Digital Transformation
cancel
Showing results for 
Search instead for 
Did you mean: 

The cloud is dead. Long live the Edge

mikeshaw747

In November last year, Peter Levine of Andreessen Horowitz venture capitalists gave a talk in which he predicted that the growth Internet of Things (IoT) and machine learning would spell the end of cloud computing.

Today, our mobile devices’ application power is delivered via cloud services. When look at the weather, or do mobile banking, or use Google Maps, or book an Uber our smartphone is acting as a “display vehicle” for a cloud service. We ask to do something, the request goes over the network to the cloud service which does the work and sends back the result - “take this route”, “these are the Ubers you could use”, “this is weather forecast for your area”, “I’ve made the payment you requested”.

Let’s jump ahead to a world where we have 20 billion (Gartner estimate for the number of connected IoT devices by 2020), 50 billion or 200 billion IoT devices. We have self-driving cars, self-driving buses, machinery that predicts when it is going to fail, self-optimizing manufacturing production lines and medical monitoring systems that can predict when an elderly person is going to suffer a health-related incident.


Characterises of IoT systems
Such systems have two characteristics.

1.  They will generate massive amounts of data. Today, the data we collect and process is human-created - it’s structured data like a sales transaction. IoT sensors sense reality, and recording reality takes a lot of data. For example, a 4k camera that might be used for an automated surveillance system generates around 4 terabytes of data an hour. So, if you have 20, 4k cameras around the perimeter of your airport, that’s 80 terabytes of raw data an hour. 

2/ We feed this data into machine learning systems. Up till now, we’ve told our computers what to do. With machine learning systems, the computer figures out what’s going on - we may give it initial training, but then it takes over the learning and inference process. For example, a machine learning system might search the video from 20 cameras, looking for someone trying to break into the perimeter fence. Or, cameras might look at a production line and work out the series of events that leads to a slowdown in production. Or, we might use the data from sensors to drive our cars for us. Or, a machine learning system might figure out how to predict when a sewage pump is going to fail.


Why “send everything over the network to the cloud” doesn’t always work

So, we have huge amounts of data that are fed into machine learning systems. The “send everything over the network to the cloud” model breaks down in these circumstances for two reasons.

1. The amount of data we need to transmit is too large, even with 5G.

2. And even if we could transmit the data, the round trip would be too slow. If my self-driving car is about to hit an obstacle, we don’t have the time for all the data to be transmitted to a cloud server and then the command to “take evasive action” be sent back to the car. Or, if there is a “strange vibration” from our sewage pump, we may want to shut it off now, not when all the data has gone up to the cloud and an answer come back.

The diagram below shows how round-trip data transmission delays would make centralized control of an oil rig non-viable.

Diagram shows how the round trip from oil rig to "the core" is too slow to take evasive actionDiagram shows how the round trip from oil rig to "the core" is too slow to take evasive action
For this reason, argues Peter Levine, we are going to see a growth in “Edge Compute” - data processing power where the sensors are - in the car, in the bus, in the pumping station, in the elderly person’s home, or at the airport perimeter.

IDC agrees with Levine. They predict that by 2021, 43% of “IoT computing” will occur at The Edge.


The architecture for Edge Compute

So, we will have an architecture where inference and action occur at the edge but where initial training and ongoing learning takes place at the core.

Inference and action at the edge, deep learning and other analytics at the core (often using high performance compute)Inference and action at the edge, deep learning and other analytics at the core (often using high performance compute) Edge compute examples : 

  • Physical surveillance
  • Predictive maintenance
  • Production control
  • Medical monitoring
  • Autonomous vehicles
  • Drilling for oil

Is there a role for the cloud in an IoT world?
Does the cloud therefore have no part to play in a world of 20 billion, 50 billion or 200 billion IoT devices? While we can't use the cloud for those situations where we have large amounts of data or where we need a very fast response, the cloud has at least three roles in an IoT world.

1: Cloud may be used where there isn't masses of data or a need for high-speed action

Firstly, there are many IoT applications that don’t involve masses of data and don’t require instantaneous action. Here are some examples:

Smart cities

  • Identifies threats and traffic, keeping citizens moving in the right direction
  • Knows when trash bins are full, and the best route
  • Locates parking spaces, sparing fuel, time and the air
  • Smart Metering drives greater efficiency

Connected Cars

  • Drivers who always have the latest information
  • Passenger comfort and entertainment, to their liking
  • Vehicles that share intelligence about congestion and conditions (known as "swarm intelligence" - c.f. Google Waze)

Smart Farming

  • Water and nutrients by the plant, not by the acre
  • Light, temperature, humidity, and pH, not by the forecast
  • Information aggregated, transmitting when needed

These applications may well use low power wide area networking (LPWA) like the LoRa network (https://www.lora-alliance.org).

And they may well use the data collection, normalisation and analysis services of an IoT cloud provider particularly if they are an small or medium sized business that don’t want to invest in IoT middleware services themselves.

2: Cloud may be used for centralized (deep) learning

Secondly, machine learning systems learn from what they see - in theory, the more data they experience, the better then become at their jobs. If we send a summarised set of data from the edge - from many cars, or many sewage pumps or many assisted living systems - back to a central system, this central system could learn from many different situations.

Machine learning systems may also learn better if they are able to cross-correlate data from different types of sources.

3: Cloud may be used to keep records of what happened

And lastly, we need to keep a record of what happened, and this will probably be stored centrally. 


The rumours of cloud’s death are greatly exaggerated
To mangle Mark Twain’s quote on his reported death, “The rumours of cloud’s death are greatly exaggerated”. IoT will create a new class of applications for which large amounts of Edge Compute will be required. But cloud will not go away.

We will simply add another dimension to hybrid computing - the dimension of “Edge versus Centre”. 


RELATED POSTS

Will cloud will keep growing and growing until it takes over everything? History says, "no" : Will cloud keep growing forever? History tells us that new innovations like cloud will reach an equilibrium with existing technologies. It's already happened with e-books and tablets, for example. 

The Cloud Cliff : two pieces of HPE research into customers who are bringing some of their workloads back from the cloud. 

Cloud myth : once in the cloud, always in the cloud : an important transition point occurs when a digitailly-fuelled application moves from its experimental to its strategic phase. The requirements of its compute platform can change quite dramatically.


Mike Shaw
Director Strategic Marketing
Hewlett Packard Enterprise

twitter.gif @mike_j_shaw
linkedin.gif Mike Shaw

Mike Shaw
Director Strategic Marketing

twitter.gif@mike_j_shaw
linkedin.gifMike Shaw

0 Kudos
About the Author

mikeshaw747

Mike has been with HPE for 30 years. Half of that time was in research and development, mainly as an architect. The other 15 years has been spent in product management, product marketing, and now, strategic marketing. .

Labels
Events
Jan 30-31, 2018
Online
Expert Days - 2018
Join us online to talk directly with our HPE experts during online Expert Days. Find information here about past, current, and upcoming Expert Days. P...
Read more
28-30 November
Madrid, Spain
Discover 2017 Madrid
Join us for Hewlett Packard Enterprise Discover 2017 Madrid, taking place 28-30 November at the Feria de Madrid Convention Center
Read more
View all