Server-centric Data Center Network Research Articles

An Improved Fault Diagnosis Algorithm for Highly Scalable Data Center Networks

Fault detection and localization are vital for ensuring the stability of data center networks (DCNs). Specifically, adaptive fault diagnosis is deemed a fundamental technology in achieving the fault tolerance of systems. The highly scalable data center network (HSDC) is a promising structure of server-centric DCNs, as it exhibits the capacity for incremental scalability, coupled with the assurance of low cost and energy consumption, low diameter, and high bisection width. In this paper, we first determine that both the connectivity and diagnosability of the m-dimensional complete HSDC, denoted by HSDCm(m), are m. Further, we propose an efficient adaptive fault diagnosis algorithm to diagnose an HSDCm(m) within three test rounds, and at most N+4m(m−2) tests with m≥3 (resp. at most nine tests with m=2), where N=m·2m is the total number of nodes in HSDCm(m). Our experimental outcomes demonstrate that this diagnosis scheme of HSDC can achieve complete diagnosis and significantly reduce the number of required tests.

The disjoint path cover in the data center network HSDC with prescribed vertices in each path

The n-dimensional HSDC with the logic graph Hn is one of the most attractive server-centric data center networks for high incremental scalability. The routing design in network topology is very important for information transmission. In particular, the application of disjoint path covers can solve various problems such as program code optimization and mapping parallel programs to parallel structures. Hn consists of 2n cliques and each two vertices in a clique are adjacent. For any vertex sets {x,u} and {y,v} in W and B, respectively, and any two disjoint vertex sets A1,A2 in some clique, where (W,B) is a bipartition of Hn, in this paper, we prove that, when |Ai|≠1 and Ai∩{x,y,u,v}=∅ for i=1,2, if |A1∪A2|≤n−3, then there exists an (x,y)-path P1 and a (u,v)-path P2 in Hn passing through A1 and A2, respectively, such that P1 and P2 have no common vertices and P1∪P2 contains all vertices of Hn. Then P1∪P2 is called the 2-disjoint path cover in Hn and we propose the efficient algorithms for the path design. Furthermore, we prove that our main result is optimal.

Constructing Multiple CISTs on BCube-Based Data Center Networks in the Occurrence of Switch Failures

The scale of data center networks (DCNs) has grown rapidly with the increasing popularity of cloud computing, data explosion, and the dramatic drop in setup costs. Thus, inevitable component failures (including switches and servers) will become more frequent. A DCN requires maintaining regular and reliable operation and providing efficient routing algorithms for transmitting data between servers. Particularly, fault-tolerant routing is necessary. Recently, constructing completely independent spanning trees (CISTs) on DCNs has received much attention as a dual-CIST (i.e., two CISTs) suffices to configure protection routing, which is a fault-tolerant routing. Moreover, the protection routing can additionally realize a secure mechanism if it is configured by more CISTs. BCube is a server-centric DCN with many advantages, and many variations were deformed from BCube with application requirements, such as RCube and RRect. In this paper, we provide a unified framework called BCube-based DCN (BDCN) that integrates the representation of the logic graphs of DCNs mentioned above, facilitating consistent algorithms' design. Then, we develop efficient algorithms to construct multiple CISTs on BDCN under the consideration of switch failures. Note that this is the first study that constructs multiple CISTs in DCNs with switch failures. Finally, using standard metrics, such as average path length (APL) and transmission failure rate (TFR), we evaluate the performance of the fault-tolerant routing through experiments.

A High-Performantal and Server-Centric Based Data Center Network

With the increasing demand of data center interconnection in the network cloud era, data center networks (DCNs) play key roles in computing and communication, becoming the center of a great deal of resources and business. Therefore, how to design a DCN with good performance has been a significant issue in networks. The hypercube is a popular interconnection network topology with low diameter, symmetry and recursive structure, and scalability. As a variant of the hypercube, the crossed cube not only retains the excellent properties of the hypercube, but also performs better in terms of Hamiltonian-connectivity, embeddability, diameter, etc. In this paper, we first propose a novel and server-centric DCN, called CSDC, which is based on the crossed cube network. The performance of CSDC can be characterized by the properties of its logical structure, so we then give its logical structure <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$C_{n}$</tex-math></inline-formula> , and determine the connectivity and edge-connectivity of <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$C_{n}$</tex-math></inline-formula> . Furthermore, we explore the fault-tolerant paths and disjoint paths between any two distinct nodes in <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$C_{n}$</tex-math></inline-formula> , and obtain the diameter of <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$C_{n}$</tex-math></inline-formula> . Moreover, we propose a method to construct two completely independent spanning trees (CISTs) in <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$C_{n}$</tex-math></inline-formula> . Extensive evaluations demonstrate that CSDC is a magnetic DCN for building large-scale data centers.

Legacy Data Center Network (DCN) Reviewed: considering Big Data Stream Mobile Computing (BDSMC) applications

Recently, Enterprises that operate over vast geographical areas uses multiple data centers to collect, store and process data in real time via energy-efficient acquisition, wirelessly transport clients or users data to the cloud. This paper seeks to technically review existing Data Center Networks (DCN) considering BDSMC applications. In this research, related works on Distributed Data Center Networks will be presented. Within the stream computing ecosystems, there are various network models but the pool of possible DCN topologies/architectures to adopt appears little and unfit for the purpose of BDSMC optimization. However, most related Data center architecture will be reviewed. Investigate efforts on both server centric and switch-centric models adaptable to BDSMC network layer. The extent of work done on distributed spine-leaf re-designed server-centric network construction so as to automatically harvest network interconnection into a ‘stellar’ dual-port server-centric SG network; how classical graph-based interconnection network translate network performance similar to generic works for BDSMC ecosystems. Review stellar transformation using the well-studied generalized hypercube family of interconnection networks for BDSMC ecosystems. The literature was searched from the databases: IEEE Xplore Digital Library, Springer Link Digital Library, and Google Scholar, IET Digital Library, Frontiers Library, ACM Digital Library repositories resulting in 98 papers after several eliminations ranging from year 2000-2022. In conclusion, state-of-the-art dual-port server-centric DCNs (FiConn, DCell, DPillar), etc, while looking at possible architectures with excellent comparative performance for BDSMC ecosystems. Research gaps are revealed for further study. Keywords: Data, Center, Network, Data Stream, Mobile Computing and applications

Completely Independent Spanning Trees on BCCC Data Center Networks With an Application to Fault-Tolerant Routing

A set of <inline-formula><tex-math notation="LaTeX">$k$</tex-math></inline-formula> spanning trees in a graph <inline-formula><tex-math notation="LaTeX">$G$</tex-math></inline-formula> are called completely independent spanning trees (CISTs for short) if the paths joining every pair of vertices <inline-formula><tex-math notation="LaTeX">$x$</tex-math></inline-formula> and <inline-formula><tex-math notation="LaTeX">$y$</tex-math></inline-formula> in any two trees have neither vertex nor edge in common, except for <inline-formula><tex-math notation="LaTeX">$x$</tex-math></inline-formula> and <inline-formula><tex-math notation="LaTeX">$y$</tex-math></inline-formula> . The existence of multiple CISTs in the underlying graph of a network has applications in fault-tolerant broadcasting and secure message distribution. In this paper, we investigate the construction of CISTs in a server-centric data center network called BCube connected crossbars (BCCC), which can provide good network performance using inexpensive commodity off-the-shelf switches and commodity servers with only two network interface card (NIC) ports. The significant advantages of BCCC are its good expandability, lower communication latency, and higher robustness in component failure. Based on the structure of compound graphs of BCCC, we provide efficient algorithms to construct <inline-formula><tex-math notation="LaTeX">$\lceil \frac{n}{4}\rceil$</tex-math></inline-formula> CISTs in the logical graph of BCCC, denoted by <inline-formula><tex-math notation="LaTeX">$L$</tex-math></inline-formula> - <inline-formula><tex-math notation="LaTeX">$BCCC(n,k)$</tex-math></inline-formula> , for <inline-formula><tex-math notation="LaTeX">$n\geqslant 5$</tex-math></inline-formula> . As a by-product, we obtain a fault-tolerant routing that takes the constructed CISTs as its routing table. We then evaluate the performance of the fault-tolerant routing through simulation results.

Transmission Failure Analysis of Multi-Protection Routing in Data Center Networks With Heterogeneous Edge-Core Servers

The recently proposed RCube network is a cube-based server-centric data center network (DCN), including two types of heterogeneous servers, called core and edge servers. Remarkably, it takes the latter as backup servers to deal with server failures and thus achieve high availability. This paper first points out that RCube is suitable as a candidate topology of DCNs for edge computing. Three transmission types are among core and edge servers based on the demand for applications' computation and instant response. We then employ protection routing to analyze the transmission failure of RCube DCNs. Unlike traditional protection routing, which only tolerates a single link or node failure, we use the multi-protection routing scheme to improve fault-tolerance capability. To configure a protection routing in a network, according to Tapolcai's suggestion, we need to construct two completely independent spanning trees (CISTs). A logic graph of RCube, denoted by $L$-$RCube(n,m,k)$, is a network with a recursive structure. Each basic building element consists of $n$ core servers and $m$ edge servers, where the order $k$ is the number of recursions applied in the structure. In this paper, we provide algorithms to construct $\min\{n,\lfloor(n+m)/2\rfloor\}$ CISTs in $L$-$RCube(n,m,k)$ for $n+m\geqslant 4$ and $n>1$. From a combination of the multiple CISTs, we can configure the desired multi-protection routing. In our simulation, we configure up to 10 protection routings for RCube DCNs. As far as we know, in past research, there were at most three protection routings developed in other network structures. Finally, we summarize some crucial analysis viewpoints about the transmission efficiency of DCNs with heterogeneous edge-core servers from the simulation results.

DACON: a reconfigurable application-centric optical network for disaggregated data center infrastructures [Invited

To solve the issues of low resource utilization and performance bottleneck in current server-centric data center networks (DCNs), we propose and experimentally demonstrate a disaggregated application-centric optical network (DACON) for data center infrastructures based on hybrid optical switches. DACON achieves flexible provision and dynamic reconfiguration of hardware nodes exploiting the softwared-define networking (SDN) orchestration plane. A four-node SDN-enabled disaggregated prototype is implemented with a field-programmable-gate-array-based controller of hardware nodes and nanosecond optical switches, performing a minimal node-to-node network latency of 378.6 ns and zero packet loss. Based on the unmodified Linux kernel and two different applications (distributed computing and a Memcached database), the application runtime of the disaggregated prototype is investigated and compared with the server-centric architecture. Experimental results show that the disaggregated prototype performs better with Memcached database applications, achieving a1.46×faster runtime than the server-centric network at a memory node access ratio of 0.9. Based on the customized control plane orchestrator and dynamic resource reallocation, the node-to-node latency is reduced by 21% when CPU nodes access memory nodes. The scalability of DACON is then numerically assessed based on experimentally measured parameters. Results show that the intra-rack node-to-node latency is less than 404.8 ns with a 6144-node network and memory node access ratio of 0.9. Finally, the cost and power consumption are also studied and compared with current DCN architectures. Results indicate DACON saves 13.4% of the cost of an interconnect network compared with current disaggregated architecture and consumes up to 31.1% less power with respect to server-centric DCN architectures.

The 3-Extra Connectivity of the Data Center Network BCube

Abstract Connectivity is a significant metric to assess the fault tolerance of a network. For a faulty vertex set $H$, the $h$-extra connectivity is defined under the assumption that every component of the network removing $H$ has at least $h+1$ fault-free vertices. Compared to the traditional connectivity, which is defined under the assumption that the network removing $H$ is disconnected or trivial, the $h$-extra connectivity can better reflect the true fault tolerance of the network. The $BCube$ is an important server-centric data center network; it has good fault tolerance and scalability. In this paper, our research focuses on the logical structure of $BCube$, named $BC_{n,k}$, which is actually a specific type of generalized hypercubes. We prove that the 3-extra connectivity of $BC_{n,k}$ is $4(k+1)(n-1)-4n$ for $k\geq 4$ and $n\geq 4$.

Parallel construction of multiple independent spanning trees on highly scalable datacenter networks

An emerging datacenter network (DCN) with high scalability called HSDC is a server-centric DCN that can help cloud computing in supporting many inherent cloud services. For example, a server-centric DCN can initiate routing for data transmission. This paper investigates the construction of independent spanning trees (ISTs for short), a set of the rooted spanning trees associated with the disjoint-path property, in HSDC. Regarding multiple spanning trees as routing protocol, ISTs have applications in data transmission, e.g., fault-tolerant broadcasting and secure message distribution. We first establish the vertex-symmetry of HSDC. Then, by the structure that n-dimensional HSDC is a compound graph of an n-dimensional hypercube Qn and n-clique Kn, we amend the algorithm constructing ISTs for Qn to obtain the algorithm required by HSDC. Unlike most algorithms of recursively constructing tree structures, our algorithm can find every node’s parent in each spanning tree directly via an easy computation relied upon only the node address and tree index. Consequently, we can implement the algorithm for constructing n ISTs in O(nN) time, where N=n2n is the number of vertices of n-dimensional HSDC; or parallelize the algorithm in O(n) time using N processors. Remarkably, the diameter of the constructed ISTs is about twice the diameter of Qn.

The Conditional Reliability Evaluation of Data Center Network BCDC

Abstract As the number of servers in a data center network (DCN) increases, the probability of server failures is significantly increased. Traditional connectivity is an important metric to measure the reliability of DCN. However, the traditional connectivity of a DCN based on the condition of arbitrary faulty servers is generally lower. Therefore, it is important to increase the connectivity of a DCN by adding some limited conditions for the faulty server set. As a result, $g$-restricted connectivity and $h$-extra connectivity, which are two crucial subjects for a DCN’s ability to tolerate faulty servers, were proposed in the literature. In this paper, we study the $g$-restricted connectivity and $h$-extra connectivity of a new server-centric DCN, called BCDC, based on crossed cube with excellent performance. We prove that the $g$-restricted connectivity of BCDC is 4 for $n=3$ and $2n+g(n-2)-2$ for $n\geq 4$, where $0\leq g\leq n-3$, and the $h$-extra connectivity of BCDC is 4 for $n=3$ and $2n+h(n-2)-2$ for $n\geq 4$, where $0\leq h\leq n-3$.

Relating the bisection width of dual-port, server-centric datacenter networks and the solution of edge isoperimetric problems in graphs

Stellar datacenter networks are a recent generic construction designed to transform a base-graph into a dual-port, server-centric datacenter network. We prove that the S-bisection width of any stellar datacenter network can be obtained from the solution of isoperimetric problems on the base-graph, provided that the base-graph is regular. We extend previous research on the stellar datacenter networks GQ⁎, instantiated with generalized hypercubes, and show that with respect to S-bisection width, GQ⁎ performs well in comparison with the dual-port datacenter network FiConn. Our work develops a strong combinatorial link between graph bisection width and throughput metrics for stellar datacenter networks.

Energy-Aware Coflow and Antenna Scheduling for Hybrid Server-Centric Data Center Networks

Data center networks (DCNs) continuously evolve to support emerging applications. For instance, server-centric architectures are becoming popular because they improve the modularity and resilience of data centers. In a server-centric DCN, switching functionalities are embedded in each server instead of in dedicated switches. Recently, wireless technologies have been introduced into the design of server-centric DCNs. Wireless interconnects provide great flexibility by creating direct flyways between servers. These wireless links help to reduce the length of paths assigned to flows so that fewer links must be powered-on to transfer data. In this paper, we investigate energy-aware scheduling of coflows and antenna directions in a server-centric DCN. Coflows are groups of parallel and high-volume flows generated by big data applications. We assume that the DCN is hybrid, i.e., that it includes both wired and wireless links. We formulate this scheduling problem as an optimization problem named energy-aware coflow and antenna scheduling (ECAS) and solve it using a mathematical programming solver. We further prove the NP-hardness of the formulated ECAS problem. We also introduce a heuristic algorithm to reduce the computational complexity. Our simulation results show that our optimal solution saves up to 48% of energy compared with representative competitors. Our heuristic algorithm achieves lower energy savings but still generates the shortest paths and lowest transferring time of flows. Further simulations on the network topology with a different size also confirm the efficiency of our proposed model and algorithm.

Mesh-of-Torus: a new topology for server-centric data center networks

Various topologies have been proposed for high-performance computing (HPC), i.e., fat-tree, Torus topology. Compared with conventional fat-tree topology, Torus performs much better when applied in HPC. Unfortunately, due to its wraparound links, Torus topology naturally has the tendency to trigger deadlock incidents inside the network. Researchers solve this problem by means of virtual channel, but this approach will also restrict the routing of message. In this paper, we propose a deadlock-free topology for HPC, called Mesh-of-Torus, which incarnates the good characteristics of Mesh and Torus topology. Comparing with mesh, Mesh-of-Torus has shorter network diameter. Furthermore, we have proposed a corresponding port assignment rules in consideration of complicated internal arbitration or scheduling mechanism incurred by the employment of virtual channel. Deadlock avoidance can be achieved when dimension-order routing algorithm and our port assignment rules are applied to Mesh-of-Torus. Finally, simulations and mathematical analysis have shown that Mesh-of-Torus outperforms Mesh in terms of average end-to-end latency and network load distribution.

Deadline Is Not Enough: Importance-Aware Transmission Control Protocol for Server-Centric Data Centers

An important challenge for modern data centers is to provide users with low-latency, high-quality, and interactive services. To improve effective application throughput, much work has been done to take into account application deadline in the design of the network flow schedule for transmission of the queried data. This raises a critical question: does application-level throughput necessarily translate to better quality of service for users? We note that queries responses typically include a set of semantically related documents (within a flow), and that while some documents are highly related to that query, others are not. This motivates us to consider the degree of a document’s semantic relatedness in the design of an efficient data transmission scheme for data centers. To understand the performance upper bound, we derive a mixed integer linear programming (MILP) model for the above problem, which can be regarded as the schedule of a network subflow problem and is hence complicated. Thus, we propose the importance-aware transmission control protocol (ITCP), a distributed event-driven rate-based delivery control protocol, for server-centric data center networks. ITCP jointly considers the subflow importance, size, and deadline, as well as the characteristics of the underlying infrastructure to maximize the goodput of the most relevant responses of a query. The results of both real and synthetic data simulations show that ITCP outperforms recent works in terms of mean average precision of top $k$ data and the sum of application-level importance.

BCDC: A High-Performance, Server-Centric Data Center Network

The capability of the data center network largely decides the performance of cloud computing. However, the number of servers in the data center network becomes increasingly huge, because of the continuous growth of the application requirements. The performance improvement of cloud computing faces great challenges of how to connect a large number of servers in building a data center network with promising performance. Traditional tree-based data center networks have issues of bandwidth bottleneck, failure of single switch, etc. Recently proposed data center networks such as DCell, FiConn, and BCube, have larger bandwidth and better fault-tolerance with respect to traditional tree-based data center networks. Nonetheless, for DCell and FiConn, the fault-tolerant length of path between servers increases in case of failure of switches; BCube requires higher performance in switches when its scale is enlarged. Based on the above considerations, we propose a new server-centric data center network, called BCDC, based on crossed cube with excellent performance. Then, we study the connectivity of BCDC networks. Furthermore, we propose communication algorithms and fault-tolerant routing algorithm of BCDC networks. Moreover, we analyze the performance and time complexities of the proposed algorithms in BCDC networks. Our research will provide the basis for design and implementation of a new family of data center networks.

RRect: A Novel Server-Centric Data Center Network with High Power Efficiency and Availability

In this paper, we propose a novel server-centric network for data centers, called RRect. Compared to existing server-centric networks, RRect provides a more graceful degradation performance in the presence of component failure. Meanwhile, RRect enjoys a linear diameter to the network order and multiple parallel paths. We present algorithms to find paths and to construct all parallel paths between any pair of servers in RRect. We also show that without using any power-aware routing algorithm, RRect behaves more power efficient in its interconnection network. To meet today's stringent high availability requirement, RRect can be configured into redundancy and failover scheme. In particular, RRect can be configured into both symmetric and asymmetric redundancy modes to cater different applications’ needs, which cannot be implemented in existing server-centric networks. Moreover, RRect gives more flexibility to adjust the network size. Given the same switches, by fine tuning the parameters, RRect provides numerous structures with different sizes, while preserving the properties of short diameter and multiple parallel paths. Our comprehensive simulations show that RRect gives much better graceful degradation performance in the presence of component failure, and RRect saves much energy in the interconnection network. Meanwhile, RRect can maintain the same performance on many critical metrics as BCube, including short diameter and excellent aggregate throughput. All these features make RRect a very empirical structure for enterprise dater center network products.

The influence of datacenter usage on symmetry in datacenter network design

We undertake the first formal analysis of the role of symmetry, interpreted broadly, in the design of server-centric datacenter networks. Although symmetry has been mentioned by other researchers, we explicitly relate it to various specific, structural, graph-theoretic properties of datacenter networks. Our analysis of symmetry is motivated by the need to ascertain the usefulness of a datacenter network as regards the support of network virtualization and prevalent communication patterns in multitenanted clouds. We argue that a number of structural concepts relating to symmetry from general interconnection networks, such as recursive-definability, the existence and dynamic construction of spanning trees, pancyclicity, and variations in Hamiltonicity, are appropriate topological metrics to use in this regard. In relation to symmetry, we highlight the relevance of algebraic properties and algebraic constructions within datacenter network design. Built upon our analysis of symmetry, we outline the first technique to embed guest datacenter networks in a host datacenter network that is specifically oriented towards server-centric datacenter networks. In short, we provide the graph-theoretic foundations for the design of server-centric datacenter networks so as to support network virtualization and communication patterns in cloud computing.

An Optimal Single-Path Routing Algorithm in the Datacenter Network DPillar

DPillar has recently been proposed as a server-centric datacenter network and is combinatorially related to (but distinct from) the well-known wrapped butterfly network. We explain the relationship between DPillar and the wrapped butterfly network before proving that the underlying graph of DPillar is a Cayley graph; hence, the datacenter network DPillar is node-symmetric. We use this symmetry property to establish a single-path routing algorithm for DPillar that computes a shortest path and has time complexity $O(k)$, where $k$ parameterizes the dimension of DPillar (we refer to the number of ports in its switches as $n$). Our analysis also enables us to calculate the diameter of DPillar exactly. Moreover, our algorithm is trivial to implement, being essentially a conditional clause of numeric tests, and improves significantly upon a routing algorithm earlier employed for DPillar. Furthermore, we provide empirical data in order to demonstrate this improvement. In particular, we empirically show that our routing algorithm improves the average length of paths found, the aggregate bottleneck throughput, and the communication latency. A secondary, yet important, effect of our work is that it emphasises that datacenter networks are amenable to a closer combinatorial scrutiny that can significantly improve their computational efficiency and performance.

Towards the Tradeoffs in Designing Data Center Network Architectures

Existing Data Center Network (DCN) architectures are classified into two categories: switch-centric and server-centric architectures. In switch-centric DCNs, routing intelligence is placed on switches; each server usually uses only one port of the Network Interface Card (NIC) to connect to the network. In server-centric DCNs, switches are only used as cross-bars, and routing intelligence is placed on servers, where multiple NIC ports may be used. In this paper, we formally introduce a new category of DCN architectures: the <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">dual-centric</i> DCN architectures, where routing intelligence can be placed on both switches and servers. The dual-centric philosophy can achieve various tradeoffs in designing DCN architectures. We propose three novel dual-centric DCN architectures: FCell, FRectangle, and FSquare, all of which are based on the folded Clos topology. FCell is a power-efficient DCN architecture, with a larger diameter and lower bisection bandwidth than FSquare and FRectangle. FSquare is a high performance DCN architecture, in which the diameter is small and the bisection bandwidth is large; however, the DCN power consumption per server in FSquare is high. FRectangle significantly reduces the DCN power consumption per server, compared to FSquare, at the sacrifice of some networking performances. By investigating FCell, FRectangle and FSquare, and by comparing them with existing architectures, we demonstrate that, the three novel dual-centric architectures enjoy the advantages of both switch-centric designs and server-centric designs, have various nice properties for practical data centers, and provide flexible tradeoff choices in designing DCN architectures.