TU Darmstadt / ULB / TUbiblio

Noise-Resilient Empirical Performance Modeling with Deep Neural Networks

Ritter, Marcus ; Geiß, Alexander ; Wehrstein, Johannes ; Calotoiu, Alexandru ; Reimann, Thorsten ; Hoefler, Torsten ; Wolf, Felix (2021)
Noise-Resilient Empirical Performance Modeling with Deep Neural Networks.
35th International Parallel and Distributed Processing Symposium (IPDPS'21). virtual Conference (17.05.2021-21.05.2021)
doi: 10.1109/IPDPS49936.2021.00012
Konferenzveröffentlichung, Bibliographie

Kurzbeschreibung (Abstract)

Empirical performance modeling is a proven instrument to analyze the scaling behavior of HPC applications. Using a set of smaller-scale experiments, it can provide important insights into application behavior at larger scales. Extra-P is an empirical modeling tool that applies linear regression to automatically generate human-readable performance models. Similar to other regression-based modeling techniques, the accuracy of the models created by Extra-P decreases as the amount of noise in the underlying data increases. This is why the performance variability observed in many contemporary systems can become a serious challenge. In this paper, we introduce a novel adaptive modeling approach that makes Extra-P more noise resilient, exploiting the ability of deep neural networks to discover the effects of numerical parameters, such as the number of processes or the problem size, on performance when dealing with noisy measurements. Using synthetic analysis and data from three different case studies, we demonstrate that our solution improves the model accuracy at high noise levels by up to 25% while increasing their predictive power by about 15%.

Typ des Eintrags: Konferenzveröffentlichung
Erschienen: 2021
Autor(en): Ritter, Marcus ; Geiß, Alexander ; Wehrstein, Johannes ; Calotoiu, Alexandru ; Reimann, Thorsten ; Hoefler, Torsten ; Wolf, Felix
Art des Eintrags: Bibliographie
Titel: Noise-Resilient Empirical Performance Modeling with Deep Neural Networks
Sprache: Englisch
Publikationsjahr: 28 Juni 2021
Ort: Piscataway, NJ
Verlag: IEEE
Buchtitel: Proceedings: 2021 35th International Parallel and Distributed Processing Symposium
Veranstaltungstitel: 35th International Parallel and Distributed Processing Symposium (IPDPS'21)
Veranstaltungsort: virtual Conference
Veranstaltungsdatum: 17.05.2021-21.05.2021
DOI: 10.1109/IPDPS49936.2021.00012
Kurzbeschreibung (Abstract):

Empirical performance modeling is a proven instrument to analyze the scaling behavior of HPC applications. Using a set of smaller-scale experiments, it can provide important insights into application behavior at larger scales. Extra-P is an empirical modeling tool that applies linear regression to automatically generate human-readable performance models. Similar to other regression-based modeling techniques, the accuracy of the models created by Extra-P decreases as the amount of noise in the underlying data increases. This is why the performance variability observed in many contemporary systems can become a serious challenge. In this paper, we introduce a novel adaptive modeling approach that makes Extra-P more noise resilient, exploiting the ability of deep neural networks to discover the effects of numerical parameters, such as the number of processes or the problem size, on performance when dealing with noisy measurements. Using synthetic analysis and data from three different case studies, we demonstrate that our solution improves the model accuracy at high noise levels by up to 25% while increasing their predictive power by about 15%.

Fachbereich(e)/-gebiet(e): 20 Fachbereich Informatik
20 Fachbereich Informatik > Data and AI Systems
20 Fachbereich Informatik > Parallele Programmierung
Zentrale Einrichtungen
Zentrale Einrichtungen > Hochschulrechenzentrum (HRZ)
Zentrale Einrichtungen > Hochschulrechenzentrum (HRZ) > Hochleistungsrechner
Hinterlegungsdatum: 02 Feb 2023 08:03
Letzte Änderung: 13 Feb 2024 15:08
PPN: 507736818
Export:
Suche nach Titel in: TUfind oder in Google
Frage zum Eintrag Frage zum Eintrag

Optionen (nur für Redakteure)
Redaktionelle Details anzeigen Redaktionelle Details anzeigen