Publication: Asynchronous Two-level Checkpointing Scheme for Large-scale Adjoints in the Spectral-Element Solver Nek5000
Introduction
Applications
Tools
Research Groups
Workshops
Publications
   List Publications
   Advanced Search
   Info
   Add Publications
My Account
About
Impress

Asynchronous Two-level Checkpointing Scheme for Large-scale Adjoints in the Spectral-Element Solver Nek5000

- Article in a journal -
 

Author(s)
Michel Schanen , Oana Marin , Hong Zhang , Mihai Anitescu

Published in
Procedia Computer Science

Year
2016

Abstract
Adjoints are an important computational tool for large-scale sensitivity evaluation, uncertainty quantification, and derivative-based optimization. An essential component of their performance is the storage/recomputation balance in which efficient adjoint checkpointing strategies play a key role. We introduce a novel asynchronous two-level adjoint checkpointing scheme for numerical time discretizations targeted at large-scale numerical simulations. The checkpointing scheme combines bandwidth-limited disk checkpointing and space-limited binomial memory checkpointing. Based on assumptions about the target petascale systems, which we later demonstrate to be realistic on the IBM Blue Gene/Q system Mira, we create a model of the predicted performance of the adjoint computation and validate it using the highly scalable Navier-Stokes spectral-element solver Nek5000 on small to moderate subsystems of the Mira supercomputer.

AD Theory and Techniques
Checkpointing

BibTeX
@ARTICLE{
         Schanen2016ATl,
       title = "Asynchronous Two-level Checkpointing Scheme for Large-scale Adjoints in the
         Spectral-Element Solver {Nek5000}",
       journal = "Procedia Computer Science",
       volume = "80",
       pages = "1147--1158",
       year = "2016",
       note = "International Conference on Computational Science 2016, ICCS 2016, 6--8 June 2016, San
         Diego, California, USA",
       issn = "1877-0509",
       doi = "10.1016/j.procs.2016.05.444",
       author = "Michel Schanen and Oana Marin and Hong Zhang and Mihai Anitescu",
       keywords = "Two-Level Checkpointing, Adjoints, Gradient, Large Scale, Nek5000, CFD",
       abstract = "Adjoints are an important computational tool for large-scale sensitivity
         evaluation, uncertainty quantification, and derivative-based optimization. An essential component of
         their performance is the storage/recomputation balance in which efficient adjoint checkpointing
         strategies play a key role. We introduce a novel asynchronous two-level adjoint checkpointing scheme
         for numerical time discretizations targeted at large-scale numerical simulations. The checkpointing
         scheme combines bandwidth-limited disk checkpointing and space-limited binomial memory
         checkpointing. Based on assumptions about the target petascale systems, which we later demonstrate
         to be realistic on the IBM Blue Gene/Q system Mira, we create a model of the predicted performance
         of the adjoint computation and validate it using the highly scalable Navier-Stokes spectral-element
         solver Nek5000 on small to moderate subsystems of the Mira supercomputer.",
       ad_theotech = "Checkpointing"
}


back
  

Contact:
autodiff.org
Username:
Password:
(lost password)