Low-delay low-cost data center network topology and construction method

Abstract

The application discloses a low-delay and low-cost data center network topology structure and a construction method, belongs to data center construction, and aims to solve the technical problem of how to design a more efficient network topology to meet the increasing demand for low delay and low cost. The method comprises the following steps: defining an element as a basic structural block of the data center network topology structure, defining the communication mode of the servers in the element and between the elements; defining the recursive mode of the data center network topology structure; defining the addresses of the servers and the switches and the connection principle between the servers and the switches; and constructing the data center network topology structure based on the definition. The element comprises m servers with two ports and a switch with m ports, in the same element, the servers communicate with each other through the first ports, and for the servers located in different elements, the servers communicate with each other through the second ports.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data center construction, in particular to a low-delay and low-cost data center network topology structure and a construction method. BACKGROUND

[0002] With the widespread application of technologies such as 5G, the Internet of Things, and artificial intelligence, network devices have reached a scale of billions of dollars, and these devices are constantly generating a large amount of data. The massive aggregation contains great value and brings new opportunities for intelligent development.

[0003] As the core node of information processing and storage, the low-delay performance of the network topology of a data center is the key to improving user experience and ensuring business continuity. The explosive growth of data has made the network delay challenge faced by data centers even more severe. Therefore, researching low-delay data center network topology is of great significance for optimizing resource allocation, improving data processing speed, and enhancing network stability. This not only helps to promote the progress of data center technology, but also provides solid technical support for the digital transformation of various industries.

[0004] Existing data center network topology structures are mainly divided into two types: switch-centered topology and server-centered topology. Server-centered topology has more advantages in construction cost and network update and upgrade. Therefore, researchers are increasingly focusing on the research of server-centered data center networks. However, traditional server-centered data center networks have poor scalability, which hinders the update and upgrade of data centers.

[0005] How to design a more efficient network topology to meet the growing demand for low delay and low cost is a technical problem that needs to be solved. SUMMARY

[0006] The technical task of the present application is to solve the technical problem of how to design a more efficient network topology to meet the growing demand for low delay and low cost by providing a low-delay and low-cost data center network topology structure and a construction method.

[0007] In a first aspect, the present application provides a construction method for a low-delay and low-cost data center network topology structure, wherein the data center network topology structure is a recursive structure composed of switches and two-port servers, and the construction method comprises the following steps:

[0008] S100, defining an element as a basic structural block of the data center network topology structure, and defining the communication mode of the servers within the element and between the elements;

[0009] S200, defining the recursive way of the data center network topology based on the communication mode of the elements and the communication mode of the servers between the elements;

[0010] S300, defining the address of the servers and the switches and the connection principle between the servers and the switches based on the recursive way of the data center network topology;

[0011] S400, constructing the data center network topology based on the definition of steps S100-S300;

[0012] Wherein, the data center network topology is represented as MBDC, the elements include m servers with two ports and one switch with m ports, and the symbol represents the MBDC with n levels, t represents the type of the MBDC, n and t are integers, n≥1, t∈{0,1}, t=0 represents the 0-type MBDC, and t=1 represents the 1-type MBDC;

[0013] In the MBDC, the servers in the same element communicate through their first ports, and the servers in different elements communicate through their second ports, and for each element, the first ports of the servers in the same element are connected through the switch, and for the servers connected to the switch, the second ports of the servers are connected to the servers in different elements.

[0014] As a preferred, for the MBDC, the recursive way includes the following two cases:

[0015] When n=1, composed of two basic elements;

[0016] When n≥2, composed of 1 and 1 .

[0017] As a preferred, for the MBDC, the address of the servers and the switches is defined as follows:

[0018] The address of the server is represented as [x0x1x2…x n ;a], wherein x i ∈{0,1}, i∈[0,n], a∈[1,m], x0=0 represents the 0-type MBDC, and x0=1 represents the 1-type MBDC;

[0019] The address of the switch is represented as [x0x1x2…x n ;0], wherein x i ∈{0,1}, i∈[0,n].

[0020] As a preferred embodiment, the address changes for servers and switches are defined as follows:

[0021] for In the context of servers and switches, the value of the 0th address bit of each server and each switch is set to t, and the values ​​of the 1st to nth address bits of each server are as follows: The values ​​of the 0th to (n-1)th address bits of the server, and the values ​​of the 1st to nth address bits of each switch are as follows: The values ​​of the 0th to (n-1)th address bits of the switch.

[0022] As a preferred option, the connection rules between the server and the switch include the following:

[0023] For elements located within the same element, with addresses [x0x1x2…x] n ;a] and addresses [x0′x1′x2′…x n The two servers [a'] are currently connected via a switch only if the following conditions are met:

[0024] x j =x j ′,j∈[0,n];

[0025] For elements located within different elements, with addresses [x0x1x2…x] n ;a] and addresses [x0′x1′x2′…x n Two servers, 'a' and 'a', are connected via a switch if the following conditions are met:

[0026] When x i-1 When = 0,

[0027] When x i-1 When = 1, And x0 = x0′, a = a′;

[0028] i∈[1,n].

[0029] In a second aspect, the present invention provides a low-latency, low-cost data center network topology, which is a data center network topology constructed by a method for constructing a low-latency, low-cost data center network topology as described in any one of the first aspects, wherein the data center network topology is a recursive structure composed of switches and two-port servers.

[0030] The data center network topology is represented as MBDC, and its elements include m servers with two ports and one switch with m ports, represented by symbols. Let t represent an MBDC with n levels, and t represent the type of MBDC. Both n and t are integers, n≥1, t∈{0,1}, t=0 represents a type 0 MBDC, and t=1 represents a type 1 MBDC.

[0031] In MBDC, servers within the same element communicate with each other through their first port, while servers located in different elements communicate with each other through their second port. For each element, the first port of the servers within the same element is connected via a switch, and the second port of the server connected to the switch is connected to the servers in different elements.

[0032] As a preferred approach, the recursive method for MBDC includes the following two cases:

[0033] When n=1, It consists of two basic elements;

[0034] When n≥2 From 1 and 1 composition.

[0035] As a preferred option, the address definitions for the server and switch in MBDC are as follows:

[0036] The server address is represented as [x0x1x2…x n ;a], where x i ∈{0,1}, i∈[0,n], a∈[1,m], x0=0 represents type 0 MBDC, x0=1 represents type 1 MBDC;

[0037] The switch's address is represented as [x0x1x2…x n ;0],where, x i ∈{0,1}, i∈[0,n].

[0038] As a preferred embodiment, the address changes for servers and switches are defined as follows:

[0039] for In the context of servers and switches, the value of the 0th address bit of each server and each switch is set to t, and the values ​​of the 1st to nth address bits of each server are as follows: The values ​​of the 0th to (n-1)th address bits of the server, and the values ​​of the 1st to nth address bits of each switch are as follows: The values ​​of the 0th to (n-1)th address bits of the switch.

[0040] As a preferred option, the connection rules between the server and the switch include the following:

[0041] For elements located within the same element, with addresses [x0x1x2…x] n ;a] and addresses [x0′x1′x2′…x n The two servers [a'] are currently connected via a switch only if the following conditions are met:

[0042] x j =x j ′,j∈[0,n];

[0043] For elements located within different elements, with addresses [x0x1x2…x] n ;a] and addresses [x0′x1′x2′…x n Two servers, 'a' and 'a', are connected via a switch if the following conditions are met:

[0044] When x i-1 When = 0,

[0045] When x i-1 When = 1, And x0 = x0′, a = a′

[0046] i∈[1,n].

[0047] The low-latency, low-cost data center network topology and construction method of the present invention have the following advantages: a recursive data center network topology consisting of a switch and m two-port servers connected to it is constructed. This network topology retains the advantages of the traditional server-centric topology and overcomes the disadvantages of the traditional server-centric topology. It has good performance in terms of diameter, scalability, fault tolerance and network scale. Attached Figure Description

[0048] To more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0049] The invention will be further described below with reference to the accompanying drawings.

[0050] Figure 1 This is a flowchart of a method for constructing a low-latency, low-cost data center network topology, as described in Example 1.

[0051] Figure 2 This is a construction method for a low-latency, low-cost data center network topology as described in Example 1. a structural block diagram of the low-latency and low-cost data center network topology structure of embodiment 1;

[0052] Figure 3 a low-latency and low-cost data center network topology structure of embodiment 1 a structural block diagram of the low-latency and low-cost data center network topology structure of embodiment 1;

[0053] Figure 4 a low-latency and low-cost data center network topology structure of embodiment 1 a structural block diagram of the low-latency and low-cost data center network topology structure of embodiment 1;

[0054] Figure 5 a low-latency and low-cost data center network topology structure of embodiment 1 DETAILED DESCRIPTION

[0055] The present application will be further described below in conjunction with the drawings and specific embodiments, so that those skilled in the art can better understand the present application and implement it, but the embodiments are not limiting to the present application, and the technical features in the embodiments and the embodiments can be combined with each other without conflict.

[0056] The embodiments of the present application provide a low-latency and low-cost data center network topology structure, which is used to solve the technical problem of how to design a more efficient network topology to meet the growing demand for low latency and low cost.

[0057] Embodiment 1:

[0058] The present application provides a low-latency and low-cost data center network topology structure, which is used to solve the technical problem of how to design a more efficient network topology to meet the growing demand for low latency and low cost.

[0059] Step S100, defining an element as a basic structure block of the data center network topology structure, and defining the communication mode of the servers in the element and between the elements;

[0060] Step S200, defining the recursive mode of the data center network topology structure based on the communication mode of the servers in the element and between the elements;

[0061] Step S300, defining the addresses of the servers and the switches and the connection principle between the servers and the switches based on the recursive mode of the data center network topology structure;

[0062] Step S400, constructing the data center network topology structure based on the definitions of steps S100-S300.

[0063] In step S100, the basic structure block of the data center network topology structure has the following definitions:

[0064] Definition 1: A data center network topology is represented as MBDC, consisting of m servers with two ports and one switch with m ports, denoted by symbols. Let t represent an MBDC with n levels, and t represent the type of MBDC. Both n and t are integers, n≥1, t∈{0,1}, t=0 represents a type 0 MBDC, and t=1 represents a type 1 MBDC.

[0065] In MBDC, servers within the same element communicate with each other through their first port, while servers located in different elements communicate with each other through their second port. For each element, the first port of the servers within the same element is connected via a switch, and the second port of the server connected to the switch is connected to the servers in different elements.

[0066] In step S200, the recursive method of MBDC has the following definition:

[0067] Definition 2: Recursive methods include the following two cases:

[0068] (1) When n = 1, It consists of two basic elements;

[0069] (2) When n≥2, From 1 and 1 composition.

[0070] In step S300, the addresses of the server and the switch are defined as follows:

[0071] Definition 3: The address of a server is represented as [x0x1x2…x n ;a], where x i ∈{0,1}, i∈[0,n], a∈[1,m], x0=0 represents type 0 MBDC, x0=1 represents type 1 MBDC;

[0072] The switch's address is represented as [x0x1x2…x n ;0],where, x i ∈{0,1}, i∈[0,n].

[0073] Correspondingly, the address changes for servers and switches are defined as follows: For In the context of servers and switches, the value of the 0th address bit of each server and each switch is set to t, and the values ​​of the 1st to nth address bits of each server are as follows: The values ​​of the 0th to (n-1)th address bits of the server, and the values ​​of the 1st to nth address bits of each switch are as follows: The values ​​of the 0th to (n-1)th address bits of the switch.

[0074] Correspondingly, the connection rules between the server and the switch include the following:

[0075] Linking rule 1: For elements located within the same element with addresses [x0x1x2…x], ... n ;a] and addresses [x0′x1′x2′…x n The two servers [a'] are currently connected via a switch only if the following conditions are met:

[0076] x j =x j ′,j∈[0,n];

[0077] Linking rule 2: For elements located in different elements with addresses [x0x1x2…x] n ;a] and addresses [x0′x1′x2′…x n Two servers, 'a' and 'a', are connected via a switch if the following conditions are met:

[0078] When x i-1 When = 0,

[0079] When x i-1 When = 1, And x0 = x0′, a = a′;

[0080] i∈[1,n].

[0081] Constructing according to the method disclosed in this embodiment, as follows Figure 4 shown The structure, as can be known from definition one, The elements consist of six two-port servers and one six-port switch. In this context, for each element, the first port of the server within the same element is connected via a switch. For servers connected to the switch, the second port of the server connects to servers in different elements. In this context, according to Definition 3, the server address can be represented as [0x1x2x3; a], where x i ∈{0,1}, i∈[1,3], a∈[1,6], the address of the switch is represented as [0x1x2x3; 0], where x i ∈{0,1}, i∈[1,3], The address representation of each server and switch in the system is as follows: Figure 2 As shown. In In the definition of 3, the server address can be represented as [1x1x2x3; a], where x i∈{0,1}, i∈[1,3], a∈[1,6]; the address of the switch is represented as [1x1x2x3; 0], wherein x i ∈{0,1}, i∈[1,3], The addresses of the servers and switches in the network are represented as shown in Figure 3 When the network is constructed, The 0th bit of each switch and server address of the network is 1, and the 1st to nth bits of each switch and server address of the network are and The 0th to n-1th bits of the switch and server addresses in the network.

[0082] In the network of Figure 4 According to the connection rule (1) between the switches and the servers, the servers in the same element, for example, the servers [1 / 0000; 1] and [1 / 0000; 1] in the element

[0000] are connected through a switch, according to the connection rule (2) between the switches and the servers, the servers [1 / 0000; 1] and [1 / 1111; 1], the servers [1 / 0101; 1] and [1 / 1010; 1], and the servers [1 / 1000; 2] and [1 / 1111; 2] are directly connected with each other.

[0083] In order to evaluate the performance of the MBDC network, comprehensive simulation experiments are performed, and comparisons are made with other topologies.

[0084] Firstly, the network diameters corresponding to different topologies are studied, and the BCube, DCell, HSDC, CSDC, HS-DCell, Hyper-BCube, 0-MBDC and 1-MBDC are simulated. The diameters of the networks are compared when the networks contain the same number of servers, and it can be seen from Figure 5 that when the number of servers is small, the diameter of the Hyper-BCube is the smallest among these topologies; the diameter of the 1-MBDC is slightly smaller than that of the BCube, but is smaller than that of the HSDC, DCell and HS-DCell; when the number of servers increases to 1024, the diameter of the 1-MBDC is almost half of that of the HSDC; compared with the 1-MBDC, the diameter of the 0-MBDC is slightly larger.

[0085] ​​Finally, the cost and energy consumption of MBDC networks are studied. In this embodiment, BCube, DCell, HSDC and MBDC are compared. For convenience, since 0-MBDC and 1-MBDC have the same number of switches, servers and links, MBDC is collectively compared. The total cost and energy consumption of servers of all networks are the same, because the servers of each network are compared using the same number of cost and energy consumption. Therefore, this embodiment only evaluates the price and energy usage of network cards and switches. In order to monitor the cost and energy utilization obviously, this embodiment constructs a data center network containing 1024, 2048 and 4096 servers respectively. The cost and energy usage of network cards and switches of the data center are summarized in Table 1. Table 1 shows that MBDC and HSDC have the same cost and energy consumption, both of which are lower than DCell, BCube and Hyper-BCube. This embodiment also points out the number of links used in these network architectures, because data center networks need to be configured with wires and cables. Table 1 shows that MBDC and HSDC have the same number of links, although DCell and BCube have more links. Based on the above data, the cost, energy usage and number of links of MBDC and HSDC are much lower than DCell, BCube and Hyper-BCube. The MBDC architecture constructed in this embodiment is more suitable for constructing a low-cost data center network with low latency, because it has better latency than HSDC.

[0086] Table 1: Cost and energy usage of five different network topologies

[0087]

[0088] Through comprehensive analysis and comprehensive comparison of MBDC with other common network topologies, it is found that MBDC has good performance in network size, network diameter, cost and energy consumption and other key indicators. The combination of these ideal characteristics makes MBDC a very promising choice for data center network communication.

[0089] Embodiment 2:

[0090] The low-latency low-cost data center network topology of the present application is a data center network topology constructed by the method disclosed in embodiment 1, which is a recursive structure composed of switches and two-port servers.

[0091] The data center network topology of this embodiment satisfies the following definition:

[0092] Definition 1: A data center network topology is represented as MBDC. Elements serve as the basic structural blocks of the data center network topology. An element includes m servers with two ports and one switch with m ports, distinguished by symbols. Let t represent an MBDC with n levels, and t represent the type of MBDC. Both n and t are integers, n≥1, t∈{0,1}, t=0 represents a type 0 MBDC, and t=1 represents a type 1 MBDC.

[0093] In MBDC, servers within the same element communicate with each other through their first port, while servers located in different elements communicate with each other through their second port. For each element, the first port of the servers within the same element is connected via a switch, and the second port of the server connected to the switch is connected to the servers in different elements.

[0094] The MBDC in this embodiment satisfies the following definition two:

[0095] (1) When n = 1, It consists of two basic elements;

[0096] (2) When n≥2, From 1 and 1 composition.

[0097] In this embodiment, the addresses of the server and the switch in the MBDC satisfy the following definition three:

[0098] (1) The server address is represented as [x0x1x2…x n ;a], where x i ∈{0,1}, i∈[0,n], a∈[1,m], x0=0 represents type 0 MBDC, x0=1 represents type 1 MBDC;

[0099] (2) The address of the switch is represented as [x0x1x2…x n ;0],where, x i ∈{0,1}, i∈[0,n].

[0100] Meanwhile, the address changes of the server and switch satisfy the following definition: For In the context of servers and switches, the value of the 0th address bit of each server and each switch is set to t, and the values ​​of the 1st to nth address bits of each server are as follows: The values ​​of the 0th to (n-1)th address bits of the server, and the values ​​of the 1st to nth address bits of each switch are as follows: The values ​​of the 0th to (n-1)th address bits of the switch.

[0101] Correspondingly, the connection rules between the server and the switch include the following:

[0102] Linking rule 1: For elements located within the same element with addresses [x0x1x2…x], ... n ;a] and addresses [x0′x1′x2′…x n The two servers [a'] are currently connected via a switch only if the following conditions are met:

[0103] x j =x j ′,j∈[0,n];

[0104] Linking rule 2: For elements located in different elements with addresses [x0x1x2…x] n ;a] and addresses [x0′x1′x2′…x n Two servers, 'a' and 'a', are connected via a switch if the following conditions are met:

[0105] When x i-1 When = 0,

[0106] When x i-1 When = 1, And x0 = x0′, a = a′;

[0107] i∈[1,n].

[0108] The present invention has been shown and described in detail above with reference to 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 will know that more embodiments of the present invention can be obtained by combining the means in the different embodiments described above, and these embodiments are also within the protection scope of the present invention.

Claims

1. A method for constructing a low-latency low-cost data center network topology, characterized by, The data center network topology is a recursive structure composed of switches and two-port servers, and the construction method comprises the following steps: S100, defining elements as the basic structure blocks of the data center network topology, and defining the communication mode of the servers in the elements and between the elements; S200, defining the recursive mode of the data center network topology based on the communication mode of the servers in the elements and between the elements; S300, defining the addresses of the servers and the switches and the connection principle between the servers and the switches based on the recursive mode of the data center network topology; S400, constructing the data center network topology based on the definitions of steps S100-S300; Wherein, the data center network topology structure is represented as MBDC, the elements include m servers with two ports and a switch with m ports, and the symbol represents the MBDC with n levels, t represents the type of the MBDC, n and t are integers, n≥1, t∈{0,1}, t=0 represents the 0-type MBDC, and t=1 represents the 1-type MBDC. In the MBDC, the servers in the same element communicate through their first ports, and the servers in different elements communicate through their second ports, and for each element, the first ports of the servers in the same element are connected through a switch, and for the servers connected to the switch, the second ports of the servers are connected to the servers in different elements; Wherein, for the MBDC, the addresses of the servers and the switches are defined as follows: The address of the server is represented as [x0x1x2…xn], wherein x n ∈{0,1}, i∈[0,n], a∈[1,m], x0=0 represents type 0 MBDC, and x0=1 represents type 1 MBDC. i The address of the server is represented as [x0x1x2…xn], wherein x n ∈{0,1}, i∈[0,n], a∈[1,m], x0=0 represents type 0 MBDC, and x0=1 represents type 1 MBDC. i The address of the server is represented as The address of the switch is represented as [x0x1x2...xn], where xi∈{0,1}, i∈[0,n]. n ;0], where xi∈{0,1}, i∈[0,n]. i ;0], where xi∈{0,1}, i∈[0,n]. Wherein, the connection rules between the servers and the switches include the following: For two servers located within the same element, with addresses [x0x1x2...x n ;a] and [x0'x1'x2'...x n ';a'] the two servers are currently connected through a switch only when the following conditions are met: x j = x j ', j e [0, n] ; For two servers located in different elements, with addresses [x0x1x2...x n ; a] and [x0'x1'x2'...x n ' ; a'] the two servers are connected through a switch when the following conditions are met: when x i-1 = 0, when x i-1 = 1, and x0= x0', a = a'. i∈[1,n].

2. The method of claim 1, wherein, For the MBDC, the recursive mode includes the following two cases: When n = 1, Consists of two basic elements; when n > 2, consists of 1 and 1 .

3. The method of claim 1, wherein, The address changes of the servers and the switches are defined as follows: for In the context of servers and switches, the value of the 0th address bit of each server and each switch is set to t, and the values ​​of the 1st to nth address bits of each server are as follows: The values ​​of the 0th to (n-1)th address bits of the server, and the values ​​of the 1st to nth address bits of each switch are as follows: The values ​​of the 0th to (n-1)th address bits of the switch.

4. A low-latency, low-cost data center network topology, comprising: The data center network topology constructed by the construction method of the low-delay low-cost data center network topology according to any one of claims 1-3, the data center network topology is a recursive structure composed of switches and two-port servers; The data center network topology is denoted as MBDC, and the elements include m servers with two ports and a switch with m ports, which is denoted by MBDC with n levels is denoted, t represents the type of MBDC, n and t are integers, n≥1, t∈{0,1}, t=0 represents type 0 MBDC, and t=1 represents type 1 MBDC. In the MBDC, the servers in the same element communicate through their first ports, and the servers in different elements communicate through their second ports, and for each element, the first ports of the servers in the same element are connected through a switch, and for the servers connected to the switch, the second ports of the servers are connected to the servers in different elements; Wherein, for the MBDC, the addresses of the servers and the switches are defined as follows: The address of the server is represented as [x0x1x2…xn], wherein x n ∈{0,1}, i∈[0,n], a∈[1,m], x0=0 represents type 0 MBDC, and x0=1 represents type 1 MBDC. i ​ The address of the switch is represented as [x0x1x2...xn], where xi∈{0,1}, i∈[0,n]. n ;0], where xi∈{0,1}, i∈[0,n]. i ;0], where xi∈{0,1}, i∈[0,n]. Wherein, the connection rules between the servers and the switches include the following: For two servers located within the same element, with addresses [x0x1x2...x n ; a] and [x0'x1'x2'...x n ' ; a'], the two servers are currently connected through a switch only when the following conditions are met: x j = x j ', j e [0, n] ; For two servers located in different elements, with addresses [x0x1x2...x n ; a] and [x0'x1'x2'...x n ' ; a'] the two servers are connected through a switch when the following conditions are met: when x i-1 = 0, when x i-1 = 1, and x0= x0', a = a'. i∈[1,n].

5. The low-latency, low-cost data center network topology of claim 4, wherein, For the MBDC, the recursive mode includes the following two cases: When n = 1, Consists of two basic elements; when n > 2, consists of 1 and 1 .

6. The low-latency, low-cost data center network topology of claim 4, wherein, The address changes of the servers and the switches are defined as follows: for In the context of servers and switches, the value of the 0th address bit of each server and each switch is set to t, and the values ​​of the 1st to nth address bits of each server are as follows: The values ​​of the 0th to (n-1)th address bits of the server, and the values ​​of the 1st to nth address bits of each switch are as follows: The values ​​of the 0th to (n-1)th address bits of the switch.

Images