Divide-and-Conquer: How Edge Processing Will Open a New Door for IoT Applications

Sergio Flores
July 13, 2018

 

It comes as no surprise to anyone in the IoT industry that we are reaching a point where continued innovation will not be possible without making fundamental changes to the approach we use to process information.

According to dozens of studies from independent consulting firms, the number of IoT-connected devices is growing at an unprecedented rate, and it is now expected to hit an all-time high in 2020. Indeed, as predicted by Gartner, Inc. back in 2017, the number of connected things will reach 20.4 billion by 2020 [1] while the majority of the use of these devices will be driven regionally by Greater China, North America and Western Europe.

 

Table 1: IoT Units Installed Base by Category (Millions of Units)
Category2016201720182020
Consumer 3,963.0 5,244.3 7,036.3 12,863.0
Business: Cross-Industry 1,102.1 1,501.0 2,132.6 4,381.4
Business: Vertical-Specific 1,316.6 1,635.4 2,027.7 3,171.0
Grand Total 6,381.8 8,380.6 11,196.6 20,415.4

Source: Gartner (January 2017)

 

Putting the situation into full perspective, since the ongoing growth of connected devices is accompanied by substantial improvements in technology, the overall volume of data that will need to be processed by cloud systems in the upcoming years is also projected to grow considerably. To take an example, in the recently booming autonomous car industry where activities like vehicle operation and in-vehicle content are essential to delivering value to users, it is expected that cars will generate 1TB of data per day [2] which will, in many cases, require real-time processing and immediate feedback to the user. The same trend can be seen in the smart home surveillance industry where the latest devices are already adopting very high-resolution 4K image sensors with the purpose of both delivering better image quality to users and enabling software to perform advanced computer vision tasks with greater accuracy. Will we be able to send all this information to the cloud and get it processed on a real-time basis? Probably not.

The biggest challenge for the IoT industry, then, comes with the fact that the evolution of sensor technology and hardware does not match up with the speed of improvements in widely-available data transportation technology. In fact, as Botta [3] highlights, it has been seen that over the last 20 years, processor power has increased by a factor of 1015, but data bandwidth capacity has only increased by a factor of 104. This not only imposes restrictions on the possible IoT applications that can be developed but also places risks in those technologies where real-time feedback is a requirement – imagine an autonomous car that takes seconds to make a maneuver. Other concerns also arise from the fact that, in the face of new data protection regulations and vigorous enforcement of data privacy, it might not be in the interest of IoT product manufacturers to start sending even higher volumes of data directly to the cloud. Indeed, a recent survey from Gemalto – focused on confirming that consumers lack confidence in IoT device security – affirmed that for about 65% of IoT consumers their most common concern is a hacker controlling their IoT devices. Nevertheless, it would be inappropriate to blindly lay blame on the cloud computing model for the challenges which face the expansion of IoT.

Looking back in time, cloud computing has been a crucial component in allowing the IoT industry to take hold and rapidly evolve since the idea of connecting devices to the internet took off – already two decades ago. Owing to the high availability of affordable cloud infrastructures, IoT devices, in general, have become more accessible to the public thanks to the reduction of processing power needed to be available on the device and the simplification of tasks handled by it – in many cases, the tasks being that of merely uploading data to the cloud. Higher volumes of data, and also the presence of heavyweight processing power resources in these cloud infrastructures, have opened up possibilities for the development of more and more complex data analytics tools and more extended compatibility of hardware and software vendors. Whether it is ideal or not, this has led most IoT solutions to be based on highly monolithic backend cloud solutions and strongly centralized processing and storage of data. Something that has worked well in most cases, so far.

However, the previously mentioned limitations urgently call for newer and better ideas on how to reduce the distance between the source of data and the place where it is processed. Edge computing, a concept motivated by the idea of decentralizing cloud computing and dealing with the current data explosion and network traffic challenges, has offered several opportunities to take the IoT industry to the next level by dividing the data processing load into smaller chunks that can be processed closer to the data sources, thereby opening up hundreds of new possibilities for IoT applications requiring low latencies. Moreover, since the adoption of this concept also implies reducing the need to stream high volumes of raw sensor data to the cloud, it can be expected that this will impact positively both on the consumer and in industrial IoT use cases where reducing connectivity costs and increasing the security of data is critical. In practice, IoT applications which explicitly require low latency like smart home cameras are already implementing this approach by offloading heavy processing tasks like object/face detection directly to the edge and consequently reducing the time needed to notify users when something relevant is happening. Similar use cases can be seen in more privacy-concerned manufacturers who are implementing mechanisms that preprocess raw video directly in the camera and blur all faces present in the video so that it can be safely uploaded to the cloud.

Overall, by stepping away from a centralized cloud approach and redirecting towards a distributed computational load approach, other additional advantages can also be expected:

  • Increased data privacy in edge device applications: for example, by reducing the amount of private information that IoT wearables send to the cloud in the form of fitness and heart monitoring data, or by guaranteeing that electricity or water usage patterns from smart homes are not at risk of being exposed to potential parties interested in knowing if there is someone at home or not.
  • Energy consumption savings in edge devices: by allowing the internal hardware to process data without the need to incur the cost of high energy-consuming communication modules, and therefore simplifying tasks for products relying on batteries – e.g. smart outdoor cameras.
  • Location-aware data processing on edge devices: by enabling location-aware devices to make data processing decisions based on location and therefore distributing data processing loads across other dimensions.

Despite the numerous benefits that a well-established edge computing concept might bring to the IoT industry, the reality is that, in practice, the potential benefits of decentralizing data processing in the cloud might not be achievable without first solving the several unknowns that still needed to be researched – such as the appropriate distribution of infrastructure resources and the management of virtual machines and containers under an edge-type model. In order to distribute resources and data processing across different infrastructures or network nodes, it is mandatory to develop standards that specify how all the various computing components should collaborate and allocate resources even when different infrastructure suppliers might be involved. More in detail, when it comes to the management of virtual machines across decentralized IoT infrastructures, for example, works in the areas of Fog computing will become extremely important since they will lead the discussion of how resources can better be distributed. In the same way, as seen in existing research [3], in order to implement edge computing successfully, situations in which virtual machines might want to entirely delegate specific tasks to other network nodes (because of the lack of resources or location-based decisions) will require VMs to understand the limits of their own containers and take the right delegation strategies accordingly. Finally, once these standards and related challenges have been well resolved, tough decisions will have to be made for already established IoT applications as to how best to start breaking down their currently deployed monolithic backend solutions and start placing centralized data processing blocks into smaller blocks on the edge.

Although it is true that there are still multiple issues that need consideration before we start building more solutions based on edge computing, it is clear that with the most recent substantial improvements in SoC, CPU, and DSP [4] technology, performing more complex data processing tasks at never-seen-before efficiencies right at the edge device will become a common scene for many more IoT devices and will give rise to an entirely new era of IoT solutions. This will undoubtedly be a great opportunity to deep dive into researching solutions to distributed computing paradigms and to be part of a new, game-changing industry.

References

[1] Gartner, Inc., "Gartner Says 8.4 Billion Connected "Things" Will Be in Use in 2017, Up 31 Percent From 2016," 7 February 2017. [Online]. Available: https://www.gartner.com/newsroom/id/3598917. [Accessed 5 July 2018].
[2] V. Madel, "Connected Vehicles and IoT Technology: Are You Ready?," Samsung, 2018. [Online]. Available: https://insights.samsung.com/2017/10/26/connected-vehicles-and-iot-technology-are-you-ready/. [Accessed 17 May 2018].
[3] A. Botta, "Integration of Cloud computing and Internet of Things: A Survey," Future Generation Computer Systems, vol. 56, pp. 684-700, 2016.
[4] R. Roman, J. Lopez, M. Mambo, "Mobile edge computing, Fog et al.: A survey and analysis of security threats and challenges," Future Generation Computer Systems, vol. 78, pp. 680-698, 2018.
[5] W. Qualcomm, "We are making on-device AI ubiquitous," Qualcomm, 2018. [Online]. Available: https://www.qualcomm.com/news/onq/2017/08/16/we-are-making-device-ai-ubiquitous. [Accessed 17 May 2018].

 


 

sergio floresSergio Flores received his degree in Electrical and Computer Engineering from Seoul National University in South Korea - one of the most prestigious university in the country. He joined Samsung Electronics in 2014 as an integrated circuit (IC) hardware engineer, where he gained the critical hardware level experience needed to initiate his career in IoT product management. Later that year, he was invited to join Samsung’s IoT R&D team where he worked as a technical product manager contributing to and leading two important Smart Home and Drone projects. He is currently acting as an IoT Product Manager at Smartfrog, a $32M funded IoT start-up headquartered in Berlin, that offers a revolutionary home security solution under a software as a service model (SaaS).