Different Requirements For Hyperscale Computing Across Vertical Application Domains

Industrial process, consumer devices, and aerospace/defense are all generating massive amounts of data every day, where data is stored and processed is changing.


As mentioned in previous posts, one of the key conversations I have with customers a lot these days is how to deal with the balance of storage, compute and connectivity as we enter the era of hyperscale computing. While there are overarching challenges that are “of similar class” across the vertical application domains—consumer, hyperscale computing, mobile, networking, aerospace/defense, automotive, industrial and healthcare—some seem to be more pronounced or even unique to certain verticals.

For a recent event in the aerospace and defense domain, I created the following illustration to compare their specific challenges in something called the “Advanced Battle Management System” with the challenges that the commercial space is facing.

The key similarities are the large number of sensors, devices and connectivity, as well as the storage of data and processing being highly distributed. As outlined in “Hyperscale and Edge Computing: The What, Where and How,” the commercial space has large volumes of data being transmitted. According to Ericsson, overall traffic through commercial networks is predicted to grow to 164 exabytes per month in 2025, up from 33 exabytes per month in 2019. Seagate/IDC predicts storage in the “global datasphere” to grow to 175 zettabytes by 2025, up from 45 zettabytes in 2019.

Associated data comes from all verticals—consumers watching videos on demand, browsing and uploading from their phones, sending step counts and fitness information, for example. To that end, 63% of the data on commercial networks was already video-based in 2019.

In automotive transportation, drivers in cars are using data to update map material and receive real-time traffic guidance. The Automotive Edge Computing Consortium estimates that autonomous vehicles could generate up to 5TB of data per hour as they require “modern cloud-based machine networks that depend on tightly integrated, low-latency hybrid architectures to effectively aggregate, analyze and transmit data in real time, ensuring safe and efficient operation on the roadways.” Previous estimates I have seen put this number at 5TB to 20TB per day. Whatever the precise number turns out to be, there is no doubt that it will be quite big.

The industrial domain – manufacturing and cyber-physical systems – not only creates significant amounts of data during production, but also during the lifetime to help with aspects like maintenance. Digital Twinning is a hot topic here, and data-management is a key foundation. For health and medical applications, privacy of data is crucial of course, but the accessibility, – the when, where and how – plays a key role too.

Like all the other application domains, the aerospace and defense vertical is dealing with enormous amounts of data coming from air, water and ground, all of which needs coordination, transmission, storage and processing.

Where will the differences be? They likely will be decided individually per application domain, driven by concerns like safety, security and privacy. Consumers may or may not be comfortable with processing of data to happen outside their devices for reasons of privacy. Life and death decisions about health may have to made locally of there is no connection, and there are key concerns about interfering with health actuators like defibrillators and pacemakers fueled by movie examples, so safety and security are key. A soldier equipped with goggles enabling augmented reality can be much more efficient given the real-time battlefield updates that VR/AR can provide. However, security of data as well as where the decision-making takes place based on the data and types of decisions to be made, will be key drivers for system partitioning, i.e., where data is stored and processed.

Case in point, In the automotive industry, one of the prevalent discussion points has been about how much networking is needed for autonomous driving. Some argue that “full self-driving should not require a cellular connection—5G, 4G or even 3G.” To me, the truth likely sits in between. For sure, the “mode of operation” will change when 5G connectivity goes away. Today, for instance, we already use a mix of “autonomous”—i.e., fully downloaded material—for route planning, with the map being fully downloaded but updated with cellular data when available. But, of course, full autonomy is a good and audacious goal.

We’ll see. Let’s get to work and figure this out!

Leave a Reply

(Note: This name will be displayed publicly)