The Design Of Wireless Data Center Networks And Their Routing Algorithms

Data center is the core infrastructure of cloud computing,and data center network is an important factor affecting the performance of data center,which plays a key role in cloud computing infrastructure.Data center network(DCN)interconnects large-scale servers,and is the basis for carrying networked storage and networked computing.The booming of the Internet and the digital transformation of the society have not only greatly promoted the development of the data centers,but also put forward many new requirements for the construction and performance of the data center.Aiming at solving the challenges faced by traditional wired data centers in terms of cabling,expansion,operation and maintenance,the research of wireless data center network is gradually rising.However,the existing wireless data center networks still have certain limitations in terms of ease of deployment and plug-and-play.This paper aims to build new wireless data center networks to solve these limitations.To this end,this paper has carried out the following research:(1)Survey the existing DCN topologies systematically.With the rise of new computing and applications such as cloud computing,Internet of Things,and big data,traditional tree structures have been unable to meet new demands.Based on different design concepts,researchers have designed a variety of data center networks to meet the high performance and sustainability requirements of data centers.This paper systematically sorts and classifies the existing data center network structures,and the typical data center network structure of each category is taken as an example to summarize the design goal of each category,and analyze their respective advantages and limitations.(2)Design inter-rack wireless DCN named WSWDC.In the process of constructing wireless DCN,to solve the limitations of current research results,this paper proposes three design rationales for wireless network construction,namely: 1)fully wireless,all inter-rack links are wireless;2)easydeployable,no need to change the existing infrastructure of the data center;3)plug-andplay,no additional centralized control and management entities.Based on these three design rationales,we have designed WSWDC.This is the first fully wireless DCN design based on visible light communication.WSWDC's wireless links are divided into regular links and random links,featuring small world networks.According to the structure characteristics of WSWDC,this paper designs an efficient routing mechanism based on identifiers to complete point-to-point wireless routing in constant time complexity.(3)Design novel hybrid wireless data center networks.In order to reduce network cost and increase network flexibility,this paper designs two hybrid wireless data center networks,named SFNet and TIO.Both of them enhance the existing wired Fat-Tree with wireless network,and realize the complementary advantages of wired structure and wireless structure.This article focuses on the construction and optimization of SFNet.In SFNet,for any pair of racks,there are wired paths,wireless paths,and hybrid paths.This paper focuses on the design of SFNet hybrid routing algorithm.In addition,in order to minimize network congestion,we model the traffic scheduling problem in SFNet network,and propose congestion-aware traffic scheduling algorithms for batched and online traffic respectively.(4)Execute experiments to evaluating our proposals.This paper mainly evaluates the proposed wireless DCNs from the aspects of topology and network performance.For WSWDC,this paper evaluates two variants,laser-enhanced WSWDC and range-limited WSWDC.By comparing the two types of WSWDC with Grid,it is proved that WSWDC outperforms Grid in different network scales and different traffic patterns.For the hybrid wireless DCN SFNet,we compare it with VLCcube and also evaluate the online flow scheduling algorithm OFS.The experimental results show that SFNet outperforms VLCcube,and the flow scheduling algorithm proposed in this paper can greatly improve the performance of SFNet.