Home5GLatency, but not as we know it – plus other industry drivers for the move to the edge

Latency, but not as we know it – plus other industry drivers for the move to the edge

Speaking last week on a webinar session with analyst house Omdia – following the release of its new industrial edge portfolio at Google Next 2021 last month, and ahead of a new joint-research report with Omdia on why mobile operators should work with hyperscalers to plant a flag at the industrial edge – Google Cloud ran through a number of useful bullet-point summaries about the Industry 4.0 market, as it sees it today.

These are neither definitive, nor exhaustive; they are handy quick-fire roundups, offered as anecdotes by one company – here, presented back by Enterprise IoT Insights as start-points, not endpoints, for discussion about the industrial edge. Look out for further entries on the key players in the edge ecosystem (‘Five on treasure island’), and why operators should climb into bed with hyperscalers (“A thousand flowers will bloom”, no less).

1 | Latency

Latency is the ‘big sell’ with 5G, invariably; each generation of cellular is marketed for its G-whizz. This perceived need-for-speed is, arguably, the only game in town for consumers and prosumers. Expensive network upgrades are sold to punters each decade with the promise of seismic gains in download speeds for streaming content, and sometimes with the more abstract allure of brand new content services.

In between times, they are sold with handset upgrades, as networks are further optimised, and devices improve. But that’s the consumer space, and the way mobile used to work. In industry, where carriers hope to gain a foothold with 5G – and a new way to expand income, grow services, and recover investments – speed is gauged differently, and is not always the king metric. For starters, download speeds matter less when connecting ‘things’.

This is important. This new industrial internet-of-things is – initially, mainly – about connecting machines and processes in the workplace to harvest data from them, in order to optimise their operation. The heavy-lifting is in the uplink channel, to stream data to the ‘cloud’, whether on the premises, in the mobile network, or in a private or public data centre – where the data is crossed and filtered for ‘insights’, and reduced and returned on the downlink side.

The other thing with latency for Industry 4.0 – brought lower with faster local-area 5G networks and nearer local-area edge compute functions – is the max speed matters not-very-much-at-all. What matters more is the average speed, and what matters most, perhaps, is the guaranteed minimum speed. In other words, industrialists want ultra-reliable low-latency coms (URLLC), which is one of the tenets of the 5G new radio (NR) standard, in Releases 16-and-up.

So, low latency in edge setups is crucial; but the theoretical headline speeds (1ms), and even the ‘real-world’ max speeds (about 20ms), used in 5G marketing are unhelpful. They refer to one-off download speeds, and imagine industrial IoT in splendid isolation – and not in the upload-chaos of smart factories. Really, reliability is the definitive measure at the industrial edge, across all the other drivers Google Cloud lists besides (see below).

Amol Phadke, managing director for global telecom industry solutions at Google Cloud, comments: “There is a common view in the industry that these use cases are driven by low latency – that they get enabled because [the] edge is closer. What we have actually found out, as we have done deep discovery [with] enterprises and CSPs, is latency is one of the key requirements, but there are two or three others that drive edge applications.”

2 | Privacy 

There are three others, actually – or, at least, three that Phadke at Google Cloud mentions. One of these others, he says, is privacy – or “privacy and security”, in fact. Both, then; so perhaps there should be four additional key drivers in total, besides industrial-minded latency. No matter; most list articles raise debate about their limits and contents, and, as above, this one is only supposed to be a start-point for further discussion / exploration.

Data privacy and data security are different, of course: the first is about how data is collected, shared, and used; the latter is about how it is protected from compromise, both from outside attacks and inside ‘jobs’. Both are moving targets for enterprises: privacy varies by region, of course, especially where personal data is concerned; security is a dynamic conflict zone, which grows wider as devices are connected, and more complex as hacking develops.

The point for Phadke, banding them together in the context of the industrial edge, is only that they are both brought under greater control as corporate threats when data is retained in-house, on the edge, in private network exchanges between sensor endpoints and local compute and analytics systems. “Having something close to the edge enables you to have that assurance around securing important and sensitive data workloads,” he says.

3 | Scalability 

Another edge-driver for Phadke, which might be defined differently or more broadly, is ‘scalability’. What he is actually talking about is the underlying economics of edge-to-cloud industrial networking, which is prohibitive enough to make the practicalities fall down. That’s the scalability issue for Industry 4.0, then – that large ‘real-time’ workloads cannot be affordably carried between the edge and the cloud, in terms of bandwidth and latency – and, where wireless backhaul is required, in terms of cost, as well.

Phadke explains: “Scale is important, as well. Oftentimes, you require enormous amounts of data to be processed in very quick-time in order to generate certain outcomes – and sending that data all the way to the cloud, whether it is to the public cloud or private cloud, is very heavy from a network economics standpoint. Which is why bringing things to the edge actually helps the economic conversation with enterprises.”

The same economics – the time to carry data, and the cost to carry it – informs the drive to bring compute power even closer to the edge, and onto industrial sensors themselves. Increasingly, data is sorted in a broad-brushed manner at the point of collection, before being ferried about to be converted into insights. This is the case even on battery-constrained IoT units, where miniaturised machine learning (tinyML) is being increasingly engaged.

4 | Simplicity

And finally, simplicity: the result of negating data privacy concerns and unburdening the backhaul network by just, well, keeping ‘things’ local. Phadke at Google Cloud appears to talk about industrial fleet management and digital infrastructure management, and suggests both of these capabilities improve with increased control over networking and compute infrastructure.

He comments: “There is tremendous focus on… certain applications such as fleet optimisation, and really ensuring your current enterprise footprint is modernised over time – and having the edge infrastructure closer allows you to do that.”

See the other entries in this series:

Latency, but not as we know it – plus other industry drivers for the move to the edge
Five on treasure island – the key players sharing the spoils at the industrial edge
‘A thousand flowers will bloom’ – should carriers climb into bed with hyperscalers?

omniverse
Previous post
Siemens Energy to use Nvidia Omniverse platform for predictive maintenance
Next post
Five on treasure island – the key players sharing the spoils at the industrial edge