Paper published in a book (Scientific congresses and symposiums)
Use of high-performance computing in animal breeding
Vandenplas, Jérémie; Gengler, Nicolas
2014In Book of Abstracts of the 65th Annual Meeting of the European Federation of Animal Science
Peer reviewed
 

Files


Full Text
Use of high-performance computing in animal breeding.pdf
Publisher postprint (421.62 kB)
Request a copy
Annexes
Use of high-performance computing in animal breeding poster.pdf
Publisher postprint (124.94 kB)
Poster presentation
Download

All documents in ORBi are protected by a user license.

Send to



Details



Abstract :
[en] High-perfomance computing facilities proposing shared-memory and distributed-memory multiprocessors are becoming available. With those clusters, parallel computing could lead to increased performances and problem sizes. However, to our knowledge and especially for variance components estimations, most software available in animal breeding, based on sparse matrices computations, do not allow parallel computing and are limited by memory accessible by the central processing unit, or allow parallel computing only for options with dense matrices computations, which limits anyway problem sizes due to storage of dense matrices. The aim was to propose simple and effective modifications for the BLUPF90 family of programs to reduce computing time with consideration of required memory. Modifications were based on academic free packages proposing solver and sparse inversion for sparse symmetric indefinite linear systems. First, modifications concerned the sparse inversion subroutine implemented in the package FSPAK. Rearrangements of 'do' loops to allow optimizations of computer operations by some compilers and addition of OpenMP directives were performed. The ordering operation was modified to more easily compare a multiple minimum degree algorithm (MMD; implemented in FSPAK) and a multilevel nested dissection algorithm (implemented in METIS 4.0.3). Second, the package PARDISO Version 5.0.0 was used instead of FSPAK. This package proposes in particular a parallel solver and sparse inversion on shared-memory multiprocessors. Modified FSPAK and PARDISO were compared to original FSPAK using MMD through REMLF90. Different models, such as univariate or bivariate (random regressions) test-day animal and single-step genomic models, were tested. All jobs were run 5 times. With an appropriate ordering algorithm, speedup for each REMLF90 iteration were up to 7.5 for modified FSPAK and up to 22.8 for PARDISO with 2 threads. With 4 threads, speedup increased to 8.3 and 32.5, respectively.
Disciplines :
Animal production & animal husbandry
Genetics & genetic processes
Author, co-author :
Vandenplas, Jérémie ;  Université de Liège - ULiège > Sciences agronomiques > Zootechnie
Gengler, Nicolas  ;  Université de Liège - ULiège > Agronomie, Bio-ingénierie et Chimie (AgroBioChem) > Ingénierie des productions animales et nutrition
Language :
English
Title :
Use of high-performance computing in animal breeding
Publication date :
28 August 2014
Event name :
65th Annual Meeting of the European Federation of Animal Science
Event organizer :
EAAP - European Federation of Animal Science
Event place :
Copenhagen, Denmark
Event date :
25-29 August 2014
Audience :
International
Main work title :
Book of Abstracts of the 65th Annual Meeting of the European Federation of Animal Science
Publisher :
Wageningen Academic Publishers, Wageningen, Netherlands
Edition :
Copenhagen 2014
ISBN/EAN :
978-90-8686-248-1
Collection name :
Nº. 20
Pages :
326
Peer reviewed :
Peer reviewed
Funders :
F.R.S.-FNRS - Fonds de la Recherche Scientifique [BE]
Available on ORBi :
since 30 August 2014

Statistics


Number of views
189 (43 by ULiège)
Number of downloads
92 (11 by ULiège)

Bibliography


Similar publications



Contact ORBi