Applications like augmented/extended reality (AR, XR), autonomous driving, control and automation procedures in Industry 4.0, have strict connectivity requirements (e.g., latency, throughput) that current networks cannot guarantee. For their successful implementation, such applications may also require computational resources (e.g., for object detection, data aggregation, inference decisions). However, the resources they require may not be available in the end devices (e.g., smartphones, sensors, actuators). Multi-access Edge Computing (MEC), among the enabling technologies for fifth generation (5G) and beyond systems, aims to support these requirements by moving cloud services to network nodes close to the end devices, so to decrease latency and the wastage of network bandwidth required to transfer input data to be processed. In this context it is necessary to define solutions able to decide how and where to allocate the services, in order to satisfy the performance requirements, taking into account the heterogeneity and dynamics of computational, storage, and network resources, distributed on the edge nodes. Moreover, Edge Intelligence (EI), i.e., the edge orchestrated by Artificial Intelligence (AI) techniques (e.g., machine learning, deep neural networks, etc.), is already regarded as one of the main missing pieces in 5G networks in order to support the performance, new unprecedented functionalities, and new challenging and demanding services of future sixth generation (6G) networks. This thesis contributes to this research area with the design of a novel centralized framework based on the synergy of innovative networking technologies, for a joint and judicious orchestration of computational and network resources at the edge. The study encompasses the design of computing task allocation strategies and their mathematical formulation as optimization algorithms, their resolution through heuristics and machine learning-based algorithms, the validation of the proposal against benchmark solutions under a variety of simulation settings
Applicazioni come la realtà aumentata/estesa (AR, XR), la guida autonoma, procedure di controllo e automazione nell’Industria 4.0, hanno requisiti di connettività stringenti (ad esempio, latenza, throughput) che le reti attuali non possono garantire. Per la loro implementazione, tali applicazioni richiedono anche risorse computazionali (ad esempio, per il rilevamento degli oggetti, l’aggregazione dei dati e le decisioni di inferenza). Le risorse richieste potrebbero non essere disponibili nei dispositivi finali (ad esempio, smartphone, sensori, attuatori). Il Multi-access Edge Computing (MEC), tra le tecnologie abilitanti per i sistemi di quinta generazione (5G), mira a supportare questi requisiti spostando i servizi cloud sui nodi di rete più vicini ai dispositivi finali, in modo da ridurre la latenza e lo spreco di banda di rete necessari per trasferire i dati da elaborare. In questo contesto è necessario definire soluzioni in grado di decidere come e dove allocare i servizi, al fine di soddisfare i requisiti e tenendo conto dell’eterogeneità e della dinamicità delle risorse computazionali, di memorizzazione e di rete, distribuite sui nodi edge. Inoltre, l’Edge Intelligence (EI), ossia l’edge orchestrato grazie all’ausilio di tecniche di Intelligenza Artificiale (AI) (ad esempio, machine learning, deep neural networks, ecc.), è già considerato uno dei principali tasselli mancanti nelle reti 5G e di supportare le prestazioni, le nuove funzionalità e i nuovi servizi delle future reti 6G. Questa tesi contribuisce in quest’area di ricerca con la progettazione di un framework centralizzato basato sulla sinergia di tecnologie di rete innovative, per un’orchestrazione congiunta di risorse computazionali e di rete in un contesto edge. Lo studio comprende la progettazione di strategie di allocazione di task computazionali all’edge, la formulazione matematica di tali strategie come algoritmi di ottimizzazione, la loro risoluzione attraverso euristiche e algoritmi basati sul machine learning, la loro validazione rispetto a soluzioni di benchmark in diversi scenari di simulazione
Task allocation in a softwarized edge network / Lia, Gianmarco. - (2023 Apr 03).
Task allocation in a softwarized edge network
Lia, Gianmarco
2023-04-03
Abstract
Applications like augmented/extended reality (AR, XR), autonomous driving, control and automation procedures in Industry 4.0, have strict connectivity requirements (e.g., latency, throughput) that current networks cannot guarantee. For their successful implementation, such applications may also require computational resources (e.g., for object detection, data aggregation, inference decisions). However, the resources they require may not be available in the end devices (e.g., smartphones, sensors, actuators). Multi-access Edge Computing (MEC), among the enabling technologies for fifth generation (5G) and beyond systems, aims to support these requirements by moving cloud services to network nodes close to the end devices, so to decrease latency and the wastage of network bandwidth required to transfer input data to be processed. In this context it is necessary to define solutions able to decide how and where to allocate the services, in order to satisfy the performance requirements, taking into account the heterogeneity and dynamics of computational, storage, and network resources, distributed on the edge nodes. Moreover, Edge Intelligence (EI), i.e., the edge orchestrated by Artificial Intelligence (AI) techniques (e.g., machine learning, deep neural networks, etc.), is already regarded as one of the main missing pieces in 5G networks in order to support the performance, new unprecedented functionalities, and new challenging and demanding services of future sixth generation (6G) networks. This thesis contributes to this research area with the design of a novel centralized framework based on the synergy of innovative networking technologies, for a joint and judicious orchestration of computational and network resources at the edge. The study encompasses the design of computing task allocation strategies and their mathematical formulation as optimization algorithms, their resolution through heuristics and machine learning-based algorithms, the validation of the proposal against benchmark solutions under a variety of simulation settingsFile | Dimensione | Formato | |
---|---|---|---|
Lia Gianmarco.pdf
accesso aperto
Tipologia:
Tesi di dottorato
Licenza:
DRM non definito
Dimensione
4.8 MB
Formato
Adobe PDF
|
4.8 MB | Adobe PDF | Visualizza/Apri |
I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.