White Paper Review: Edge Computing

Forrester surveyed 300 IT and OT decision makers from diverse companies revealing to what extent edge analytics is utilized as part of IoT deployments and found that around half has either already implemented or plan to implement edge analytics within a year. This post reports on their findings as presented in a January 2019 whitepaper and identifies issues requiring further examination for practical adoption.

Edge analytics in IoT is a relevant topic as more Internet connected edge devices are added every day and as data quantities keep increasing. The topic is at the forefront of IoT implementations these days as evident by the fact that major cloud providers offer solutions for it. Amazon Web Services Lambda, Microsoft Azure Functions, and Google Cloud Functions are prominent examples. It is also an established academic research topic with some work advocating for a move from cloud backends to edge and fog computing (Schooler, et al., 2017).

In terms of methodology and ethical disclosure Forrester’s paper does a decent job at presenting the facts. The paper was commissioned by Dell Technologies and VMware to report a survey conducted during October-November 2018. Methodology highlights are provided over three appendices and are at a level of detail one could expect from a business whitepaper.

Edge Analytics has been a focus area for Forrester. These days for instance they look into the increase in edge analytics implementations for content delivery networks (CDNs) in light of COVID-19 induced rise in content consumption (Staten & Stutzman, 2020). Edge analytics certainly could support operational IoT initiatives, but this example suggests it can have direct user experience impact as well.

The paper starts by listing IoT-enabled use-cases reported by survey participants as: security and surveillance, tracking and tracing, energy management, automatic operations, predictive maintenance, and various other use-cases, some industry-specific. The appendix indicates use-case categorization was provided to participants upfront which could miss some fine details. Since not all methodology details are shared it is hard to fully judge. In any case these results are fairly consistent with other findings surveyed here (Horev, 2020).

The whitepaper builds the case for edge analytics by reporting 40 to 49 percent of participants citing security, high costs and accessibility as potentially limiting factors for data analytics in the cloud. Furthermore, it seems half are either expanding, implementing, or planning to implement edge analytics within 12 months. Driving factors were found to be growth of edge generated data, security, cost efficiency in data transportation, reduced latency, and regulatory reasons. Overall, the report is informative, well written, and logically structured.

The best part of the paper is kept for last. In it, the authors move from results reporting into making one primary recommendation: organizations are encouraged to use specific criteria for selecting IoT use-cases best suited for edge analytics. For example, the authors suggest that by looking for IoT use-cases characterized by large volumes of data as well as low latency requirements the benefit potential of edge analytics could be maximized. The use-case identification theme is kept throughout the remainder of the paper with some rather generic set of recommendation. The question of where it is best to put our edge analytics efforts is a great practical question, and the idea of criteria combinations is a strong one. The paper only scratches the surface here and further research should be well received.

And thus, the whitepaper describes a reality of edge analytics being increasingly adopted. If so, one could have hoped for the report to go beyond incentivizing to offering implementation recommendations and alerting on expected challenges. But this is not the case. The savvy reader might not be surprised by that. Industry whitepapers tend to stop at the point at which the prospective customer would want to seek further guidance. Avnet’s whitepaper on the exact same topic serves as a second excellent example of that (Avnet, 2018).

Turning to academia, I could not find academic surveys over multiple industry cases for identification of successful patterns of edge analytics implementation. Most academic papers are either domain-specific (for example, Ferdowsi et al. (2019) on intelligent transportation systems) or theoretically driven (for example, Harth et al. (2018) on predictive intelligence algorithmic efficiency). The power of surveys such as Forrester’s could be in the finer real-world insights but those are lacking.

The paper is in fact not concerned at all with what it takes to implement edge analytics. Consider for instance the edge device itself. Research has shown that choice of machine learning algorithm as run on a Raspberry-Pi platform affects efficiency and accuracy across multiple datasets (Mahmut, et al., 2018). Not cautioning the reader as to at least some of the main concerns could be perceived as detracting from the whitepaper’s credibility.

Edge analytics does not necessarily mean running computations directly on the sensing device but rather across multiple devices at the proximity network. The industry term for that is Fog Computing and discussions on its role in IoT can be traced back to 2012 (Bonomi, et al., 2012). Unfortunately, the whitepaper does not mention the term let alone report its role in surveyed companies’ implementation.

Moreover, academic researchers have suggested an approach for using publish/subscribe systems to organize edge data analytics (Florian & Neagu, 2018) in fog computing scenarios. Publish/subscribe technology is indeed heavily utilized in the industry. When surveying companies, a lot can be gained by going into some implementation detail. It could make the whitepaper much more practically insightful.

In summary, Forrester’s white paper rightfully identifies edge analytics as a major industry trend. It provides insight into its drivers and benefits for business. And general guidelines are provided for selecting the right IoT use-case for adoption. However, the paper falls short of illuminating any practical aspects of edge data analytics implementation. Being able to ask hundreds of industry decision makers on a hot topic, one could have hoped Forrester will opt to extract practical insights on implementation trends, but this will have to wait for another time.


Avnet, 2018. AI at the Edge: The next frontier of the Internet of Things. [Online]
Available at: https://www.avnet.com/wps/wcm/connect/onesite/d3f21447-6f42-4d77-a367-736694e6c5ed/ai-at-the-edge-whitepaper.pdf?MOD=AJPERES&attachment=false&id=1552600562643
[Accessed June 2020].

Bonomi, F., Milito, R., Zhu, J. & Addepalli, S., 2012. Fog computing and its role in the internet of things. Proceedings of the first edition of the MCC workshop on mobile cloud computing, pp. 13-16.

Ferdowsi, A., Challita, U. & Saad, W., 2019. Deep Learning for Reliable Mobile Edge Analytics in Intelligent Transportation Systems: An Overview. IEEE Vehicular Technology Magazine, 14(1), pp. 62-70.

Florian, V. & Neagu, G., 2018. Towards an IoT platform with edge intelligence capabilities. Studies in Informatics and Control, 27(1), pp. 65-72.

Harth, N., Anagnostopoulos, C. & Pezaros, D., 2018. Predictive intelligence to the edge: impact on edge analytics. Evolving Systems, Volume 9, pp. 95-118.

Horev, B., 2020. Webinar Review: IoT goals, applications and challenges with emphasis on security and private networks. [Online]
Available at: https://unearth.blog/2020/06/04/webinar/

Mahmut, T., Basurra, S. & Mohamed, M., 2018. Edge machine learning: Enabling smart internet of things applications. Big Data and Cognitive Computing, 2(3).

Schooler, E. et al., 2017. An Architectural Vision for a Data-Centric IoT: Rethinking Things, Trust and Clouds. IEEE 37th International Conference on Distributed Computing Systems (ICDCS), pp. 1717-1728.

Staten, J. & Stutzman, E., 2020. IoT Analytics Create New Edge Computing Value Props For Content Delivery Networks. [Online]
Available at: https://go.forrester.com/blogs/iot-analytics-create-new-edge-computing-value-props-for-content-delivery-networks-cdns/

One-Minute Paper: Information

This is a one-minute paper to promote thinking based on the content of the very first lecture in an Information Systems course. As I listen to the lecture, I look for points relevant to my field of interest, the Internet of Things. The introductory session surveys several different topics and I will focus on just one – information.

First, a distinction is made between data and information and what strikes me as particularly important is the fact that it is meaning which separates the two. I suppose it can be deduced also from the fact that the same data can generate different kinds of information, depending on how it is analyzed and to what purpose. Through data analysis one creates a specific meaning which is relevant to one’s context of investigation.

Continuing along this theme, the lecture surveys information characteristics (timeliness, frequency, accuracy, etc.). Those characteristics are viewed in the context of their impact on business. The discussion is brief and there is no dwelling on priority or comparative significance. Perhaps rightfully so since different businesses have different needs. Still, I believe just as meaning is key to our understanding of the essence of information, so do the appropriateness and rarity characteristics of information are key for deriving business value out of information.

IoT research has attempted to deal with appropriateness of information in several ways. Some research is aimed at providing models for generating reliable information from sensors data (Lewandowski & Thoben, 2017). While others attempt to provide a taxonomy for types of data analytics techniques used in across a wide range of IoT applications in different verticals (Siow, et al., 2018).

I think that IoT systems are sometimes viewed as data mining systems leveraging new connectivity technologies to haul data from edge to center. This view does not hold promise for generating surprising or rare information, not even for appropriate information, and frankly not even for any information at all. In other words, IoT platforms run the risk of creating efficient data pipeline with very minimal effectiveness as measured by information.

There is a mentality of collect the data first, worry about what to do with it later. Since we have the technologies to create vast data lakes we are tempted to do so. We also say: well, we may have this initial application in mind, but after we learn some more, surely, we will find other applications for the data we have already accumulated. Thus, we better start collecting now.

Nowhere is this approach more challenged than in consumer applications of data. When it comes to back-office decisions, sure you can gather telemetry data about your customers and figure out new ways to analyze it to improve operations and cut costs. But when it comes to value driven by information presented to the end user, we better have a better plan.

For example, using big data techniques, a service can offer “recommendations” for more of something: books, podcasts, things to buy, content to consume. It is obvious though this is not highly personalized. If big data is used to solve a personal problem, then the value is much more apparent.

Information appropriateness raises the issue of data privacy and trust. The greater the value provided to the user the more trust they have with the system and the more willing they may be to relinquish data in exchange for more value. But if users see that their data is applied in generic or business self-serving ways then trust could be broken. One attempt to deal with privacy in IoT is to contextualize the requirements – to differentiate the privacy concerns between groups and individuals in a way that benefits both the user and business (Zhou & Piramuthu, 2015).

In IoT, data efficiency considerations enhance edge computing trends. Instead of just collecting data and sending everything to the cloud, it is now possible to apply machine learning models at the edge for early information generation. This is more efficient and can better serve the user. Additionally, it can address trust issues by not requiring some of the data to be transferred over the Internet and to be centrally stored. Edge analytics is covered in Lalitha (2017).

These musings only scratch the surface, but I wanted to keep this post short. I find HBR’s Analytics 3.0 an interesting read.


Lalitha, B., 2017. Edge Analytics on Internet of Things: A Survey. i-Manager’s Journal on Computer Science, 5(4), pp. 36-40.

Lewandowski, M. & Thoben, K., 2017. Deriving Information from Sensor Data. 14th IFIP International Conference on Product Lifecycle Management (PLM), Volume 51, pp. 623-631.

Siow, E., Tiropanis, T. & Hall, W., 2018. Analytics for the Internet of Things: A Survey. ACM Computing Surveys, 51(4), pp. 1-36.

Zhou, W. & Piramuthu, S., 2015. Information Relevance Model of Customized Privacy for IoT. Journal of Business Ethics, 131(1), pp. 19-30.