We are building adaptable, intelligent factories, hospitals, and smart cities—Xilinx SoCs power scalable Industrial & Healthcare IoT platforms integrating safety, security, communications, control, vision, and AI. We will be exhibitors at SPS IPC Drives 26-28 November 2019, H4-558 and we welcome the visitors to stop by.
Data Gravity or Anti-Gravity
For the past few years, the technical community has been attempting to espouse counter arguments to the concept of “Data Gravity” introduced by Dave McRory. The theory, simply put, is data has mass and has led many to conclude that it’s less expensive to move processing to the data than it is to move data to processing. Cloud computing offers a plethora of applications and services, but it comes at a cost of data transmission, storage, and computation, which can easily be in the tens of thousands of euros per month for even a small factory. Of course, if data insights can earn you or save you hundreds of thousands of euros per month, the economics are simple—but still, there is something enticing Industrial and Healthcare IoT architects to act. Specifically, in traditional machine vision applications, the status quo means that data is captured in a camera, transferred to a frame grabber, and processed in an Industrial PC or perhaps in the cloud. Alveo™ acceleration cards from Xilinx, such as the U50, enable you to create smart frame grabbers with capabilities beyond those found in most industrial PCs. With a simple addition of the PCIe card and some C++ language or TensorFlow configuration of the hardware accelerated algorithms, machine learning neural networks or otherwise, you can bring cloud capability on premises. Going a step further in the name of data gravity, moving processing upstream to the camera enables a more efficient use case of sending just the calculated outcomes to a human machine interface (HMI) rather than lazily sending everything upstream and letting the PC figure it out. This dramatically simplifies bandwidth requirements, complexity of cabling, and the amount of equipment required for data analysis, not to mention speeding up the entire process and saving cost. Indeed, the recognition that most data has a shelf life and, like fresh fish that has been flown in from the coast in the morning, there is a market for something more actionable than the alternative. For those that have the most to gain from the Industry 4.0 revolution—Industrial, Medical, and Security applications among others, latency is what drives processing and the consumption of insights toward the data source. If you can save cost and monetise a rarer form of data, that’s more than enough justification.
Let Physics Dictate the Solution to the Problem
A further frustration to the Data Anti-Gravity theorists defending their position is the fact that cloud computing is migrating to what is known as the “infrastructure edge”. The “edge” – and all its variants – is a murky term, with point-of-view and application dictating exactly where this mythical place is, if it is even a place at all. For most IIoT and HcIoT purists, the edge is the most extreme place you can put data, the analog-digital boundary, where the physical world meets the digital world—in a machine vision or embedded vision use case, this is the camera. The cloud is slowly moving in this direction because it can conquer much of the consumer requirements from its traditional home base, but as it sets its sights on the next horizon, deeper integration with factories, hospitals and the like, it must mobilise to access latency-critical data. Rather than an “either-or” mindset, let’s adopt a more ambidextrous approach to solving the problem. Let’s put processing near the analog-digital boundary and let’s also make processing available in a centralised place and allow applications and services to migrate to where they are most valuable. A can of beans will be just as satisfying in a month if that is what your appetite calls for, but when looking at the fresh fish example, this certainly is not the case. This way, the problem that needs to be solved dictates the resources that are utilised for the highest performing, best monetising, and lowest cost approach. Amazon IoT with their Greengrass offering, Microsoft with their Azure Edge offering, and others are enabling this today. A great example of this is what Kutleng Engineering is enabling to stop poaching of the endangered African Rhino. According to Benjamin Hlophe, Director Technology Operations at Kutleng, “In the last decade, about 9,000 African Rhinos have been lost to poaching. Xilinx and AWS are enabling Kutleng to combat poachers in Kruger National Park through a new series of AI cameras powered by AWS IoT Greengrass and SageMaker Neo running on Zynq® UltraScale+™ SoCs.” This is edge AI being used for good.
Pivot Faster to Respond to the Industry 4.0 Potential
The world we live in is ever changing, and certainly the questions asked are always evolving. Pivot faster to answer new questions at the edge with a scalable embedded edge platform. The traditional approach was to build custom electronics for each new product—with disparate processor architectures, operating systems, and connectivity. Due to the complexity of IoT, the progressive choice is to choose an embedded electronics system-on-chip platform that is adaptable to serve the needs of a variety of applications, can streamline the proliferation of insights for cloud application developers via a common architecture across multiple product lines, and has the performance/watt attributes to credibly transform data into insights at the edge. Especially in Industrial and Healthcare IoT, this platform needs to perform operational tasks like functional safety, real-time vision and control, industrial networking, machine learning and IT tasks like cloud connectivity, cybersecurity, deploying containerised applications, and others.
Built to Last
Unlike traditional cloud computing equipment, which has a typical shelf life of 3-6 years, most industrial and healthcare assets are expected to perform for 15-25 years. This time-in-market creates another set of challenges that a scalable embedded platform needs to address: reliability over harsh conditions, longevity of supply, and adaptability to respond to evolving cybersecurity threats. Documented performance and a proven track record is the best assurances for the former, but for the latter, the ability to adapt both software and hardware offer the best response to the uncertainty of the future—a future that clearly relies on edge computing in the camera.
For more information on Xilinx’s Industrial Solutions across Edge and Cloud Visit: https://www.xilinx.com/applications/industrial.html