Task placement optimization in cloud-edge-fog environments is a challenging problem that requires balancing multiple objectives, such as minimizing latency and energy consumption, while adhering to resource constraints. This paper proposes a framework that integrates Deep Reinforcement Learning with Graph Neural Networks to address these challenges. Specifically, we explore the effectiveness of such architectures, including Graph Convolutional Networks and Message Parsing Neural Networks, within a DRL agent for task allocation. The framework is evaluated on synthetic task flow graphs, representing parallel workflows of varying complexities ($\mathbf{1 0}$ and $\mathbf{1 0 0}$ tasks), and benchmarked against traditional methods such as Genetic Algorithms and a Random Agent. Our results demonstrate that the RL Agent with GCN layers outperforms the MPNN-based RL Agent, GA, and Random Agent in small-scale scenarios while it performs equivalently with the MPNN-based RL Agent in largescale scenarios in which both surpass the heuristic approach and the Random agent.
Wind energy plays a pivotal role in the global shift toward sustainable energy systems. However, the maintenance of wind turbines remains a significant challenge due to their distributed nature, harsh environmental exposure, and the high cost of unplanned downtime. In this work, a novel architecture for predictive maintenance of wind turbines based on continuous acoustic monitoring is presented, based upon OASEES—a decentralized, intelligent, and programmable edge framework that spans the full computing continuum. The proposed system leverages low-cost recording equipment to capture turbine-generated sound data, which are processed locally at the edge using Federated Learning, thus preserving data privacy and reducing communication overhead. A pre-trained deep learning model based on wav2vec is fine-tuned to classify turbine operational states, using labeled acoustic datasets. The effectiveness of the architecture, which, to the best of the authors' knowledge, is among the first to utilize the said distributed learning paradigm for acoustic-based wind turbine predictive maintenance, is validated in a proof-of-concept experimental setting using a publicly available relevant dataset, where both centralized and federated training methods are evaluated. The results demonstrate promising classification accuracy, with the federated model achieving over 78 % accuracy, closely matching the centralized baseline.
Centralized solutions offered by cloud providers are increasingly prevalent in the Internet, despite being single points of failure and threatening the privacy of users by giving access of large amounts of data to a few powerful entities. Decentralized peer-to-peer systems have garnered attention in opposition. However, their limitations, such as high latency and message complexity in searching content constitute a barrier to widespread adoption. This is the case of InterPlanetary File System (IPFS), where the algorithm used to search content comprises two steps: an optimistic and a fallback step. The optimistic step relies on the Bitswap protocol and operates on an unstructured network that tries to find content in a single hop. This solution has low latency but also low success rate. Should a search fail with Bitswap the fallback alternative resorts to the use of a Distributed Hash Table (DHT) to complete the search with a significantly increased latency cost. We propose, implement, and evaluate Goose, an alternative to the Bitswap protocol that exploits principles of informed search to introduce a distributed lightweight indexing scheme and an indirection layer in the unstructured network; increasing the success rate of searches when compared with Bitswap, significantly reducing the reliance on the DHT and improving the median access latency in IPFS by 27% and reducing the number of messages exchanged by up to 67%.
We revisit the concept of order fairness in blockchain systems. Order fairness imposes additional constraints on the actual order of the transactions, preventing adversaries from manipulating the order of transactions to gain undue advantages. While numerous notions of fairness and corresponding fair ordering protocols have been proposed, these efforts typically focus on designing a dedicated protocol tailored to each specific fairness notion. Consequently, it remains unclear whether multiple fairness notions can be realized simultaneously within a single framework. In this work, we propose “2-hop receiver fairness”, a new and unified fairness notion that simultaneously captures approximate sender fairness, block order fairness, and consequence transaction fairness. To achieve 2-hop receiver fairness, we propose TxSort, a generic framework that reduces the fair ordering problem to the asynchronous common subset (ACS), an extensively studied primitive for asynchronous consensus and multiparty computation. In this way, we can integrate the state-of-theart ACS protocols to provide a communication-efficient and round-optimal fair ordering protocol. Specifically, our protocol achieves an amortized communication complexity per transaction (CCpT) of $O\left(n^{2}\right)$, while existing state-of-the-art fair ordering protocols achieve $O\left(n^{3}\right)$ CCpT. As a crucial building block of our framework, we introduce a novel and computation-efficient local algorithm called “pivot quick sort”, which might be of independent interest.
Land registry systems maintain public records of land ownership, transactions, and property boundaries, offering legal documentation and information on titles and associated rights. However, there are challenges in the security, accuracy, efficiency, and transparency of the current registries. The paper proposes the Decentralised Land Registry (DLR), a system that leverages a permissioned blockchain, Hyperledger Fabric, enabling a secure and controlled environment for land registration and transaction validation. The use of smart contracts ensures consensus and immutability and allows support for legitimate property transactions overseen by registry authorities. Decentralised identities and ownership proofs are issued using a Self-Sovereign Identity approach, leveraging Hyperledger Indy as the ledger and Hyperledger Aries as the agent representing government entities and issuing these credentials. This solution would mark a pivotal shift in land registry systems, revolutionising traditional practices by addressing their limitations and fostering a transparent, reliable, and efficient landscape for managing property rights and transactions.
