Optically connected memory for disaggregated data centers

Jorge Gonzalez; Mauricio G. Palma; Maarten Hattink; Ruth Rubio-Noriega; Lois Orosa; Onur Mutlu; Keren Bergman; Rodolfo Azevedo

doi:10.1016/j.jpdc.2022.01.013

Optically connected memory for disaggregated data centers

Jorge Gonzalez, Mauricio G. Palma, Maarten Hattink, Ruth Rubio-Noriega, Lois Orosa, Onur Mutlu, Keren Bergman, Rodolfo Azevedo

Department of Computer Science

Research output: Contribution to journal › Article › peer-review

7 Scopus citations

Abstract

Recent advances in integrated photonics enable the implementation of reconfigurable, high-bandwidth, and low energy-per-bit interconnects in next-generation data centers. We propose and evaluate an Optically Connected Memory (OCM) architecture that disaggregates the main memory from the computation nodes in data centers. OCM is based on micro-ring resonators (MRRs), and it does not require any modification to the DRAM memory modules. We calculate energy consumption from real photonic devices and integrate them into a system simulator to evaluate performance. Our results show that (1) OCM is capable of interconnecting four DDR4 memory channels to a computing node using two fibers with 1.02 pJ energy-per-bit consumption and (2) OCM performs up to 5.5× faster than a disaggregated memory with 40G PCIe NIC connectors to computing nodes.

Original language	English
Pages (from-to)	300-312
Number of pages	13
Journal	Journal of Parallel and Distributed Computing
Volume	163
DOIs	https://doi.org/10.1016/j.jpdc.2022.01.013
State	Published - May 2022

Keywords

DRAM
Data-centers
Disaggregated memory
Memory systems
Photonics

Access to Document

10.1016/j.jpdc.2022.01.013

Cite this

@article{f645dc1d95aa46c7b3d267e59276f958,

title = "Optically connected memory for disaggregated data centers",

abstract = "Recent advances in integrated photonics enable the implementation of reconfigurable, high-bandwidth, and low energy-per-bit interconnects in next-generation data centers. We propose and evaluate an Optically Connected Memory (OCM) architecture that disaggregates the main memory from the computation nodes in data centers. OCM is based on micro-ring resonators (MRRs), and it does not require any modification to the DRAM memory modules. We calculate energy consumption from real photonic devices and integrate them into a system simulator to evaluate performance. Our results show that (1) OCM is capable of interconnecting four DDR4 memory channels to a computing node using two fibers with 1.02 pJ energy-per-bit consumption and (2) OCM performs up to 5.5× faster than a disaggregated memory with 40G PCIe NIC connectors to computing nodes.",

keywords = "DRAM, Data-centers, Disaggregated memory, Memory systems, Photonics",

author = "Jorge Gonzalez and {G. Palma}, Mauricio and Maarten Hattink and Ruth Rubio-Noriega and Lois Orosa and Onur Mutlu and Keren Bergman and Rodolfo Azevedo",

note = "Publisher Copyright: {\textcopyright} 2022 Elsevier Inc.",

year = "2022",

month = may,

doi = "10.1016/j.jpdc.2022.01.013",

language = "English",

volume = "163",

pages = "300--312",

journal = "Journal of Parallel and Distributed Computing",

issn = "0743-7315",

}

TY - JOUR

T1 - Optically connected memory for disaggregated data centers

AU - Gonzalez, Jorge

AU - G. Palma, Mauricio

AU - Hattink, Maarten

AU - Rubio-Noriega, Ruth

AU - Orosa, Lois

AU - Mutlu, Onur

AU - Bergman, Keren

AU - Azevedo, Rodolfo

PY - 2022/5

Y1 - 2022/5

N2 - Recent advances in integrated photonics enable the implementation of reconfigurable, high-bandwidth, and low energy-per-bit interconnects in next-generation data centers. We propose and evaluate an Optically Connected Memory (OCM) architecture that disaggregates the main memory from the computation nodes in data centers. OCM is based on micro-ring resonators (MRRs), and it does not require any modification to the DRAM memory modules. We calculate energy consumption from real photonic devices and integrate them into a system simulator to evaluate performance. Our results show that (1) OCM is capable of interconnecting four DDR4 memory channels to a computing node using two fibers with 1.02 pJ energy-per-bit consumption and (2) OCM performs up to 5.5× faster than a disaggregated memory with 40G PCIe NIC connectors to computing nodes.

AB - Recent advances in integrated photonics enable the implementation of reconfigurable, high-bandwidth, and low energy-per-bit interconnects in next-generation data centers. We propose and evaluate an Optically Connected Memory (OCM) architecture that disaggregates the main memory from the computation nodes in data centers. OCM is based on micro-ring resonators (MRRs), and it does not require any modification to the DRAM memory modules. We calculate energy consumption from real photonic devices and integrate them into a system simulator to evaluate performance. Our results show that (1) OCM is capable of interconnecting four DDR4 memory channels to a computing node using two fibers with 1.02 pJ energy-per-bit consumption and (2) OCM performs up to 5.5× faster than a disaggregated memory with 40G PCIe NIC connectors to computing nodes.

KW - DRAM

KW - Data-centers

KW - Disaggregated memory

KW - Memory systems

KW - Photonics

UR - http://www.scopus.com/inward/record.url?scp=85124986604&partnerID=8YFLogxK

U2 - 10.1016/j.jpdc.2022.01.013

DO - 10.1016/j.jpdc.2022.01.013

M3 - Article

AN - SCOPUS:85124986604

SN - 0743-7315

VL - 163

SP - 300

EP - 312

JO - Journal of Parallel and Distributed Computing

JF - Journal of Parallel and Distributed Computing

ER -

Optically connected memory for disaggregated data centers

Abstract

Keywords

Access to Document

Other files and links

Fingerprint

Cite this