-
E. Agullo,
M. Faverge,
L. Giraud,
A. Guermouche,
P. Ramet,
and J. Roman.
Toward parallel scalable linear solvers suited for large scale hierarchical parallel platforms.
In WCCM-ECCM-ECFD workshop on Enabling Technologies and their Application for Advancing Computational Mechanics,
Barcelona, Spain,
July 2014.
Keyword(s): Sparse.
@InProceedings{C:LaBRI::ecfd14a,
author = {Agullo, E. and Faverge, M. and Giraud, L. and Guermouche, A. and Ramet, P. and Roman, J.},
title = {Toward parallel scalable linear solvers suited for large scale hierarchical parallel platforms},
OPTcrossref = {},
OPTkey = {},
booktitle = {WCCM-ECCM-ECFD workshop on "Enabling Technologies and their Application for Advancing Computational Mechanics"},
OPTpages = {},
year = {2014},
OPTeditor = {},
OPTvolume = {},
OPTnumber = {},
OPTseries = {},
address = {Barcelona, Spain},
month = jul,
OPTorganization = {},
OPTpublisher = {},
OPTnote = {},
OPTannote = {},
URL = {http://www.labri.fr/~ramet/restricted/a836.pdf},
KEYWORDS = "Sparse"
}
-
A. Casadei,
P. Ramet,
and J. Roman.
An improved recursive graph bipartitioning algorithm for well balanced domain decomposition.
In 21st IEEE International Conference on High Performance Computing (HiPC),
Goa, India,
pages 1-10,
December 2014.
Keyword(s): Sparse.
Abstract: |
In the context of hybrid sparse linear solvers based on domain decomposition and Schur complement approaches, getting a domain decomposition tool leading to a good balancing of both the internal node set size and the interface node set size for all the domains is a critical point for load balancing and efficiency issues in a parallel computation context. For this purpose, we revisit the original algorithm introduced by Lipton, Rose and Tarjan which performed the recursion for nested dissection in a particular manner. From this specific recursive strategy, we propose in this paper several variations of the existing algorithms in the multilevel Scotch partitioner that take into account these multiple criteria and we illustrate the improved results on a collection of graphs corresponding to finite element meshes used in numerical scientific applications. |
@InProceedings{C:LaBRI::hipc14,
author = {Casadei, A. and Ramet, P. and Roman, J.},
title = {An improved recursive graph bipartitioning algorithm for well balanced domain decomposition},
OPTcrossref = {},
OPTkey = {},
booktitle = {21st IEEE International Conference on High Performance Computing (HiPC)},
pages = {1-10},
year = {2014},
OPTeditor = {},
OPTvolume = {},
OPTnumber = {},
OPTseries = {},
address = {Goa, India},
month = dec,
OPTorganization = {},
OPTpublisher = {},
OPTnote = {},
OPTannote = {},
DOI = {10.1109/HiPC.2014.7116878},
KEYWORDS = "Sparse",
URL = {http://www.labri.fr/~ramet/restricted/hipc14.pdf},
ABSTRACT = { In the context of hybrid sparse linear solvers based on domain decomposition and Schur complement approaches, getting a domain decomposition tool leading to a good balancing of both the internal node set size and the interface node set size for all the domains is a critical point for load balancing and efficiency issues in a parallel computation context. For this purpose, we revisit the original algorithm introduced by Lipton, Rose and Tarjan which performed the recursion for nested dissection in a particular manner. From this specific recursive strategy, we propose in this paper several variations of the existing algorithms in the multilevel Scotch partitioner that take into account these multiple criteria and we illustrate the improved results on a collection of graphs corresponding to finite element meshes used in numerical scientific applications. }
}
-
A. Casadei,
P. Ramet,
and J. Roman.
Nested Dissection with Balanced Halo.
In SIAM Workshop on Combinatorial Scientific Computing,
Lyon, France,
July 2014.
Keyword(s): Sparse.
@InProceedings{C:LaBRI::CSC14,
author = {Casadei, A. and Ramet, P. and Roman, J.},
title = {Nested Dissection with Balanced Halo},
OPTcrossref = {},
OPTkey = {},
booktitle = {SIAM Workshop on Combinatorial Scientific Computing},
OPTpages = {},
year = {2014},
OPTeditor = {},
OPTvolume = {},
OPTnumber = {},
OPTseries = {},
address = {Lyon, France},
month = jul,
OPTorganization = {},
OPTpublisher = {},
OPTnote = {},
OPTannote = {},
KEYWORDS = "Sparse",
URL = {http://www.labri.fr/~ramet/restricted/csc14.pdf},
}
-
C. Dudley,
E. Darve,
S. Ambikasaran,
and A. H. Aminfar.
Fast Algorithms for Dense Linear Algebra.
In Proceedings of PMAA'2014,
Lugano, Swiss,
July 2014.
Keyword(s): Sparse.
Abstract: |
In recent years there has been a resurgence in direct methods to solve linear systems. These methods can have many advantages compared to iterative solvers; in particular their accuracy and performance is less sensitive to the distribution of eigenvalues. However, they typically have a larger computational cost in cases where iterative solvers converge in few iterations. We will discuss a recent trend of methods that address this cost and can make these direct solvers competitive. Techniques involved include hierarchical matrices, hierarchically semi-separable matrices, fast multipole method, etc. |
@InProceedings{C:LaBRI::PMAA2014,
author = {Dudley, C. and Darve, E. and S. Ambikasaran and Aminfar, A.~H.},
title = {Fast Algorithms for Dense Linear Algebra},
OPTcrossref = {},
OPTkey = {},
booktitle = {Proceedings of {PMAA}'2014},
OPTpages = {},
year = {2014},
OPTeditor = {},
OPTvolume = {},
OPTnumber = {},
OPTseries = {},
address = {Lugano, Swiss},
month = jul,
OPTorganization = {},
OPTpublisher = {},
OPTnote = {},
OPTannote = {},
KEYWORDS = "Sparse",
ABSTRACT = {In recent years there has been a resurgence in direct methods to solve linear systems. These methods can have many advantages compared to iterative solvers; in particular their accuracy and performance is less sensitive to the distribution of eigenvalues. However, they typically have a larger computational cost in cases where iterative solvers converge in few iterations. We will discuss a recent trend of methods that address this cost and can make these direct solvers competitive. Techniques involved include hierarchical matrices, hierarchically semi-separable matrices, fast multipole method, etc.}
}
-
C. Dudley,
E. Darve,
S. Ambikasaran,
and A. H. Aminfar.
Fast direct linear solvers for the boundary element method.
In WCCM-ECCM-ECFD workshop on Fast Direct Solvers,
Barcelona, Spain,
July 2014.
Keyword(s): Sparse.
@InProceedings{C:LaBRI::ecfd14b,
author = {Dudley, C. and Darve, E. and S. Ambikasaran and Aminfar, A.~H.},
title = {Fast direct linear solvers for the boundary element method},
OPTcrossref = {},
OPTkey = {},
booktitle = {WCCM-ECCM-ECFD workshop on "Fast Direct Solvers"},
OPTpages = {},
year = {2014},
OPTeditor = {},
OPTvolume = {},
OPTnumber = {},
OPTseries = {},
address = {Barcelona, Spain},
month = jul,
OPTorganization = {},
OPTpublisher = {},
OPTnote = {},
OPTannote = {},
URL = {http://www.labri.fr/~ramet/restricted/a1916.pdf},
KEYWORDS = "Sparse"
}
-
X. Lacoste,
M. Faverge,
P. Ramet,
S. Thibault,
and G. Bosilca.
Taking advantage of hybrid systems for sparse direct solvers via task-based runtimes.
In Proceedings of HCW'2014 workshop of IPDPS,
Phoenix, USA,
pages 29-38,
May 2014.
Keyword(s): Sparse.
Abstract: |
The ongoing hardware evolution exhibits an es- calation in the number, as well as in the heterogeneity, of computing resources. The pressure to maintain reasonable levels of performance and portability forces application de- velopers to leave the traditional programming paradigms and explore alternative solutions. PASTIX is a parallel sparse direct solver, based on a dynamic scheduler for modern hierarchical manycore architectures. In this paper, we study the benefits and limits of replacing the highly specialized internal scheduler of the PASTIX solver with two generic runtime systems: PARSEC and STARPU. The tasks graph of the factorization step is made available to the two runtimes, providing them the opportunity to process and optimize its traversal in order to maximize the algorithm efficiency for the targeted hardware platform. A comparative study of the performance of the PASTIX solver on top of its native internal scheduler, PARSEC, and STARPU frameworks, on different execution environments, is performed. The analysis highlights that these generic task-based runtimes achieve comparable results to the application-optimized embed- ded scheduler on homogeneous platforms. Furthermore, they are able to significantly speed up the solver on heterogeneous environments by taking advantage of the accelerators while hiding the complexity of their efficient manipulation from the programmer. |
@InProceedings{C:LaBRI::hcw14,
author = {Lacoste, X. and Faverge, M. and Ramet, P. and Thibault, S. and Bosilca, G.},
title = {{Taking advantage of hybrid systems for sparse direct solvers via task-based runtimes}},
booktitle = {Proceedings of {HCW}'2014 workshop of IPDPS},
OPTcrossref = {},
OPTkey = {},
pages = {29-38},
year = 2014,
OPTeditor = {},
OPTnumber = {},
OPTvolume = {},
OPTseries = {},
address = {Phoenix, USA},
month = may,
OPTorganization = {},
OPTpublisher = {},
OPTnote = {},
OPTannote = {},
DOI={10.1109/IPDPSW.2014.9},
URL = {http://www.labri.fr/~ramet/restricted/hcw14.pdf},
KEYWORDS = "Sparse",
ABSTRACT = {The ongoing hardware evolution exhibits an es- calation in the number, as well as in the heterogeneity, of computing resources. The pressure to maintain reasonable levels of performance and portability forces application de- velopers to leave the traditional programming paradigms and explore alternative solutions. PASTIX is a parallel sparse direct solver, based on a dynamic scheduler for modern hierarchical manycore architectures. In this paper, we study the benefits and limits of replacing the highly specialized internal scheduler of the PASTIX solver with two generic runtime systems: PARSEC and STARPU. The tasks graph of the factorization step is made available to the two runtimes, providing them the opportunity to process and optimize its traversal in order to maximize the algorithm efficiency for the targeted hardware platform. A comparative study of the performance of the PASTIX solver on top of its native internal scheduler, PARSEC, and STARPU frameworks, on different execution environments, is performed. The analysis highlights that these generic task-based runtimes achieve comparable results to the application-optimized embed- ded scheduler on homogeneous platforms. Furthermore, they are able to significantly speed up the solver on heterogeneous environments by taking advantage of the accelerators while hiding the complexity of their efficient manipulation from the programmer.}
}
-
S. Moustafa,
M. Faverge,
L. Plagne,
and P. Ramet.
Parallel 3D Sweep Kernel with PARSEC.
In 16th IEEE International Conference on High Performance and Communications, workshop on HPC-CFD in Energy/Transport Domains,
Paris, France,
August 2014.
Keyword(s): Neutron.
@InProceedings{C:LaBRI::sweep2013,
author = {Moustafa, S. and Faverge, M. and Plagne, L. and Ramet, P.},
title = {Parallel 3D Sweep Kernel with PARSEC},
OPTcrossref = {},
OPTkey = {},
booktitle = {16th IEEE International Conference on High Performance and Communications, workshop on HPC-CFD in Energy/Transport Domains},
year = 2014,
OPTeditor = {},
OPTvolume = {},
OPTnumber = {},
OPTseries = {},
OPTpages = {},
month = aug,
address = {Paris, France},
OPTorganization = {},
OPTpublisher = {},
OPTnote = {},
OPTannote = {},
URL = {http://www.labri.fr/~ramet/restricted/sweep-parsec.pdf},
KEYWORDS = "Neutron"
}
-
A. Casadei,
P. Ramet,
and J. Roman.
An improved recursive graph bipartitioning algorithm for well balanced domain decomposition.
Research Report RR-8582,
August 2014.
Keyword(s): Sparse.
@techreport{casadei:hal-01056749,
TITLE = {{An improved recursive graph bipartitioning algorithm for well balanced domain decomposition}},
AUTHOR = {Casadei, A. and Ramet, P. and Roman, J.},
URL = {https://hal.inria.fr/hal-01056749},
TYPE = {Research Report},
NUMBER = {RR-8582},
PAGES = {29},
YEAR = {2014},
MONTH = Aug,
keywords = {Sparse},
HAL_ID = {hal-01056749},
HAL_VERSION = {v1},
}
-
X. Lacoste,
M. Faverge,
P. Ramet,
S. Thibault,
and G. Bosilca.
Taking advantage of hybrid systems for sparse direct solvers via task-based runtimes.
Research Report RR-8446,
INRIA,
January 2014.
Keyword(s): Sparse.
@techreport{lacoste:hal-00925017,
AUTHOR = {Lacoste, X. and Faverge, M. and Ramet, P. and Thibault, S. and Bosilca, G.},
TITLE = {{Taking advantage of hybrid systems for sparse direct solvers via task-based runtimes}},
TYPE = {Research Report},
PAGES = {25},
YEAR = {2014},
MONTH = Jan,
KEYWORDS = "Sparse",
INSTITUTION = {INRIA},
NUMBER = {RR-8446},
URL = {http://hal.inria.fr/hal-00925017}
}
-
H. Sellama,
G. Huijsmans,
and P. Ramet.
Adaptive mesh refinement for numerical simulation of MHD instabilities in tokamaks: JOREK code.
Research Report RR-8635,
INRIA Bordeaux,
November 2014.
@techreport{sellama:hal-01088094,
TITLE = {{Adaptive mesh refinement for numerical simulation of MHD instabilities in tokamaks: JOREK code}},
AUTHOR = {Sellama, H. and Huijsmans, G. and Ramet, P.},
URL = {https://hal.inria.fr/hal-01088094},
TYPE = {Research Report},
NUMBER = {RR-8635},
PAGES = {18},
INSTITUTION = {{INRIA Bordeaux}},
YEAR = {2014},
MONTH = Nov,
HAL_ID = {hal-01088094},
HAL_VERSION = {v1},
}
-
E. Agullo,
M. Faverge,
L. Giraud,
A. Guermouche,
P. Ramet,
and J. Roman.
Toward parallel scalable linear solvers suited for large scale hierarchical parallel platforms.
Workshop INRIA-CNPq, HOSCAR meeting, Gramado, Brazil,
September 2014.
Keyword(s): Sparse.
Abstract: |
In this talk we will discuss the current and future research activities on the design of parallel scalable linear systems for large scale problems that range from dense linear algebra, to parallel sparse direct solver and hybrid iterative-direct approaches that attempt to go beyond the best capabilities one can expect from sparse direct solvers. In particular we will describe the current activities on the implementations designed on top of runtime systems that should provide both code and performance portabilities across different parallel platforms. Finally, we will present some preliminary results to address the resilience issues on extreme scale computers; for that purpose we consider numerical alternatives that do not intensively rely on checkpoint restart mechanisms. |
@Misc{c:LaBRI::HOSCAR2014,
author = {Agullo, E. and Faverge, M. and Giraud, L. and Guermouche, A. and Ramet, P. and Roman, J.},
title = {Toward parallel scalable linear solvers suited for large scale hierarchical parallel platforms},
OPTcrossref = {},
OPTkey = {},
howpublished = {Workshop INRIA-CNPq, HOSCAR meeting, Gramado, Brazil},
OPTpages = {},
year = {2014},
OPTeditor = {},
OPTvolume = {},
OPTnumber = {},
OPTseries = {},
month = sep,
OPTorganization = {},
OPTpublisher = {},
OPTnote = {},
OPTannote = {},
KEYWORDS = "Sparse",
ABSTRACT = { In this talk we will discuss the current and future research activities on the design of parallel scalable linear systems for large scale problems that range from dense linear algebra, to parallel sparse direct solver and hybrid iterative-direct approaches that attempt to go beyond the best capabilities one can expect from sparse direct solvers. In particular we will describe the current activities on the implementations designed on top of runtime systems that should provide both code and performance portabilities across different parallel platforms. Finally, we will present some preliminary results to address the resilience issues on extreme scale computers; for that purpose we consider numerical alternatives that do not intensively rely on checkpoint restart mechanisms. }
}
-
E. Agullo and P. Ramet.
Task-based linear solvers for modern architectures.
7th ITER International School, High Performance Computing in Fusion Science, Aix-en-Provence, France,
August 2014.
@Misc{c:LaBRI::ITER14,
author = {Agullo, E. and Ramet, P.},
title = {Task-based linear solvers for modern architectures},
month = aug,
year = {2014},
howpublished = {7th ITER International School, High Performance Computing in Fusion Science, Aix-en-Provence, France},
ABSTACT={ Modern supercomputers have reached a very high degree of hardware complexity. They are often composed of multicore processors relying on multiple memory banks and possibly accelerated with GPUs or co-processors. Programming these machines may be extremely hard, achieving high performance across such machines even harder. In this presentation, we discuss a methodology for programming numerical algorithms independently of the architectures and use runtime systems to exploit the potential of the underlying architectures. To do so, the numerical algorithm (and code) is written in terms of graph of tasks. We will show how high-performance dense linear algebra libraries (think LAPACK and ScaLAPACK) have been rewritten accordingly. We will also present simular results for a sparse direct solver (PaStiX), which is used in the Jorek plasma physics code. We will finally briefly present a few direct, iterative and hybrid open source solvers based on these building blocks. }
}
-
E. Darve,
A. H. Aminfar,
C. Dudley,
P. Ramet,
and M. Faverge.
Fast Algorithms for Dense Linear Algebra.
CPU - Cluster of excellence, Bordeaux, France,
July 2014.
@Misc{c:LaBRI::CPUb,
OPTkey = {},
author = {Darve, E. and Aminfar, A.~H. and Dudley, C. and Ramet, P. and Faverge, M.},
title = {Fast Algorithms for Dense Linear Algebra},
howpublished = {CPU - Cluster of excellence, Bordeaux, France},
month = jul,
year = 2014,
OPTnote = {},
OPTannote = {}
}
-
X. Lacoste,
M. Faverge,
and P. Ramet.
Distributed sparse matrix factorization on top of tasks based runtime systems.
SOLHAR meeting, Bordeaux, France,
November 2014.
Keyword(s): Sparse.
@Misc{c:LaBRI::pastix-solhar2,
OPTkey = {},
author = {Lacoste, X. and Faverge, M. and Ramet, P.},
title = {Distributed sparse matrix factorization on top of tasks based runtime systems},
howpublished = {SOLHAR meeting, Bordeaux, France},
month = nov,
year = 2014,
OPTnote = {},
OPTannote = {},
URL = {http://www.labri.fr/~ramet/restricted/pastix-solhar2.pdf},
KEYWORDS = "Sparse"
}
-
S. Moustafa,
M. Faverge,
L. Plagne,
and P. Ramet.
3D Cartesian Transport Sweep for Massively Parallel Architectures on top of PaRSEC.
SOLHAR meeting, Toulouse, France,
June 2014.
Keyword(s): Neutron.
@Misc{c:LaBRI::sweep-solhar,
OPTkey = {},
author = {Moustafa, S. and Faverge, M. and Plagne, L. and Ramet, P.},
title = {3D Cartesian Transport Sweep for Massively Parallel Architectures on top of PaRSEC},
howpublished = {SOLHAR meeting, Toulouse, France},
month = jun,
year = 2014,
OPTnote = {},
OPTannote = {},
URL = {http://www.labri.fr/~ramet/restricted/sweep-solhar.pdf},
KEYWORDS = "Neutron"
}
-
S. Moustafa,
M. Faverge,
L. Plagne,
and P. Ramet.
3D Cartesian Transport Sweep for Massively Parallel Architectures on top of PaRSEC.
9th Scheduling for Large Scale Systems Workshop, Lyon, France,
July 2014.
Keyword(s): Neutron.
@Misc{c:LaBRI::sweep-scheduling,
OPTkey = {},
author = {Moustafa, S. and Faverge, M. and Plagne, L. and Ramet, P.},
title = {3D Cartesian Transport Sweep for Massively Parallel Architectures on top of PaRSEC},
howpublished = {9th Scheduling for Large Scale Systems Workshop, Lyon, France},
month = jul,
year = 2014,
OPTnote = {},
OPTannote = {},
URL = {http://www.labri.fr/~ramet/restricted/schedworkshop.pdf},
KEYWORDS = "Neutron"
}
-
P. Ramet.
From hybrid architectures to hybrid solvers.
CPU - Cluster of excellence, Bordeaux, France,
July 2014.
@Misc{c:LaBRI::CPUa,
OPTkey = {},
author = {Ramet, P.},
title = {From hybrid architectures to hybrid solvers},
howpublished = {CPU - Cluster of excellence, Bordeaux, France},
month = jul,
year = 2014,
OPTnote = {},
OPTannote = {}
}
-
P. Ramet.
Hybrid methods, Hybrid architectures, Hybrid compressions for sparse direct solvers.
Seminar at MCIA (Mésocentre de Calcul Intensif Aquitain),
February 2014.
@Misc{c:LaBRI::MCIA14,
OPTkey = {},
author = {Ramet, P.},
title = {Hybrid methods, Hybrid architectures, Hybrid compressions for sparse direct solvers},
howpublished = {Seminar at MCIA (M\'esocentre de Calcul Intensif Aquitain)},
month = feb,
year = 2014,
OPTnote = {},
OPTannote = {}
}
-
P. Ramet.
Solveurs Directs.
Maison de la Simulation, Formation PATC, Algèbre Linéaire Creuse Parallèle, Paris, France,
March 2014.
@Misc{c:LaBRI::MDS14,
author = {Ramet, P.},
title = {Solveurs Directs},
month = mar,
year = {2014},
howpublished = {Maison de la Simulation, Formation PATC, Alg\`ebre Lin\'eaire Creuse Parall\`ele, Paris, France}
}