The original paper is in English. Non-English content has been machine-translated and may contain typographical errors or mistranslations. ex. Some numerals are expressed as "XNUMX".
Copyrights notice
The original paper is in English. Non-English content has been machine-translated and may contain typographical errors or mistranslations. Copyrights notice
O crescimento do tráfego e da escala do datacenter está impulsionando inovações na construção de instalações fortemente acopladas com comunicação de baixa latência para diferentes aplicações específicas. Um design personalizado famoso é a computação em escala de rack (RS), que reúne os principais componentes de recursos do servidor em diferentes pools de recursos. Essa implementação de agrupamento de recursos requer uma nova pilha de software para gerenciar a descoberta de recursos, a alocação de recursos e a comunicação de dados. A reconfiguração das redes de interligação em seus componentes é potencialmente necessária para suportar a demanda acima no RS. Neste contexto, como uma evolução da arquitetura RS original, foi proposta a arquitetura inter-rackscale (IRS), que desagrega os componentes de hardware em diferentes racks de acordo com suas próprias áreas. O coração do IRS é usar um número limitado de canais ópticos de espaço livre (FSO) para conexões sem fio entre diferentes racks de recursos, por meio dos quais pares selecionados de racks podem se comunicar diretamente e, assim, os requisitos de agrupamento de recursos são atendidos sem gerenciamento de software adicional. Neste estudo avaliamos as influências dos links FSO nas redes IRS. Os resultados da avaliação mostram que os links FSO reduzem a contagem média de saltos de comunicação para trabalhos de usuário, que está próxima do melhor valor possível de 2 saltos e, portanto, fornece desempenho de referência comparável ao da arquitetura RS homóloga. Além disso, se forem permitidos quatro terminais FSO por rack, a latência de interconexão CPU/SSD (GPU) é reduzida em 25.99% no Fat-tree e em 67.14% no 2-D Torus. Também apresentamos a vantagem de um sistema IRS equipado com FSO no tempo médio de resposta dos trabalhos despachados para determinados conjuntos de cargas de trabalho de referência.
Yao HU
National Institute of Informatics
Michihiro KOIBUCHI
National Institute of Informatics
The copyright of the original papers published on this site belongs to IEICE. Unauthorized use of the original or translated papers is prohibited. See IEICE Provisions on Copyright for details.
Copiar
Yao HU, Michihiro KOIBUCHI, "Enhancing Job Scheduling on Inter-Rackscale Datacenters with Free-Space Optical Links" in IEICE TRANSACTIONS on Information,
vol. E101-D, no. 12, pp. 2922-2932, December 2018, doi: 10.1587/transinf.2018PAP0010.
Abstract: Datacenter growth in traffic and scale is driving innovations in constructing tightly-coupled facilities with low-latency communication for different specific applications. A famous custom design is rackscale (RS) computing by gathering key server resource components into different resource pools. Such a resource-pooling implementation requires a new software stack to manage resource discovery, resource allocation and data communication. The reconfiguration of interconnection networks on their components is potentially needed to support the above demand in RS. In this context as an evolution of the original RS architecture the inter-rackscale (IRS) architecture, which disaggregates hardware components into different racks according to their own areas, has been proposed. The heart of IRS is to use a limited number of free-space optics (FSO) channels for wireless connections between different resource racks, via which selected pairs of racks can communicate directly and thus resource-pooling requirements are met without additional software management. In this study we evaluate the influences of FSO links on IRS networks. Evaluation results show that FSO links reduce average communication hop count for user jobs, which is close to the best possible value of 2 hops and thus provides comparable benchmark performance to that of the counterpart RS architecture. In addition, if four FSO terminals per rack are allowed, the CPU/SSD (GPU) interconnection latency is reduced by 25.99% over Fat-tree and by 67.14% over 2-D Torus. We also present the advantage of an FSO-equipped IRS system in average turnaround time of dispatched jobs for given sets of benchmark workloads.
URL: https://global.ieice.org/en_transactions/information/10.1587/transinf.2018PAP0010/_p
Copiar
@ARTICLE{e101-d_12_2922,
author={Yao HU, Michihiro KOIBUCHI, },
journal={IEICE TRANSACTIONS on Information},
title={Enhancing Job Scheduling on Inter-Rackscale Datacenters with Free-Space Optical Links},
year={2018},
volume={E101-D},
number={12},
pages={2922-2932},
abstract={Datacenter growth in traffic and scale is driving innovations in constructing tightly-coupled facilities with low-latency communication for different specific applications. A famous custom design is rackscale (RS) computing by gathering key server resource components into different resource pools. Such a resource-pooling implementation requires a new software stack to manage resource discovery, resource allocation and data communication. The reconfiguration of interconnection networks on their components is potentially needed to support the above demand in RS. In this context as an evolution of the original RS architecture the inter-rackscale (IRS) architecture, which disaggregates hardware components into different racks according to their own areas, has been proposed. The heart of IRS is to use a limited number of free-space optics (FSO) channels for wireless connections between different resource racks, via which selected pairs of racks can communicate directly and thus resource-pooling requirements are met without additional software management. In this study we evaluate the influences of FSO links on IRS networks. Evaluation results show that FSO links reduce average communication hop count for user jobs, which is close to the best possible value of 2 hops and thus provides comparable benchmark performance to that of the counterpart RS architecture. In addition, if four FSO terminals per rack are allowed, the CPU/SSD (GPU) interconnection latency is reduced by 25.99% over Fat-tree and by 67.14% over 2-D Torus. We also present the advantage of an FSO-equipped IRS system in average turnaround time of dispatched jobs for given sets of benchmark workloads.},
keywords={},
doi={10.1587/transinf.2018PAP0010},
ISSN={1745-1361},
month={December},}
Copiar
TY - JOUR
TI - Enhancing Job Scheduling on Inter-Rackscale Datacenters with Free-Space Optical Links
T2 - IEICE TRANSACTIONS on Information
SP - 2922
EP - 2932
AU - Yao HU
AU - Michihiro KOIBUCHI
PY - 2018
DO - 10.1587/transinf.2018PAP0010
JO - IEICE TRANSACTIONS on Information
SN - 1745-1361
VL - E101-D
IS - 12
JA - IEICE TRANSACTIONS on Information
Y1 - December 2018
AB - Datacenter growth in traffic and scale is driving innovations in constructing tightly-coupled facilities with low-latency communication for different specific applications. A famous custom design is rackscale (RS) computing by gathering key server resource components into different resource pools. Such a resource-pooling implementation requires a new software stack to manage resource discovery, resource allocation and data communication. The reconfiguration of interconnection networks on their components is potentially needed to support the above demand in RS. In this context as an evolution of the original RS architecture the inter-rackscale (IRS) architecture, which disaggregates hardware components into different racks according to their own areas, has been proposed. The heart of IRS is to use a limited number of free-space optics (FSO) channels for wireless connections between different resource racks, via which selected pairs of racks can communicate directly and thus resource-pooling requirements are met without additional software management. In this study we evaluate the influences of FSO links on IRS networks. Evaluation results show that FSO links reduce average communication hop count for user jobs, which is close to the best possible value of 2 hops and thus provides comparable benchmark performance to that of the counterpart RS architecture. In addition, if four FSO terminals per rack are allowed, the CPU/SSD (GPU) interconnection latency is reduced by 25.99% over Fat-tree and by 67.14% over 2-D Torus. We also present the advantage of an FSO-equipped IRS system in average turnaround time of dispatched jobs for given sets of benchmark workloads.
ER -