layer congregates data from IoT devices and jumps onto the internet or terminates in a data acquisition system. Analog-to-digital conversion of IoT data often happens in this layer.
- Platform layer. This pathway connects clients and operators, often terminating in a cloud or a data center.
These three layers don’t include several resources that IoT needs. Organizations need better tools for local data analysis and routing. IoT generates more data than traditional networks by orders of magnitude; it is prudent to screen unnecessary data before dumping it in cloud or data center storage. It’s also necessary to figure out what data goes where as it’s created if the data is going more than one place, rather than sending it to one destination and then moving it again to another.
Many IoT networks require enhanced processing to respond to IoT applications in real time. There isn’t time for data to make a round trip to a cloud. Examples include support of driverless vehicles and facial recognition. Local processing resources make this technology practical.
Increasingly, IoT networks become intelligent and respond to activity in the real world. Here, too, there isn’t always time for cloud traffic. It’s more effective to have machine learning and AI resources embedded in the architecture, especially when the data for machine learning doesn’t have a separate use in the cloud.
To meet these needs, a new IoT architecture layer is currently emerging: the edge layer. Sandwiched between the gateway and platform layers, the edge is a new innovation in IoT architecture that provides all of the resources above.
Edge computing is the practice of placing servers in the field — often beyond the enterprise firewall and certainly beyond the physical server farms that host clouds — in the proximity of the IoT devices they support. This distributed computing paradigm is not new but represents an innovative answer to IoT problems. Edge nodes can provide in-the-moment data cleansing and routing, as well as real-time turnaround in complex applications with far less latency, and they make it possible to place machine learning where IoT lives. Nodes can also serve as a bulwark against IoT’s expanded attack surface, increasing the security of the gateway layer.
How to choose the right IoT topologies
Within this basic architecture model, an organization can use a number of topologies for distribution and interconnection of IoT elements. It’s important to make good choices here because these IoT topologies are specifically engineered to accommodate networks with specific uses. All of them interact with the gateway layer in different ways, and a bad design could limit the network’s performance, compromise security or even render some applications impossible. The most versatile topologies are point-to-point, star and mesh.
A point-to-point network has a one-to-one connection between nodes; communication only happens between the two points. This is the simplest and cheapest configuration. The downside is that point-to-point configurations aren’t scalable. No redundancy is possible, so there is no graceful degradation. Typically, this is just a simple connection, such as an earpiece to a cellphone or a single device accessing the internet at a single point.
In a star network, many nodes connect to a central hub. The cardinality of the hub is one-to-many. But none of the nodes connect to each other; they only connect to the hub. This configuration tends to have low latency and be consistent. Network tools can easily detect and isolate faults. On the downside, although reliability is generally high, there is no rerouting if interference occurs. In addition, the hub represents a single point of failure for the entire network. For example, home Wi-Fi has many devices reaching the internet through a single router.
A mesh network includes multiple device, gateway and router nodes. Mesh networks have high scalability and redundancy, as well as exceptional fault tolerance. The downside of mesh configurations is considerable complexity and a high maintenance requirement with increased latency from the multiple hops data packets must make. Mesh networks include industrial automation, large-scale fire monitoring and security, and energy management systems.
To decide which IoT topologies will work best, consider the complexity that can be supported and prioritize latency, fault tolerance, reliability and whether the network needs to scale.
IoT architecture layers and topologies make a well-organized mess
Once an organization decides on an IoT network’s architecture and topology, additional questions will arise. What are the best practices involved in managing the network? How does the new network affect the security of the systems it is interacting with? Who owns what?
IoT network administrative standards and best practices have been established, but like IoT architecture itself, they constantly evolve. The truth is there’s a make-it-up-as-you-go factor in IoT. On the technology side, this method is considered a good thing. Siloed, do-it-all platforms have had their day, and the organic nature of IoT portends a more versatile digital landscape. It calls for greater planning, diligence and innovation than have ever been required before.