High-fault-tolerance ultra-large data center network topology structure and construction method

Abstract

The invention discloses a high-fault-tolerance ultra-large data center network topological structure and a construction method, belongs to the technical field of data center network construction, and aims to solve the technical problem of how to reasonably design the topological structure of the ultra-large data center network and improve the high fault tolerance of the data center network. Comprising the following steps: defining clusters as basic structure blocks of a data center network topology structure, and defining communication modes of servers in the clusters and among the clusters; defining a recursion mode of the HFDC; defining a server address and a switch address, and defining a connection rule between the server and the switch; and constructing a data center network topology structure. The cluster is composed of n2n three-port servers and 2n n-port switches, first ports of the servers in the same cluster are connected through internal switches, and for the servers connected with the internal switches, second ports of the servers are connected with other servers which are not connected through the same internal switch in the same cluster, and the servers are expanded through third ports of the servers.

Description

Technical Field

[0001] The present invention relates to the technical field of data center network construction, and in particular to a highly fault-tolerant super-large data center network topology structure and a construction method thereof. Background Art

[0002] The rapid growth of the digital economy has driven the demand for high-performance computing resources and given rise to a new trend in computing networks. As the core of computing networks, the performance of data centers is directly related to the efficiency of the entire network and the healthy development of the digital economy. As the scale of data center networks continues to expand, network failures have become increasingly prominent, making the reliability of data center networks particularly prominent.

[0003] Data center network topologies are mainly divided into switch-centric topologies and server-centric topologies. The server-centric topology can significantly reduce network equipment costs, improve processing efficiency, and promote network technology innovation. Therefore, server-centric data center networks are gradually becoming the focus of attention.

[0004] Data centers integrate computing, storage, and network functions. They are the core carriers of high-performance computing capabilities and the key driving force for the development of the digital economy. The performance of data centers is significantly affected by the topological characteristics of the Internet. Since data center networks directly or indirectly affect the service quality of mobile communications and computing, research on data center networks is very necessary. High-performance data center network topologies can promote the high-quality construction of data centers, thereby accelerating the pace of building an integrated computing network.

[0005] How to reasonably design the topology of ultra-large data center networks and improve the high fault tolerance of data center networks are technical problems that need to be solved. Summary of the invention

[0006] The technical task of the present invention is to address the above shortcomings and provide a highly fault-tolerant ultra-large data center network topology structure and construction method to solve the technical problem of how to reasonably design the topology structure of an ultra-large data center network and improve the high fault tolerance of the data center network.

[0007] In a first aspect, the present invention provides a method for constructing a highly fault-tolerant ultra-large data center network topology structure, wherein the data center network topology structure is a recursive structure consisting of a three-port server and two types of switches;

[0008] The construction method comprises the following steps:

[0009] S100, defining a cluster as a basic structural block of a data center network topology, and defining a communication method for servers within a cluster and between clusters;

[0010] S200, defining a recursive method of HFDC based on the communication method of servers within a cluster and between clusters;

[0011] S300, based on the recursive method of HFDC, define the server address and the switch address, and define the connection rules between the server and the switch;

[0012] S400, constructing a data center network topology structure based on the definitions of steps S100 to S300;

[0013] The data center network topology is represented as HFDC, and the cluster is represented by n2 n A three-port server and 2 n The network consists of n-port switches. HFDC(n,k) represents a k-dimensional HFDC. The cluster in HFDC(n,0) is the base cluster, and the other clusters outside the base cluster in HFDC are additional clusters. n represents the number of servers connected to the switches in each cluster, and k represents the total number of recursions when constructing the data center network topology. Both n and k are integers, n≥2, k≥0.

[0014] In HFDC, an n-port switch is defined as an internal switch and a k+1-port switch is defined as an external switch. For each cluster, each server in the cluster is connected to the internal switch in the same cluster through its first port. For other servers connected to the internal switch, the server is connected to other servers in the same cluster that are not connected to the internal switch through its second port. Each server is extended and connected to servers in different clusters through the external switch through its third port.

[0015] Preferably, for HFDC, the recursive method includes the following two cases:

[0016] When n ≥ 2 and k = 0, HFDC (n, 0) consists of a base cluster and n2 n An external switch;

[0017] When n≥2 and k>0, HFDC(n,k) is n2 n HFDC (n, k-1) and n k (2 n ) k Additional clusters are composed of HFDC(n,k), which has k+1 layers. The first layer consists of n k (2 n ) k HFDC(n,0), and each of the remaining k layers consists of n k (2 n ) k For HFDC(n,k), n2n The HFDCs (n, k-1) are numbered Num.p, p∈[0,n2 n -1], n k (2 n ) k Additional clusters are divided into n2 n modules, and number each module as Num.q, where q∈[0,n2 n -1].

[0018] Preferably, the addresses of the server, external switch and internal switch are defined as follows:

[0019] Each three-port server has an address of k+2 bits, denoted as a k+1 a k a k-1 ...a0, where a0∈[0,k], a i ∈[0,n2 n -1], i∈[1,k+1], HFDC(n,k) has k+1 layers, if the server is located in the cth layer, a0=c, where c∈[0,k];

[0020] Each external switch has an address of k+1 bits, denoted as s k s k-1 ...s0, where s0∈[0,n2 n -1],s j ∈[0,n2 n -1], j∈[1,k], address s j represents the index of HFDC(n,j-1) in HFDC(n,k), index p∈[0,n2 n -1];

[0021] Each internal switch has an address of k+1 bits, denoted as s k s k-1 ...s0, where s0∈[n2 n ,2 n (n+k+1)-1],s j ∈[0,n2 n -1], j∈[1,k], address s j represents the index of HFDC(n,j-1) in HFDC(n,k), index p∈[0,n2 n -1], if the internal switch is located in a cluster in the a0th layer, then s0∈[2 n (n+a0),2 n (n+a0+1)-1], a0∈[0,k].

[0022] Preferably, the address change of the server and the address change of the switch are defined as follows:

[0023] For HFDC(n,k), the addresses of switches and servers in the first k layers of HFDC(n,k) are obtained by adding the index number p of HFDC(n,k-1) to the left of the switch address and server address in HFDC(n,k-1). The address of switches in the first k layers of HFDC(n,k) is expressed as s k s k-1 ...s0, the server address is represented by a k+1 a k a k-1 ...a0;

[0024] When k = 1, in HFDC (n, 1), n2 n The address of the additional cluster server is obtained by changing the address of the server in the original HFDC (n, 0). When k>1, in HFDC (n, k), the address of the additional cluster server in the k+1th layer network is obtained by changing the address of the additional cluster server in the kth layer network of the original HFDC (n, k-1). When changing the server address, the following is performed: the value of position a0 is changed to k, the values ​​of the other positions remain unchanged, and the module index number q of the current server is added between the first and second leftmost bits of the server address, where q∈[0,n2 n -1], the address of the additional cluster server in the k+1th layer of HFDC(n,k) is represented by a k+1 a k a k-1 ...a0;

[0025] When k = 1, in HFDC (n, 1), n2 n The address of the additional intra-cluster switch is obtained by changing the address of the switch in the original HFDC (n, 0). When k>1, in HFDC (n, k), the address of the additional intra-cluster switch in the k+1th layer network is obtained by changing the address of the additional intra-cluster switch in the kth layer network of the original HFDC (n, k-1). When changing the switch address, the following is performed: modify the value of the s0 bit, keep the values ​​of the other positions unchanged, and add the module index number q of the current switch to the leftmost of the switch address, where s0∈[2 n (n+k),2 n (n+k+1)-1],q∈[0,n2 n -1], the address of the additional intra-cluster switch in the k+1th layer network in HFDC(n,k) is denoted as s k s k-1 ……s0.

[0026] Preferably, the connection rules between the server and the switch are defined as follows:

[0027] First connection rule: When integers n≥2 and k≥0, the server in HFDC(n,k) is connected to the internal switch if and only if the following conditions are met:

[0028] s0∈[2 n (n+a0),2 n (n+a0+1)-1],

[0029] a k+1 a k a k-1 ……a i+1 a i-1 ...a1 = s k s k-1 ...s1,

[0030] a i ∈[mn,(m+1)n-1],

[0031] Where i = a0+1, a0∈[0,k], m = s0mod 2 n (n+a0);

[0032] Second connection rule: When the integers n≥2 and k≥0, the two servers a in HFDC(n,k) are only connected if the following conditions are met k+1 a k a k-1 ...a0 and a′ k+1 a′ k a′ k-1 ...a′0 is directly connected:

[0033] a j =a′ j ,

[0034] Where i = a0 + 1, m = a i modn,a0∈[0,k],j∈[0,k+1]\{i},a i ,a j ∈[0,2 n -1];

[0035] Third connection rule: When the integers n≥2 and k≥0, the server in HFDC(n,k) is connected to the external switch only when the following conditions are met:

[0036] a k+1 a k a k-1 ...a1 = s k s k-1 ……s0.

[0037] In a second aspect, the present invention provides a highly fault-tolerant ultra-large data center network topology structure, which is a server-centric data center network topology structure constructed by a construction method for a highly fault-tolerant ultra-large data center network topology structure as described in any one of the first aspects, wherein the data center network topology structure is a recursive structure consisting of a three-port server and two types of switches;

[0038] The data center network topology is represented as HFDC, and the cluster is represented by n2 n A three-port server and 2 n The network consists of n-port switches. HFDC(n,k) represents a k-dimensional HFDC. The cluster in HFDC(n,0) is the base cluster, and the other clusters outside the base cluster in HFDC are additional clusters. n represents the number of servers connected to the switches in each cluster, and k represents the total number of recursions when constructing the data center network topology. Both n and k are integers, n≥2, k≥0.

[0039] In HFDC, an n-port switch is defined as an internal switch and a k+1-port switch is defined as an external switch. For each cluster, each server in the cluster is connected to the internal switch in the same cluster through its first port. For other servers connected to the internal switch, the server is connected to other servers in the same cluster that are not connected to the internal switch through its second port. Each server is extended and connected to servers in different clusters through the external switch through its third port.

[0040] Preferably, for HFDC, the recursive method includes the following two cases:

[0041] When n ≥ 2 and k = 0, HFDC (n, 0) consists of a base cluster and n2 n An external switch;

[0042] When n≥2 and k>0, HFDC(n,k) is n2 n HFDC (n, k-1) and n k (2 n ) k Additional clusters are composed of HFDC(n,k), which has k+1 layers. The first layer consists of n k (2 n ) k HFDC(n,0), and each of the remaining k layers consists of n k (2 n ) k For HFDC(n,k), n2 nThe HFDCs (n, k-1) are numbered Num.p, p∈[0,n2 n -1], n k (2 n ) k The additional clusters are divided into n2 n modules, and number each module as Num.q, where q∈[0,n2 n -1].

[0043] Preferably, the addresses of the server, external switch and internal switch in HFDC are as follows:

[0044] Each three-port server has an address of k+2 bits, denoted as a k+1 a k a k-1 ...a0, where a0∈[0,k], a i ∈[0,n2 n -1], i∈[1,k+1], HFDC(n,k) has k+1 layers, if the server is located in the cth layer, a0=c, where c∈[0,k];

[0045] Each external switch has an address of k+1 bits, denoted as s k s k-1 ...s0, where s0∈[0,n2 n -1],s j ∈[0,n2 n -1], j∈[1,k], address s j represents the index of HFDC(n,j-1) in HFDC(n,k), index p∈[0,n2 n -1];

[0046] Each internal switch has an address of k+1 bits, denoted as s k s k-1 ...s0, where s0∈[n2 n ,2 n (n+k+1)-1],s j ∈[0,n2 n -1], j∈[1,k], address s j represents the index of HFDC(n,j-1) in HFDC(n,k), index p∈[0,n2 n -1], if the internal switch is located in a cluster in the a0th layer, then s0∈[2 n (n+a0),2 n (n+a0+1)-1], a0∈[0,k].

[0047] Preferably, the address change of the server and the address change of the switch in the HFDC follows the following rules:

[0048] For HFDC(n,k), the addresses of switches and servers in the first k layers of HFDC(n,k) are obtained by adding the index number p of HFDC(n,k-1) to the left of the switch address and server address in HFDC(n,k-1). The address of switches in the first k layers of HFDC(n,k) is expressed as s k s k-1 ...s0, the server address is represented by a k+1 a k a k-1 ...a0;

[0049] When k = 1, in HFDC (n, 1), n2 n The address of the additional cluster server is obtained by changing the address of the server in the original HFDC (n, 0). When k>1, in HFDC (n, k), the address of the additional cluster server in the k+1th layer network is obtained by changing the address of the additional cluster server in the kth layer network of the original HFDC (n, k-1). When changing the server address, the following is performed: the value of position a0 is changed to k, the values ​​of the other positions remain unchanged, and the module index number q of the current server is added between the first and second leftmost bits of the server address, where q∈[0,n2 n -1], the address of the additional cluster server in the k+1th layer of HFDC(n,k) is represented by a k+1 a k a k-1 ...a0;

[0050] When k = 1, in HFDC (n, 1), n2 n The address of the additional intra-cluster switch is obtained by changing the address of the switch in the original HFDC (n, 0). When k>1, in HFDC (n, k), the address of the additional intra-cluster switch in the k+1th layer network is obtained by changing the address of the additional intra-cluster switch in the kth layer network of the original HFDC (n, k-1). When changing the switch address, the following is performed: modify the value of the s0 bit, keep the values ​​of the other positions unchanged, and add the module index number q of the current switch to the leftmost of the switch address, where s0∈[2 n (n+k),2 n (n+k+1)-1],q∈[0,n2 n -1], the address of the additional intra-cluster switch in the k+1th layer network in HFDC(n,k) is denoted as s k s k-1 ……s0.

[0051] Preferably, the connection rules between the server and the switch in HFDC include the following:

[0052] First connection rule: When integers n≥2 and k≥0, the server in HFDC(n,k) is connected to the internal switch if and only if the following conditions are met:

[0053] s0∈[2 n (n+a0),2 n (n+a0+1)-1],

[0054] a k+1 a k a k-1 ……a i+1 a i-1 ...a1 = s k s k-1 ...s1,

[0055] a i ∈[mn,(m+1)n-1],

[0056] Where i = a0+1, a0∈[0,k], m = s0mod 2 n (n+a0);

[0057] Second connection rule: When the integers n≥2 and k≥0, the two servers a in HFDC(n,k) are only connected if the following conditions are met k+1 a k a k-1 ...a0 and a′ k+1 a′ k a′ k-1 ...a′0 is directly connected:

[0058]

[0059] a j =a′ j ,

[0060] Where i = a0 + 1, m = a i modn,a0∈[0,k],j∈[0,k+1]\{i},a i ,a j ∈[0,2 n -1];

[0061] Third connection rule: When the integers n≥2 and k≥0, the server in HFDC(n,k) is connected to the external switch only when the following conditions are met:

[0062] a k+1 a k a k-1...a1 = s k s k-1 ……s0.

[0063] The highly fault-tolerant super-large data center network topology structure and construction method of the present invention have the following advantages: a recursive data center network topology structure consisting of a three-port server and two types of switches is constructed, which retains the advantages of the traditional server-centric topology structure and overcomes the disadvantages of the traditional server-centric topology structure. The new topology structure has good performance in terms of scalability, network scale, binary bandwidth and fault tolerance. BRIEF DESCRIPTION OF THE DRAWINGS

[0064] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0065] The present invention is further described below in conjunction with the accompanying drawings.

[0066] Figure 1 This is a flowchart of a method for constructing a highly fault-tolerant ultra-large data center network topology structure according to Example 1;

[0067] Figure 2 A structural block diagram of HFDC (2,1) constructed by a method for constructing a highly fault-tolerant ultra-large data center network topology structure in Example 1;

[0068] Figure 3 A structural block diagram of HFDC (2,0) constructed by a method for constructing a highly fault-tolerant ultra-large data center network topology structure in Example 1;

[0069] Figure 4 The number of servers included in the network under different configurations in the method for constructing a highly fault-tolerant super-large data center network topology structure in Example 1;

[0070] Figure 5 In a method for constructing a highly fault-tolerant super-large data center network topology structure in Example 1, under the premise that the number of servers is the same, the network dimension k and the network diameter corresponding to each structure are found;

[0071] Figure 6 In a method for constructing a highly fault-tolerant super-large data center network topology structure in Example 1, under the premise that the number of servers is the same, find the binary bandwidth corresponding to each structure;

[0072] Figure 7The fault tolerance of the HFDC network constructed by the construction method of a highly fault-tolerant ultra-large data center network topology structure in Example 1 is shown.

[0073] Figure 2 and Figure 3 In the example, internal switch refers to the internal switch, external switch refers to the external switch, and server refers to the server. DETAILED DESCRIPTION

[0074] The present invention is further described below in conjunction with the accompanying drawings and specific embodiments so that those skilled in the art can better understand the present invention and implement it. However, the embodiments are not intended to limit the present invention. In the absence of conflict, the embodiments of the present invention and the technical features in the embodiments may be combined with each other.

[0075] The embodiments of the present invention provide a highly fault-tolerant super-large data center network topology structure and a construction method, which are used to solve the technical problem of how to reasonably design the topology structure of a super-large data center network and improve the high fault tolerance of the data center network.

[0076] Embodiment 1:

[0077] The present invention discloses a method for constructing a highly fault-tolerant super-large data center network topology structure, wherein the data center network topology structure is a recursive structure composed of a three-port server and two types of switches, and the construction method comprises the following steps:

[0078] S100, defining a cluster as a basic structural block of a data center network topology, and defining a communication method for servers within a cluster and between clusters;

[0079] S200, defining a recursive method of HFDC based on the communication method of servers within a cluster and between clusters;

[0080] S300, based on the recursive method of HFDC, define the server address and the switch address, and define the connection rules between the server and the switch;

[0081] S400: Construct a data center network topology structure based on the definitions of steps S100 to S300.

[0082] In step S100, the communication mode of the cluster and the servers within and between clusters is defined as definition 1, and definition 1 includes the following contents:

[0083] (1) The data center network topology is represented as HFDC, where the cluster is represented by n2 n A three-port server and 2 nThe network consists of n-port switches. HFDC(n,k) represents a k-dimensional HFDC. The cluster in HFDC(n,0) is the base cluster, and the other clusters outside the base cluster in HFDC are additional clusters. n represents the number of servers connected to the switches in each cluster, and k represents the total number of recursions when constructing the data center network topology. Both n and k are integers, n≥2, k≥0.

[0084] (2) In HFDC, an n-port switch is defined as an internal switch and a k+1-port switch is defined as an external switch. For each cluster, each server in the cluster is connected to the internal switch in the same cluster through its first port. For other servers connected to the internal switch, the server is connected to other servers in the same cluster that are not connected to the internal switch through its second port. Each server is extended through its third port and connected to servers in different clusters through the external switch.

[0085] In step S200, the recursive method is defined as definition 2, and the recursive method includes the following two cases:

[0086] (1) When n ≥ 2 and k = 0, HFDC(n,0) consists of a base cluster and n2 n An external switch;

[0087] (2) When n≥2 and k>0, HFDC(n,k) is n2 n HFDC (n, k-1) and n k (2 n ) k Additional clusters are composed of HFDC(n,k), which has k+1 layers. The first layer consists of n k (2 n ) k HFDC(n,0), and each of the remaining k layers consists of n k (2 n ) k For HFDC(n,k), n2 n The HFDCs (n, k-1) are numbered Num.p, p∈[0,n2 n -1], n k (2 n ) k Additional clusters are divided into n2 n modules, and number each module as Num.q, where q∈[0,n2 n -1].

[0088] In step S300, the addresses of the server and the switch and the connection rules between the server and the switch are used as definition three, and definition three involves the addresses of the server and the switch, address changes of the server and the switch and the connection rules between the server and the switch.

[0089] The addresses of the server, external switch, and internal switch are defined as follows:

[0090] (1) Each three-port server has an address of k+2 bits, represented by a k+1 a k a k-1 ...a0, where a0∈[0,k], a i ∈[0,n2 n -1], i∈[1,k+1], HFDC(n,k) has k+1 layers, if the server is located in the cth layer, a0=c, where c∈[0,k];

[0091] (2) Each external switch has an address of k+1 bits, denoted as s k s k-1 ...s0, where s0∈[0,n2 n -1],s j ∈[0,n2 n -1], j∈[1,k], address s j represents the index of HFDC(n,j-1) in HFDC(n,k), index p∈[0,n2 n -1];

[0092] (3) Each internal switch has an address of k+1 bits, denoted as s k s k-1 ...s0, where s0∈[n2 n ,2 n (n+k+1)-1],s j ∈[0,n2 n -1], j∈[1,k], address s j represents the index of HFDC(n,j-1) in HFDC(n,k), index p∈[0,n2 n -1], if the internal switch is located in a cluster in the a0th layer, then s0∈[2 n (n+a0),2 n (n+a0+1)-1], a0∈[0,k].

[0093] The address changes of the server and the switch are defined as follows:

[0094] (1) For HFDC(n,k), the addresses of switches and servers in the first k layers of HFDC(n,k) are obtained by adding the index number p of HFDC(n,k-1) to the leftmost of the switch address and server address in HFDC(n,k-1). The address of the switch in the first k layers of HFDC(n,k) is denoted as s k s k-1 ...s0, the server address is represented by a k+1 a k a k-1 ...a0;

[0095] (2) When k = 1, in HFDC (n, 1), n2 n The address of the additional cluster server is obtained by changing the address of the server in the original HFDC (n, 0). When k>1, in HFDC (n, k), the address of the additional cluster server in the k+1th layer network is obtained by changing the address of the additional cluster server in the kth layer network of the original HFDC (n, k-1). When changing the server address, the following is performed: the value of position a0 is changed to k, the values ​​of the other positions remain unchanged, and the module index number q of the current server is added between the first and second leftmost bits of the server address, where q∈[0,n2 n -1], the address of the additional cluster server in the k+1th layer of HFDC(n,k) is represented by a k+1 a k a k-1 ...a0;

[0096] (3) When k = 1, in HFDC (n, 1), n2 n The address of the additional intra-cluster switch is obtained by changing the address of the switch in the original HFDC (n, 0). When k>1, in HFDC (n, k), the address of the additional intra-cluster switch in the k+1th layer network is obtained by changing the address of the additional intra-cluster switch in the kth layer network of the original HFDC (n, k-1). When changing the switch address, the following is performed: modify the value of the s0 bit, keep the values ​​of the other positions unchanged, and add the module index number q of the current switch to the leftmost of the switch address, where s0∈[2 n (n+k),2 n (n+k+1)-1],q∈[0,n2 n -1], the address of the additional intra-cluster switch in the k+1th layer network in HFDC(n,k) is denoted as s k s k-1 ……s0.

[0097] The connection rules between the server and the switch are defined as follows:

[0098] (1) First connection rule: When the integers n≥2 and k≥0, the server in HFDC(n,k) is connected to the internal switch if and only if the following conditions are met:

[0099] s0∈[2 n (n+a0),2 n (n+a0+1)-1],

[0100] a k+1 a k a k-1 ……a i+1 a i-1 ...a1 = s k s k-1 ...s1,

[0101] a i ∈[mn,(m+1)n-1],

[0102] Where i = a0+1, a0∈[0,k], m = s0mod 2 n (n+a0);

[0103] (2) Second connection rule: When the integers n≥2 and k≥0, the two servers a in HFDC(n,k) are only connected if the following conditions are met k+1 a k a k-1 ...a0 and a′ k+1 a′ k a′ k-1 ...a′0 is directly connected:

[0104]

[0105] a j =a′ j ,

[0106] Where i = a0 + 1, m = a i modn,a0∈[0,k],j∈[0,k+1]\{i},a i ,a j ∈[0,2 n -1];

[0107] (3) Third connection rule: When the integers n≥2 and k≥0, the server in HFDC(n,k) is connected to the external switch only when the following conditions are met:

[0108] a k+1 a k a k-1 ...a1 = s k s k-1 ……s0.

[0109] According to the method disclosed in this embodiment, Figure 2 The structure of HFDC (2, 1) shown in FIG. 1 shows that the cluster in HFDC (2, 1) consists of eight three-port servers and four two-port switches according to definition 1. Obviously, the base cluster and the additional cluster have the same structure. Whether a cluster is a base cluster or an additional cluster depends only on whether the cluster is in HFDC (2, 0). According to the definition of the recursive method of step S300, HFDC (2, 1) consists of eight HFDC (2, 0) and eight additional clusters. The eight HFDC (2, 0) are numbered from Num. 0 to Num. 7 in sequence. The eight additional clusters are divided into eight modules. Each module is numbered from Num. 0 to Num. 7 in sequence. Each module contains an additional cluster.

[0110] In HFDC (2,0), according to the definition of the server address and the switch address in step S400, the server address can be expressed as a1a0, where a0=0, a1∈[0,7]. Each internal switch has a 1-bit address, and its address can be expressed as s0, where s0∈[8,11]. In addition, each external switch has a 1-bit address, and its address can be expressed as s0, where s0∈[0,7]. The addresses of each server and switch in HFDC (2,0) are expressed as follows Figure 3 The server addresses are 00, 10, 20, 30, 40, 50, 60, and 70. The internal switch addresses are 8, 9, 10, and 11. The external switch addresses are 0, 1, 2, 3, 4, 5, 6, and 7.

[0111] In HFDC(2,1), the addresses of switches and servers in the previous network layer are obtained by adding the index number p of HFDC(2,0) to the leftmost of the switch and server addresses in the original HFDC(2,0), where p∈[0,7].

[0112] In HFDC (2,1), the addresses of the servers in the eight additional clusters are obtained by changing the addresses of the servers in the original HFDC (2,0). In this embodiment, a0 is first modified to 1, and then a module index number q of the current server is added between a1 and a0, where q∈[0,7]. In HFDC (2,1), the addresses of the switches in the eight additional clusters are obtained by changing the addresses of the switches in the original HFDC (2,0). First, s0 is modified, where s0∈[12,15]. Then, a module index number q of the current switch is added to the leftmost of the switch address, where q∈[0,7]. Figure 2In the figure, the newly added bits in the server address and the switch address are indicated by red fonts, the s0 bit of the switch address is indicated by green fonts, and the a0 bit in the server address is indicated by black fonts.

[0113] In HFDC(2,0) numbered as HFDC(2,1), according to the first connection rule, it can be known that servers 000 and 010 are connected to internal switch 08, servers 020 and 030 are connected to internal switch 09, servers 040 and 050 are connected to internal switch 010, and servers 060 and 070 are connected to internal switch 011. According to the second connection rule, it can be known that servers 000 and 020 are directly connected, servers 040 and 060 are directly connected, servers 010 and 050 are directly connected, and servers 030 and 070 are directly connected. According to the third connection rule, it can be known that servers 000, 010, 020, 030, 040, 050, 060 and 070 are respectively connected to external switches 00, 01, 02, 03, 04, 05, 06 and 07 in sequence and servers 001, 011, 021, 031, 041, 051, 061 and 071 are also respectively connected to external switches 00, 01, 02, 03, 04, 05, 06 and 07 in sequence.

[0114] In order to evaluate the performance of the HFDC network, this embodiment conducts a comprehensive simulation experiment and compares it with other topologies.

[0115] First, this embodiment studies the scalability of HFDC. The scalability of the data center network is the key to ensuring system performance. The expansion of the data center network is usually achieved by adding new components rather than replacing old components. This embodiment mainly considers three factors that affect scalability: the port number limit of the server and switch, and whether the existing structure needs to be changed when the network is expanded. As shown in Table 1, not all data center networks are suitable for expansion to higher dimensions. The network expansion of BCube is limited by the number of server ports. As the number of servers increases, more NIC ports are required and may cause changes in the network structure. In contrast, the scalability of Ficonn is not limited by server or switch ports, and there is no need to change the network structure. The expansion of the full HSDC, BCCC, HS-DCell and HFDC is limited by the number of switch ports. When port resources are exhausted, it is difficult for these networks to continue to add servers. The solution is to use switches with more ports to build a data center network. Since HFDC uses cheap commercial switches to expand the network to a higher dimensional network, it does not add a lot of additional costs.

[0116] Table 1. Comparison of various network scalability factors

[0117] factor BCube Ficonn HSDC BCCC HS-DCell HFDC Is it limited by the number of server ports? yes no no no no no Is it limited by the number of switch ports? no no yes yes yes yes Does the existing structure need to be changed when the network expands? yes yes no no no no

[0118] Secondly, this embodiment studies the distribution of the number of servers under different configurations. Figure 4 (a) to Figure 4 (d) shows the impact of the number of switch ports n on the number of servers when the network dimension k is 2, 3, 4, and 5, respectively, where the value range of n is {2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12}. Figure 4 (e) to Figure 4 (h) shows the impact of the network dimension k on the number of servers when the number of switch ports n is fixed to 2, 4, 6, and 8, respectively, where the value range of k is {1, 2, 3, 4, 5, 6}. Figure 4 It can be clearly seen that the number of servers in the HFDC network is always significantly larger than that in other networks, and the growth rate of its server number is also the fastest.

[0119] Then, this embodiment respectively performs BCube(2,k), complete HSDC k The network diameters in (k), BCCC(2,k), FiConn(6,k), HS-DCell(2,k) and HFDC(2,k) are simulated. The parameter k is adjusted so that each network contains the same number of servers. When the k value of HFDC takes {1,2,3,4,5,6} in sequence, the k values ​​of other networks are adjusted to make the number of servers in these networks equal to the number of servers in HFDC, and the corresponding diameters are obtained. Figure 5 In the figure, the x-axis represents the number of servers, the y-axis represents the network dimension of each network, and the z-axis represents the network diameter of each network. Figure 5 It can be seen that the diameter of FiConn grows the fastest, while the diameters of the other five structures grow slowly, and the k value of HFDC is relatively small. When the number of servers is the same, the diameter of BCube is the smallest, and the diameter of HFDC is smaller.

[0120] Next, we compared the binary bandwidth of each network under the same server. The comparison results are as follows: Figure 6 As shown. Figure 6 In Figure 2, we can see that BCube has the best bisection bandwidth, FiConn has the worst bisection bandwidth, and HFDC has a higher bisection bandwidth. HFDC is inherently fault-tolerant, and it provides the possibility of designing multipath routing on it.

[0121] Finally, the fault tolerance of the HFDC network is studied. The number of disjoint paths in the network is a key indicator to measure its fault tolerance. Sufficient backup paths can maintain communication in the event of a failure and reduce service interruptions. At the same time, the number of servers in the network is also related to fault tolerance. More servers mean more redundancy and load balancing mechanisms can be provided. Therefore, the more disjoint paths there are and the larger the network size, the higher the fault tolerance of the network. By comparing network structures such as BCube and Ficonn, it is known that the HFDC network not only has a larger number of disjoint paths under the same configuration, but also can accommodate more servers. Therefore, HFDC has higher fault tolerance.

[0122] In order to better evaluate the fault tolerance of HFDC, this embodiment conducts a simulation experiment. The experiment assumes that the data packet transmission is not affected by congestion to ensure that the data arrives reliably. By randomly selecting the source server, the target server, and the faulty server, this embodiment calculates the average number of disjoint paths in different network scales. HFDC(n,k) is simulated, where n∈{4,5}, k∈{1,2,3,4,5,6} and the number of server failures is {100,300,500,700,900}. The simulation results show that HFDC still maintains a high fault tolerance even in the case of up to 900 server failures. Figure 7 As shown in Figure 2, the x-axis represents the number of failed servers, and the y-axis represents the average number of disjoint paths, i.e., the fault tolerance of the network. Figure 7 It can be seen that HFDC has high fault tolerance, which confirms its advantage in building a super-large data center network. Therefore, HFDC is more suitable for building a highly fault-tolerant super-large data center network.

[0123] Through comprehensive analysis and comparison of HFDC and other common network topologies, it is found that HFDC has good performance in key indicators such as scalability, network scale, network diameter, binary bandwidth, fault tolerance, etc. The combination of these ideal characteristics makes HFDC a promising choice for network communications in ultra-large data centers.

[0124] Embodiment 2:

[0125] The present invention discloses a highly fault-tolerant ultra-large data center network topology structure, which is a server-centric data center network topology structure constructed by the method disclosed in Example 1. The data center network topology structure is a recursive structure composed of a three-port server and two types of switches.

[0126] The data center network topology is represented as HFDC, which satisfies the following definition 1:

[0127] (1) The data center network topology is represented as HFDC, where the cluster is represented by n2 nA three-port server and 2 n The network consists of n-port switches. HFDC(n,k) represents a k-dimensional HFDC. The cluster in HFDC(n,0) is the base cluster, and the other clusters outside the base cluster in HFDC are additional clusters. n represents the number of servers connected to the switches in each cluster, and k represents the total number of recursions when constructing the data center network topology. Both n and k are integers, n≥2, k≥0.

[0128] (2) In HFDC, an n-port switch is defined as an internal switch and a k+1-port switch is defined as an external switch. For each cluster, each server in the cluster is connected to the internal switch in the same cluster through its first port. For other servers connected to the internal switch, the server is connected to other servers in the same cluster that are not connected to the internal switch through its second port. Each server is extended through its third port and connected to servers in different clusters through the external switch.

[0129] Definition 2: The recursive method of HFDC includes the following two cases:

[0130] (1) When n ≥ 2 and k = 0, HFDC(n,0) consists of a base cluster and n2 n An external switch;

[0131] (2) When n≥2 and k>0, HFDC(n,k) is n2 n HFDC (n, k-1) and n k (2 n ) k Additional clusters are composed of HFDC(n,k), which has k+1 layers. The first layer consists of n k (2 n ) k HFDC(n,0), and each of the remaining k layers consists of n k (2 n ) k For HFDC(n,k), n2 n The HFDCs (n, k-1) are numbered Num.p, p∈[0,n2 n -1], n k (2 n ) k Additional clusters are divided into n2 n modules, and number each module as Num.q, where q∈[0,n2 n -1].

[0132] Definition 3: The addresses of the server, external switch and internal switch in HFDC are as follows:

[0133] (1) Each three-port server has an address of k+2 bits, represented by a k+1 a k a k-1 ...a0, where a0∈[0,k], a i ∈[0,n2 n -1], i∈[1,k+1], HFDC(n,k) has k+1 layers, if the server is located in the cth layer, a0=c, where c∈[0,k];

[0134] (2) Each external switch has an address of k+1 bits, denoted as s k s k-1 ...s0, where s0∈[0,n2 n -1],s j ∈[0,n2 n -1], j∈[1,k], address s j represents the index of HFDC(n,j-1) in HFDC(n,k), index p∈[0,n2 n -1];

[0135] (3) Each internal switch has an address of k+1 bits, denoted as s k s k-1 ...s0, where s0∈[n2 n ,2 n (n+k+1)-1],s j ∈[0,n2 n -1], j∈[1,k], address s j represents the index of HFDC(n,j-1) in HFDC(n,k), index p∈[0,n2 n -1], if the internal switch is located in a cluster in the a0th layer, then s0∈[2 n (n+a0),2 n (n+a0+1)-1], a0∈[0,k].

[0136] The address changes of the server and the switch in HFDC follow the following rules:

[0137] (1) For HFDC(n,k), the addresses of switches and servers in the first k layers of HFDC(n,k) are obtained by adding the index number p of HFDC(n,k-1) to the leftmost of the switch address and server address in HFDC(n,k-1). The address of the switch in the first k layers of HFDC(n,k) is denoted as s k s k-1 ...s0, the server address is represented by a k+1 a ka k-1 ...a0;

[0138] (2) When k = 1, in HFDC (n, 1), n2 n The address of the additional cluster server is obtained by changing the address of the server in the original HFDC (n, 0). When k>1, in HFDC (n, k), the address of the additional cluster server in the k+1th layer network is obtained by changing the address of the additional cluster server in the kth layer network of the original HFDC (n, k-1). When changing the server address, the following is performed: the value of position a0 is changed to k, the values ​​of the other positions remain unchanged, and the module index number q of the current server is added between the first and second leftmost bits of the server address, where q∈[0,n2 n -1], the address of the additional cluster server in the k+1th layer of HFDC(n,k) is represented by a k+1 a k a k-1 ...a0;

[0139] (3) When k = 1, in HFDC (n, 1), n2 n The address of the additional intra-cluster switch is obtained by changing the address of the switch in the original HFDC (n, 0). When k>1, in HFDC (n, k), the address of the additional intra-cluster switch in the k+1th layer network is obtained by changing the address of the additional intra-cluster switch in the kth layer network of the original HFDC (n, k-1). When changing the switch address, the following is performed: modify the value of the s0 bit, keep the values ​​of the other positions unchanged, and add the module index number q of the current switch to the leftmost of the switch address, where s0∈[2 n (n+k),2 n (n+k+1)-1],q∈[0,n2 n -1], the address of the additional intra-cluster switch in the k+1th layer network in HFDC(n,k) is denoted as s k s k-1 ……s0.

[0140] The connection rules between servers and switches in HFDC include the following:

[0141] (1) First connection rule: When the integers n≥2 and k≥0, the server in HFDC(n,k) is connected to the internal switch if and only if the following conditions are met:

[0142] s0∈[2 n (n+a0),2 n (n+a0+1)-1],

[0143] a k+1 ak a k-1 ……a i+1 a i-1 ...a1 = s k s k-1 ...s1,

[0144] a i ∈[mn,(m+1)n-1],

[0145] Where i = a0+1, a0∈[0,k], m = s0mod 2 n (n+a0);

[0146] (2) Second connection rule: When the integers n≥2 and k≥0, the two servers a in HFDC(n,k) are only connected if the following conditions are met k+1 a k a k-1 ...a0 and a′ k+1 a′ k a′ k-1 ...a′0 is directly connected:

[0147]

[0148] a j =a′ j ,

[0149] Where i = a0 + 1, m = a i modn,a0∈[0,k],j∈[0,k+1]\{i},a i ,a j ∈[0,2 n -1];

[0150] (3) Third connection rule: When the integers n≥2 and k≥0, the server in HFDC(n,k) is connected to the external switch only when the following conditions are met:

[0151] a k+1 a k a k-1 ...a1 = s k s k-1 ……s0.

[0152] The present invention has been shown and described in detail above through the accompanying drawings and preferred embodiments. However, the present invention is not limited to these disclosed embodiments. Based on the above multiple embodiments, those skilled in the art can know that the means in the above different embodiments can be combined to obtain more embodiments of the present invention, and these embodiments are also within the protection scope of the present invention.

Claims

1. A method for constructing a highly fault-tolerant ultra-large data center network topology structure, characterized in that: The data center network topology is a recursive structure consisting of a three-port server and two types of switches; The construction method comprises the following steps: S100, defining a cluster as a basic structural block of a data center network topology, and defining a communication method for servers within a cluster and between clusters; S200, defining a recursive method of HFDC based on the communication method of servers within a cluster and between clusters; S300, defining a server address and a switch address based on a recursive method of HFDC, and defining a connection rule between the server and the switch; S400, constructing a data center network topology structure based on the definitions of steps S100 to S300; The data center network topology is represented as HFDC, and the cluster is represented by n2 n A three-port server and 2 n The network consists of n-port switches. HFDC(n,k) represents a k-dimensional HFDC. The cluster in HFDC(n,0) is the base cluster, and the other clusters outside the base cluster in HFDC are additional clusters. n represents the number of servers connected to the switches in each cluster, and k represents the total number of recursions when constructing the data center network topology. Both n and k are integers, n≥2, k≥0. In HFDC, an n-port switch is defined as an internal switch and a k+1-port switch is defined as an external switch. For each cluster, each server in the cluster is connected to the internal switch in the same cluster through its first port. For other servers connected to the internal switch, the server is connected to other servers in the same cluster that are not connected to the internal switch through its second port. Each server is extended and connected to servers in different clusters through the external switch through its third port.

2. The method for constructing a highly fault-tolerant ultra-large data center network topology structure according to claim 1, characterized in that: For HFDC, the recursive method includes the following two cases: When n ≥ 2 and k = 0, HFDC (n, 0) consists of a base cluster and n2 n An external switch; When n≥2 and k>0, HFDC(n,k) is n2 n HFDC (n, k-1) and n k (2 n ) k Additional clusters are composed of HFDC(n,k), which has k+1 layers. The first layer consists of n k (2 n ) k HFDC(n,0), and each of the remaining k layers consists of n k (2 n ) k For HFDC(n,k), n2 n The HFDCs (n, k-1) are numbered Num.p, p∈[0,n2 n -1], n k (2 n ) k Additional clusters are divided into n2 n modules, and number each module as Num.q, where q∈[0,n2 n -1].

3. The method for constructing a highly fault-tolerant ultra-large data center network topology structure according to claim 1, characterized in that: The addresses of the server, external switch, and internal switch are defined as follows: Each three-port server has an address of k+2 bits, denoted as a k+1 a k a k-1 ...a0, where a0∈[0,k], a i ∈[0,n2 n -1], i∈[1,k+1], HFDC(n,k) has k+1 layers, if the server is located in the cth layer, a0=c, where c∈[0,k]; Each external switch has an address of k+1 bits, denoted as s k s k-1 ...s0, where s0∈[0,n2 n -1],s j ∈[0,n2 n -1], j∈[1,k], address s j represents the index of HFDC(n,j-1) in HFDC(n,k), index p∈[0,n2 n -1]; Each internal switch has an address of k+1 bits, denoted as s k s k-1 ...s0, where s0∈[n2 n ,2 n (n+k+1)-1],s j ∈[0,n2 n -1], j∈[1,k], address s j represents the index of HFDC(n,j-1) in HFDC(n,k), index p∈[0,n2 n -1], if the internal switch is located in a cluster in the a0th layer, then s0∈[2 n (n+a0),2 n (n+a0+1)-1], a0∈[0,k].

4. The method for constructing a highly fault-tolerant ultra-large data center network topology structure according to claim 3, characterized in that: The address changes of the server and the switch are defined as follows: For HFDC(n,k), the addresses of switches and servers in the first k layers of HFDC(n,k) are obtained by adding the index number p of HFDC(n,k-1) to the left of the switch address and server address in HFDC(n,k-1). The address of switches in the first k layers of HFDC(n,k) is expressed as s k s k-1 ...s0, the server address is represented by a k+1 a k a k-1 ...a0; When k = 1, in HFDC (n, 1), n2 n The address of the additional cluster server is obtained by changing the address of the server in the original HFDC (n, 0). When k>1, in HFDC (n, k), the address of the additional cluster server in the k+1th layer network is obtained by changing the address of the additional cluster server in the kth layer network of the original HFDC (n, k-1). When changing the server address, the following is performed: the value of position a0 is changed to k, the values ​​of the other positions remain unchanged, and the module index number q of the current server is added between the first and second leftmost bits of the server address, where q∈[0,n2 n -1], the address of the additional cluster server in the k+1th layer of HFDC(n,k) is represented by a k+1 a k a k-1 ...a0; When k = 1, in HFDC (n, 1), n2 n The address of the additional intra-cluster switch is obtained by changing the address of the switch in the original HFDC (n, 0). When k>1, in HFDC (n, k), the address of the additional intra-cluster switch in the k+1th layer network is obtained by changing the address of the additional intra-cluster switch in the kth layer network of the original HFDC (n, k-1). When changing the switch address, the following is performed: modify the value of the s0 bit, keep the values ​​of the other positions unchanged, and add the module index number q of the current switch to the leftmost of the switch address, where s0∈[2 n (n+k),2 n (n+k+1)-1],q∈[0,n2 n -1], the address of the additional intra-cluster switch in the k+1th layer network in HFDC(n,k) is denoted as s k s k-1 ……s0.

5. The method for constructing a highly fault-tolerant ultra-large data center network topology structure according to claim 4, characterized in that: The connection rules between the server and the switch are as follows: First connection rule: When integers n≥2 and k≥0, the server in HFDC(n,k) is connected to the internal switch if and only if the following conditions are met: s0∈[2 n (n+a0),2 n (n+a0+1)-1], a k+1 a k a k-1 ……a i+1 a i-1 ……a1=s k s k-1 ……s1, a i ∈[mn,(m+1)n-1], Where i = a0+1, a0∈[0,k], m = s0mod 2 n (n+a0); Second connection rule: When the integers n≥2 and k≥0, the two servers a in HFDC(n,k) are only connected if the following conditions are met k+1 a k a k-1 ...a0 and a′ k+1 a′ k a′ k-1 ...a′0 is directly connected: a j =a′ j , Among them, i=a0+1, m=a i modn, a0∈[0,k], j∈[0,k+1]\{i}, a i ,a j ∈[0,2 n -1]; Third connection rule: When the integers n≥2 and k≥0, the server in HFDC(n,k) is connected to the external switch only when the following conditions are met: a k+1 a k a k-1 ……a1=s k s k-1 ……s0。 6. A highly fault-tolerant super-large data center network topology structure, characterized in that: A server-centric data center network topology structure constructed by a method for constructing a highly fault-tolerant ultra-large data center network topology structure according to any one of claims 1 to 5, wherein the data center network topology structure is a recursive structure consisting of a three-port server and two types of switches; The data center network topology is represented as HFDC, and the cluster is represented by n2 n A three-port server and 2 n The network consists of n-port switches. HFDC(n,k) represents a k-dimensional HFDC. The cluster in HFDC(n,0) is the base cluster, and the other clusters outside the base cluster in HFDC are additional clusters. n represents the number of servers connected to the switches in each cluster, and k represents the total number of recursions when constructing the data center network topology. Both n and k are integers, n≥2, k≥0. In HFDC, an n-port switch is defined as an internal switch and a k+1-port switch is defined as an external switch. For each cluster, each server in the cluster is connected to the internal switch in the same cluster through its first port. For other servers connected to the internal switch, the server is connected to other servers in the same cluster that are not connected to the internal switch through its second port. Each server is extended and connected to servers in different clusters through the external switch through its third port.

7. The highly fault-tolerant ultra-large data center network topology structure according to claim 6, characterized in that: For HFDC, the recursive method includes the following two cases: When n ≥ 2 and k = 0, HFDC (n, 0) consists of a base cluster and n2 n An external switch; When n≥2 and k>0, HFDC(n,k) is n2 n HFDC (n, k-1) and n k (2 n ) k Additional clusters are composed of HFDC(n,k), which has k+1 layers. The first layer consists of n k (2 n ) k HFDC(n,0), and each of the remaining k layers consists of n k (2 n ) k For HFDC(n,k), n2 n The HFDCs (n, k-1) are numbered Num.p, p∈[0,n2 n -1], n k (2 n ) k Additional clusters are divided into n2 n modules, and number each module as Num.q, where q∈[0,n2 n -1].

8. The highly fault-tolerant ultra-large data center network topology structure according to claim 6, characterized in that: The addresses of the server, external switch, and internal switch in HFDC are as follows: Each three-port server has an address of k+2 bits, denoted as a k+1 a k a k-1 ...a0, where a0∈[0,k], a i ∈[0,n2 n -1], i∈[1,k+1], HFDC(n,k) has k+1 layers, if the server is located in the cth layer, a0=c, where c∈[0,k]; Each external switch has an address of k+1 bits, denoted as s k s k-1 ...s0, where s0∈[0,n2 n -1],s j ∈[0,n2 n -1], j∈[1,k], address s j represents the index of HFDC(n,j-1) in HFDC(n,k), index p∈[0,n2 n -1]; Each internal switch has an address of k+1 bits, denoted as s k s k-1 ...s0, where s0∈[n2 n ,2 n (n+k+1)-1],s j ∈[0,n2 n -1], j∈[1,k], address s j represents the index of HFDC(n,j-1) in HFDC(n,k), index p∈[0,n2 n -1], if the internal switch is located in a cluster in the a0th layer, then s0∈[2 n (n+a0),2 n (n+a0+1)-1], a0∈[0,k].

9. The highly fault-tolerant ultra-large data center network topology structure according to claim 7, characterized in that: The address changes of the server and the switch in HFDC follow the following rules: For HFDC(n,k), the addresses of switches and servers in the first k layers of HFDC(n,k) are obtained by adding the index number p of HFDC(n,k-1) to the left of the switch address and server address in HFDC(n,k-1). The address of switches in the first k layers of HFDC(n,k) is expressed as s k s k-1 ...s0, the server address is represented by a k+1 a k a k-1 ...a0; When k = 1, in HFDC (n, 1), n2 n The address of the additional cluster server is obtained by changing the address of the server in the original HFDC (n, 0). When k>1, in HFDC (n, k), the address of the additional cluster server in the k+1th layer network is obtained by changing the address of the additional cluster server in the kth layer network of the original HFDC (n, k-1). When changing the server address, the following is performed: the value of position a0 is changed to k, the values ​​of the other positions remain unchanged, and the module index number q of the current server is added between the first and second leftmost bits of the server address, where q∈[0,n2 n -1], the address of the additional cluster server in the k+1th layer of HFDC(n,k) is represented by a k+1 a k a k-1 ...a0; When k = 1, in HFDC (n, 1), n2 n The address of the additional intra-cluster switch is obtained by changing the address of the switch in the original HFDC (n, 0). When k>1, in HFDC (n, k), the address of the additional intra-cluster switch in the k+1th layer network is obtained by changing the address of the additional intra-cluster switch in the kth layer network of the original HFDC (n, k-1). When changing the switch address, the following is performed: modify the value of the s0 bit, keep the values ​​of the other positions unchanged, and add the module index number q of the current switch to the leftmost of the switch address, where s0∈[2 n (n+k),2 n (n+k+1)-1],q∈[0,n2 n -1], the address of the additional intra-cluster switch in the k+1th layer network in HFDC(n,k) is denoted as s k s k-1 ……s0.

10. The highly fault-tolerant ultra-large data center network topology structure according to claim 9, characterized in that: The connection rules between servers and switches in HFDC include the following: First connection rule: When integers n≥2 and k≥0, the server in HFDC(n,k) is connected to the internal switch if and only if the following conditions are met: s0∈[2 n (n+a0),2 n (n+a0+1)-1], a k+1 a k a k-1 ……a i+1 a i-1 ……a1=s k s k-1 ……s1, a i ∈[mn,(m+1)n-1], Where i = a0+1, a0∈[0,k], m = s0mod 2 n (n+a0); Second connection rule: When the integers n≥2 and k≥0, the two servers a in HFDC(n,k) are only connected if the following conditions are met k+1 a k a k-1 ...a0 and a′ k+1 a′ k a′ k-1 ...a′0 is directly connected: a j =a′ j , Among them, i=a0+1, m=a i modn, a0∈[0,k], j∈[0,k+1]\{i}, a i ,a j ∈[0,2 n -1]; Third connection rule: When the integers n≥2 and k≥0, the server in HFDC(n,k) is connected to the external switch only when the following conditions are met: a k+1 a k a k-1 ……a1=s k s k-1 ……s0。

Images