The Internet of Things (IoT) is in a nascent form at the moment, but it won’t be long before the world becomes enmeshed in an immense network of connected, intelligent devices. The complexity of this infrastructure, not to mention the workloads it supports, all but guarantees that bandwidth, throughput and flexibility will drive IoT networking going forward.
In this light, there is a lot at stake in the way the enterprise lays the groundwork for its future virtual networking infrastructure, both within the data center and over the wide area. The function of networking, after all, is to connect disparate data points. But what happens when those data points pop into existence in a flash and then just as quickly disappear into the ether? And how should intelligent devices interact with an intelligent network to support intelligent applications and services?
The IoT will most likely make good use of software-defined networking (SDN), network functions virtualization (NFV) and other forms of virtual networking, says RCR Wireless’ Nathan Crawford. He believes this will go a long way toward redefining the human/machine relationship. A recent report by Dell EMC noted that SDN will be critical in the evolution of artificial intelligence (AI) from today’s novelty to tomorrow’s behind-the-scenes technology. Only through a flexible and automated network topology will smart devices and personal digital assistants be able to take on the myriad mundane tasks that make up a typical day and allow humans to assume the role of “digital conductor.”
The key question at this point, says The Next Web’s Ben Dickson, is how to bring artificial intelligence from the cloud to the edge. Only in the past few years has compute and storage infrastructure evolved to the point where AI can be supported outside of massive systems at universities and scientific research organizations. Pushing this level of number-crunching to the IoT edge will require something along the lines of a blockchain-like distributed computing platform that harnesses the power of legions of idle devices. As well, we will likely see a strong push for AI coprocessors throughout the IoT data chain, and possibly a new generation of AI algorithms that dramatically lessen the data load.
Microsoft, for instance, recently announced Project Brainwave that seeks to deploy the Intel Stratix 10 FPGA on data center networks in support of real-time AI processing. The idea is to enable deep neural network (DNN) microservices on the hardware level, which removes much of the processing overhead from CPUs in the server. This “soft DNN” approach is expected to exceed the performance of hard-coded DPUs because the FPGA is able to process AI requests as fast as the network can stream them.
An even better approach, however, is to implement AI on the IoT device itself. Qualcomm recently acquired a Dutch company called Scyfer BV, which specializes in AI platforms for a range of industry verticals. Qualcomm is hoping to leverage Scyfer’s expertise to implement AI on smartphones, smart cars, robotics and other end points in order to provide AI functionality without networking or even a WiFi connection. By pushing AI to the device, Qualcomm says it can provide immediate response to queries or changing conditions, as well as better privacy protection and more efficient use of network bandwidth.
Networking is the heart of virtually all data architectures, since it is the exchange of information that drives real value in most digital processes. But the intelligent IoT is unlike any networking challenge that has come before, both in scale and complexity.
The technologies under development now are a good start, but it’s going to take a lot more effort and ingenuity to realize the all-encompassing digital ecosystem that today’s IoT platform providers are promising for the future.
Arthur Cole is a freelance journalist with more than 25 years’ experience covering enterprise IT, telecommunications and other high-tech industries.