This paper explores an extended applications' cost function to model the willingness of Edge data centres to accommodate additional users in decentralized edge computing environments. By enhancing the Marginal Computing Cost per User (MCU) concept, we introduce a dynamic cost factor influenced by the number of users currently served. Through extensive simulations conducted on the PureEdgeSim platform, we evaluate the impact of this variable MCU on system performance across various configurations. The results reveal a critical trade-off between cost sensitivity (i.e., collaboration willingness) of Edge data centres and optimization potential. This work offers insights into user allocation strategies in heterogeneous edge systems and sets the stage for future research into non-linear MCU configurations and diverse application workloads.
Testing serverless functions differs significantly from that of conventional software products. Their distributed nature and event-driven architecture make test development inherently complex. Further, the development of functions and their tests often relies on using tools and software development kits managed by cloud providers, resulting in minimal control over the test execution environment and a lack of debugging tools. As such, we have developed the Icarus RESTful, open-source Testing-as-a-Service tool with a transparent and predictable operation, supplying the user with full control over test creation and execution. Icarus does not burden the user with details of a function's deployment across different providers, does not require the user to write any line of testing code and relies solely on using familiar open-source and well-known tools to automatically compose and execute functional and non-functional tests. Its experimental evaluation shows that it scales well with the user and workload increase, rapidly delivering test results to the user.
This extended keynote abstract outlines research on a new paradigm of data sharing designed to recover privacy for data providers, while reducing costs and risks of service providers, who collect personal sensitive data. This paradigm is based on the novel idea of coordinated data sharing within a community, the data collective, which makes automated and scalable collective arrangements of how to share data using decentralized artificial intelligence.
Digital Twins (DTs) are digital representations of physical objects or processes that can be used for their computer-based analysis. This technique has been used in different fields to analyze, simulate, and optimize various scenarios in real time without interfering with the real twin. Using Serverless Computing as a use case, this paper discusses the underlying costs of using DTs when applied to analyze and optimize the management of computational resources for cloud-continuum applications. We argue that, although feasible, using DTs for Digital Systems can be prohibitively expensive. Defining the ideal DT fidelity for a given application is challenging, as this impacts both system management and performance through heavy monitoring as well as the DT running costs.
This study introduces FaaS-Utility, a novel approach aimed at optimizing Function-as-a-Service (FaaS) systems by addressing the critical issue of cold starts, which significantly impede system performance. By introducing a utility function informed by customer preferences and pricing goals, our methodology prioritizes resource allocation to enhance service quality effectively. We implement this strategy within Apache OpenWhisk, demonstrating its integration into a real-world FaaS platform. Our evaluation reveals that the proposed approach notably improves system performance, particularly in over-provisioned states, by reducing latency up to 2.37 times with a maximum additional cost of only 30
Despite technological advancements, ensuring aircraft safety remains a challenge, however, Machine learning (ML)-based approaches for predicting future incidents play a crucial role in addressing flight safety. As ML models increase in complexity, their decision-making process becomes less transparent, posing significant challenges to trustworthiness. While simpler models demonstrate lower accuracy, more intricate models such as deep neural networks achieve higher accuracy but sacrifice interpretability. In this study, we enhance trustworthiness in aircraft safety prediction by leveraging a dataset of past accidents and incidents to prevent similar accidents from occurring in the future. To achieve this, we apply Random Forest and Extreme Gradient Boosting models to classify different categories of aircraft incidents. Additionally, we apply two powerful explainable artificial intelligence (XAI) techniques: Local Interpretable Model-Agnostic Explanations (LIME) and Shapley Additive exPlanations (SHAP) to provide insights into both local and global predictions made by the models. Notably, our results reveal high accuracy in these predictions while maintaining trustworthiness. This research contributes to the advancement of XAI and offers valuable insights for safety-critical applications and decision support systems.
Stream Processing is a very effective predominant paradigm for data processing. It provides an efficient approach to extract information from new data, as the data arrives. However, spikes in data throughput, can impact the accuracy and latency guarantees stream processing systems provide. This work proposes data sampling, a type of data reduction, as a solution to this problem. It provides a user-transparent implementation of two sampling methods in the Apache Spark Streaming framework. The results show a reduced amount of input data, leading to decreased processing time, but retaining a good accuracy in the extracted information.
Data centers are increasingly becoming significant energy consumers worldwide. To reduce the amount of electricity they consume, power capping may be used to set a limit to the maximum power they can use at some given point in time. In this situation, an interesting problem is how to make best use of the available power by throttling the CPU frequency of different servers. As different tasks assigned to each of these servers may not be impacted the same way when changing a server's CPU frequency, one problem that arises is how to select CPU frequencies for each of the servers running tasks with specific characteristics in such a way that the total execution time of all these tasks is minimized while the overall power cap for all the servers is respected. The paper presents an approach that models this problem as an optimization problem and shows how to find an optimal solution in different cases. This work can provide the basis to find economical solutions to operate large data centers under power capping efficiently.
Fault tolerance is essential for the correct execution of large distributed simulations of discrete event systems, as the likelihood of faults increases with the size of the cloud infrastructure used. Achieving optimal performance and cost in a fault-tolerant distributed simulation remains a challenge. In this paper, we propose a replication-based approach in a conservative distributed simulation strategy that is specifically designed to minimize latency introduced by fault tolerance mechanisms. Unlike traditional replication methods, our method is tailored for conservative simulation, leveraging simulation messages and timing to maintain consistency while decoupling replica execution. As a result, our approach reduces the need for messaging and synchronization and maintains eventual consistency windows with low latency overhead, achieving near-nominal simulation performance in the absence of faults. If replicas have similar performance, memory usage can be lower compared to optimistic approaches, and recovery can be fast following a node failure, despite asynchronous replication. Experimental results show that without faults, the performance of a distributed simulator with fault management is similar to one without it. Recovery from a fault reveals that the main overhead is in replica provisioning, with minimal overhead for synchronization.
Function-as-a-Service (FaaS) is an emerging cloud computing model ideal for processing vast amounts of data generated by the Internet of Things. However, existing FaaS approaches struggle to leverage resources efficiently on distributed edge devices. Our work presents FaaS@Edge, a solution that employs volunteered resources from edge nodes, discovered through the IPFS network, to deploy functions using the Apache OpenWhisk framework and enhancing the system’s scalability and efficiency. This approach supports various language runtimes, ensuring near-universal deployability on edge devices. Our evaluation demonstrates that FaaS@Edge introduces a latency overhead for function submission but achieves similar invocation times compared to a local OpenWhisk deployment. FaaS@Edge maintains high request success rates, with overall request success rates around 98
In edge computing, task offloading involves transferring computational tasks from the "far-edge", which includes end-user devices or less powerful edge devices, to the "near-edge", comprising more capable edge servers, or to the 'core' cloud infrastructure. This practice optimizes performance, reduces latency, and enhances overall efficiency. Energy efficiency in particular has recently become a high-priority criterion for task offloading. A prominent technique for making offloading decisions in edge computing environments is Deep Reinforcement Learning (DRL), known for its ability to adapt to complex environments and excel in multi-objective optimization tasks in terms of decision quality and speed. This paper explores the details of DRL approaches, providing an overview of recent research developments in this field. To simplify the literature analysis, we classify DRL approaches for energy-efficient task offloading between two "computing continua": the far/near-edge continuum, and the (far-)edge-cloud continuum.
Explainability is a growing concern in many machine learning applications. Machine learning platforms now typically provide explanations accompanying their model output. However, this may be considered as just a first step towards defining explainability as a service in itself, which would allow users to get more control over the kind of explainability technique they wish to employ. In this paper, we first provide a survey of the current offer of machine learning platforms, observing their pricing models and the explainability features they possibly offer. In order to progress towards Explainability-as-a-Service (XaaS), we propose to base its definition on the REST paradigm, considering three major examples of explainability techniques, relying on either feature scoring, surrogate linear models, or internal state observation. We also show that XaaS is dependent on machine-learning model provisioning, and the two services are linked by a one-way essential complement relationship, where ML provisioning plays the role of the essential component and XaaS is the complement option. We also suggest that vertical integration is the natural arrangement for companies offering either service, given their mutual relationship.
The paper investigates the prediction capabilities of a Machine Learning model in real-time scheduling applications on Kubernetes in a serverless computing environment with the aim to achieve a degree of energy efficiency. A highly pluggable framework for integrating a learning-based model into the Kubernetes scheduler is proposed and evaluated in a serverless setup on OpenFaaS. The experimental results in a cloud-native deployment demonstrate that, while maintaining Quality of Service for the application, an overall 8
This paper presents a groundbreaking approach to modernizing the Letter-of-Credit (L/C) process in trade finance by integrating RSA-Encrypted Non-Fungible Tokens (NFTs). Addressing the prevalent challenges of fraud, privacy concerns, and operational inefficiencies in traditional L/C mechanisms, our research proposes a robust, transparent, and efficient framework. Building upon the advancements in blockchain technology and smart contracts, we introduce a dual-layered security strategy combining RSA encryption with the unique characteristics of NFTs. This integration ensures the secure encryption of trade documents, accessible only to authorized parties, and leverages blockchain's immutability to create a transparent, tamper-evident record of document ownership and verification. Our approach not only enhances the security and privacy aspects of trade finance but also introduces significant improvements in efficiency through the automation of verification and settlement processes via smart contracts. This paper discusses how our innovative framework sets a new standard in international trade finance, promising a more secure, efficient, and reliable L/C transaction.
The rise of 5G fuels multi-access edge computing (MEC), a transformative computing paradigm that leverages edge resources for low-latency mobile access and complex service execution. Deploying services across geographically distributed edge nodes challenges providers to optimize performance metrics like latency and resource efficiency, impacting user experience, operational cost, and environmental footprint. In the context of service scheduling with data flow dependencies, we propose heuristic-based service placement algorithms that balance minimizing latency and maximizing resource efficiency. Our algorithms, evaluated in a simulated environment using state-of-the-art workload benchmarks, achieve significant energy consumption improvements while maintaining comparable latency.
This paper discusses a framework for supply chain management in the handicraft industry, aiming to address the challenges of transparency and ethical sourcing. We introduce a system that integrates blockchain technology, smart contracts, Non-Fungible Tokens (NFTs), and the InterPlanetary File System (IPFS) to enhance the traceability and integrity of handicraft supply chains from creation to consumer. The study reviews related work that demonstrates the importance of blockchain for trust and transparent practices within supply chains. Our methodology was tested across multiple EVM-compatible platforms, examining the operational performance and economic feasibility of the system. The results of these tests inform our analysis of the system's functionality in terms of transaction speed, cost efficiency, and reliability in data management. Our findings indicate that the proposed framework can significantly contribute to the handicraft industry by providing a transparent, reliable, and efficient approach to supply chain management.
One of the main challenges facing businesses migrating to the cloud is getting an estimate of their costs in advance. The estimators available to date allow companies to compare the different virtual machine offerings from each operator, but only venture very slightly into estimating the overall cost, which includes operational and network costs. Existing estimators include operational costs in their estimates, but almost no one considers the network cost, which is a complex but far from negligible component. In this paper, we seek to address this issue by proposing a new estimator called PricingTheCloud. It is an estimator that enables companies to have an accurate estimate of their costs in advance. Unlike other estimators, PricingTheCloud considers network costs in the cost estimation. Its evaluation shows an average accuracy of 86.73% for compute costs and 65.44% for network costs in different AWS-to-AWS scenarios as compared to AWS invoices and shows the effectiveness of the proposed estimator compared to three other cloud costs estimators namely, Cloudorado, Holori, and Vantage.
The rapid growth of e-commerce, particularly on platforms that merge socializing and shopping functions like TikTok, has led to the rise of the influencer economy, where influencers drive significant sales through their online presence. This paper aims to forecast the prominent factors influencing influencer sales on TikTok, motivated by the need to better understand this evolving economic model. We establish an original dataset containing data from 100 influencers over a three-month period on TikTok's e-commerce platform. A comprehensive descriptive analysis is conducted to identify variations among influencers, followed by a data mining process to extract key characteristics based on their behavior across different levels, including daily activities, marketing strategies, and basic demographic information. A predictive model is developed to assess influencers' sales levels, revealing that the root mean square error (RMSE) of the model is close to 13. In particular, the study identifies the top 10 most influential sales features. These findings contribute to a deeper understanding of the factors driving influencer sales, offering valuable insights for both influencers and marketers in optimizing their strategies for success on TikTok.
The specification of experiments expressed as Complex Analytics Workflows is a complex task that involves many decision-making steps with various degrees of complexity. The use of the context, the expert knowledge, and the potential for its sharing and reuse in the context of experiment specification have not been addressed sufficiently until now. Moreover, to make such knowledge instrumental, it should be coupled with specific probabilistic measures, such as particular assurances, ranking, and verification of various options. The paper aims to present a novel semantic model for probabilistic reasoning in any experimentation context coupled with a functional system for knowledge generation, reuse, and sharing. The result of our work can be used within existing experimentation engines.
Cybersecurity threats information (CTI) sharing protects firms and stakeholders from cyberattacks and avoid security vulnerabilities. However, despite these benefits of CTI sharing, firms are still unwilling to share due to barriers and challenges related to a lack of trust. Some studies explored the significance of trust in sharing cyber security information, but further studies are required to determine what dimensions compose trust, which processes support trust, and what trust building policies have been enacted to foster the sharing of information in cybersecurity ecosystems, which is the main purpose of this review. The deliverables from this review present 25 trust dimensions, 6 main processes supporting trust, and 30 trust government policies enacted to foster trust and sharing in cybersecurity. These outcomes enable the creation of a framework for building trust in cybersecurity ecosystems and facilitating the cyberthreat information sharing.
