ABSTRACT. TThe rapid proliferation of electric vehicles (EVs) necessitates advanced charging infrastructure, which is increasingly reliant on cloud-based technologies and the Internet of Things (IoT). However, these systems are vulnerable to cyberattacks that could have severe repercussions, including power grid failures. This paper addresses the security vulnerabilities inherent in the EV charging system. We propose a novel hybrid security solution tailored for cloud-based EV charging systems that integrates time-series-based regression models with classification algorithms to detect and mitigate both power consumption anomalies and network intrusions effectively. Our approach includes a comprehensive analysis to identify vulnerabilities and threats for cloud-based EV systems, a dual-model system for anomaly and intrusion detection, and a feedback-based threshold adjustment mechanism to assist overflow and anomaly identification. We provide a detailed analysis of data communication threats from a cloud perspective, design a robust security model, and evaluate various models to select the most effective ones for real-time security management. The finally decided models present excellent accuracy and practical value. Our findings contribute to enhancing the resilience of EV charging systems against cyber-physical threats, ensuring more reliable and secure operations.

ABSTRACT. Short-term water usage demand forecasting is crucial for improving water management and controlling water supply for residential and industrial purposes. Several studies have depicted the usefulness of water usage demand forecasting in making smart cities sustainable. However, the real-world translation of such forecasting systems still needs to be discussed. In this research, we have developed +10 days short-term demand forecasting framework that utilizes the existing state-of-the-art statistical and deep learning models. The designed framework has been mapped to the utility company’s legacy water usage demand forecasting process to promote digitization and sustainability. The experimental results demonstrate the successful integration of the data-driven system into the existing system with improved forecast accuracy.

ABSTRACT. This study examines blockchain technology's potential to enhance the stock market by offering a decentralized alternative to the traditional system. The study explores the challenges of the current traditional stock market, such as long settlement times, weak transparency, dependency on centralized clearing houses, and heightened fees. The proposed novel system leverages blockchain technology to create a fully decentralized stock market encompassing all fundamental operations from Initial Public Offerings (IPOs) to stock trading and extending to distributing profits or dividends. It connects all market participants through an Ethereum consortium blockchain fuelled by smart contracts. This approach replaces traditional intermediaries like brokers with automated regulations, ensuring more rapid transaction settlements, reduced operational errors and fees, and strengthened security measures. The proposed blockchain-based system, consisting of four smart contracts for IPO, stock exchange, dividend distribution, and participant management, demonstrates remarkable performance, showing promise in efficiently handling the current load of the largest stock markets like the London Stock Exchange (LSE). Through theoretical and practical testing, the system achieved a remarkable throughput of 120 transactions per second and latency or finality of just 4 seconds, underscoring its capability to significantly enhance the efficiency of stock market operations.

ABSTRACT. In a multiple Cloud environment, the placement of execution environments is crucial and may be subject to data constraints. Data could be anchored to some environments due to regulatory compliance, data sovereignty issues, or performance optimization. Consequently, applications and microservices must be designed to operate efficiently with these constraints. This enforces specific placement strategies to obtain good performances and scalability. In this paper, we propose a technique to enforce a constrained data-centric deployment placement in a federated multi-cloud environment. This algorithm analyzes a graph model of microservices interaction, considering communication with data storage and adopting ``anchors" for implementing a data-centric placement strategy. The results show how a better placement based on data position in multiple Clouds improves performance in terms of overall system response time. This also allows microservices to offload near data sources for multi-cloud environments, improving overall system performance without violating data movement constraints.

ABSTRACT. Microservices architecture, known for its agility and efficiency, is ideal for cloud-based software development and deployment. Often paired with containers and orchestration systems, it benefits from tools that automate resource management. Cloud platforms like AWS offer transient pricing options, such as Spot Pricing, which can significantly reduce costs. However, the dynamic demand and abrupt termination of spot VMs pose challenges. Containerization and orchestration systems can optimize microservices deployment costs using spot pricing while meeting consumer goals. We introduce SpotKube, an open-source, Kubernetes-based solution that uses a genetic algorithm for cost optimization. SpotKube autoscales clusters for microservices applications on public clouds with spot pricing, analyzing application characteristics to recommend optimal resource allocation. It features an elastic cluster autoscaler that ensures cost- effective deployments, maintains performance requirements, and handles node terminations gracefully, minimizing system availability impact. Evaluations with real public cloud setups show SpotKube’s effectiveness compared to alternative optimization strategies.

ABSTRACT. Many clouds provide containers as lightweight virtualized environments inside virtual machines (VMs). Containers can be migrated between source and destination VMs for various reasons such as load balancing. However, the performance of container migration is largely degraded by the load and virtualization overhead of VMs because the migration mechanism runs inside VMs. Conversely, the performance of containers is largely affected by the load of container migration. This paper proposes OVmigrate to enable out-of-band container migration, which migrates a container running inside a VM from the outside of the VM. OVmigrate analyzes and obtains the states of a container in the memory of the source VM using a technique called VM introspection. It enables the migration mechanism outside a VM to independently save the states of a container running inside the VM. We have implemented OVmigrate for Linux and KVM and compared the performance of state saving with the existing tool called CRIU running inside the VM.

ABSTRACT. The rise of more complex and sizeable Machine Learning (ML) models challenges traditional cloud computing models with respect to the high volume of incoming data which results in increased bandwidth usage and network congestion, as well as delays in inference. Such ML models are being rapidly developed thanks to advances in computing platforms and the real-time computing demands of ML-driven applications such as autonomous vehicles and video processing. To mitigate these challenges, ML model distribution and inference offloading to computing devices close to data sources have been explored, especially through \textit{partitioning} the models across the IoT-Edge-Cloud continuum. Existing efforts in this area have not successfully mastered the fully automatic and efficient determination of optimal partition points, along with the incorporation of Early-Exit layers which enable early termination of model inference at earlier stages when possible. In this paper, we introduce a novel partitioning algorithm designed to distribute ML models across edge devices with the goal of \textit{reducing response time when facing high-rate input data streams}. Our proposed approach leverages the principles of Dynamic Programming to determine optimal partition points and establish appropriate exit thresholds for Early-Exit layers. Our evaluation results reveal that, in the context of continuous, high-rate input data, our method consistently lowers the maximum round-trip time for processing inference requests compared to state-of-the-art methods such as NeuroSurgeon and Genetic.

ABSTRACT. In the context of modern cloud computing, the integration of specialized hardware accelerators, including Graphics Processing Units (GPUs), Tensor Processing Units (TPUs), and Field-Programmable Gate Arrays (FPGAs), is pivotal for achieving high-performance computing (HPC) and supporting the intensive computational demands of artificial intelligence (AI) and machine learning (ML) workloads. Extensive work has been done in virtualization technology to support the use of these devices in cloud architectures. However, current research is focused on concrete technologies and devices. HPC and AI work benefit from the use of different accelerator devices. There is the need to generalize and account for the heterogeneity of devices and vendors. This paper proposes a conceptual model for a heterogeneous cloud architecture that schedules HPC and AI tasks across various accelerator devices. We propose the ``running rounds'' (RR) indicator, an analytical measure used in previous work, generalized for multiple GPU programming technologies, adapted for Intel GPUs, to predict the performance of jobs on different sets of resources. The efficacy of the model is demonstrated through experiments that validate the capability of the RR indicator to approximate performance, although with noted discrepancies that highlight the need for further refinement. The study underscores the potential and challenges of heterogeneous cloud environments.

ABSTRACT. Cloud benchmarks suffer from performance fluctuations caused by resource contention, network latency, hardware heterogeneity, and other factors along with decisions taken in the benchmark design. Particularly, the sampling strategy that benchmark designers employ may significantly influence benchmark results. Despite this well-known fact, no systematic approach has been devised so far to make sampling results comparable and guide benchmark designers in choosing their sampling strategy for use within benchmarks. To identify systematic problems, we critically review sampling in recent cloud computing research. Our analysis identifies concerning trends: (i) a high prevalence of non-probability sampling, (ii) over-reliance on a single benchmark, and (iii) restricted access to samples. To address these issues, we propose methodological guidelines for sampling in cloud benchmarking. We hope that our work contributes to improving the generalizability, reproducibility, and reliability of cloud computing research.

ABSTRACT. The Berkeley Packet Filter (BPF) is increasingly used for a variety of use cases including auditing, security, monitoring, networking, etc. However, BPF lacks improved and effective tools, which makes the integration of programs written for BPF quite hard in current automated testing solutions and continue integration pipelines. We present tBPF, a library for integration testing of BPF programs that allows automated testing of arbitrary programs in a kernel agnostic manner without superuser privileges. We show that our solution can be integrated in existing development workflows and pipelines to enable reproducible testing and auditing of BPF programs. In this article, we describe our approach and compare it against other approaches in relevant work and literature.

ABSTRACT. In the era of cloud computing, accessing virtual computing resources has become increasingly convenient for users to meet their demands. Cloud providers offer two primary payment plans for virtual resource provisioning: reservation and on-demand. The reservation plan requires users to reserve resources and pay upfront, making it more cost-effective for long-term requirements despite demand uncertainty. Conversely, the on-demand plan charges based on actual resource usage, making it more expensive but suitable for short-term needs. Efficient resource provisioning is crucial to balance user demands and costs, as inefficient provisioning can lead to high costs. A key challenge is determining the optimal number of resources to reserve to accommodate unpredictable demands while minimizing costs. This paper addresses the resource reservation problem in cloud environments by focusing on the optimal reservation of virtual machines (VMs). We propose a linear regression approach that fits a linear function to features such as past demands and previously reserved instances that are still available to determine the quantity of VMs to reserve. Our model assigns specific weights to these features to predict reserved instances, minimizing the overall cost, including both the expense of reserving resources and renting additional on-demand resources as needed. Our evaluation, based on real standard workload traces, demonstrates the effectiveness of our approach in achieving cost-efficient resource provisioning, reducing the total cost by efficiently balancing reserved and on-demand resources.

ABSTRACT. In this paper, we propose an auto-adaptive ML model approach to mitigate the negative impact of data drift on model performance in cloud systems, leveraging the principles of generalization bounds and domain adaptation. Our method builds a knowledge base of time series data batches from historical cloud operational data, each linked to a trained ML model and clustered using the HDBSCAN algorithm. When performance degradation is detected, the system matches the latest inference data to a similar batch using Dynamic Time Warping (DTW) and applies the matched model's hyperparameters to the new data stream.

Extensive experiments on two real-world cloud system traces show that the proposed approach maintains high model accuracy while minimizing retraining overhead. In scenarios with frequent data drift, such as the Wikipedia trace, our framework reduces retraining overhead by 22.9% compared to drift detection-based retraining and by 97% compared to incremental retraining. In more stable environments, like the Google cluster trace, retraining costs are reduced by 96.3% and 88.9%, respectively. Scalability tests with large data inputs further show that the proposed approach efficiently handles large-scale data within a limited timeframe, making it well-suited for dynamic cloud environments.