Edge Computing: Part 2 - Design Patterns - Recipes & Patterns for Value Creation
Ryan Anderson
IBM CTO for Palo Alto Networks; IBM Architect in Residence, San Francisco; Cambridge University; VC Investor and Advisor
Design Patterns - Recipes & Patterns for Value Creation is part two of a four part series on Edge Computing.
*Please note it's a draft / work in progress - but figured I'd share the current state of my thinking. It's a bit heavy on IOT/Industrial at the moment - and needs more in the rapidly evolving space of enterprise hybrid cloud and 5G.
PART 1 was Basic Overview of Edge, discussed how the definition of edge (and edge value) was very stakeholder dependent - and introduced a ‘four corners of edge’ framework that includes (a) IOT/Industrial, (b)Enterprise Network Cloud, (c) Telco/5G and (d) Consumer, Retail etc..
Now we explore DESIGN PATTERNS
What are design patterns? Design patterns can be thought of as as meta-use-cases, bundles of use cases, or repeatable recipes - all for composing technologies for useful purposes.
The formal (Wikipedia) definition of a software design pattern is “a general, reusable solution to a commonly occurring problem … a description or template for how to solve common problems when designing a solution”
Solution Types (Big Buckets)
Drilling Deeper
Here are a few use cases to add some context...
And let's explore a few Patterns in more detail
INDUSTRIAL & IOT
Challenge: I have equipment I care about and want to understand status, health or stop bad things happening - like catastrophic failures or unplanned maintenance shutdowns.
There are a number of design patterns that touch on Industrial and IOT. Data consumed might be structured telemetry data across a time domain – for example, a vibration sensor attached to an expensive cooing pump in a factory.
Many Edge-IOT use cases deliver value by producing a degree of “Sensemaking” and “Situational Awareness” about the things we care about – the longevity and operational efficiency of the cooling pump. Without it, bad and expensive things happen. Edge allows the intelligence to happen near the device and near the action. Low latency, and not dependent on connectivity.
Vibration and Acoustic Analysis are used to extract signal and meaning from the data. This could be something as simple as an amplitude a frequency (normal hum level) or a bit more fancy (a Fourier spectral analysis (FFT) of multiple spectra of signals, to understand if any deviation from the norm; all the way up to machine learning (literally) using Tensorflow Lite for on-device machine learning inference.
Below are some raw videos I recorded from my home lab.
The key story here is not the FFT time-to-frequency conversion that's being done on a $30 Raspberry Pi using Python (even thought that's sort of cool) - as this sort of thing was being done in the 80's and 90's in energy sector and the math is even older.
The key story is that COMPUTE and ANALYTICS can be easily deployed to the EDGES - it's cheap and powerful now, like it has never been before. And with Machine Learning and Deep Learning - it's a much smarter and accessible system.
Brownfield Analytics (Retrofits)
Challenge: I have really old but still mission critical assets. Many are decades old. They are non-instrumented or weakly instrumented. As they continue to age, I need more data and better data to understand the state of my aging asset/.
Today, there billions of dollars in assets still in operation today. Assets deployed before Ronald Reagan was president. Assets with cathode ray tube displays. Assets communicating with pre-internet protocols like RS-232 (introduced in 1960), RS-485 and SCADA / Modbus RTU. Assets running Windows 95.
(Side note - There's also a heap of National Infrastructure built out as part of Eisenhower-era expansion, and like some baby boomers from the same era - is starting to show its age)
Most of these assets are not instrumented or weakly instrumented.
“Brownfield Analytics” refers to applying new sensors and instrumentation to old, but still useful plants and machinery. Putting on the dirty boots and scraping off a layer of dirt in order to apply a sensor or an edge signal processing module - with an eye on applying modern analytics, compute and communications to the system.
Brownfield is a cheap way to get data, information, signal and actionable intelligence from assets already in operation. Avoid Downtime. Predict faults.
As above - we have common threads
- transducers
- some compute on the edge
- power and connectivity (intermittent OK)
- human part of the wider 'sense-making systems'
Other Patterns (developing...)
Imagine: You are a manufacturer of electric bicycles trying to reduce downtime...
Imagine: You are a bank security team responsible for protecting ATM equipment & customers
Imagine: You manage a retail chain that sells groceries and manages facilities
Imagine: You need your augmented reality & VR headsets to offload compute to the 5G Network Edge
below is (partial) list of Design Patterns for expansion over next few months.. (as I edit this posting)
- Enterprise Network Evolution – Compute at the Edge
- Enterprise Network Evolution – Critical Data and AI ‘on-prem’
- Edge IOT Gateways
- Digital Twins
- Verbal Interrogation of Assets
- Voice Command and Control / NLU (industrial)
- Voice Driven Analytics / Operations / Business Intelligence
- Visual Models (Situational Awareness)
- People Counting - Retail and Events
- Machine Vision - Dynamic Configuration / Agile manufacturing
- Worker Insights / Eldercare / Healthcare
- Vehicle Telemetry Analytics
- Automotive/Drones
- Digital Modeled Behavior
- AR/VR - 5G NR low latency base stations (compute offload)
- Edge Video Processing, Gaming, E-sports
Open Source Patterns (Linux Foundation)
The Linux foundation LFEdge project has several projects that are developing BLUEPRINTS (Patterns) for Edge Computing. More info below on:
- Akraino Network Cloud blueprint family
- Connected Vehicle Blueprint
- Edge Video Processing
- ELIOT: Edge Lightweight and IoT Blueprint Family
- Integrated Edge Cloud (IEC) Blueprint Family
- Kubernetes-Native Infrastructure for Edge (KNI-Edge) Family
- Micro-MEC
- Radio Edge Cloud
- StarlingX Far Edge Distributed Cloud
- Time-Critical Edge Compute
... more to come! this blog is a work in progress – will be updating over time - thanks for reading, and I welcome your comments!
--- Appendix ---
Open Source Patterns (Blueprints)
Part of LFEdge - Akraino Edge Stack aims to create an open source software stack that supports high-availability cloud services optimized for edge computing systems and applications. The Akraino Edge Stack is designed to improve the state of edge cloud infrastructure for enterprise edge, OTT edge, and carrier edge networks.
and for Approved Blueprints:
EdgeX Foundry (also LFEdge) is a vendor-neutral open source software platform at the edge of the network, that interacts with the physical, every day working world of devices, sensors, actuators, and other IoT objects. The intent is to build a common framework for Industrial IoT edge computing.
---
This article is my opinion and do not necessarily reflect views of my employer or LF Edge
Very Good read, Is it possible to cover defence a) aviation, weapons where the edge computing can come in very handy with more 'situational awareness' cases involving matter of life and death b) telemetry though covered, is there any low cost, at scale , solution proposed which addresses the real problems like traffic management (think about peak, unruly traffic) in mega cities like Delhi, Jakarta, Cairo, Karanchi c) Any use-case in compliance ? its a killer needed to weed out low scale high volume corruption the developing countries are plagued with - think of how to curb cash payments, quick situational outputs to agencies thanks