Managing Response Time Tails by Sharding
Matrix analytic methods are developed to compute the probability distribution of response times (i.e., data access times) in distributed storage systems protected by erasure coding, which is implemented by sharding a data object into N fragments, only K less than N of which are required to reconstru...
- Autores:
- Tipo de recurso:
- Fecha de publicación:
- 2019
- Institución:
- Universidad del Rosario
- Repositorio:
- Repositorio EdocUR - U. Rosario
- Idioma:
- eng
- OAI Identifier:
- oai:repository.urosario.edu.co:10336/23233
- Acceso en línea:
- https://doi.org/10.1145/3300143
https://repository.urosario.edu.co/handle/10336/23233
- Palabra clave:
- Multiprocessing systems
Probability distributions
Quality of service
Response time (computer systems)
Distributed storage system
Matrix analytic methods
Mean response time
Numerical results
Parallel task
Performance
Sharding
Workload intensities
Digital storage
Parallel task processing
Performance
Quality of service
Response time
Sharding
- Rights
- License
- Abierto (Texto Completo)
id |
EDOCUR2_87ce6d0102aee5952804e4123fa4f2c4 |
---|---|
oai_identifier_str |
oai:repository.urosario.edu.co:10336/23233 |
network_acronym_str |
EDOCUR2 |
network_name_str |
Repositorio EdocUR - U. Rosario |
repository_id_str |
|
spelling |
c6b727dd-a371-4d51-b960-3a8d758399e18cfccc85-3a7d-4edf-9b82-cc930c598dd480035202600fc4c8af7-473a-473f-8114-951ac1db1cae2020-05-26T00:00:31Z2020-05-26T00:00:31Z2019Matrix analytic methods are developed to compute the probability distribution of response times (i.e., data access times) in distributed storage systems protected by erasure coding, which is implemented by sharding a data object into N fragments, only K less than N of which are required to reconstruct the object. This leads to a partial-fork-join model with a choice of canceling policies for the redundant N - K tasks. The accuracy of the analytical model is supported by tests against simulation in a broad range of setups. At increasing workload intensities, numerical results show the extent to which increasing the redundancy level reduces the mean response time of storage reads and significantly flattens the tail of their distribution; this is demonstrated at medium-high quantiles, up to the 99th. The quantitative reduction in response time achieved by two policies for canceling redundant tasks is also shown: For cancel-at-finish and cancel-at-start, which limits the additional load introduced whilst losing the benefit of selectivity amongst fragment service times. © 2019 Copyright held by the owner/author(s).application/pdfhttps://doi.org/10.1145/3300143https://repository.urosario.edu.co/handle/10336/23233engAssociation for Computing MachineryNo. 1ACM Transactions on Modeling and Performance Evaluation of Computing SystemsVol. 4ACM Transactions on Modeling and Performance Evaluation of Computing Systems, Vol.4, No.1 (2019)https://www.scopus.com/inward/record.uri?eid=2-s2.0-85074676254&doi=10.1145%2f3300143&partnerID=40&md5=241dbfc47e2d8a85dd826cf3508f0f13Abierto (Texto Completo)http://purl.org/coar/access_right/c_abf2instname:Universidad del Rosarioreponame:Repositorio Institucional EdocURMultiprocessing systemsProbability distributionsQuality of serviceResponse time (computer systems)Distributed storage systemMatrix analytic methodsMean response timeNumerical resultsParallel taskPerformanceShardingWorkload intensitiesDigital storageParallel task processingPerformanceQuality of serviceResponse timeShardingManaging Response Time Tails by ShardingarticleArtículohttp://purl.org/coar/version/c_970fb48d4fbd8a85http://purl.org/coar/resource_type/c_6501Harrison, P. G.Patel, N. M.Pérez, Juan F.Qiu, Z.10336/23233oai:repository.urosario.edu.co:10336/232332022-05-02 07:37:17.073938https://repository.urosario.edu.coRepositorio institucional EdocURedocur@urosario.edu.co |
dc.title.spa.fl_str_mv |
Managing Response Time Tails by Sharding |
title |
Managing Response Time Tails by Sharding |
spellingShingle |
Managing Response Time Tails by Sharding Multiprocessing systems Probability distributions Quality of service Response time (computer systems) Distributed storage system Matrix analytic methods Mean response time Numerical results Parallel task Performance Sharding Workload intensities Digital storage Parallel task processing Performance Quality of service Response time Sharding |
title_short |
Managing Response Time Tails by Sharding |
title_full |
Managing Response Time Tails by Sharding |
title_fullStr |
Managing Response Time Tails by Sharding |
title_full_unstemmed |
Managing Response Time Tails by Sharding |
title_sort |
Managing Response Time Tails by Sharding |
dc.subject.keyword.spa.fl_str_mv |
Multiprocessing systems Probability distributions Quality of service Response time (computer systems) Distributed storage system Matrix analytic methods Mean response time Numerical results Parallel task Performance Sharding Workload intensities Digital storage Parallel task processing Performance Quality of service Response time Sharding |
topic |
Multiprocessing systems Probability distributions Quality of service Response time (computer systems) Distributed storage system Matrix analytic methods Mean response time Numerical results Parallel task Performance Sharding Workload intensities Digital storage Parallel task processing Performance Quality of service Response time Sharding |
description |
Matrix analytic methods are developed to compute the probability distribution of response times (i.e., data access times) in distributed storage systems protected by erasure coding, which is implemented by sharding a data object into N fragments, only K less than N of which are required to reconstruct the object. This leads to a partial-fork-join model with a choice of canceling policies for the redundant N - K tasks. The accuracy of the analytical model is supported by tests against simulation in a broad range of setups. At increasing workload intensities, numerical results show the extent to which increasing the redundancy level reduces the mean response time of storage reads and significantly flattens the tail of their distribution; this is demonstrated at medium-high quantiles, up to the 99th. The quantitative reduction in response time achieved by two policies for canceling redundant tasks is also shown: For cancel-at-finish and cancel-at-start, which limits the additional load introduced whilst losing the benefit of selectivity amongst fragment service times. © 2019 Copyright held by the owner/author(s). |
publishDate |
2019 |
dc.date.created.spa.fl_str_mv |
2019 |
dc.date.accessioned.none.fl_str_mv |
2020-05-26T00:00:31Z |
dc.date.available.none.fl_str_mv |
2020-05-26T00:00:31Z |
dc.type.eng.fl_str_mv |
article |
dc.type.coarversion.fl_str_mv |
http://purl.org/coar/version/c_970fb48d4fbd8a85 |
dc.type.coar.fl_str_mv |
http://purl.org/coar/resource_type/c_6501 |
dc.type.spa.spa.fl_str_mv |
Artículo |
dc.identifier.doi.none.fl_str_mv |
https://doi.org/10.1145/3300143 |
dc.identifier.uri.none.fl_str_mv |
https://repository.urosario.edu.co/handle/10336/23233 |
url |
https://doi.org/10.1145/3300143 https://repository.urosario.edu.co/handle/10336/23233 |
dc.language.iso.spa.fl_str_mv |
eng |
language |
eng |
dc.relation.citationIssue.none.fl_str_mv |
No. 1 |
dc.relation.citationTitle.none.fl_str_mv |
ACM Transactions on Modeling and Performance Evaluation of Computing Systems |
dc.relation.citationVolume.none.fl_str_mv |
Vol. 4 |
dc.relation.ispartof.spa.fl_str_mv |
ACM Transactions on Modeling and Performance Evaluation of Computing Systems, Vol.4, No.1 (2019) |
dc.relation.uri.spa.fl_str_mv |
https://www.scopus.com/inward/record.uri?eid=2-s2.0-85074676254&doi=10.1145%2f3300143&partnerID=40&md5=241dbfc47e2d8a85dd826cf3508f0f13 |
dc.rights.coar.fl_str_mv |
http://purl.org/coar/access_right/c_abf2 |
dc.rights.acceso.spa.fl_str_mv |
Abierto (Texto Completo) |
rights_invalid_str_mv |
Abierto (Texto Completo) http://purl.org/coar/access_right/c_abf2 |
dc.format.mimetype.none.fl_str_mv |
application/pdf |
dc.publisher.spa.fl_str_mv |
Association for Computing Machinery |
institution |
Universidad del Rosario |
dc.source.instname.spa.fl_str_mv |
instname:Universidad del Rosario |
dc.source.reponame.spa.fl_str_mv |
reponame:Repositorio Institucional EdocUR |
repository.name.fl_str_mv |
Repositorio institucional EdocUR |
repository.mail.fl_str_mv |
edocur@urosario.edu.co |
_version_ |
1814167609694748672 |