电气/光混合数据中心网络中的资源分配策略

97 浏览量更新于2024-08-28 收藏 1.01MB PDF 举报

在当前和未来数据中心网络（DCNs）的设计中，电气/光学混合交换技术结合了电子包交换的灵活性和高速光电路交换的效率，被认为是一种具有成本效益的解决方案。本文的焦点在于资源分配，这是构建混合开关DCN的关键要素，旨在平衡光学电路交换与电子包交换的资源投入，同时满足性能需求，如减少整体系统建设成本或电力消耗。论文《Resource Allocation in Electrical/Optical Hybrid Switching Data Center Networks》由张晓峰、孙伟强、朱杰、邵俊义和胡蔚生等人提出。他们提出了一个名为Blockading Loss Curve (BLOC)的新工具，这是一个用于描绘混合开关DCN特性的模型。BLOC有助于整合和研究混合系统中的三个相互作用组件：资源分配、流量分割以及系统性能。通过BLOC，研究人员能够分析不同资源分配和流量分配组合对系统的影响，找到既满足性能指标又能优化成本或能耗的可行方案。这种方法将传统的资源管理和流量管理策略与实际网络性能紧密联系起来，使得设计者能够在兼顾效率和经济性的同时，做出更加明智的决策。 BLOC的优势在于它能够直观地展示不同资源分配策略下可能遇到的阻塞损失，即因资源不足导致的数据包丢失率。这有助于网络设计者识别瓶颈并进行调整，确保在网络流量增长的同时，系统仍能保持高效稳定运行。此外，BLOC还可能为优化网络拓扑、选择合适的技术栈和设备配置提供有价值的信息，从而实现DCN的可持续发展和高效运营。这篇研究论文不仅提供了混合交换DCN中资源分配的理论框架，还引入了一种实用工具——BLOC，这对于理解和优化这类复杂网络的性能、成本和能源效率具有重要意义。通过深入理解BLOC的工作原理，业界可以更好地规划和实施未来的数据中心网络架构，确保在竞争激烈的市场环境中保持竞争优势。

Resource Allocation in Electrical/Optical

Hybrid Switching Data Center Networks

Zhangxiao Feng, Weiqiang Sun, Jie Zhu, Junyi Shao, and Weisheng Hu

Abstract—Hybrid switching combines the strengths of

electronic packet switching and high-speed optical circuit

switching and is thus widely believed to be a cost-effective

solution for current and future data center networks

(DCNs). In designing a hybrid switching DCN, it is essential

to know how many resources should be devoted to optical

circuit switching and electronic packet switching so the

performance requirements in both planes can be satisfied,

and, at the same time, objectives such as minimizing the

overall system building cost or the power consumption

can be realized. In this paper, we introduce BLOC, the

Blocking LOss Curve, as a tool to characterize hybrid

switching DCNs. With BLOC, the three interacting compo-

nents in hybrid switching systems—resource allocation,

traffic partitioning, and system performance—can be natu-

rally integrated and studied together. We show how BLOC

can be used to obtain feasible resource allocation/traffic

partition combinations. We also show how system parame-

ters, such as flow arrival characteristics, may affect re-

source allocation and system costs. Our study reveals

that the total amount of resource required to satisfy 95th

percentile traffic can be as low as 60% of that required

for 99.9th percentile traffic.

Index Terms—Data center networking; Hybrid switching

networks; Network performance analysis; Network re-

source allocation; Optical circuit switching; Traffic

scheduling.

I. INTRODUCTION

o meet the increasing demands for cloud computing,

streaming video, and other emerging cloud applica-

tions, data center networks (DCNs) have been constructed

that can accommodate hundreds of thousands of servers.

The data-intensive applications hosted in the servers

require significant link capacity for data transmission.

Annual global data center traffic is expected to grow at a

rate of 27% over the next few years and will reach 15.3

zettabytes by the end of 2020 [1]. Cisco reports that 77%

of data center traffic remains within data centers [1].

This huge rack-to-rack traffic presents major challenges

for high-speed and low-cost network interconnection

between top-of-rack (ToR) switches.

To meet the bandwidth demands in data centers, hierar-

chical multi-layer tree topologies based on conventional

electronic packet switching (EPS) have been proposed,

such as Fat-Tree [2] and VL2 [3]. However, the number

of required switches increases as the size of the data center

itself increases, and EPS-based DCN architectures con-

sume a large amount of power and have extremely complex

cabling. Hybrid switching architectures that combine opti-

cal circuit switching (OCS) and EPS have recently been

introduced into DCNs [4–10]. By offloading large flows

or relatively stable portions of traffic to slow-changing

and coarser-granular OCS, the cost-effectiveness of deliver-

ing the same amount of traffic in DCNs can be improved. In

[11], the author claimed that using hybrid switching may

bring about 2.8×, 6×, and 4.7× reductions in cost, power,

and cabling complexity, respectively.

In designing a hybrid DCN with both OCS and EPS, it is

essential to know how many resources should be allocated

to each switching plane so the performance requirements

in both planes can be satisfied and objectives such as min-

imizing the overall system power consumption can be real-

ized. Here, resource can refer to the overall budget for

realizing the network or the available cabling/interfaces

used to interconnect the ToRs. This problem is complicated

by the fact that, in hybrid switching systems, the portions

of traffic to be delivered by each switching plane may not be

static or known in advance. It is further complicated by the

fact that the performance requirements in the two switch-

ing planes are different, and, given a cost constraint, the

resource allocation is essentially a tradeoff between the

performances in the two switching planes. In summary,

the interactions between the three components, i.e.,

resource allocation, traffic partitioning, and system perfor-

mance, are essential in the design of hybrid switching

DCNs and must therefore be studied together.

In this paper, we study the design of DCNs with a par-

ticular emphasis on resource allocation. We assume the de-

sign problem is subject to a budget constraint, denoted by

the number of ToR uplink interfaces. We also assume the

performances in the two switching planes are bounded by

maximal allowable values. The overall traffic demand,

comprising both mice and elephant flows, is assumed to

be served by EPS in part and the remainder by OCS,

depending on a system-wide flow size threshold.

Our research makes the following contributions:

• We introduce a framework called BLOC (Blocking LOss

Curve), into which the three components in hybrid

switching DCN design can be naturally integrated and

studied.

https://doi.org/10.1364/JOCN.9.000648

Manuscript received January 18, 2017; revised May 19, 2017; accepted

June 6, 2017; published July 25, 2017 (Doc. ID 284910).

The authors are with the State Key Laboratory of Advanced Optical

Communication Systems and Networks, Shanghai Jiao Tong University,

Shanghai 200240, China (e-mail: sunwq@sjtu.edu. cn).

648 J. OPT. COMMUN. NETW./VOL. 9, NO. 8/AUGUST 2017 Feng et al.

下载后可阅读完整内容，剩余9页未读，立即下载

weixin_38529239

粉丝: 4
资源: 927

电气/光混合数据中心网络中的资源分配策略

"PLC在万能铣床控制系统设计与应用研究

面向任务的多传感器资源预分配建模与求解：优化整体性能与资源利用【摘要】

"Apache Storm的资源感知调度研究与实践

Sharing Ratio Based Manycast Resource Allocation in Elastic Optical Networks

Performance evaluation of a WDM/OCDM based hybrid optical switch utilizing efficient resource allocation

Interference aware resource allocation for hybrid hierarchical wireless networks

A Relaxation Algorithm for Decentralized Resource Allocation in Femtocell Networks

Joint resource allocation for hybrid NOMA-assisted MEC in 6G networks.pdf

QoE and Energy Aware Resource Allocation in Small Cell Networks with Power Selection, Load Management and Channel Allocation

Resource Allocation for Spectrum Underlay in cognitive radio networks

最新资源