Where is the data? Challenges for edge computing
Professor Diego Klabjan invited me to deliver the inaugural seminar at his Center for Deep Learning at Northwestern University. What follows is based on the seminar, which was broken into three TED-sized talks. Part One focuses on the five characteristics of a new class on enterprise applications: analytic applications. One of the key characteristics of analytics applications is they will be created from large data sets. So where will this data come from? While much has been mined from the Internet of People, there is much more data in the Internet of Things. While this has been obvious for a while, there remain numerous challenges to connecting to existing machines and collecting the data. Part Two is based on work McKinsey presented at their annual Sundance conference and highlights five key challenges. This is Part Two.
So where is the data to feed these new enterprise analytic applications? Some of it will be generated with the Internet of People, but far more of it will come from Things. My last in-person keynote speech was at the annual partner meeting of Morgan Lewis, a global law firm. My talk centered around 5G. One of the specifications for 5G is support of up to 1,000,000 devices in a square kilometer. So it got me to wondering, where is the densest population of people in the world? Maybe it should be no surprise, but it’s Mumbai, with a density of 30,000 people per square kilometer. Let’s assume each of them has a cell phones (probably not a big leap), then what are the other 970,000 connections going to be used for?
The only answer has to be Things. Things include cars, compressors, ultrasounds, HVACs, blood analyzers, front loaders, water quality instrumentation, industrial printers, etc. And by the way the Things have lots of data. This data includes machine data such as error codes, location, temperature in the room or the laser power level in the gene sequencer. Machines/Things also have nomic data, data the machine generates, genomic data or the gene sequence, agronomic data, the boron level in the soil. With machine and nomic data we could detect Covid from the wastewater, predict elevator failures, diagnose pneumonia in Africa, or detect Fusarium Wilt in Korean radish farms.
While the path is clear, the implementations have been difficult. Even today very few machines are connected. While it might be true that new machines are built to be connected, the challenge is how do you connect the existing 600B Things? The answer is to add a computer, what we’ll call and edge server, which can be programmed to talk to the machine (tractor, ultrasound of HVAC), can do some local processing and then communicate with either a public or private cloud service.
A few years ago I was invited to an annual conference held by McKinsey, focused on the Internet of Things, held at Sundance. It’s a small conference so you actually get some time to meet people. The 2020 conference was of course held online and among other presentations there was one, which focused on the challenges of connecting and collecting data from machines/Things. These challenges all resonated with me and the experiences I’ve had.
Challenge 1: Upfront Capex Cost
If you’re going to connect and collect data from things you’re going to need some kind of computer to connect to the Thing, collect the data and send it out, whether to a public or private cloud. The hardware to do this represents an up front capital expenditure, which of course slows progress and often results in the R&D organization spinning from low cost Arduino, Raspberry PI implementations to more supported hardware from classic hardware providers like Dell or HP. Just to make a point, as a manufacturer of “hardware” they often ignore the cost of software, and the cost of management.
Challenge 2: Connectivity
Wi-Fi can be rocky, what can be done about this? Connectivity or networking of these edge servers is the next big challenge. Will you use Wi-Fi? How will passwords be provisioned within the hospital, utility or construction site? Can you use cellular? How much will it cost? Do you have enough bandwidth? Echocardiograms are about 1GB of data and many sites do 5–10 per day? Can you afford to use cellular? And how do you get to multi-mode, switching from one communication method to another?
Finally, while thinking about one edge server has its challenges many OEMs have 100,000+ machines in the field, and many hospitals have 10,000 assets. How will you manage a network with 10,000, 100,000 or 1,000,000 edge servers?
Challenge 3: Interoperability
Each supplier has a different IoT solution; how do you share data? For those of you old timers you might remember the days of Netware, DOS, Unix, AS/400s where each ecosystem was an island. Re-typing the document more easily did file transfer. The same is true for many of the manufacturers of machines — each of them has their own ecosystem, and trying to combine data from multiple suppliers is as bad as the old days of computing.
Challenge 4: Security & Privacy
Some will argue that security and privacy are being used by those who are standing in the way of change, but it doesn’t take too much research to see how security and privacy, in particular ransom ware have become big challenges. Furthermore many of the security breaches (e.g. Target Stores) were exploiting vulnerabilities in the HVAC and not laptops. So any edge server solutions will have to be engineered to both have security features and privacy features.
Challenge 5: Talent
Even in fintech and consumer tech there is a fight for talent, in particular software development talent. Companies have built open floor plan, Silicon Valley style buildings to hopefully entice the millennial into selecting their place of work. I do wonder what this means post-Pandemic. While this might work in the more glamorous industries, what does this mean if you’re in the water, power or food industries? How will you find and retain the software talent to build next generation edge applications?
Many, including me, have written about how connecting machines, collecting data and learning from that data could change shrimp farming, power generation, textile manufacturing, construction, healthcare and the buildings we live in. However, unless we solve these five challenges the vision will remain just a vision and never become a reality.
