Edge and Fog Computing along with 5G are changing the analytics landscape

Edge and fog analytics are on the rise as a result of an unprecedented increase in data capture and data flow from the Internet of Things devices. These IoT devices are spreading rapidly with the expansion of cloud-to-edge new range of machine learning and artificial intelligence (ML/AI) chips that are built-in to accommodate high-performance computing (HPC) for processing streaming data. [1] ML and AI at the edges include deep learning (DL) and reinforcing learning (RL) to process data flow and carry out analytics on real-time. This change in the analytics landscape is creating a new analytics interface supported by new cloud-to-edge architecture.[2]

Changing and evolving analytics landscape (Peter Levine 2017)

The need to carry out real-time analytics at the edge of networks is driven by the rapid proliferation of sensors and also the increase of machine to machine communication. It is estimated that 40 percent of all data will be generated from sensors by 2020 and most of it as stream and massive data. A single jet engine for example may generate up to 1 terabyte of data in a single flight.[3]

Gartner predicts that by 2022, 75% of enterprise-generated data will be created and processed outside a centralized data center, “lake” or cloud. [4] DataAge 2025 study also reported that by 2025, 20 percent of all data created will be real-time in nature that would no longer require sending it to the core of the network, to the cloud as raw data for processing. [5]

An interesting example where edge analytics was used is the double-engine boat “SilverHook” with a speed of 200 miles/hour. The data generated was received and instantly analyzed to assess the boat’s performance during the race. An algorithm generates data to be returned to the crew in seconds. The challenge is that the double powerful engine must operate at its highest level without overheating while increasing the chances to beat world records. The real-time feedback allows the release of pressure just at the right time based on real-time analytics embedded in the boat. [6]

In this changing analytics landscape, there is a shift from cloud analytics to edge analytics to address latency as well as security issues.[7] With an unprecedented growth of already voluminous data, the time that it will take for data to transit from a IoT device to “negotiate” a complex network topology at the cloud, it could be too lengthy for effective device management at the edges. By the time data reaches the cloud for analysis, the opportunity to act on it may be missed, especially in remote locations.[8]

For data to travel from the edge where it’s generated to a cloud provider and back, it may take between 150 to 200 milliseconds, while at the edges it may take only between 2 and 5 milliseconds for data to transit back-and-forth locally. The implications can be enormous, when dealing of image recognition or signal processing to assess the quality of products and improve overall productivity.

Edge and fog analytics can help tremendously today to overcome latency and the strains caused by data massive and continuous flow at the edges. Real time analytics can allow to take critical decision in remote locations, on a timely manner. The analytics at the edges can also scale processing and analytics capabilities by decentralizing effective device management to the locations where the data is being generated.[10]

In terms of privacy and security, edge and fog analytics are also more prone to keep up and rapidly catch up with defenders and breaches, on a timely manner. Privacy remain a major concern, and ensuring the protection of sensitive information is necessary in the swaths of Big Data that is generated by a never-ending and growing number of IoT devices at the edges across sectors. [11] ML/AI embedded in the edges can help to process information rapidly and locally without the need of sending valuable data over the networks to the cloud or to other processors. Precise authentication can be carried out instantly by the internet of things (IoT) devices, such as smartphones, at the edges with embedded ML/AI.[12]

Edge and fog analytics allow a decentralized approach where ML and AI can hoard information in bulk and distribute it across multiple edge devices, which could help develop systems for privacy. The edge’s capability to carry out on real time ML/AI processing, locally, without sending data over networks or to other processors, offers the opportunity of protection. This an essential feature in a time when most of our sensor-based work and activities are driven by smartphones and internet of things (IoT) devices. [13] No longer there will be a need of sending, even accidentally, individual data or other data private points all the way out to a cloud, data lake or data centre.

IoT devices are getting closer and closer to the ability to do their analysis, locally with simpler architectures and more responsive systems. [14] These devices at the edges are becoming the new analytics and authentication interface between the cloud and the real world to protect, process and derive real time information to solve problems at the edges, where data originates and privacy matters the most.

ML neural networks (Chapter 6) such as Generative Adversarial Networks (GANs), which is a type of artificial intelligence algorithm with discriminative capability that can distinguish between intrusive data and real local data. While a cyber defender can transmit intrusive messages to manipulate the outcome, GAN’s discriminator capability can distinguish the intrusive messages generated by the defender from the actual data at the edges and on real time. In addition to GANs, the Long short-term memory (LSTM) recurrent neural network can also trace the data sequences and detect immediately information that do not match the authentication. [15]

ML or AI such as deep reinforcement learning (Chapter 6) algorithm can also be used to dynamically predict the state of unauthenticated IoTDs and allow the cloud to decide on which IoTDs to authenticate. [16] ML reinforcement learning, which is requiring little or no “ground truth” training data, is expected to account for 50% of envisaged ML-based edge applications in the near future.[17]

Embedded systems include also object detection (Chapter 3) processors and machine learning (Chapter 4) that filter instantly information for identification purposes. The filtering of information at that the edge has also the advantage of reducing the bulk of data transmitted to the cloud.

Today, the IoT devices are more appealing to ML and AI applications because of the increase in their processing power. The Apple-designed 64 bit Cortex A10 ARM architecture has more than 1 billion of transistors with multicore processor to process approximately ~3.5 billion instructions per second. A mobile phone CPU running at speeds of more than 1 GHz can execute millions of calculations each second. When compared with the processing rate of instructions of computers used to guide spacecraft to the moon 45 years ago, the mobile CPU today is more than ten thousands of times faster . [18, 19]

With the arrival of the fifth generation telecommunications (5G), there will be even more dramatic changes at the edges. This fifth generation will make possible more billions of new instantaneous connections, while ensuring compliance with ethical policies and that the connections are secure. It is expected that the 5G technology will likely change things more profoundly than 3G and 4G as on 2019 onwards. It has been likened to what electricity or automobile did in long past. [20]

Edge analytics refers to the processing and analysis of information closer to the point of origin. Edge analytics is also known as distributed analytics performed on raw data, at the data collection point or close to it. [20] Gartner refers to edge computing/analytics as solutions that facilitate data processing and analytics at or near the source of data generation, close to the thing or person generating it. [22]

The term Fog, which is coined by CISCO (2015), refers to a form of cloud computing and analytics closer to the network edge. Wen many edges are involved requiring to ensure a degree of orchestration among them, fog platform becomes much important or as important as cloud analytics. [23] While fog analytics and computing focus on infrastructure and communicational (Chapter 2) edge computing focuses on things at the edges. [24]

Open Fog Consortium defines fog computing as “a system level horizontal architecture that distributes resources and services of computing, storage, control and networking anywhere, along the continuum from cloud to Thing”. By analogy to Hadoop ecosystem (Chapter 2), which is as distributed paradigm to process large data sets, fog computing is another distributed paradigm providing distributed analytics platform which may not be supported by centralized cloud computing distributed paradigm.

Edge and fog analytics shares many similarities as both allow analytics to be carried out closer to the devices that produce data. Edge analytics aims to harness the power of the IoT device itself and fog analytics aims at the infrastructure and architecture to rally together the networked devices such as radio access networks (RAN), which was proposed by ETSI as a platform that pushes cloud computing capabilities closer to mobile devices.[24] ETSI was established, in 1988, to address barrier by establishing common standards in telecommunications, broadcasting and other electronic communications networks and services, including Multi-access Edge Computing (MEC).[25]

This period of transition of cloud-to-edge computing and analytics is also referred to as the “Cambrian Explosion” period, by analogy to the sudden appearance of all major animal body plans some 500 million years ago, leading to changes in earth’s ecosystems.

Edge analytics can go beyond handling data analytics. The ability of edge and fog analytics to manage the amount of data traffic at the edges and of addressing energy-efficiency to save power in remote battery-based nodes can help in extending operational lifespan in the absence of a grid system. Their ability to encompass more sophisticated algorithms to carry out digital signal processing (DSP) can increase their memory capacity to allow data to be buffered for longer low-power states.[26]

It has been reported by the World Forum that a signal, from its base station can undergo a loss of up to 130 decibels before it reaches a mobile phone. In terms of loss it is an impressive and gigantesque loss as it is almost the equivalence of transmitting signal power of roughly the size of the Earth, of which only a tiny fraction, the size of a bacteria, is received at the edge. That is a tremendous loss of power requiring extraordinary engineering to compensate for the effect of the loss on the things we sent and receive across the airwaves in terms of words, pictures and videos.[28]

Applications at the edges using machine-learning algorithms that run directly on IoT devices, and only interact with the Cloud, if needed, for example, to train machine learning models continuously. This can help in addressing the loss of information and ensures that applications on remote locations are safe and not disrupted in case of limited or intermittent network connectivity.[29]

ML and AI at the edges can also be combined with other approaches such power-law like approaches or fractal geometry (Chapter 3) to address intermittent phenomena, such as wind or solar energy, due to changing environment conditions and reduce costs. These power-law like approaches can capture better natural fluctuating non-linear phenomena. SAMAWATT company, in Sweden, leverages fractal geometry to reduce in Sweden parameters uncertainty in sensors, leading to a 30% reduction, on average, in grid imbalance penalties for their clients. The fractal-based algorithm was used, at the edges, on wind and solar farm operators to solve the high grid imbalance challenges due to intermittent power production and uncertain weather conditions.[30]

“Smart edge devices and sensors are popping up everywhere. Now, there’s a new fog rolling in that may give that trend a major boost.” [31] Instead of transmitting data up to the cloud, to the source intelligence, edge and fog analytics can carry out the intelligence at the source of data. According to Automation World [32] : “Both edge analytics and fog analytics involve pushing intelligence and processing capabilities down closer to where the data originates.”

— -

This article is the chapter from the book titled “Edge and Fog Analytics — The New Analytics Interface”.

OperAI develops IoTs with Math and AI Embedded Solutions to speed up and streamline operational processes at the edges of the cloud.