• We propose EXASAGE, the first ODA operational data analysis assistant for data centers. To the best of our knowledge, this is the first prototype of a Large Language Model (LLM)-based tool that provides an AI-driven interoperable layer designed to interact with data collected at data center facilities, serving as an on-demand data access assistant that generates graph database query codes for timely, non-critical operational analysis. • The proposed framework leverages a Knowledge Graph (KG) approach instead of a standard NoSQL database at a data center. To achieve this, we provide a formal representation of the data collected at the data center using a Resource Description Framework (RDF) ontology. • We evaluated the framework in a real-world setting using 1,000 complex queries representative of the daily tasks performed by facility managers and engineers. The framework achieved a 93.6% accuracy for correctly generated and executed graph queries, compared to only 25% accuracy for standard NoSQL query generation, demonstrating the benefits of combining LLMs and KG. • We address the significant storage challenges caused by time-series data conversion into a KG, which results in a storage size increase of than 745x compared to NoSQL database storage, using virtualization of KGs. This results in a max storage overhead of just 52.62 MiB over all the 1000 user input queries. Data centers increasingly depend on Operational Data Analytics (ODA) for real-time insights from vast streams of telemetry data. They typically utilize NoSQL databases for scalability and data diversity, which leads to unstructured data representation and presents significant challenges for the data interoperability. Indeed, the lack of standardization, combined with schema flexibility and complex data structures, makes it difficult for system administrators to write and execute queries, ultimately complicating the automation of data retrieval tasks. Pre-trained Large Language Models (LLMs), with their latent knowledge, promise a ready-to-use AI-driven data interoperability layer, enabling data retrieval through natural language input. However, they often generate inaccurate or hallucinated query code when handling heterogeneous data sources and complex data structures. In this paper we present EXASAGE, the first ODA operational data analysis assistant that leverages a Knowledge Graph (KG)-based approach, addressing these LLM limitations and simplifying data retrieval tasks in data center facilities through a prototype implementation. EXASAGE employs an LLM based query generator as an interoperable layer to convert natural language into SPARQL queries (native to KGs), executed at a graph database endpoint, along with a virtual KG approach that retrieves only the data relevant to the user input query. In evaluations on 1,000 user input queries, EXASAGE achieved a 93.6% accuracy in generating correct SPARQL code and retrieving correct answers, significantly outperforming the 25% accuracy of NoSQL/SQLite queries, which frequently exhibited hallucinations. Furthermore, SPARQL queries are generally more concise and demonstrate shorter inference and execution times compared to compared to NoSQL/SQLite queries. For EXASAGE, the average end-to-end time for a single execution cycle is 12.77 seconds, which is suitable for interactive, non-critical operational data analysis tasks. The maximum observed storage overhead across all generated virtual KGs is just 52.62 MiB.

Cooperative perception extends a Connected and Autonomous Vehicle (CAV) capability to detect obstacles beyond its physical sensors towards a collective coverage by enabling nearby CAVs to sense and share their respective perception data. Unfortunately, the process is energy hungry, both in terms of hardware considering multiple on-board LiDARs and Cellular V2X (C-V2X) communication devices as well as computationally intensive algorithms for object detection. This paper proposes an energy-aware cooperative perception platform EECP, leveraging heterogeneous sensors along multiple object detection algorithms such as Vision Transformers (ViTs) to optimize energy consumption to reduce compute/communication resource usage while satisfying accuracy and timeliness constraints using a real testbed. The EECP Proof-of-concept development reveals several challenges towards achieving energy efficiency and timeliness for cooperative perception such as high computing/communication resources consumption of complex models, end-to-end delays along other calibration/synchronization issues.

Connected and Autonomous Vehicles (CAVs) can share their future trajectories with nodes around them as the intended navigation path, for nearby nodes to avoid crashing into them. However, trust must be established on the shared trajectories where the nearby nodes can verify the truthfulness of the shared trajectories in an efficient and timely manner. This paper proposes FLOWER, a federated learning based approach as a distributed zero trust security protocol for nearby nodes to verify the trajectories shared among CAVs by employing a machine learning algorithm to predict the corresponding future trajectories and verify the truthfulness of the data shared by the CAV via a blockchain based consensus. We employ several machine learning algorithms including transformer models on realistic trajectories from New York City to achieve this and results have shown that simple time series algorithms (RNN, LSTM, GRUs) achieved similar performance without additional complexity for real-time verification of CAV trajectories.

Modeling and understanding an expensive next-generation data center operating at a sustainable exascale performance remains a challenge yet to solve. The paper presents the approach taken by the Graph-Massivizer project, funded by the European Union, towards a sustainable data center, targeting a massive graph representation and analysis of its digital twin. We introduce five interoperable open-source tools that support this undertaking, creating an automated, sustainable loop of graph creation, analytics, optimization, sustainable resource management, and operation, emphasizing state-of-the-art progress. We plan to employ the tools for designing a massive data center graph, representing a digital twin describing spatial, semantic, and temporal relationships between the monitoring metrics, hardware nodes, cooling equipment, and jobs. The project aims to strengthen Bologna Technopole as a leading European supercomputing and big data hub offering sustainable green computing for improved societally relevant science throughput.

Trajectory data from mobile, micro-mobility devices (e-scooter, e-bikes, etc) and vehicles need validation regarding its trustworthiness for utility in different applications. Sharing of false trajectories from compromised devices can lead to potentially fatal consequences for safety-related applications. There is no scalable method to assess the truthfulness of trajectory data in real-time, therefore, this paper proposes FLOATING, leveraging federated reinforcement learning to automate trajectory validation on a private-by-design blockchain. FLOATING employs a three-tier consensus process for nodes in each others vicinity to endorse trajectories in real-time. We evaluate FLOATING using NS-3 and it shows to achieve lower delays and network overhead for a network size of up to 50 nodes participating in the consensus, while reducing network resource utilization by 10 times.

Using local caches is becoming a necessity to alleviate bandwidth pressure on cellular links, and a number of caching approaches advocate caching popular content at nodes with high centrality, which quantifies how well connected nodes are. These approaches have been shown to outperform caching policies unrelated to node connectivity. However, caching content at highly connected nodes places poorly connected nodes with low centrality at a disadvantage: in addition to their poor connectivity, popular content is placed far from them at the more central nodes. We propose reversing the way in which node connectivity is used for the placement of content in caching networks, and introduce a Low-Centrality High-Popularity (LoCHiP) caching algorithm that populates poorly connected nodes with popular content. We conduct a thorough evaluation of LoCHiP against other centrality-based caching policies and traditional caching methods using hit rate, and hop-count to content as performance metrics. The results show that LoCHiP outperforms significantly the other methods.

Vehicular Cloud (VC) has gained popularity today allowing mobile users to access a variety of on demand resources while on the move using low cost Vehicular Network. VC enables vehicles with sufficient resources to act as mobile cloud servers and provide their computing, communication and caching resources to nearby vehicles. However, due to high mobility and intermittent connectivity, it is challenging for mobile users to efficiently discover providers' services before request targeted services from them. Therefore, service advertising is of great interest with which offered services by Provider Vehicles (PVs) in the vehicular cloud can be fast propagated into the network. Given PVs' limited budget for renting advertiser vehicles, how to achieve the maximum service advertising coverage within a given period of time for a given budget requirements is NP-hard. This work aims to propose a new Centrality-based approach, PUBLISH, for PVs' services advertising in the vehicular cloud. We exploit the centrality score of both services and vehicles to find the best set of 'appropriate' vehicles as services advertisers. Results from scalable simulations show that PUBLISH efficiently identify the best services advertisers in comparison to other schemes in the literature.

Vehicular Cloud Computing (VCC) exploits resources at vehicles, such as computing, storage and internet connectivity to provide services for applications supporting different ITS (Intelligent Transportation System) services. Current Vehicular Cloud (VC) systems allow Consumer Vehicles (CVs) to discover and consume offered services by nearby mobile cloud servers (vehicles). However, to consume the required services, the CVs must first select the most suitable service provider, given that each of providers is characterized by specific features, limitations and prices. To the best of our knowledge, no work to date addresses the critical question of how to select the best provider fitting the quality of services and costs requirements of the consumer vehicles. Similarly, Provider Vehicles (PVs) should adjust the provided services' features and prices under certain conditions such as the rate of consumers' requests which makes this issue even harder. In this paper, we propose GSS-VC as a new distributed game theory-based approach to manage the service provisioning in vehicular cloud. Our approach takes into account the benefit of each player and allows the CVs to find the most suitable PV based on the probability interaction between them. Simulation results are carried out using urban mobility model and illustrate the effectiveness of the proposed approach to answer the raised questions: what is the best condition under which the CVs may request the PVs for services? and how to select the best service with respect to the CV preferences? Results from extensive simulations on up to 1, 500 vehicles show that GSS-VC is a an efficient and reliable service selection scheme while achieving high QoS.

Energy efficient neighbor discovery for multiple mobile devices in each others proximity is a challenge along duty cycling where low power devices are inactive for a large fraction of time. Existing schemes allow each device to employ a schedule to become active and send periodic messages or listen to neighboring devices to ensure a neighbor discovery in a bounded delay. However, collisions can occur due to simultaneous transmission of messages from multiple devices resulting in failure of neighbor discovery. We propose to reduce the number of message transmissions in a neighbor discovery process to avoid collisions and in result enhance the number of devices discovered.To do so, in this paper, we propose Welcome, a low latency and energy efficient neighbor discovery scheme. Instead of all nodes transmitting messages, only a single node can become a delegate to discover the nodes in vicinity and provide the neighborhood information to its neighbors. A node first finds its eligibility to become delegate based on its residual energy and association to the neighborhood. It then declares itself a delegate and listens to messages from its neighbors. Finally, it broadcasts the information regarding its neighbors to the devices in its communication range. Moreover, delegates can be rotated among neighbors where a node with high eligibility can content to become delegate. Welcome is compared with seven existing neighbor discovery schemes and it successfully discovers 100% of neighbors with low energy consumption and low latency for a neighborhood size of upto 100 nodes.

Content distribution in vehicular networks is greatly impaired by high mobility and intermittent connectivity. Social-aware content distribution schemes based on typical centrality metrics address the challenge, however they suffer due to their network-centric nature instead of information centric. We suggest to exploit the recently proposed information-centric networking architecture which cater the issue by decoupling host-user and support in-network caching at intermediate nodes. In this paper, we propose a novel information-centric social-aware content distribution protocol, STRIVE confining the broadcast nature of interest/content by routing it selectively towards high centrality information facilitator vehicles. We use a three-tier forwarding strategy to discover potential local and global information facilitators in an urban environment for efficient content delivery. The performance evaluation implements a scalable simulation environment deploying up to 2986 vehicles using realistic vehicular mobility traces. Simulation results show that our proposed novel vehicle centrality based content distribution protocol, STRIVE outperforms existing social content distribution metrics used in the literature.

Modern cars are transforming towards autonomous cars capable to make intelligent decisions to facilitate our travel comfort and safety. Such "Smart Vehicles" are equipped with various sensor platforms and cameras that are capable to constantly sense tremendous amount of heterogeneous data from urban streets. This paper aims to identify the appropriate vehicles, important to be selected as information hubs for the efficient collection, storage and distribution of such massive data. Therefore, we propose an Information-Centric algorithm, "GRank" for vehicles to autonomously find their importance based on their reachability for different location-aware information in a collaborative manner, without relying on any infrastructure network. GRank is the first step to identify socially important information hubs to be used in the network. Results from scalable simulations using realistic vehicular mobility traces show that GRank is an efficient ranking algorithm to find important vehicles in comparison to other ranking metrics in the literature.

Future cars are becoming powerful sensor platforms capable to collect, store and share large amount of sensory data by constant monitoring of urban streets. It is quite challenging to upload such data from all vehicles to the infrastructure due to limited bandwidth resources and high upload cost. This invoke the need to identify the appropriate vehicles within the Vehicular Ad-hoc Network, that are important for different urban sensing tasks based on their natural mobility and availability. This paper address this problem leveraging the self-decision making ability of a "Smart Vehicle" regarding its importance in the network. To do so, we present Car Rank, an Information-Centric algorithm for a vehicle to first rank different location-aware information. It then uses the information importance, its spatio-temporal availability and neighborhood topology to analytically find its relative importance in the network. Car Rank is the first step towards identifying the best set of information hubs to be used in the network for the efficient collection, storage and distribution of urban sensory information. We evaluate Car Rank under a scalable simulation environment using realistic vehicular mobility traces. Results show that Car Rank is an efficient ranking algorithm to identify socially important vehicles in comparison to other ranking metrics used in the literature.

Modern cars are transforming towards autonomous cars capable to make intelligent decisions to facilitate our travel comfort and safety. Such "Smart Vehicles" are equipped with various sensor platforms and cameras to collect, store and share tremendous amount of heterogeneous data from urban streets. This paper addresses the efficient collection and distribution of such massive data by allowing a popular Smart Vehicle to autonomously decide its user relevant importance in the vehicular network without relying on the infrastructure network. Therefore, we propose an Information-Centric algorithm, "InfoRank" for a vehicle to rank different location- dependent information associated to it. It then uses the information importance to analytically find its influence in the network. InfoRank is the first step towards identifying the best information hubs to be used in the network for the efficient collection, storage and distribution of urban sensory information. Results from scalable simulations using realistic vehicular mobility traces show that InfoRank is an efficient ranking algorithm to find top information facilitator vehicles in comparison to other ranking metrics in the literature.

Energy provisioning trend in Wireless Sensor Networks (WSNs) is shifted towards alternate sources by utilizing available ambient energy, of which solar irradiance harvesting is considered a viable alternative to fixed batteries. However, the energy storage buffer for harvested solar energy should be adaptive to the sporadic nature of the diurnal solar radiation availability. We believe that the typical fixed battery models no longer apply in harvesting enabled sensors. Therefore, we propose a random walk based stochastic model namely; Trinomial Random Walk (TRW) model for the storage capacity of harvesting enabled sensors. We then apply the proposed model on a comprehensive solar radiation data set of four different locations around the globe. Our performance evaluation demonstrates that the proposed model better analyze the sporadic nature of the diurnal solar radiation availability for estimating the required storage capacity. We further investigate an optimal power consumption value for a given energy store size, such that the utilization of harvested energy is maximized and the probability of energy depletion is minimized. For a given energy harvesting scenario, our model better approximates the optimal load with probability of up to a maximum of 98%, compared to a maximum of 37% for the binomial random walk model.