Edge Computing: A deep dive into Edge IoT Analytics

The central portion of edge computing is the seamless and strong integration between the cloud and the IoT. Photo Courtesy of 123rf.com | Vasin Leenanuruksa

Innovations in embedded systems-on-a-chip (SoCs) have opened doors to several commercial devices that are strong enough to run fully fledged complex algorithms and operating systems. The devices include a rich set of various sensors (for example, GPS, microphone or cameras), integrating more than one option for connectivity. One such example is the Raspberry Pi, but more alternatives are arriving in the market with various costs, computing capacities, power footprints and form factors.

These trends are increasing the potential of the Internet of Things (IoT). While many of the earlier IoT applications were about gathering data from “things” and sending them for analysis elsewhere, the evolving computing capacity of “things” now enables complex computation to be carried out on-site, without leaving the physical world.

Edge Computing: Integrating IoT and the Cloud

Most people refer to edge computing as a way to emphasize that portion of the work that is carried out at the edge of the network where the IoT links the physical world to the cloud. But edge computing is way more than just having data processing and computation on IoT devices. The central portion of it is the seamless and strong integration between the cloud and the IoT.

An edge computing application utilizes the IoT devices’ processing power to score, aggregate, pre-process or filter IoT data. It utilizes the flexibility and power of cloud services to execute complex analytics on that data and, in a feedback loop, support actions and decisions on and about the physical world.

Consider the major motivating factors for utilizing edge computing:

Decrease latency. The flexibility and power of cloud computing has enabled various scenarios that were not possible before. Think about how voice or image recognition algorithms’ accuracy has enhanced in recent years. Nevertheless, there is a price for this accuracy – the time required to get a piece of audio or an image recognized is affected significantly by the non-negligible yet inevitable delays in the network because data is shipped to the cloud where results are computed and then sent back to the edge. When low-latency results are required, edge computing applications can implement machine-learning algorithms that directly run on IoT devices, and they only communicate with the cloud off the critical path, for instance, to train machine learning continuously using captured data.

Add robustness to connectivity. Designing applications as part of the computation on the edge directly not only decreases latency, but makes sure that applications are not disrupted in the event of intermittent or limited network connectivity. This is quite useful in the case when applications are installed on remote locations where the network coverage is not good or even to decrease costs coming from expensive connectivity such as cellular technologies.

Preserve privacy. The data captured by IoT devices consists of private or sensitive information (microphones, streams from cameras or GPS data). While an application might utilize this information for running complex analytics in the cloud platform, it is highly important that the sensitive content’s privacy is preserved whenever data leaves the place where it is generated. With the help of edge computing, an application can ensure the data that is sensitive is pre-processed on-site, and only the privacy compliant data is moved to the cloud for additional analysis after passing via a first layer of anonymizing aggregation.

Edge Computing Architecture

The physical world is divided into locations, and the location is a geographical unit where the deployment of one or more IoT devices is done. The devices can be of one of three types based on their role in an edge architecture: edge devices, edge sensors and actuators, and edge gateways.

Edge sensors and actuators are special purpose devices that do not have operating systems or general-purpose processors. They are linked to edge gateways or devices directly or through low-power radio technologies.

Edge devices are general-purpose devices that can run fully fledged operating systems and are often powered by a battery. Devices that run iOS, Android or Linux, for example, can qualify as edge devices. The edge intelligence is run by edge devices (this implies the edge devices run computation on data they get from sensors, and they send commands to actuators). They are connected to the cloud platform either directly or via the mediation of an edge gateway.

Like edge devices, edge gateways run fully fledged operating systems, but they generally have more storage and memory, CPU power and an unconstrained power supply. Gateways can act as intermediaries between edge devices and the cloud, possibly providing extra location management services.

Both edge devices and gateways forward chosen subsets of pre-processed or raw IoT data to services that are running in the cloud, such as machine learning or analytics services and storage services, and they get commands from the cloud symmetrically, such as machine learning models, data queries or configurations to store IoT data locally.

Figure 1 shows a four-level hierarchy, with edge gateways as the intermediate tier above edge sensors and actuators and edge devices, arbitrary graphs or trees are also possible, aiding more levels of computation with hybrid cloud configurations or at the edge, for example.

Figure 1: Edge computing architecture example. Source: IBM

Figure 1: Edge computing architecture example.
Source: IBM

An edge computing application is comprised of several modules, each one running at different places in the hierarchy. For instance, an analytics module may run in the cloud for analyzing data that comes from edge devices and gateways; a dashboard module may be deployed in the cloud to offer a query interface or a global data view; and a machine learning module may be deployed on edge gateways for interfacing with modules that run on edge devices and score pre-aggregated data that comes from them. An edge computing application must explain how modules communicate and interact by clearly defining data flows between components, subsequently defining privacy and business rules, as well as visibility restrictions.

IBM Edge IoT Analytics – The Prototype

A 2009 Genesys survey revealed that unsatisfactory customer service cost companies $338.5 billion around the world in a single year in lost business. This is worsened by the fact that attaining a high level of customer satisfaction is often difficult. Most customers do not fill out paper or online forms, and those who do aren’t a fully representative sample of the entire customer base.

Focusing on customer satisfaction, IBM wanted to determine the potential of bringing together the benefits of edge computing and the cognitive capabilities of Watson in the following scenario that targets hospitality services.

A huge global hotel chain looks to enhance the process used to gather customer satisfaction. It understands the fact that spontaneous information about customer satisfaction is exchanged daily at its hotel premises but is lost systematically. Guests continuously communicate with the employees of the hotel at the reception desk, and all these interactions, explicitly or implicitly, convey a tone or a mood that directly connects to their satisfaction.

What if this information could somehow be seized and examined to obtain better customer insights? Moreover, what if real-time customer satisfactions from all the chain’s hotels around the world could be queried and examined very easily by the hotel management board via a simple query interface and a visual dashboard?

Understanding the Conversation

IBM’s Watson Cloud already has most of the intelligence that is crucial to extract valuable information from conversations. The conversation tones can be extracted from textual transcripts by Watson Tone Analyzer, and the recorded spoken conversations are easily converted into text by Watson Speech-to-Text .

However, there are major limitations in the hotel scenario, which make the utilization of such services impossible in the cloud. For example, there is every chance that guests’ conversations that are captured will have highly sensitive data, and passing this data to a cloud service has major privacy implications. Additionally, the transmission of continuous audio streams can be a costly affair and subject to issues related to connectivity.

In order to address these challenges, IBM developed an initial proof of concept that is targeted to the scenario described, known as Edge IoT Analytics. IBM used Watson Tone Analyzer and Watson Speech-to-Text as modules for edge gateway, and a distributed IoT query engine functions as a cloud module.

In the prototype, IBM shows how inexpensive devices like a Raspberry Pi that are equipped with a microphone can be placed at the hotels’ reception desks and function as edge gateways.

As outlined schematically in Figure 2, the edge application modules that run on gateways capture the audio of conversations and run tone and speech-to-text analysis directly on the device, thus fully avoiding sending any data that is sensitive to the cloud. The edge gateways store only the tone analysis results that are just numerical scores for a fixed set of tone attributes, such as anger, sadness or happiness, on a local database.

http://analytics-magazine.org/wp-content/uploads/2018/01/FTR-04-Deep-Dive-Figure-2-200x133.jpg 200w, http://analytics-magazine.org/wp-content/uploads/2018/01/FTR-04-Deep-Dive-Figure-2-300x200.jpg300w" sizes="(max-width: 600px) 100vw, 600px">

Figure 2: IBM Edge IoT Analytics schematic workflow.
Source: IBM

To allow the general hotel management access to this global information, an additional module is being run by Edge IoT Analytics on the cloud as a Bluemix service. This service implements a visual dashboard with a query engine for exploring the IoT information (Figure 3), which makes it possible to visualize what devices the application is running by viewing the map on the dashboard. A subset of standard SQL with extensions is implemented by the IoT query engine (we call it as EdgeSQL). This will make the IoT query engine suitable to query data that is stored on edge gateways.

Figure 3: IBM Edge IoT Analytics web dashboard screenshot. Source: IBM

Figure 3: IBM Edge IoT Analytics web dashboard screenshot.
Source: IBM

The hotel management can develop and execute queries such as:

  • Which hotel, among the chain of hotels, had the most angry interactions with guests?
  • Within the last three months, what was the saddest conversation in Athens?”
  • In Rome, what is the average happiness of customer interactions?

All the communication between edge computing and the cloud leverage the services that are provided by the Watson IoT Platform.

Future Directions

IBM is working on extending its prototype to a generic runtime platform and framework in order to aid developers and engineers in deploying the applications of edge computing on heterogeneous scenarios. From the developer’s point of view, IBM is extending its prototype to include: common edge-to-cloud and on-edge data exchange abilities with clear-cut visibility of what leaves the edge and what stays on it, and APIs and streaming data analytics and batch frameworks that unify and streamline data analytics across edge and cloud. IBM aims at developing the infrastructure management platform in order to deploy as well as control edge computing applications all through their complete lifecycle.

Edge computing is the key in exploiting the complete potential of IoT. The future goal of organizations that are carrying out research in the field of Edge IoT Analytics is to make it easy for everyone to develop edge computing applications and enable cognitive computing and analytics on the edge.

Savaram Ravindra ( Địa chỉ email này đã được bảo vệ từ spam bots, bạn cần kích hoạt Javascript để xem nó. "> Địa chỉ email này đã được bảo vệ từ spam bots, bạn cần kích hoạt Javascript để xem nó. ) is a content contributor at Mindmajix.com. He previously worked as a programmer analyst at Cognizant Technology Solutions. He holds a master’s degree in nanotechnology from Vellore Institute of Technology.

Bài liên quan