In this thesis, a general-purpose, efficient, and portable data structure is presented, with direct applications in high-energy physics, particularly within the CMS experiment at CERN. A review of the data structure literature was conducted to enhance the existing Structure of Arrays (SoA) implementation in CMSSW. The new design increases abstraction and introduces new layout representations, including an Array of Structures (AoS) and automatic conversion mechanisms. Further extensions, such as SoA blocks and advanced composition tools, enable greater flexibility and performance portability between CPUs and GPUs through the Alpaka framework. The developed solution provides a user-friendly and flexible interface while maintaining high efficiency and zero overhead. A comparison with existing academic and industrial implementations has been performed, outlining future directions for data structure development within CMSSW and similar high-performance computing environments. The work is part of the Next Generation Trigger project, aiming to improve the computational efficiency of the CMS reconstruction software (CMSSW) across heterogeneous architectures.
In questa tesi viene presentata una struttura dati generica, efficiente e portabile, con applicazioni dirette nell’ambito della fisica delle alte energie, in particolare nell’esperimento CMS al CERN. È stata condotta un’analisi della letteratura sulle strutture dati con l’obiettivo di migliorare l’implementazione esistente della struttura SoA (Structure of Arrays) all’interno di CMSSW. Il nuovo design aumenta il livello di astrazione e introduce nuove rappresentazioni di layout, tra cui l’Array of Structures (AoS) e meccanismi di conversione automatica tra i due modelli. Ulteriori estensioni, come i SoA Blocks e strumenti avanzati di composizione, consentono una maggiore flessibilità e garantiscono portabilità delle prestazioni tra CPU e GPU tramite il framework Alpaka. La soluzione sviluppata fornisce un’interfaccia flessibile e facile da utilizzare, mantenendo al contempo alta efficienza e assenza di overhead. È stata inoltre effettuata una comparazione con implementazioni accademiche e industriali esistenti, delineando le possibili direzioni future per l’evoluzione delle strutture dati all’interno di CMSSW e in altri contesti di calcolo ad alte prestazioni. Il lavoro si inserisce nel progetto Next Generation Trigger, con l’obiettivo di migliorare l’efficienza computazionale del software di ricostruzione di CMS su architetture eterogenee.
Efficient data structure for heterogeneous reconstruction at the CMS experiment at CERN
Beltrame, Leonardo
2024/2025
Abstract
In this thesis, a general-purpose, efficient, and portable data structure is presented, with direct applications in high-energy physics, particularly within the CMS experiment at CERN. A review of the data structure literature was conducted to enhance the existing Structure of Arrays (SoA) implementation in CMSSW. The new design increases abstraction and introduces new layout representations, including an Array of Structures (AoS) and automatic conversion mechanisms. Further extensions, such as SoA blocks and advanced composition tools, enable greater flexibility and performance portability between CPUs and GPUs through the Alpaka framework. The developed solution provides a user-friendly and flexible interface while maintaining high efficiency and zero overhead. A comparison with existing academic and industrial implementations has been performed, outlining future directions for data structure development within CMSSW and similar high-performance computing environments. The work is part of the Next Generation Trigger project, aiming to improve the computational efficiency of the CMS reconstruction software (CMSSW) across heterogeneous architectures.| File | Dimensione | Formato | |
|---|---|---|---|
|
2025_12_Beltrame.pdf
accessibile in internet per tutti
Descrizione: Master Thesis
Dimensione
4.99 MB
Formato
Adobe PDF
|
4.99 MB | Adobe PDF | Visualizza/Apri |
|
2025_12_Beltrame_Summary.pdf
accessibile in internet per tutti
Descrizione: Executive Summary
Dimensione
1.28 MB
Formato
Adobe PDF
|
1.28 MB | Adobe PDF | Visualizza/Apri |
I documenti in POLITesi sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/10589/246405