Towards a Malleable Tensorflow Implementation

Autores
Libutti, Leandro Ariel; Igual, Francisco; Piñuel, Luis; De Giusti, Laura Cristina; Naiouf, Marcelo; Rucci, Enzo; Naiouf, Marcelo; Chichizola, Franco; De Giusti, Laura Cristina
Año de publicación
2020
Idioma
inglés
Tipo de recurso
parte de libro
Estado
versión publicada
Descripción
The TensorFlow framework was designed since its inception to provide multi-thread capabilities, extended with hardware accelerator support to leverage the potential of modern architectures. The amount of parallelism in current versions of the framework can be selected at multiple levels (intra- and inter-paralellism) under demand. However, this selection is fixed, and cannot vary during the execution of training/inference sessions. This heavily restricts the flexibility and elasticity of the framework, especially in scenarios in which multiple TensorFlow instances co-exist in a parallel architecture. In this work, we propose the necessary modifications within TensorFlow to support dynamic selection of threads, in order to provide transparent malleability to the infrastructure. Experimental results show that this approach is effective in the variation of parallelism, and paves the road towards future co-scheduling techniques for multi-TensorFlow scenarios.
Instituto de Investigación en Informática
Materia
Ciencias Informáticas
TensorFlow
Malleability
Containers
Resource management
Co-scheduling
Nivel de accesibilidad
acceso abierto
Condiciones de uso
http://creativecommons.org/licenses/by-nc-sa/4.0/
Repositorio
SEDICI (UNLP)
Institución
Universidad Nacional de La Plata
OAI Identificador
oai:sedici.unlp.edu.ar:10915/145222
Acceder
id SEDICI_19029ca215ed51acfccafa916652c8cd
oai_identifier_str oai:sedici.unlp.edu.ar:10915/145222
network_acronym_str SEDICI
repository_id_str 1329
network_name_str SEDICI (UNLP)
spelling Towards a Malleable Tensorflow ImplementationLibutti, Leandro ArielIgual, FranciscoPiñuel, LuisDe Giusti, Laura CristinaNaiouf, MarceloRucci, EnzoNaiouf, MarceloChichizola, FrancoDe Giusti, Laura CristinaCiencias InformáticasTensorFlowMalleabilityContainersResource managementCo-schedulingThe TensorFlow framework was designed since its inception to provide multi-thread capabilities, extended with hardware accelerator support to leverage the potential of modern architectures. The amount of parallelism in current versions of the framework can be selected at multiple levels (intra- and inter-paralellism) under demand. However, this selection is fixed, and cannot vary during the execution of training/inference sessions. This heavily restricts the flexibility and elasticity of the framework, especially in scenarios in which multiple TensorFlow instances co-exist in a parallel architecture. In this work, we propose the necessary modifications within TensorFlow to support dynamic selection of threads, in order to provide transparent malleability to the infrastructure. Experimental results show that this approach is effective in the variation of parallelism, and paves the road towards future co-scheduling techniques for multi-TensorFlow scenarios.Instituto de Investigación en InformáticaSpringer2020-10-24info:eu-repo/semantics/bookPartinfo:eu-repo/semantics/publishedVersionCapitulo de librohttp://purl.org/coar/resource_type/c_3248info:ar-repo/semantics/parteDeLibroapplication/pdf30-40http://sedici.unlp.edu.ar/handle/10915/145222enginfo:eu-repo/semantics/altIdentifier/isbn/978-3-030-61218-4info:eu-repo/semantics/altIdentifier/issn/1865-0929info:eu-repo/semantics/altIdentifier/issn/1865-0937info:eu-repo/semantics/altIdentifier/doi/10.1007/978-3-030-61218-4_3info:eu-repo/semantics/openAccesshttp://creativecommons.org/licenses/by-nc-sa/4.0/Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International (CC BY-NC-SA 4.0)reponame:SEDICI (UNLP)instname:Universidad Nacional de La Platainstacron:UNLP2025-09-17T10:15:00Zoai:sedici.unlp.edu.ar:10915/145222Institucionalhttp://sedici.unlp.edu.ar/Universidad públicaNo correspondehttp://sedici.unlp.edu.ar/oai/snrdalira@sedici.unlp.edu.arArgentinaNo correspondeNo correspondeNo correspondeopendoar:13292025-09-17 10:15:00.598SEDICI (UNLP) - Universidad Nacional de La Platafalse
dc.title.none.fl_str_mv Towards a Malleable Tensorflow Implementation
title Towards a Malleable Tensorflow Implementation
spellingShingle Towards a Malleable Tensorflow Implementation
Libutti, Leandro Ariel
Ciencias Informáticas
TensorFlow
Malleability
Containers
Resource management
Co-scheduling
title_short Towards a Malleable Tensorflow Implementation
title_full Towards a Malleable Tensorflow Implementation
title_fullStr Towards a Malleable Tensorflow Implementation
title_full_unstemmed Towards a Malleable Tensorflow Implementation
title_sort Towards a Malleable Tensorflow Implementation
dc.creator.none.fl_str_mv Libutti, Leandro Ariel
Igual, Francisco
Piñuel, Luis
De Giusti, Laura Cristina
Naiouf, Marcelo
Rucci, Enzo
Naiouf, Marcelo
Chichizola, Franco
De Giusti, Laura Cristina
author Libutti, Leandro Ariel
author_facet Libutti, Leandro Ariel
Igual, Francisco
Piñuel, Luis
De Giusti, Laura Cristina
Naiouf, Marcelo
Rucci, Enzo
Chichizola, Franco
author_role author
author2 Igual, Francisco
Piñuel, Luis
De Giusti, Laura Cristina
Naiouf, Marcelo
Rucci, Enzo
Chichizola, Franco
author2_role author
author
author
author
author
author
dc.subject.none.fl_str_mv Ciencias Informáticas
TensorFlow
Malleability
Containers
Resource management
Co-scheduling
topic Ciencias Informáticas
TensorFlow
Malleability
Containers
Resource management
Co-scheduling
dc.description.none.fl_txt_mv The TensorFlow framework was designed since its inception to provide multi-thread capabilities, extended with hardware accelerator support to leverage the potential of modern architectures. The amount of parallelism in current versions of the framework can be selected at multiple levels (intra- and inter-paralellism) under demand. However, this selection is fixed, and cannot vary during the execution of training/inference sessions. This heavily restricts the flexibility and elasticity of the framework, especially in scenarios in which multiple TensorFlow instances co-exist in a parallel architecture. In this work, we propose the necessary modifications within TensorFlow to support dynamic selection of threads, in order to provide transparent malleability to the infrastructure. Experimental results show that this approach is effective in the variation of parallelism, and paves the road towards future co-scheduling techniques for multi-TensorFlow scenarios.
Instituto de Investigación en Informática
description The TensorFlow framework was designed since its inception to provide multi-thread capabilities, extended with hardware accelerator support to leverage the potential of modern architectures. The amount of parallelism in current versions of the framework can be selected at multiple levels (intra- and inter-paralellism) under demand. However, this selection is fixed, and cannot vary during the execution of training/inference sessions. This heavily restricts the flexibility and elasticity of the framework, especially in scenarios in which multiple TensorFlow instances co-exist in a parallel architecture. In this work, we propose the necessary modifications within TensorFlow to support dynamic selection of threads, in order to provide transparent malleability to the infrastructure. Experimental results show that this approach is effective in the variation of parallelism, and paves the road towards future co-scheduling techniques for multi-TensorFlow scenarios.
publishDate 2020
dc.date.none.fl_str_mv 2020-10-24
dc.type.none.fl_str_mv info:eu-repo/semantics/bookPart
info:eu-repo/semantics/publishedVersion
Capitulo de libro
http://purl.org/coar/resource_type/c_3248
info:ar-repo/semantics/parteDeLibro
format bookPart
status_str publishedVersion
dc.identifier.none.fl_str_mv http://sedici.unlp.edu.ar/handle/10915/145222
url http://sedici.unlp.edu.ar/handle/10915/145222
dc.language.none.fl_str_mv eng
language eng
dc.relation.none.fl_str_mv info:eu-repo/semantics/altIdentifier/isbn/978-3-030-61218-4
info:eu-repo/semantics/altIdentifier/issn/1865-0929
info:eu-repo/semantics/altIdentifier/issn/1865-0937
info:eu-repo/semantics/altIdentifier/doi/10.1007/978-3-030-61218-4_3
dc.rights.none.fl_str_mv info:eu-repo/semantics/openAccess
http://creativecommons.org/licenses/by-nc-sa/4.0/
Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International (CC BY-NC-SA 4.0)
eu_rights_str_mv openAccess
rights_invalid_str_mv http://creativecommons.org/licenses/by-nc-sa/4.0/
Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International (CC BY-NC-SA 4.0)
dc.format.none.fl_str_mv application/pdf
30-40
dc.publisher.none.fl_str_mv Springer
publisher.none.fl_str_mv Springer
dc.source.none.fl_str_mv reponame:SEDICI (UNLP)
instname:Universidad Nacional de La Plata
instacron:UNLP
reponame_str SEDICI (UNLP)
collection SEDICI (UNLP)
instname_str Universidad Nacional de La Plata
instacron_str UNLP
institution UNLP
repository.name.fl_str_mv SEDICI (UNLP) - Universidad Nacional de La Plata
repository.mail.fl_str_mv alira@sedici.unlp.edu.ar
_version_ 1843532790675537920
score 13.001348

Publicaciones similares