Online gaming is undergoing a phase of strong expansion, driven by increasing connectivity, the spread of cross-play platforms, and advancements in cloud technologies. In this context, the ability to effectively scale server infrastructures is essential to ensure stable performance and a smooth gaming experience, even during traffic peaks. Optimal resource management therefore represents a strategic challenge for companies in the gaming industry. This thesis focuses on forecasting the number of active players on Rocket League, one of the most representative multiplayer titles, with the goal of effectively estimating the number of servers needed to meet demand while balancing operational costs and service quality. To this end, a high-frequency dataset covering the period from December 2017 to August 2020 is used, allowing for a detailed analysis of demand trends across different time scales. The Prophet model is employed to generate forecasts over hourly, weekly, and monthly horizons, thereby supporting both short-term and long-term strategic decisions. The analysis also considers the impact of extraordinary events, such as the COVID-19 pandemic, by evaluating scenarios where such shocks are either included or excluded from the predictive model, to understand their influence on forecast accuracy. The results obtained with Prophet are then compared to those of the ARIMA model, to analyze differences in terms of accuracy and adaptability. The work concludes with an estimate of the optimal number of servers to keep active based on the predicted demand, showing how modeling results can support informed decision-making. The aim is to offer a replicable methodological framework for optimizing scalability in online gaming environments, contributing to smarter and more efficient resource management.
Il gaming online sta vivendo una fase di forte espansione, spinta dalla crescente connettività, dalla diffusione delle piattaforme cross-play e dai progressi nelle tecnologie cloud. In questo contesto, la capacità di scalare efficacemente le infrastrutture server è fondamentale per garantire prestazioni stabili e un’esperienza di gioco fluida, anche durante i picchi di traffico. La gestione ottimale delle risorse rappresenta quindi una sfida strategica per le aziende del settore videoludico. Questa tesi si concentra sulla previsione del numero di giocatori attivi su Rocket League, uno dei titoli multiplayer più rappresentativi, con l’obiettivo di stimare in modo efficace la quantità di server necessari per soddisfare la domanda, bilanciando costi operativi e qualità del servizio. A tal fine, viene utilizzato un dataset ad alta frequenza che copre il periodo da dicembre 2017 ad agosto 2020, permettendo di analizzare l’evoluzione della domanda in modo dettagliato e su diverse scale temporali. Viene impiegato il modello Prophet per generare previsioni su orizzonti orari, settimanali e mensili, supportando così decisioni strategiche nel breve e lungo termine. L’analisi considera anche l’impatto di eventi straordinari, come la pandemia di COVID-19, valutando scenari in cui tale shock viene incluso o escluso dal modello predittivo, per comprenderne l’influenza sulla qualità delle stime. I risultati ottenuti con Prophet sono infine confrontati con quelli del modello ARIMA, al fine di analizzare le differenze in termini di accuratezza e adattabilità. Il lavoro si conclude con una stima del numero ottimale di server da mantenere attivi in funzione della domanda prevista, mostrando come i risultati modellistici possano supportare decisioni informate. L’obiettivo è offrire una base metodologica replicabile per l’ottimizzazione della scalabilità nei contesti di gioco online, contribuendo a una gestione più intelligente ed efficiente delle risorse.
Previsione del numero di giocatori attivi per la scalabilità dei server attraverso il modello Prophet.
CALORI, DAKOTA DAVIDE
2024/2025
Abstract
Online gaming is undergoing a phase of strong expansion, driven by increasing connectivity, the spread of cross-play platforms, and advancements in cloud technologies. In this context, the ability to effectively scale server infrastructures is essential to ensure stable performance and a smooth gaming experience, even during traffic peaks. Optimal resource management therefore represents a strategic challenge for companies in the gaming industry. This thesis focuses on forecasting the number of active players on Rocket League, one of the most representative multiplayer titles, with the goal of effectively estimating the number of servers needed to meet demand while balancing operational costs and service quality. To this end, a high-frequency dataset covering the period from December 2017 to August 2020 is used, allowing for a detailed analysis of demand trends across different time scales. The Prophet model is employed to generate forecasts over hourly, weekly, and monthly horizons, thereby supporting both short-term and long-term strategic decisions. The analysis also considers the impact of extraordinary events, such as the COVID-19 pandemic, by evaluating scenarios where such shocks are either included or excluded from the predictive model, to understand their influence on forecast accuracy. The results obtained with Prophet are then compared to those of the ARIMA model, to analyze differences in terms of accuracy and adaptability. The work concludes with an estimate of the optimal number of servers to keep active based on the predicted demand, showing how modeling results can support informed decision-making. The aim is to offer a replicable methodological framework for optimizing scalability in online gaming environments, contributing to smarter and more efficient resource management.| File | Dimensione | Formato | |
|---|---|---|---|
|
Calori_Dakota.pdf
accesso aperto
Dimensione
895.6 kB
Formato
Adobe PDF
|
895.6 kB | Adobe PDF | Visualizza/Apri |
The text of this website © Università degli studi di Padova. Full Text are published under a non-exclusive license. Metadata are under a CC0 License
https://hdl.handle.net/20.500.12608/92933