Title
Optically connected memory for disaggregated data centers
Date Issued
01 September 2020
Access level
open access
Resource Type
conference paper
Author(s)
Gazman A.
Hattink M.
Palma M.G.
Bahadori M.
Orosa L.
Glick M.
Mutlu O.
Bergman K.
Azevedo R.
Universidad Estatal de Campinas
Publisher(s)
IEEE Computer Society
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 1.07 pJ energy-per-bit consumption and (2) OCM performs up to 5.5x faster than a disaggregated memory with 40G PCIe NIC connectors to computing nodes.
Start page
43
End page
50
Volume
2020-September
Language
English
OCDE Knowledge area
Ciencias de la computación
Telecomunicaciones
Subjects
Scopus EID
2-s2.0-85095864848
Source
Proceedings - Symposium on Computer Architecture and High Performance Computing
Resource of which it is part
Proceedings - Symposium on Computer Architecture and High Performance Computing
ISSN of the container
15506533
ISBN of the container
9781728199245
Conference
32nd IEEE International Symposium on Computer Architecture and High Performance Computing, SBAC-PAD 2020 Virtual, Porto 8 September 2020 through 11 September 2020
Sponsor(s)
This work is supported by the LPS Advanced Computing Systems (ACS) Research Program (contract HD TAT DO 7 (HT 15-1158)), the Department of Energy (DOE) Small Business Innovation Research (SBIR) ASCR Program (contract DE-SC0017182), the Sao Paulo Research Foundation (FAPESP) (fellowships 2013/08293-7 and 2014/01642-9), CAPES (fellowships 2013/08293-7 and 88882.329108/2019-01), and CNPq (fellowships 438445/2018-0, 309794/2017-0 and 142016/2020-9).
Sources of information:
Directorio de Producción Científica
Scopus