Device-Edge-Cloud Continuum

Cloud-edge continuum is a recently recreated term used to refer to solutions that aim at taking advantage of a distributed cloud-edge infrastructure to run applications where they best fit at a given moment, which is dependent on infrastructure system, application, and constraints, eventually adapting that according to a change on those factors. Until now, there not exists a universal approach to build continuum native applications, and therefore different frameworks have been proposed focusing on the context and scenario where they are applied. In this work, we try to provide a reference architecture that allows to build relocable continuum native applications by the use of the function-as-a-service (FaaS) model, adapting the concept of scientific workflow, to build instead FaaS continuum workflows. By doing that, we will introduce a dictionary of terms and architecture guidelines to follow to get a continuum infrastructure. We will validate this work by describing a compliant implementation of this infrastructure called OpenWolf, a recently born open-source project that lets to design FaaS-based workflows for heterogeneous Kubernetes clusters. Moreover, we will also validate this project by designing and testing a continuous learning workflow used to keep security and safety in smart city.

IoT sensors in the agri-food industry are always more in use day by day. Different advanced solutions are considered for monitoring some stages of the supply chain, particularly the agricultural or farming phases. A solution that is able to monitor the whole supply chain and all the different products along it does not actually exist.

Therefore, this work is mainly focused on the entire agri-food supply chain in order to prevent fraud and damage to the food products. In this perspective, the IoT sensors – installed into the critical points of the supply chain – provided valuable data to record on the Hyperledger Fabric blockchain platform. IoT sensors could also be installed as edge devices to provide automatic generation of the list of devices present in the system, without adding information manually. Data was collected in a cloud environment, in order to make the system further flexible, reducing the cost of an eventual IT infrastructure. Through the use of a specific smart contract, called “Foodchain” written in the Go language, it was possible to analyze a more general solution. The blockchain network proposed takes into account all possible actors of the agri-food supply chain, from farm to fork and vice versa. In particular, this work evaluated the network’s performance in terms of throughput, latency, and scalability. Some interesting results, above all regarding scalability, have been achieved with the use of Google virtual machines, which represented the different actors of the supply chain. Eventually, it was also possible to add an actor to the blockchain network, without the reset and restarting it. In this way, it is easy to imagine that all agri-food products are possible to track and trace. The problem is the impact of the network workload. It will have to be analyzed for each specific agri-food supply chain or a specific product.

Nowadays, the number of people living with cancer is constantly increasing. Numerous multidisciplinary research teams are working on development of powerful intelligent systems that will support medical decisions and help patients with critical diseases, including cancer, to maintain and even increase their quality of life (QoL). ASCAPE (Artificial intelligence Supporting CAncer Patients across Europe) is an H2020 project whose main objective is to use powerful techniques in big data, artificial intelligence, and machine learning in processing cancer (breast and prostate) patients’ data in order to support their health status. A key result of the project is the implementation of an artificial intelligence/machine learning (AI/ML) infrastructure. It will allow the deployment and execution of AI/ML algorithms locally in a hospital on patients’ private data, producing new knowledge. Newly generated knowledge will be sent back to the infrastructure and will be available to other users of the system keeping private patients’ data locally in hospitals. In this chapter, we will briefly present the structure of an open AI/ML infrastructure and how federated learning (FL) is employed in it.

Multiple applications use autonomous drones to perform data collection and processing missions, which may involve heavyweight computations that need to be performed at runtime, rather than post-mission. The processing time (and thus also the total mission time) can be reduced by offloading such computations to nearby edge servers. However, these servers may have limited resources so that it is not possible to serve all offloading requests at the same time. In this case, the edge resources must be shared among the drones operating in the wider area in a fair way, so that every drone gets a share of the available processing capacity to reduce its mission time. In this work, we present different path and offload planning heuristics, which we evaluate for a wide range of mission scenarios with varying degrees of contention and for drones that are heterogeneous in terms of their flight and onboard processing capabilities. We show that the mission time of each drone can be significantly reduced, compared to the default case where all computations are performed onboard, while producing fair schedules that respect the heterogeneity of the drones and their missions.

Mobile edge computing provides an effective approach to reducing the workload of smart devices and the network delay induced by data transfer through deploying computational resources in the proximity of the devices. In a mobile edge computing system, it is of great importance to improve the quality of experience of users and reduce the deployment cost for service providers. This chapter investigates a joint cloudlet deployment and task offloading problem with the objectives of minimizing energy consumption and task response delay of users and the number of deployed cloudlets. Since it is a multi-objective optimization problem, a set of trade-off solutions ought to be found. After formulating this problem as a mixed integer nonlinear program and proving its NP-completeness, we propose a modified guided population archive whale optimization algorithm to solve it. The superiority of our devised algorithm over other methods is confirmed through extensive simulations.

Recently, ML tasks that have been traditionally associated with high-performance CPUs and GPUs have started to be performed also on highly constrained devices at the far edge. This shift toward the devices, often named TinyML, has many well-recognized advantages such as lower bandwidth requirements and energy consumption, cheaper prices, increased privacy, and scalability. However, it also poses serious challenges: first of all, it requires handling even complex ML tasks with microcontrollers (MCUs) equipped with small memories, low-performance processors, and limited power supply; moreover, TinyML has to face the additional security threats that can specifically affect small devices, which usually have to rely on less support from the hardware and the OS to implement security, and once deployed in the field, can be exposed to physical threats. A first contribution of this work is to provide a thorough review of related literature to help delineate the state of the art and classify existing approaches based on their scope, goals, and employed technical solutions. A second contribution is to delineate a research program to advance such state of the art, with a special focus on secure and energy-efficient ML applications, in the context of a standardized component-based architecture recently proposed by the MPAI organization, which applies in particular to far edge AI applications.

In the era of the Internet of Things (IoT), conventional cloud-based solutions struggle to handle the huge amount, high velocity, and heterogeneity of data generated at the network edge. In this context, the edge-to-cloud compute continuum has emerged as an effective solution to reduce bandwidth consumption and latency in large-scale applications, through seamless integration of edge computing with cloud services and features. In this chapter, we show how the compute continuum can be effectively leveraged in the context of smart agriculture, with the aim of supporting greenhouse monitoring and management. We also analyze how long short-term memory (LSTM) neural networks can be integrated into the system to cope with the presence of missing and anomalous sensor data. A thorough experimental evaluation is performed to assess the LSTM performance, also showing how the application deployment at the compute continuum can ensure higher scalability in terms of bandwidth and latency, compared to a conventional cloud-based solution. Our findings show how the joint use of the compute continuum and deep learning can enable the development of a green-aware solution that fosters sustainable and efficient agricultural practices.

The edge-to-cloud compute continuum has become increasingly popular in recent years for effectively collecting and analyzing data generated by Internet of Things (IoT) devices at the network edge, ensuring low latency, high scalability, and privacy preservation. This continuum of computing resources, features, and services, which spans from the edge to the cloud, can be effectively leveraged in various application domains like smart cities, industrial IoT, and smart healthcare. However, many unexplored scenarios still exist where this technology can be successfully applied. This chapter investigates how the compute continuum can support speaker tracking in smart spaces, such as smart homes, offices, and public venues, especially focusing on multimodal systems that leverage both audio and visual data. The effectiveness of the edge-to-cloud continuum in supporting such systems was assessed through a simulation-based experimental evaluation performed with the iFogSim toolkit. Our findings reveal that edge-cloud integration improves application performance in terms of network usage and latency, compared to a centralized solution that solely relies on cloud computing.

This chapter proposes a new routing strategy, based on Deep Reinforcement Learning, to improve traffic flow and decrease congestion in the context of smart cities. The idea is to perform routing decisions in real time while considering the actual traffic situation on the overall road network. These high-level routing actions are then translated into set-points for set-theoretic receding horizon controllers, which are in charge of computing the more adequate control action for each involved vehicle. To adequately analyze the performance of the resulting control architecture, the SUMO and MATLAB environments are used to implement complex operating scenarios where road maps data and vehicle state trajectories can be shared and exchanged. Finally, some simulations, performed considering a real city district, show that the proposed algorithm can successfully realize real-time routing decisions and reduce waiting time despite dynamic changes within the road environment.

Healthcare is becoming increasingly technologically advanced today, and the Internet of Things (IoT) is an integral part of personal healthcare systems. In a typical personal healthcare system, vital parameters are acquired from users and stored on a cloud platform for further analysis. Such systems analyze specific diseases through the acquisition of biomedical signals, including electroencephalography (EEG), electrocardiography (ECG), electromyography (EMG), and electrodermal activity (EDA). Wireless sensors are used to capture these physiological signals, which consume more power for signal acquisition and sampling, as well as large bandwidth during real-time data transmission. Therefore, real-time data compression is necessary to consume less power and channel bandwidth. Compressed sensing is a new paradigm that exploits signal sparsity and has attracted significant interest from researchers due to its ability to faithfully reconstruct signals from only a few measurements (less than the Nyquist sampling frequency). It also enables feature extraction directly from compressed measurements. This chapter provides the framework of compressed sensing for Internet of Medical Things (IoMT)-based healthcare devices, discussing various technical aspects of research components. A key aspect of this study is to investigate optimal sensing methods, reconstruction algorithms, and reconstruction-free approaches.

In the contemporary era, a substantial portion of global energy consumption is allocated to residential and office buildings. Regrettably, a considerable amount of this energy is squandered due to inefficient utilization of electrical systems. One of the recognized approaches to curbing this wastage involves the detection, learning, and prediction of user presence within buildings, enabling proactive measures based on these forecasts.

Forecasting the presence and occupancy of individuals within building environments can yield substantial energy savings and contribute to maintaining optimal comfort levels for the occupants. Additionally, such predictions hold significant potential for enhancing security and ensuring the safety of individuals within the buildings.

Given the aforementioned aspects, this chapter aims to provide an overview of the prominent research conducted on occupancy forecasting in building environments. Initially, we will examine the key monitoring methods, based on Internet of Things technologies, employed for assessing presence in buildings. Subsequently, we will delve into some machine learning and deep learning algorithms utilized for predicting occupancy. Finally, we will explore potential future directions in this field.