Computer Science - Doctoral Theses

Permanent URI for this collection

https://hdl.handle.net/10468/511

Browse

Now showing 1 - 5 of 102

Packetponder: a multi-domain, software-defined, packet-optical convergence node at the edge
(University College Cork, 2025-09-01) Raulin, Julie; Gunning, Fatima; Sreenan, Cormac J.; Gunning, Paul; Science Foundation Ireland
The exponential growth of traffic from cloud services, Artificial Intelligence (AI) workloads, and latency-sensitive edge applications (such as those running close to end-users or devices) is placing unprecedented demands on the underpinning telecommunications infrastructure. Supporting such use cases requires networks that not only can provide for high-capacity, but are also dynamic, resilient, and more efficient. Yet, despite progress, development and research in disaggregation and open networking, today’s infrastructure remains highly fragmented, with separate hardware platforms and control systems for different network domains — access, metro and core — leading to operational silos, complicated end-to-end orchestration, duplicated functions, and underused capabilities. To answer these challenges, this thesis introduces the packetponder concept: it relies on a fully open and programmable whitebox switch capable of accommodating a wide range of modular, common-form-factor pluggable transceivers. While traditionally designed to support electronic switching functions, the packetponder's open software stack and programmable hardware enable not only layer 1 and layer 2 convergence, but also position it as a strong candidate for broader applications spanning access, metro, and core networks, while enabling unified management across network layers. Indeed, by consolidating packet and optical functions into a single node, the packetponder reduces the need for multiple dedicated devices such as Optical Line Terminals (OLTs), transponders, and aggregation switches. This consolidation not only simplifies network architectures but also lowers operational complexity and, in turn, energy consumption. Moreover, by exposing all functions to a centralised controller through open and standardised APIs, the packetponder allows end-to-end visibility and control, enabling more flexible, resilient, and vendor-agnostic networks. As a proof of concept, this thesis demonstrates, for the first time, the feasibility of thepacketponder across multiple network domains. In the metro/core domain, it bridges the gap between packet and optical layers, a convergence that gained industry traction during the course of this PhD with the rise of coherent pluggable transceivers. However, managing such a device remains challenging, as operational responsibility are traditionally split between packet and optical teams. With a packetponder, the thesis proposes one approach based on common, standardised APIs and open-source software, to enable vendor-independent integration. Specifically, an open-source Network Operating System (NOS) originally designed for packet switching was extended to support the optical-layer control, including dynamic configuration of tunable and coherent pluggable transceivers. This enhancement provides a foundation for multi-domain, multi-vendor interoperability. Consequently, the control plane was re-designed to reflect the packetponder’s hybrid role: an advanced Routing and Wavelength Assignment (RWA) algorithm was developed and embedded in an SDN controller, leveraging the wavelength tunability of packetponder nodes to improve resource utilisation and fault recovery. Simulations over realistic topologies demonstrated the ability to identify and configure additional lightpaths in failure scenarios, even after conventional recovery strategies were applied. Finally, the feasibility of extending the packetponder to the network edge was experimentally validated, integrating mixed access-domain traffic via standard form-factor pluggables and open software interfaces where possible, and successfully transmitting an aggregated high-capacity signal of 99.56 Gbits/s over metro-scale distances. By co-locating packet switching with 3R optical-electrical-optical wavelength conversion, the packetponder introduces a new architectural paradigm with significant implications for the evolution of TELCO networks. Through these contributions, this thesis shows that the packetponder is a strong candidate to enable true node consolidation and end-to-end simplification of fragmented infrastructures, delivering vendor-agnostic flexibility under a unified SDN control framework. As data growth accelerates with AI and other emerging applications, such convergence is not just desirable but essential to ensure networks remain scalable, efficient, and resilient in the years ahead.
The colour of sound: enhancing pitch-colour cross-modal associations in virtual reality
(University College Cork, 2025-07-31) O'Toole, Patrick; Maye, Laura; Pitt, Ian; Mancini, Maurizio; Research Ireland
This thesis investigates how Virtual Reality (VR) can enhance cross-modal associations between pitch and colour. While previous research has shown that people often associate higher pitches with brighter colours and lower pitches with darker ones, it remains unclear whether immersive VR environments can provide a more meaningful experience that could support, personalise, or strengthen these associations. The potential of VR is how key immersive features, such as multisensory integration, embodied interaction, spatial audio, and affective environments, may influence the formation of pitch-colour associations in a more engaging and meaningful way. To address this, three mixed-methods studies were conducted, each progressing the understanding of pitch-colour cross-modality in VR. Study One tested whether cross-modal consistency scores differ between desktop and VR conditions. Study Two explored the role of immersive VR features, such as environment design, interaction, and spatial audio, in shaping users' engagement and experience of pitch-colour associations. Study Three evaluated whether personalised pitch-colour cross-modal mappings, created by participants, offered any potential learning benefits over pre-defined mappings in a music theory task. Findings from across the studies indicate that VR is a viable platform for investigating pitch-colour associations, and that design elements such as multisensory integration, interactions, embodiment, and affective environment cues can support user engagement and perceptual clarity. While consistency scores did not differ significantly between environments in Study One, participants expressed a strong preference for VR, citing greater immersion and focus. Studies Two and Three highlighted the use of minimalistic and affective environments aided task engagement, and that participants benefited from interactive tasks that supported embodied cognition. Taken together, the studies suggest that immersive VR environments, when designed with attention to cross-modal coherence, interactivity, and affect, can support a deeper understanding of pitch-colour associations. This has potential for application in learning contexts, such as the music theory training explored in Study Three. The findings have implications for the design of future VR tools in music education, multisensory learning, and accessible interface design, and also point towards further research into adaptive personalisation in virtual learning environments.
Computing the policy parameters for perishable inventory management and cold supply chains under stochastic demand
(University College Cork, 2025-10-31) Gulecyuz, Suheyl; O'Sullivan, Barry; Visentin, Andrea; Tarim, S. Armagan; Science Foundation Ireland
The objective of the work in this dissertation to introduce techniques to compute the near-optimal policy parameters for the perishable inventory problem and sustainable cold food supply chains with environmental concerns under stochastic demand, as well as tackling and dealing with the demand uncertainty and fluctuations arising from the nature of demand patterns. The solutions we present in this dissertation for the perishability problem hybridises the well-known Wagner-Whitin algorithm, the Silver-Meal heuristic on a network graph setting. Our solution for sustainable cold supply chains employs and combines the inventory routing problem and several formulations found in the literature to calculate environmental effects. We consider a perishable inventory system under a finite planning horizon, periodic review, non-stationary stochastic demand, zero lead time, first in first out issuing policy, and a fixed shelf life. The inventory system has a fixed setup cost and linear ordering, holding, and outdating costs per item. We consider the penalty cost and the service level cases separately and develop two different versions for each case. In addition, we propose a mixed-integer programming model for cold food supply chains to solve a multi-period inventory routing problem under non-stationary stochastic demand, route-dependent costs, and environmental concerns. The study aims to find the replenishment and vehicle routing plans to minimise the total expected cost while producing a minimum amount of CO₂ emissions. The first contribution presented herein is a mixed-integer linear programming model that computes near-optimal (R, S) policy parameters for the sustainable cold food supply chain problem with environmental concerns, while maintaining flexibility in ordering decisions under a pre-determined replenishment schedule. Our numerical experiments show that the (R, S) policy reduces inventory costs significantly, since it solves the excess inventory issue caused by higher excess ending inventory levels arising from the pre-determined and inflexible order quantities in the (R, Q) policy. However, the CO₂ emission levels and routing costs remain similar in both models. The reduced inventory costs make the (R, S) policy significantly reduce the total cost. In addition, our numerical experiments show that the difference between the cumulative ending inventory levels for the (R,Q) and (R,S) policies increasingly grows as the time horizon gets longer, and it results in increasingly larger differences in the total cost values for both policies. The second contribution is a heuristic for the penalty cost case of the perishable inventory problem. The problem can be solved to optimality by the stochastic dynamic programming technique; however, due to the increasing dimensionality of the inventory state which is defined by the inventory levels of each item age, it becomes impractical for longer shelf lives. Our heuristic represents the inventory model as a network graph, and then calculates the shortest path in the graph in a recursive way, based on the hybridisation of the Wagner-Whitin algorithm and the Silver-Meal heuristic. It firstly determines the replenishment periods and cycles using the deterministic-equivalent shortest path approach. Using the replenishment plan determined in the first step, it calculates the order quantities based on the demand observations as a second step. We hereby employ Bookbinder and Tan's static-dynamic uncertainty strategy to respond to demand fluctuations. Our extensive numerical studies show that the computation time is significantly reduced and the heuristic finds near-optimal policy parameters. The final contribution is the modification of our proposed heuristic for the service-level case. To our knowledge, there is no procedure in the literature that calculates the optimal solution. Our extensive computational experiments that use the near-optimal stochastic dynamic programming solution as a benchmark show that the heuristic has a similar performance in terms of near-optimality and significantly reduced computation times.
Optimised charging scheduling and predictive modelling to reduce emissions from battery electric buses
(University College Cork, 2025-07-31) Jarvis, Padraigh; Arbelaez, Alejandro; Climent, Laura; Brown, Kenneth; Sustainable Energy Authority of Ireland; Insight SFI Research Centre for Data Analytics
Battery Electric Vehicles provide an opportunity to decarbonise the transportation sector, which has proven to be a large contributor to Greenhouse Gases. This decarbonisation proves even more effective for Battery Electric Buses, as public transport services expand worldwide to help mitigate climate change. However, while Battery Electric Buses do not produce emissions themselves, the electricity used to power such vehicles may come from polluting sources, such as diesel or natural gas, resulting in lifecycle emissions. Reduction of these lifecycle emissions can aid in meeting carbon emission targets set by countries across the world. The Battery Electric Bus Charging Schedule Problem aims to create charging schedules for Battery Electric Bus to enable the continued operation of established bus routes while considering the restrictions tied to such vehicles, such as a lower operational range. The focus of this thesis is the modelling of this problem to prioritise charging of Battery Electric Buses using low-emission electricity, thus lowering lifecycle emissions. Furthermore, the impact on service quality is also considered to reduce any potential delays caused by charging. The first contribution made by this thesis is the evaluation of Deep Learning predictive model configurations for imbalanced regression problems to estimate the availability of low-emission electricity. Six Deep Learning models are considered, along with three resampling strategies and three loss metrics. The resulting 54 configurations are empirically evaluated using multiple test metrics. It was determined that Convolution Neural Networks trained with a Squared Error Relevance Area loss function performed best when estimating the availability of low-emission energy, such as excess wind-generated electricity. However, the same configuration using an Inverse-Weighted Mean Squared Error loss function appears to perform well in estimating extremes in excess wind-generated electricity. The second contribution creates a Battery Electric Bus Charging Schedule Problem to reduce the consumption of non-low-emission energy, utilising predictive models. The performance of the models was compared to a naïve scenario, and an ideal scenario with 100% accurate information, low-emission energy information. Use of the predictive models showed improvement of up to 10.04% compared to the naïve scenario. Furthermore, there was only an average discrepancy of 4.89% between the predictive and ideal scenarios. The final contribution made in this dissertation is the extension of the previously developed Battery Electric Bus Charging Schedule Problem to consider potential negative service impact. The problem aims to reduce the daily operational costs of a bus fleet, considering fuel costs, carbon emissions, and passengers’ Value of Time expressed as a financial value. Furthermore, a combination of aspects which have not been modelled before for a Battery Electric Bus Charging Schedule Problem is used to accurately represent the problem. Evaluation on a large dataset with over 1000 buses showed that carbon emissions can be reduced by up to 64%, with savings in fuel costs reaching 50% per day, while only inducing an average delay of 8.8 seconds.
Design and implementation of an n-tier metacomputer with decentralised fault tolerance
(University College Cork, 2004-03-31) Kennedy, James J.; Morrisson, John
The provision of fault tolerance and fault survival in a large scale dynamic system is complex and heretofore has been the responsibility of the application developer. The exponential increase in complexity associated with present day applications means that this traditional approach is not scalable. In addition, modern dynamic computing platforms, such as the Grid, make it impossible for application programmers to make any static assumptions on resource availability. The WebCom approach is designed to alleviate the application programmer from this, and other implementation level constraints, by providing an abstract machine behind which these implementation details are hidden. This thesis presents the design and implementation of such a machine and shows how fault tolerance can be handled automatically by the Webcom metacomputer, independently of the application.

Browse

Recent Submissions