Towards tightly-coupled datacenter with free-space optical links

Yao Hu, Hiroki Matsutani, Hiroaki Hara, Hideharu Amano, Ikki Fujiwara, Michihiro Koibuchi

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

1 Citation (Scopus)

Abstract

Clean slate design of computing system is an emerging topic for continuing growth of warehouse-scale computers. A famous custom design is rackscale (RS) computing by considering a single rack as a computer that consists of a number of processors, storages and accelerators customized to a target application. In RS, each user is expected to occupy a single or more than one rack. However, new users frequently appear and the users often change their application scales and parameters that would require different numbers of processors, storages and accelerators in a rack. The reconfiguration of interconnection networks on their components is potentially needed to support the above demand in RS. In this context, we propose the inter-rackscale (IRS) architecture that disaggregates various hardware resources into different racks according to their own areas. The heart of IRS is to use free-space optics (FSO) for tightly-coupled connections between processors, storages and GPUS distributed in different racks, by swapping endpoints of FSO links to change network topologies. Through a large IRS system simulation, we show that by utilizing FSO links for interconnection between racks, the FSO-equipped IRS architecture can provide comparable communication latency between heterogeneous resources to that of the counterpart RS architecture. A utilization of 3 FSO terminals per rack can improve at least 87.34% of inter-CPU/SSD(GPU) communication over Fat-tree and improve at least 92.18% of that over 2-D Torus. We verify the advantages of IRS over RS in job scheduling performance..

Original languageEnglish
Title of host publication2017 International Conference on Cloud and Big Data Computing, ICCBDC 2017
PublisherAssociation for Computing Machinery
Pages33-39
Number of pages7
ISBN (Electronic)9781450353434
DOIs
Publication statusPublished - 2017 Sep 17
Event2017 International Conference on Cloud and Big Data Computing, ICCBDC 2017 - London, United Kingdom
Duration: 2017 Sep 172017 Sep 19

Other

Other2017 International Conference on Cloud and Big Data Computing, ICCBDC 2017
CountryUnited Kingdom
CityLondon
Period17/9/1717/9/19

Fingerprint

Space optics
Optical links
Particle accelerators
Slate
Communication
Warehouses
Oils and fats
Program processors
Scheduling
Topology
Hardware

Keywords

  • Datacenter
  • Free-space optics
  • Interconnection network
  • Job scheduling
  • Rackscale architecture

ASJC Scopus subject areas

  • Human-Computer Interaction
  • Computer Networks and Communications
  • Computer Vision and Pattern Recognition
  • Software

Cite this

Hu, Y., Matsutani, H., Hara, H., Amano, H., Fujiwara, I., & Koibuchi, M. (2017). Towards tightly-coupled datacenter with free-space optical links. In 2017 International Conference on Cloud and Big Data Computing, ICCBDC 2017 (pp. 33-39). Association for Computing Machinery. https://doi.org/10.1145/3141128.3141130

Towards tightly-coupled datacenter with free-space optical links. / Hu, Yao; Matsutani, Hiroki; Hara, Hiroaki; Amano, Hideharu; Fujiwara, Ikki; Koibuchi, Michihiro.

2017 International Conference on Cloud and Big Data Computing, ICCBDC 2017. Association for Computing Machinery, 2017. p. 33-39.

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

Hu, Y, Matsutani, H, Hara, H, Amano, H, Fujiwara, I & Koibuchi, M 2017, Towards tightly-coupled datacenter with free-space optical links. in 2017 International Conference on Cloud and Big Data Computing, ICCBDC 2017. Association for Computing Machinery, pp. 33-39, 2017 International Conference on Cloud and Big Data Computing, ICCBDC 2017, London, United Kingdom, 17/9/17. https://doi.org/10.1145/3141128.3141130
Hu Y, Matsutani H, Hara H, Amano H, Fujiwara I, Koibuchi M. Towards tightly-coupled datacenter with free-space optical links. In 2017 International Conference on Cloud and Big Data Computing, ICCBDC 2017. Association for Computing Machinery. 2017. p. 33-39 https://doi.org/10.1145/3141128.3141130
Hu, Yao ; Matsutani, Hiroki ; Hara, Hiroaki ; Amano, Hideharu ; Fujiwara, Ikki ; Koibuchi, Michihiro. / Towards tightly-coupled datacenter with free-space optical links. 2017 International Conference on Cloud and Big Data Computing, ICCBDC 2017. Association for Computing Machinery, 2017. pp. 33-39
@inproceedings{99a52e3aa8f14a6f8e8f4eff81641264,
title = "Towards tightly-coupled datacenter with free-space optical links",
abstract = "Clean slate design of computing system is an emerging topic for continuing growth of warehouse-scale computers. A famous custom design is rackscale (RS) computing by considering a single rack as a computer that consists of a number of processors, storages and accelerators customized to a target application. In RS, each user is expected to occupy a single or more than one rack. However, new users frequently appear and the users often change their application scales and parameters that would require different numbers of processors, storages and accelerators in a rack. The reconfiguration of interconnection networks on their components is potentially needed to support the above demand in RS. In this context, we propose the inter-rackscale (IRS) architecture that disaggregates various hardware resources into different racks according to their own areas. The heart of IRS is to use free-space optics (FSO) for tightly-coupled connections between processors, storages and GPUS distributed in different racks, by swapping endpoints of FSO links to change network topologies. Through a large IRS system simulation, we show that by utilizing FSO links for interconnection between racks, the FSO-equipped IRS architecture can provide comparable communication latency between heterogeneous resources to that of the counterpart RS architecture. A utilization of 3 FSO terminals per rack can improve at least 87.34{\%} of inter-CPU/SSD(GPU) communication over Fat-tree and improve at least 92.18{\%} of that over 2-D Torus. We verify the advantages of IRS over RS in job scheduling performance..",
keywords = "Datacenter, Free-space optics, Interconnection network, Job scheduling, Rackscale architecture",
author = "Yao Hu and Hiroki Matsutani and Hiroaki Hara and Hideharu Amano and Ikki Fujiwara and Michihiro Koibuchi",
year = "2017",
month = "9",
day = "17",
doi = "10.1145/3141128.3141130",
language = "English",
pages = "33--39",
booktitle = "2017 International Conference on Cloud and Big Data Computing, ICCBDC 2017",
publisher = "Association for Computing Machinery",

}

TY - GEN

T1 - Towards tightly-coupled datacenter with free-space optical links

AU - Hu, Yao

AU - Matsutani, Hiroki

AU - Hara, Hiroaki

AU - Amano, Hideharu

AU - Fujiwara, Ikki

AU - Koibuchi, Michihiro

PY - 2017/9/17

Y1 - 2017/9/17

N2 - Clean slate design of computing system is an emerging topic for continuing growth of warehouse-scale computers. A famous custom design is rackscale (RS) computing by considering a single rack as a computer that consists of a number of processors, storages and accelerators customized to a target application. In RS, each user is expected to occupy a single or more than one rack. However, new users frequently appear and the users often change their application scales and parameters that would require different numbers of processors, storages and accelerators in a rack. The reconfiguration of interconnection networks on their components is potentially needed to support the above demand in RS. In this context, we propose the inter-rackscale (IRS) architecture that disaggregates various hardware resources into different racks according to their own areas. The heart of IRS is to use free-space optics (FSO) for tightly-coupled connections between processors, storages and GPUS distributed in different racks, by swapping endpoints of FSO links to change network topologies. Through a large IRS system simulation, we show that by utilizing FSO links for interconnection between racks, the FSO-equipped IRS architecture can provide comparable communication latency between heterogeneous resources to that of the counterpart RS architecture. A utilization of 3 FSO terminals per rack can improve at least 87.34% of inter-CPU/SSD(GPU) communication over Fat-tree and improve at least 92.18% of that over 2-D Torus. We verify the advantages of IRS over RS in job scheduling performance..

AB - Clean slate design of computing system is an emerging topic for continuing growth of warehouse-scale computers. A famous custom design is rackscale (RS) computing by considering a single rack as a computer that consists of a number of processors, storages and accelerators customized to a target application. In RS, each user is expected to occupy a single or more than one rack. However, new users frequently appear and the users often change their application scales and parameters that would require different numbers of processors, storages and accelerators in a rack. The reconfiguration of interconnection networks on their components is potentially needed to support the above demand in RS. In this context, we propose the inter-rackscale (IRS) architecture that disaggregates various hardware resources into different racks according to their own areas. The heart of IRS is to use free-space optics (FSO) for tightly-coupled connections between processors, storages and GPUS distributed in different racks, by swapping endpoints of FSO links to change network topologies. Through a large IRS system simulation, we show that by utilizing FSO links for interconnection between racks, the FSO-equipped IRS architecture can provide comparable communication latency between heterogeneous resources to that of the counterpart RS architecture. A utilization of 3 FSO terminals per rack can improve at least 87.34% of inter-CPU/SSD(GPU) communication over Fat-tree and improve at least 92.18% of that over 2-D Torus. We verify the advantages of IRS over RS in job scheduling performance..

KW - Datacenter

KW - Free-space optics

KW - Interconnection network

KW - Job scheduling

KW - Rackscale architecture

UR - http://www.scopus.com/inward/record.url?scp=85045750710&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85045750710&partnerID=8YFLogxK

U2 - 10.1145/3141128.3141130

DO - 10.1145/3141128.3141130

M3 - Conference contribution

AN - SCOPUS:85045750710

SP - 33

EP - 39

BT - 2017 International Conference on Cloud and Big Data Computing, ICCBDC 2017

PB - Association for Computing Machinery

ER -