Coupling software are the backbone of modern climate models. So called Coupled General Circulation Models, employed in climate simulations, are the result of the integration between several, optimized numerical codes modeling individual components of the Earth’s climate system. OASIS3-MCT (CERFACS & CNRS) is currently one of the most popular coupler software in the community for this goal, enabling an efficient communication paradigm between several models on HPC architectures. XIOS (CEA/LSCE - IPSL), on the other hand, is an I/O library that efficiently manages post-processing and I/O of models data on supercomputers file systems. Due to the significant overlap in their functionalities, particularly in parallel communication, domain partitioning, model data operations, and grid-to-grid interpolation, as well as their common use within the software stacks of CGCMs, the two tools are progressively undergoing convergence within the framework of the TRACCS project. In this work, we performed a feasibility study on the latest experimental XIOS coupling functionalities, performing a qualitative and quantitative analysis on the software. To achieve this, an extensive and detailed testsuite was produced in order to highlight the potentialities and the current limitations of XIOS as a coupler for the research community. Benchmarks on up to 1024 CPU cores have been conducted to evaluate communication overhead and the weights generation routine for grid-to-grid interpolations. The resulting work is intended to serve both as a starting point for the future development of XIOS in this direction and as a guide to assist current OASIS users in migrating to XIOS. Furthermore, the work includes a comprehensive review of coupling strategies relevant to this class of problems, extending beyond climate modeling, and provides some theoretical and experimental justification.
I couplers rappresentano la spina dorsale dei modelli climatici moderni. I cosiddetti Coupled General Circulation Models, utilizzati nelle simulazioni climatiche, sono il risultato dell’integrazione di diversi codici numerici ottimizzati che modellano i singoli componenti del sistema climatico terrestre. OASIS3-MCT (CERFACS & CNRS) è attualmente uno dei coupling software più diffusi nella comunità a questo scopo, poiché abilita un paradigma di comunicazione efficiente tra diversi modelli su architetture HPC. XIOS (CEA/LSCE - IPSL), invece, è una libreria di I/O che gestisce in modo efficiente il postprocessing e l’I/O dei dati dei modelli sui file system dei supercomputer. A causa della significativa sovrapposizione nelle loro funzionalità, in particolare nella comunicazione parallela, nella partizione del dominio su più processi, nelle operazioni sui dati dei modelli e nelle interpolazioni grid-to-grid, così come del loro utilizzo comune negli stack software dei CGCMs, una analisi riguardo alla eventuale convergenza dei due strumenti è in corso grazie al progetto di ricerca francese TRACCS. In questo lavoro abbiamo condotto uno studio di fattibilità sulle più recenti funzionalità sperimentali di coupling di XIOS, realizzando un’analisi qualitativa e quantitativa del software. A tale scopo è stata prodotta un’ampia e dettagliata testsuite, per mettere in evidenza le potenzialità e le attuali limitazioni di XIOS come coupler per la comunità scientifica. Sono stati eseguiti benchmark fino a 1024 CPU cores per valutare l’overhead di comunicazione e la routine di generazione dei pesi per le interpolazioni grid-to-grid. Il lavoro risultante è inteso sia come punto di partenza per lo sviluppo futuro di XIOS in questa direzione, sia come guida per assistere gli attuali utenti di OASIS nella migrazione verso XIOS. Inoltre, il lavoro include una breve introduzione alle strategie di coupling rilevanti per questa classe di problemi, che non si limitano alla climatologia, al fine di fornire alcune giustificazioni teoriche e sperimentali all’approccio in analisi.
Towards new community-oriented coupling software in climate modeling: evaluating XIOS beyond I/O
Ferrario, Daniele
2024/2025
Abstract
Coupling software are the backbone of modern climate models. So called Coupled General Circulation Models, employed in climate simulations, are the result of the integration between several, optimized numerical codes modeling individual components of the Earth’s climate system. OASIS3-MCT (CERFACS & CNRS) is currently one of the most popular coupler software in the community for this goal, enabling an efficient communication paradigm between several models on HPC architectures. XIOS (CEA/LSCE - IPSL), on the other hand, is an I/O library that efficiently manages post-processing and I/O of models data on supercomputers file systems. Due to the significant overlap in their functionalities, particularly in parallel communication, domain partitioning, model data operations, and grid-to-grid interpolation, as well as their common use within the software stacks of CGCMs, the two tools are progressively undergoing convergence within the framework of the TRACCS project. In this work, we performed a feasibility study on the latest experimental XIOS coupling functionalities, performing a qualitative and quantitative analysis on the software. To achieve this, an extensive and detailed testsuite was produced in order to highlight the potentialities and the current limitations of XIOS as a coupler for the research community. Benchmarks on up to 1024 CPU cores have been conducted to evaluate communication overhead and the weights generation routine for grid-to-grid interpolations. The resulting work is intended to serve both as a starting point for the future development of XIOS in this direction and as a guide to assist current OASIS users in migrating to XIOS. Furthermore, the work includes a comprehensive review of coupling strategies relevant to this class of problems, extending beyond climate modeling, and provides some theoretical and experimental justification.| File | Dimensione | Formato | |
|---|---|---|---|
|
2025_10_DanieleFerrario_v1.pdf
accessibile in internet per tutti
Dimensione
7.41 MB
Formato
Adobe PDF
|
7.41 MB | Adobe PDF | Visualizza/Apri |
|
2025_10_DanieleFerrario_ExecutiveSummary_v1.pdf
accessibile in internet per tutti
Dimensione
2.41 MB
Formato
Adobe PDF
|
2.41 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/10589/243466