Memory Efficient One-Sided Communucation Library "aCP" in Globary Memory on Raspberry Pi 2

Yoshiyuki Morie, Hiroaki Honda, Takeshi Nanri, Taizo Kobayashi, Hidetomo Shibamura, Ryutaro Susukita, Yuichiro Ajima

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Abstract

Previously, communications in parallel programs forHigh Performance Computing (HPC) and Distributed Computing(DC) are mostly written with two-sided communicationinterfaces that are based on a pair of operations, Send andReceive. Since such interface requires explicit synchronizationbetween both sides of the communication, techniquesfor communication optimization such as overlapping are notefficiently described in many cases. On the other hand, onesidedcommunication interface is becoming important as amethod to describe asynchronous communications to enablehighly overlapped communication with computation. As oneof such interface, in this demonstration, Advanced CommunicationPrimitives (ACP) is introduced. ACP is a portableinterface that supports UDP, IBverbs of InfiniBand and Tofulibrary of K Computer. In addition to that, it is designed tobe memory efficient. For example, with 10 thousand processes, the memory consumption of ACP over UDP is estimated to beless than 1MB. Since the number of computational elements isincreasing more rapidly than the amount of available memory, this memory efficiency is becoming one of the keys for parallelprograms in HPC and DC. To show this characteristics, we runACP library on Raspberry Pi 2, and examine its performanceand memory consumption.

Original languageEnglish
Title of host publicationProceedings - 2016 IEEE 36th International Conference on Distributed Computing Systems, ICDCS 2016
PublisherInstitute of Electrical and Electronics Engineers Inc.
Pages761-762
Number of pages2
ISBN (Electronic)9781509014828
DOIs
Publication statusPublished - Aug 8 2016
Event36th IEEE International Conference on Distributed Computing Systems, ICDCS 2016 - Nara, Japan
Duration: Jun 27 2016Jun 30 2016

Publication series

NameProceedings - International Conference on Distributed Computing Systems
Volume2016-August

Other

Other36th IEEE International Conference on Distributed Computing Systems, ICDCS 2016
CountryJapan
CityNara
Period6/27/166/30/16

Fingerprint

Data storage equipment
Communication
Distributed computer systems
Demonstrations

All Science Journal Classification (ASJC) codes

  • Computer Networks and Communications
  • Hardware and Architecture
  • Software

Cite this

Morie, Y., Honda, H., Nanri, T., Kobayashi, T., Shibamura, H., Susukita, R., & Ajima, Y. (2016). Memory Efficient One-Sided Communucation Library "aCP" in Globary Memory on Raspberry Pi 2. In Proceedings - 2016 IEEE 36th International Conference on Distributed Computing Systems, ICDCS 2016 (pp. 761-762). [7536590] (Proceedings - International Conference on Distributed Computing Systems; Vol. 2016-August). Institute of Electrical and Electronics Engineers Inc.. https://doi.org/10.1109/ICDCS.2016.54

Memory Efficient One-Sided Communucation Library "aCP" in Globary Memory on Raspberry Pi 2. / Morie, Yoshiyuki; Honda, Hiroaki; Nanri, Takeshi; Kobayashi, Taizo; Shibamura, Hidetomo; Susukita, Ryutaro; Ajima, Yuichiro.

Proceedings - 2016 IEEE 36th International Conference on Distributed Computing Systems, ICDCS 2016. Institute of Electrical and Electronics Engineers Inc., 2016. p. 761-762 7536590 (Proceedings - International Conference on Distributed Computing Systems; Vol. 2016-August).

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Morie, Y, Honda, H, Nanri, T, Kobayashi, T, Shibamura, H, Susukita, R & Ajima, Y 2016, Memory Efficient One-Sided Communucation Library "aCP" in Globary Memory on Raspberry Pi 2. in Proceedings - 2016 IEEE 36th International Conference on Distributed Computing Systems, ICDCS 2016., 7536590, Proceedings - International Conference on Distributed Computing Systems, vol. 2016-August, Institute of Electrical and Electronics Engineers Inc., pp. 761-762, 36th IEEE International Conference on Distributed Computing Systems, ICDCS 2016, Nara, Japan, 6/27/16. https://doi.org/10.1109/ICDCS.2016.54
Morie Y, Honda H, Nanri T, Kobayashi T, Shibamura H, Susukita R et al. Memory Efficient One-Sided Communucation Library "aCP" in Globary Memory on Raspberry Pi 2. In Proceedings - 2016 IEEE 36th International Conference on Distributed Computing Systems, ICDCS 2016. Institute of Electrical and Electronics Engineers Inc. 2016. p. 761-762. 7536590. (Proceedings - International Conference on Distributed Computing Systems). https://doi.org/10.1109/ICDCS.2016.54
Morie, Yoshiyuki ; Honda, Hiroaki ; Nanri, Takeshi ; Kobayashi, Taizo ; Shibamura, Hidetomo ; Susukita, Ryutaro ; Ajima, Yuichiro. / Memory Efficient One-Sided Communucation Library "aCP" in Globary Memory on Raspberry Pi 2. Proceedings - 2016 IEEE 36th International Conference on Distributed Computing Systems, ICDCS 2016. Institute of Electrical and Electronics Engineers Inc., 2016. pp. 761-762 (Proceedings - International Conference on Distributed Computing Systems).
@inproceedings{9ddc205bf6084d55b2424335d71af29d,
title = "Memory Efficient One-Sided Communucation Library {"}aCP{"} in Globary Memory on Raspberry Pi 2",
abstract = "Previously, communications in parallel programs forHigh Performance Computing (HPC) and Distributed Computing(DC) are mostly written with two-sided communicationinterfaces that are based on a pair of operations, Send andReceive. Since such interface requires explicit synchronizationbetween both sides of the communication, techniquesfor communication optimization such as overlapping are notefficiently described in many cases. On the other hand, onesidedcommunication interface is becoming important as amethod to describe asynchronous communications to enablehighly overlapped communication with computation. As oneof such interface, in this demonstration, Advanced CommunicationPrimitives (ACP) is introduced. ACP is a portableinterface that supports UDP, IBverbs of InfiniBand and Tofulibrary of K Computer. In addition to that, it is designed tobe memory efficient. For example, with 10 thousand processes, the memory consumption of ACP over UDP is estimated to beless than 1MB. Since the number of computational elements isincreasing more rapidly than the amount of available memory, this memory efficiency is becoming one of the keys for parallelprograms in HPC and DC. To show this characteristics, we runACP library on Raspberry Pi 2, and examine its performanceand memory consumption.",
author = "Yoshiyuki Morie and Hiroaki Honda and Takeshi Nanri and Taizo Kobayashi and Hidetomo Shibamura and Ryutaro Susukita and Yuichiro Ajima",
year = "2016",
month = "8",
day = "8",
doi = "10.1109/ICDCS.2016.54",
language = "English",
series = "Proceedings - International Conference on Distributed Computing Systems",
publisher = "Institute of Electrical and Electronics Engineers Inc.",
pages = "761--762",
booktitle = "Proceedings - 2016 IEEE 36th International Conference on Distributed Computing Systems, ICDCS 2016",
address = "United States",

}

TY - GEN

T1 - Memory Efficient One-Sided Communucation Library "aCP" in Globary Memory on Raspberry Pi 2

AU - Morie, Yoshiyuki

AU - Honda, Hiroaki

AU - Nanri, Takeshi

AU - Kobayashi, Taizo

AU - Shibamura, Hidetomo

AU - Susukita, Ryutaro

AU - Ajima, Yuichiro

PY - 2016/8/8

Y1 - 2016/8/8

N2 - Previously, communications in parallel programs forHigh Performance Computing (HPC) and Distributed Computing(DC) are mostly written with two-sided communicationinterfaces that are based on a pair of operations, Send andReceive. Since such interface requires explicit synchronizationbetween both sides of the communication, techniquesfor communication optimization such as overlapping are notefficiently described in many cases. On the other hand, onesidedcommunication interface is becoming important as amethod to describe asynchronous communications to enablehighly overlapped communication with computation. As oneof such interface, in this demonstration, Advanced CommunicationPrimitives (ACP) is introduced. ACP is a portableinterface that supports UDP, IBverbs of InfiniBand and Tofulibrary of K Computer. In addition to that, it is designed tobe memory efficient. For example, with 10 thousand processes, the memory consumption of ACP over UDP is estimated to beless than 1MB. Since the number of computational elements isincreasing more rapidly than the amount of available memory, this memory efficiency is becoming one of the keys for parallelprograms in HPC and DC. To show this characteristics, we runACP library on Raspberry Pi 2, and examine its performanceand memory consumption.

AB - Previously, communications in parallel programs forHigh Performance Computing (HPC) and Distributed Computing(DC) are mostly written with two-sided communicationinterfaces that are based on a pair of operations, Send andReceive. Since such interface requires explicit synchronizationbetween both sides of the communication, techniquesfor communication optimization such as overlapping are notefficiently described in many cases. On the other hand, onesidedcommunication interface is becoming important as amethod to describe asynchronous communications to enablehighly overlapped communication with computation. As oneof such interface, in this demonstration, Advanced CommunicationPrimitives (ACP) is introduced. ACP is a portableinterface that supports UDP, IBverbs of InfiniBand and Tofulibrary of K Computer. In addition to that, it is designed tobe memory efficient. For example, with 10 thousand processes, the memory consumption of ACP over UDP is estimated to beless than 1MB. Since the number of computational elements isincreasing more rapidly than the amount of available memory, this memory efficiency is becoming one of the keys for parallelprograms in HPC and DC. To show this characteristics, we runACP library on Raspberry Pi 2, and examine its performanceand memory consumption.

UR - http://www.scopus.com/inward/record.url?scp=84985945032&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=84985945032&partnerID=8YFLogxK

U2 - 10.1109/ICDCS.2016.54

DO - 10.1109/ICDCS.2016.54

M3 - Conference contribution

AN - SCOPUS:84985945032

T3 - Proceedings - International Conference on Distributed Computing Systems

SP - 761

EP - 762

BT - Proceedings - 2016 IEEE 36th International Conference on Distributed Computing Systems, ICDCS 2016

PB - Institute of Electrical and Electronics Engineers Inc.

ER -