Towards a Malleable Tensorflow Implementation

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...

Descripción completa

Guardado en:
Detalles Bibliográficos
Autores principales: Libutti, Leandro Ariel, Igual, Francisco, Piñuel, Luis, De Giusti, Laura Cristina, Naiouf, Marcelo, Rucci, Enzo, Chichizola, Franco
Formato: Libro Capitulo de libro
Lenguaje:Inglés
Publicado: Springer 2020
Materias:
Ciencias Informáticas
TensorFlow
Malleability
Containers
Resource management
Co-scheduling
Acceso en línea:http://sedici.unlp.edu.ar/handle/10915/145222
Aporte de:
SEDICI (UNLP) de Universidad Nacional de La Plata
id I19-R120-10915-145222
record_format dspace
spelling I19-R120-10915-1452222023-05-31T13:40:10Z http://sedici.unlp.edu.ar/handle/10915/145222 issn:1865-0929 issn:1865-0937 isbn:978-3-030-61218-4 Towards a Malleable Tensorflow Implementation Libutti, Leandro Ariel Igual, Francisco Piñuel, Luis De Giusti, Laura Cristina Naiouf, Marcelo 2020-10-24 2022-11-04T17:15:23Z Rucci, Enzo Naiouf, Marcelo Chichizola, Franco De Giusti, Laura Cristina Springer en Ciencias Informáticas TensorFlow Malleability Containers Resource management Co-scheduling 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 Libro Capitulo de libro http://creativecommons.org/licenses/by-nc-sa/4.0/ Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International (CC BY-NC-SA 4.0) application/pdf 30-40
institution Universidad Nacional de La Plata
institution_str I-19
repository_str R-120
collection SEDICI (UNLP)
language Inglés
topic Ciencias Informáticas
TensorFlow
Malleability
Containers
Resource management
Co-scheduling
spellingShingle Ciencias Informáticas
TensorFlow
Malleability
Containers
Resource management
Co-scheduling
Libutti, Leandro Ariel
Igual, Francisco
Piñuel, Luis
De Giusti, Laura Cristina
Naiouf, Marcelo
Rucci, Enzo
Naiouf, Marcelo
Chichizola, Franco
De Giusti, Laura Cristina
Towards a Malleable Tensorflow Implementation
topic_facet Ciencias Informáticas
TensorFlow
Malleability
Containers
Resource management
Co-scheduling
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.
format Libro
Capitulo de libro
author Libutti, Leandro Ariel
Igual, Francisco
Piñuel, Luis
De Giusti, Laura Cristina
Naiouf, Marcelo
Rucci, Enzo
Naiouf, Marcelo
Chichizola, Franco
De Giusti, Laura Cristina
author_facet Libutti, Leandro Ariel
Igual, Francisco
Piñuel, Luis
De Giusti, Laura Cristina
Naiouf, Marcelo
Rucci, Enzo
Naiouf, Marcelo
Chichizola, Franco
De Giusti, Laura Cristina
author_sort Libutti, Leandro Ariel
title Towards a Malleable Tensorflow Implementation
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
publisher Springer
publishDate 2020
url http://sedici.unlp.edu.ar/handle/10915/145222
work_keys_str_mv AT libuttileandroariel towardsamalleabletensorflowimplementation
AT igualfrancisco towardsamalleabletensorflowimplementation
AT pinuelluis towardsamalleabletensorflowimplementation
AT degiustilauracristina towardsamalleabletensorflowimplementation
AT naioufmarcelo towardsamalleabletensorflowimplementation
AT ruccienzo towardsamalleabletensorflowimplementation
AT naioufmarcelo towardsamalleabletensorflowimplementation
AT chichizolafranco towardsamalleabletensorflowimplementation
AT degiustilauracristina towardsamalleabletensorflowimplementation
_version_ 1767633957110677504

Ejemplares similares