SmartApprox: learning-based configuration of approximate memories for energy-efficient execution
Autor(a) principal: | |
---|---|
Data de Publicação: | 2022 |
Outros Autores: | , |
Tipo de documento: | Artigo |
Idioma: | eng |
Título da fonte: | Repositório Institucional da UTFPR (da Universidade Tecnológica Federal do Paraná (RIUT)) |
Texto Completo: | http://repositorio.utfpr.edu.br/jspui/handle/1/29764 https://doi.org/10.1016/j.suscom.2022.100701 |
Resumo: | Approximate memories reduce power and increase energy efficiency, at the expense of errors in stored data. These errors may be tolerated, up to a point, by many applications with negligible impact on the quality of results. Uncontrolled errors in memory may, however, lead to crashes or broken outputs. Error rates are determined by fabrication and operation parameters, and error tolerance depends on algorithms, implementation, and inputs. An ideal configuration features parameters for approximate memory that minimize energy while allowing applications to produce acceptable results. This work introduces SmartApprox, a framework that configures approximation levels based on features of applications. In SmartApprox, a training phase executes a set of applications under different approximation settings, building a knowledge base that correlates application features (e.g., types of instructions and cache efficiency) with suitable approximate memory configurations. At runtime, features of new applications are sampled and approximation knobs are adjusted to correspond to the predicted error tolerance, according to existing knowledge and the current error scenario, in consonance with hardware characterization. In this work, we list and discuss sets of features that influence the approximation results and measure their impact on the error tolerance or applications. We evaluate SmartApprox on different voltage-scaled DRAM scenarios using a knowledge base of 26 applications, wherein energy savings of 36% are possible with acceptable output. An evaluation using a combined energy and quality metric shows that SmartApprox scores 97% of an exhaustive search for ideal configurations, with significantly lower effort and without application-specific quality evaluation. |
id |
UTFPR-12_c0e56ba1d9d81f540cd474d0abd333cc |
---|---|
oai_identifier_str |
oai:repositorio.utfpr.edu.br:1/29764 |
network_acronym_str |
UTFPR-12 |
network_name_str |
Repositório Institucional da UTFPR (da Universidade Tecnológica Federal do Paraná (RIUT)) |
repository_id_str |
|
spelling |
2022-09-28T13:03:15Z50002022-09-28T13:03:15Z2022-04FABRÍCIO FILHO, João; FELZMANN, Isaías; WANNER, Lucas. SmartApprox: learning-based configuration of approximate memories. Sustainable Computing: Informatics and Systems, v. 34, 100701, abr. 2022. DOI: https://doi.org/10.1016/j.suscom.2022.100701. Disponível em: https://www.sciencedirect.com/science/article/pii/S2210537922000427. Acesso em: 09 jun. 2022.2210-5379http://repositorio.utfpr.edu.br/jspui/handle/1/29764https://doi.org/10.1016/j.suscom.2022.100701Approximate memories reduce power and increase energy efficiency, at the expense of errors in stored data. These errors may be tolerated, up to a point, by many applications with negligible impact on the quality of results. Uncontrolled errors in memory may, however, lead to crashes or broken outputs. Error rates are determined by fabrication and operation parameters, and error tolerance depends on algorithms, implementation, and inputs. An ideal configuration features parameters for approximate memory that minimize energy while allowing applications to produce acceptable results. This work introduces SmartApprox, a framework that configures approximation levels based on features of applications. In SmartApprox, a training phase executes a set of applications under different approximation settings, building a knowledge base that correlates application features (e.g., types of instructions and cache efficiency) with suitable approximate memory configurations. At runtime, features of new applications are sampled and approximation knobs are adjusted to correspond to the predicted error tolerance, according to existing knowledge and the current error scenario, in consonance with hardware characterization. In this work, we list and discuss sets of features that influence the approximation results and measure their impact on the error tolerance or applications. We evaluate SmartApprox on different voltage-scaled DRAM scenarios using a knowledge base of 26 applications, wherein energy savings of 36% are possible with acceptable output. An evaluation using a combined energy and quality metric shows that SmartApprox scores 97% of an exhaustive search for ideal configurations, with significantly lower effort and without application-specific quality evaluation.engSustainable Computing: Informatics and Systemshttps://www.sciencedirect.com/science/article/pii/S2210537922000427https://s100.copyright.com/AppDispatchServlet?publisherName=ELS&contentID=S2210537922000427&orderBeanReset=trueinfo:eu-repo/semantics/embargoedAccessCNPQ::CIENCIAS EXATAS E DA TERRA::CIENCIA DA COMPUTACAOSistemas de memória de computadorFalhas de sistemas de computaçãoEnergia - ConsumoComputer storage devicesComputer system failuresEnergy consumptionSmartApprox: learning-based configuration of approximate memories for energy-efficient executioninfo:eu-repo/semantics/publishedVersioninfo:eu-repo/semantics/articleCampo MouraoBrasil344Fabrício Filho, JoãoFelzmann, Isaías BittencourtWanner, Lucas Franciscoreponame:Repositório Institucional da UTFPR (da Universidade Tecnológica Federal do Paraná (RIUT))instname:Universidade Tecnológica Federal do Paraná (UTFPR)instacron:UTFPRORIGINALsmartapproxapproximatememories.pdfsmartapproxapproximatememories.pdfapplication/pdf7798086http://repositorio.utfpr.edu.br:8080/jspui/bitstream/1/29764/1/smartapproxapproximatememories.pdf58e5123b2e6ae663789e5b5c7d1bc3bdMD51LICENSElicense.txtlicense.txttext/plain; charset=utf-81290http://repositorio.utfpr.edu.br:8080/jspui/bitstream/1/29764/2/license.txtb9d82215ab23456fa2d8b49c5df1b95bMD52TEXTsmartapproxapproximatememories.pdf.txtsmartapproxapproximatememories.pdf.txtExtracted texttext/plain12http://repositorio.utfpr.edu.br:8080/jspui/bitstream/1/29764/3/smartapproxapproximatememories.pdf.txt36c6f0b2061da514c400c0bc2749b5cfMD53THUMBNAILsmartapproxapproximatememories.pdf.jpgsmartapproxapproximatememories.pdf.jpgGenerated Thumbnailimage/jpeg1708http://repositorio.utfpr.edu.br:8080/jspui/bitstream/1/29764/4/smartapproxapproximatememories.pdf.jpg1fa63cf2b77bd128316c5abed3bc81e2MD541/297642022-09-29 03:07:41.068oai:repositorio.utfpr.edu.br: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ório de PublicaçõesPUBhttp://repositorio.utfpr.edu.br:8080/oai/requestopendoar:2022-09-29T06:07:41Repositório Institucional da UTFPR (da Universidade Tecnológica Federal do Paraná (RIUT)) - Universidade Tecnológica Federal do Paraná (UTFPR)false |
dc.title.pt_BR.fl_str_mv |
SmartApprox: learning-based configuration of approximate memories for energy-efficient execution |
title |
SmartApprox: learning-based configuration of approximate memories for energy-efficient execution |
spellingShingle |
SmartApprox: learning-based configuration of approximate memories for energy-efficient execution Fabrício Filho, João CNPQ::CIENCIAS EXATAS E DA TERRA::CIENCIA DA COMPUTACAO Sistemas de memória de computador Falhas de sistemas de computação Energia - Consumo Computer storage devices Computer system failures Energy consumption |
title_short |
SmartApprox: learning-based configuration of approximate memories for energy-efficient execution |
title_full |
SmartApprox: learning-based configuration of approximate memories for energy-efficient execution |
title_fullStr |
SmartApprox: learning-based configuration of approximate memories for energy-efficient execution |
title_full_unstemmed |
SmartApprox: learning-based configuration of approximate memories for energy-efficient execution |
title_sort |
SmartApprox: learning-based configuration of approximate memories for energy-efficient execution |
author |
Fabrício Filho, João |
author_facet |
Fabrício Filho, João Felzmann, Isaías Bittencourt Wanner, Lucas Francisco |
author_role |
author |
author2 |
Felzmann, Isaías Bittencourt Wanner, Lucas Francisco |
author2_role |
author author |
dc.contributor.author.fl_str_mv |
Fabrício Filho, João Felzmann, Isaías Bittencourt Wanner, Lucas Francisco |
dc.subject.cnpq.fl_str_mv |
CNPQ::CIENCIAS EXATAS E DA TERRA::CIENCIA DA COMPUTACAO |
topic |
CNPQ::CIENCIAS EXATAS E DA TERRA::CIENCIA DA COMPUTACAO Sistemas de memória de computador Falhas de sistemas de computação Energia - Consumo Computer storage devices Computer system failures Energy consumption |
dc.subject.por.fl_str_mv |
Sistemas de memória de computador Falhas de sistemas de computação Energia - Consumo Computer storage devices Computer system failures Energy consumption |
description |
Approximate memories reduce power and increase energy efficiency, at the expense of errors in stored data. These errors may be tolerated, up to a point, by many applications with negligible impact on the quality of results. Uncontrolled errors in memory may, however, lead to crashes or broken outputs. Error rates are determined by fabrication and operation parameters, and error tolerance depends on algorithms, implementation, and inputs. An ideal configuration features parameters for approximate memory that minimize energy while allowing applications to produce acceptable results. This work introduces SmartApprox, a framework that configures approximation levels based on features of applications. In SmartApprox, a training phase executes a set of applications under different approximation settings, building a knowledge base that correlates application features (e.g., types of instructions and cache efficiency) with suitable approximate memory configurations. At runtime, features of new applications are sampled and approximation knobs are adjusted to correspond to the predicted error tolerance, according to existing knowledge and the current error scenario, in consonance with hardware characterization. In this work, we list and discuss sets of features that influence the approximation results and measure their impact on the error tolerance or applications. We evaluate SmartApprox on different voltage-scaled DRAM scenarios using a knowledge base of 26 applications, wherein energy savings of 36% are possible with acceptable output. An evaluation using a combined energy and quality metric shows that SmartApprox scores 97% of an exhaustive search for ideal configurations, with significantly lower effort and without application-specific quality evaluation. |
publishDate |
2022 |
dc.date.accessioned.fl_str_mv |
2022-09-28T13:03:15Z |
dc.date.available.fl_str_mv |
2022-09-28T13:03:15Z 5000 |
dc.date.issued.fl_str_mv |
2022-04 |
dc.type.status.fl_str_mv |
info:eu-repo/semantics/publishedVersion |
dc.type.driver.fl_str_mv |
info:eu-repo/semantics/article |
format |
article |
status_str |
publishedVersion |
dc.identifier.citation.fl_str_mv |
FABRÍCIO FILHO, João; FELZMANN, Isaías; WANNER, Lucas. SmartApprox: learning-based configuration of approximate memories. Sustainable Computing: Informatics and Systems, v. 34, 100701, abr. 2022. DOI: https://doi.org/10.1016/j.suscom.2022.100701. Disponível em: https://www.sciencedirect.com/science/article/pii/S2210537922000427. Acesso em: 09 jun. 2022. |
dc.identifier.uri.fl_str_mv |
http://repositorio.utfpr.edu.br/jspui/handle/1/29764 |
dc.identifier.issn.pt_BR.fl_str_mv |
2210-5379 |
dc.identifier.doi.pt_BR.fl_str_mv |
https://doi.org/10.1016/j.suscom.2022.100701 |
identifier_str_mv |
FABRÍCIO FILHO, João; FELZMANN, Isaías; WANNER, Lucas. SmartApprox: learning-based configuration of approximate memories. Sustainable Computing: Informatics and Systems, v. 34, 100701, abr. 2022. DOI: https://doi.org/10.1016/j.suscom.2022.100701. Disponível em: https://www.sciencedirect.com/science/article/pii/S2210537922000427. Acesso em: 09 jun. 2022. 2210-5379 |
url |
http://repositorio.utfpr.edu.br/jspui/handle/1/29764 https://doi.org/10.1016/j.suscom.2022.100701 |
dc.language.iso.fl_str_mv |
eng |
language |
eng |
dc.relation.ispartof.pt_BR.fl_str_mv |
Sustainable Computing: Informatics and Systems |
dc.relation.uri.pt_BR.fl_str_mv |
https://www.sciencedirect.com/science/article/pii/S2210537922000427 |
dc.rights.driver.fl_str_mv |
info:eu-repo/semantics/embargoedAccess |
eu_rights_str_mv |
embargoedAccess |
dc.publisher.none.fl_str_mv |
Campo Mourao |
dc.publisher.country.fl_str_mv |
Brasil |
publisher.none.fl_str_mv |
Campo Mourao |
dc.source.none.fl_str_mv |
reponame:Repositório Institucional da UTFPR (da Universidade Tecnológica Federal do Paraná (RIUT)) instname:Universidade Tecnológica Federal do Paraná (UTFPR) instacron:UTFPR |
instname_str |
Universidade Tecnológica Federal do Paraná (UTFPR) |
instacron_str |
UTFPR |
institution |
UTFPR |
reponame_str |
Repositório Institucional da UTFPR (da Universidade Tecnológica Federal do Paraná (RIUT)) |
collection |
Repositório Institucional da UTFPR (da Universidade Tecnológica Federal do Paraná (RIUT)) |
bitstream.url.fl_str_mv |
http://repositorio.utfpr.edu.br:8080/jspui/bitstream/1/29764/1/smartapproxapproximatememories.pdf http://repositorio.utfpr.edu.br:8080/jspui/bitstream/1/29764/2/license.txt http://repositorio.utfpr.edu.br:8080/jspui/bitstream/1/29764/3/smartapproxapproximatememories.pdf.txt http://repositorio.utfpr.edu.br:8080/jspui/bitstream/1/29764/4/smartapproxapproximatememories.pdf.jpg |
bitstream.checksum.fl_str_mv |
58e5123b2e6ae663789e5b5c7d1bc3bd b9d82215ab23456fa2d8b49c5df1b95b 36c6f0b2061da514c400c0bc2749b5cf 1fa63cf2b77bd128316c5abed3bc81e2 |
bitstream.checksumAlgorithm.fl_str_mv |
MD5 MD5 MD5 MD5 |
repository.name.fl_str_mv |
Repositório Institucional da UTFPR (da Universidade Tecnológica Federal do Paraná (RIUT)) - Universidade Tecnológica Federal do Paraná (UTFPR) |
repository.mail.fl_str_mv |
|
_version_ |
1805923010361688064 |