CN113810286B - Computer network system and routing method - Google Patents
Computer network system and routing method Download PDFInfo
- Publication number
- CN113810286B CN113810286B CN202111043324.XA CN202111043324A CN113810286B CN 113810286 B CN113810286 B CN 113810286B CN 202111043324 A CN202111043324 A CN 202111043324A CN 113810286 B CN113810286 B CN 113810286B
- Authority
- CN
- China
- Prior art keywords
- switch
- switches
- clusters
- computer network
- cluster
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L45/00—Routing or path finding of packets in data switching networks
- H04L45/14—Routing performance; Theoretical aspects
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L12/00—Data switching networks
- H04L12/28—Data switching networks characterised by path configuration, e.g. LAN [Local Area Networks] or WAN [Wide Area Networks]
- H04L12/42—Loop networks
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L45/00—Routing or path finding of packets in data switching networks
- H04L45/74—Address processing for routing
Landscapes
- Engineering & Computer Science (AREA)
- Computer Networks & Wireless Communication (AREA)
- Signal Processing (AREA)
- Data Exchanges In Wide-Area Networks (AREA)
- Multi Processors (AREA)
Abstract
The present application relates to the field of computer networks, and in particular, to a computer network system and a routing method. The computer network system provided by the embodiment of the application comprises a plurality of switch clusters. The inside of each switch cluster in a plurality of switch clusters is provided with a multi-layer switch, and the multi-layer switch builds a top-of-rack topology network through a fat tree structure. The top-layer switch arranged in each switch cluster in the plurality of switch clusters is respectively connected with the top-layer switch arranged in other adjacent switch clusters, and a ring topology network is built to form a computer network system comprising the plurality of switch clusters. In the deployment process, the computer network system provided by the embodiment of the application does not involve a large number of long-line connections, and has lower construction cost compared with the existing computer network system.
Description
Technical Field
The present application relates to the field of computer networks, and in particular, to a computer network system and a routing method.
Background
For a high-performance computer network system, because the scale is very large, different network building modes are selected, so that the space, layout and building cost of the computer network system are greatly influenced, and the influence even exceeds the influence on the performance and other technical indexes. With the high-performance computer network system entering the 200G era, although the cost of a chip is not remarkably improved following moore's law, the cost of a cable, particularly the price of an optical fiber is greatly increased, and meanwhile, the maximum length supported by a low-cost copper cable which can be used for substitution is greatly reduced, so that the construction cost of the existing computer network system is higher and higher.
However, the existing computer network system is usually built in a fat tree structure, and the performance of the existing computer network system is best, but a large number of long-line connections are involved in the deployment process of the existing computer network system due to the addition of a switch cluster for switching, so that the building cost is high, and particularly after the computer network system is expanded to tens of thousands of nodes on a large scale, the cost is increased sharply.
Disclosure of Invention
An object of the present application is to provide a computer network system and a routing method, so as to solve the above-mentioned problems.
In a first aspect, a computer network system provided in an embodiment of the present application includes a plurality of switch clusters;
a multi-layer switch is arranged in each switch cluster in the plurality of switch clusters, and the multi-layer switch builds a top-of-rack topology network in a fat tree structure mode;
the top-layer switch arranged in each switch cluster in the plurality of switch clusters is respectively connected with the top-layer switch arranged in other adjacent switch clusters, and a ring topology network is built to form a computer network system comprising the plurality of switch clusters.
The computer network system provided by the embodiment of the application comprises a plurality of switch clusters, wherein a multi-layer switch is arranged in each switch cluster in the plurality of switch clusters, a top-mounted topology network is built through a fat tree structure of the multi-layer switch, top-layer switches arranged in each switch cluster in the plurality of switch clusters are respectively connected with top-layer switches arranged in other adjacent switch clusters, and a ring topology network is built to form the computer network system comprising the plurality of switch clusters. Compared with a computer network system which is built in a fat tree structure completely in the prior art, the switch cluster for switching is omitted, and a large number of long wires used for connecting a plurality of switch clusters with the switching switch cluster respectively are omitted.
With reference to the first aspect, the embodiment of the present application further provides a first optional implementation manner of the first aspect, in the multiple switch clusters, a bottom layer switch set inside each switch cluster includes M first switches, and a top layer switch set inside the switch cluster includes N second switches, where M is greater than or equal to 1, and is an integer, N is greater than or equal to 1, and is an integer;
for each first switch of the M first switches, the first switch is connected with each second switch of the N second switches through at least one uplink to form a top-of-rack topology network.
In the above embodiment, for each of the M first switches, the first switch is connected with each of the N second switches through at least one uplink line to form a top-of-rack topology network, based on which, in a subsequent routing process, the first switch can directly communicate with each of the N second switches without compensation by a routing algorithm, thereby reducing difficulty in developing a routing algorithm and improving routing efficiency.
With reference to the first aspect, or the first optional implementation manner of the first aspect, the embodiment of the present application further provides a second optional implementation manner of the first aspect, at least three layers of switches are disposed inside each of the plurality of switch clusters, and the at least three layers of switches build a top-of-rack topology network in a fat tree structure.
In the above embodiment, at least three layers of switches are provided in each of the plurality of switch clusters, and the at least three layers of switches build a top-of-rack topology network in a fat tree structure, so that the requirement of expanding switch nodes can be met under the condition of insufficient switch ports.
With reference to the first aspect, the embodiment of the present application further provides a third optional implementation manner of the first aspect, in the multiple switch clusters, a top-level switch set inside each switch cluster includes N second switches, where N is greater than or equal to 1 and is an integer;
n second switches arranged in each switch cluster in the plurality of switch clusters are respectively connected with N second switches arranged in other adjacent switch clusters in a one-to-one correspondence mode so as to form a ring topology network.
In the above embodiment, since the N second switches disposed in each switch cluster in the plurality of switch clusters are respectively connected with the N second switches disposed in the adjacent other switch clusters in a one-to-one correspondence manner, in the subsequent routing process, compensation by a routing algorithm is not needed, so that the difficulty in developing the routing algorithm is reduced, and the routing efficiency is also improved.
With reference to the third optional implementation manner of the first aspect, the embodiment of the present application further provides a fourth optional implementation manner of the first aspect, where N second switches set inside each switch cluster in the plurality of switch clusters are respectively connected with N second switches set inside adjacent other switch clusters in a one-to-one correspondence manner according to a setting position of the switch, so as to form a ring topology network.
In the above embodiment, since the N second switches arranged in each switch cluster in the plurality of switch clusters are connected in one-to-one correspondence with the N second switches arranged in the adjacent other switch clusters according to the setting positions of the switches, the connection operation is more convenient, and the efficiency of building the computer network can be further improved.
With reference to the first aspect, the embodiment of the present application further provides a fifth optional implementation manner of the first aspect, in the plurality of switch clusters, a top-level switch set inside each switch cluster includes N second switches, where N is greater than or equal to 1 and is an integer;
for each of the plurality of switch clusters, the N second switches arranged inside the switch cluster comprise a plurality of groups of second switch groups, and for each group of second switch groups in the plurality of groups of second switch groups, the plurality of second switches included in the second switch groups are respectively connected with the plurality of second switches included in the N second switches arranged inside other switch clusters adjacent to the target dimension in a one-to-one correspondence manner so as to form a ring topology network, wherein the target dimension comprises at least one dimension direction of the ring topology network.
In the above embodiment, for each of the plurality of switch clusters, the N second switches provided inside the switch cluster are not completely connected to the N second switches provided inside the adjacent other switch clusters, and therefore, the long-line connection can be further reduced, and the cost of constructing the computer network system can be further reduced as compared with the existing computer network system.
With reference to the fifth optional implementation manner of the first aspect, the embodiment of the present application further provides a sixth optional implementation manner of the first aspect, for each group of second switch groups in the multiple groups of second switch groups, a plurality of second switches included in the second switch group are connected in a one-to-one correspondence with a plurality of second switches included in N second switches that are set inside other switch groups that are adjacent in the target dimension according to a setting position of the switch, so as to form a ring topology network.
In the above embodiment, for each group of the plurality of groups of second switch groups, the plurality of second switches included in the second switch group are connected in one-to-one correspondence with the plurality of second switches included in the N second switches that are set inside the other switch group that is adjacent in the target dimension, according to the setting positions of the switches, so that the connection operation is more convenient, and the efficiency of building the computer network can be further improved.
With reference to the first aspect, the embodiment of the present application further provides a seventh optional implementation manner of the first aspect, where each switch cluster in the plurality of switch clusters is installed inside the same first cabinet.
In the above embodiment, each of the plurality of switch clusters is installed inside the same first cabinet, so that a large number of long-line connections are not involved inside each first cabinet, and compared with the existing computer network system, the construction cost of the computer network system can be further reduced.
With reference to the seventh optional implementation manner of the first aspect, the embodiment of the present application further provides an eighth optional implementation manner of the first aspect, wherein a plurality of layers of switches disposed inside each of the plurality of switch clusters are connected by using copper cables;
the top-layer switch arranged in each switch cluster in the switch clusters is connected with the top-layer switch arranged in other adjacent switch clusters by adopting optical fibers.
In the above embodiment, the multi-layer switch set inside each switch cluster in the plurality of switch clusters is connected by using a copper cable, and only the top layer switch set inside each switch cluster in the plurality of switch clusters is connected with the top layer switch set inside the adjacent other switch clusters by using an optical fiber, so that the copper cable has lower cost, and the optical fiber consumption is less, thereby further reducing the construction cost of the computer network system.
With reference to the first aspect, the embodiment of the present application further provides a ninth optional implementation manner of the first aspect, in the multiple switch clusters, a top-level switch set inside each switch cluster includes N second switches, where N is greater than or equal to 1 and is an integer;
in the switch clusters, a plurality of second switches with matched positions are arranged in the same second cabinet;
for each of the plurality of switch clusters, the non-top-level switches disposed within the switch cluster are mounted within the same third enclosure.
In the above embodiment, in the plurality of switch clusters, the top-level switch set in each switch cluster includes N second switches, where N is greater than or equal to 1 and is an integer. Under this condition, can install in a plurality of switch clusters, the position is matchd a plurality of second switches in same second rack is inside, and to each switch cluster in a plurality of switch clusters, the inside non-top layer switch that sets up of this switch cluster is installed in same third rack is inside, so, every second rack is inside all can adopt the stub to connect, in order to reduce the construction cost of computer network system, and to each switch cluster in a plurality of switch clusters, the non-top layer switch that its inside set up adopts the long line to connect, need not to consider the distance restriction problem, therefore, the application scenario is more general.
With reference to the ninth optional implementation manner of the first aspect, the embodiment of the present application further provides a tenth optional implementation manner of the first aspect, wherein in the plurality of switch clusters, the plurality of second switches with matched positions are connected by using copper cables;
the non-top-layer switch and the top-layer switch which are arranged in each switch cluster in the switch clusters are connected by adopting optical fibers.
In the above embodiment, in the plurality of switch clusters, the plurality of second switches with matched positions are connected by using copper cables, and the non-top-layer switch and the top-layer switch which are arranged inside each switch cluster in the plurality of switch clusters are connected by using optical fibers, and compared with the existing computer network system, the copper cable quantity is reduced, so that the computer network system provided by the embodiment of the application has lower construction cost.
In a second aspect, an embodiment of the present application further provides a routing method, which is applied to the computer network system provided in the first aspect, or any optional implementation manner of the first aspect, where the routing method includes:
for each switch cluster included in the computer network system, judging a target port pointed by a destination address of a message to be sent after any switch arranged in the switch cluster receives the message to be sent;
if the target port pointed by the target address is positioned on other switches arranged in the switch cluster, sending the message to be sent to the target port through an intra-group routing algorithm;
if the target port pointed by the target address is positioned on a switch arranged in other switch clusters in the system, the message to be sent is sent to the target port through an inter-group routing algorithm.
In the routing method provided by the embodiment of the application, the intra-group routing algorithm is adopted in the switch clusters, and the inter-group routing algorithm is adopted between the switch clusters, so that the risk of deadlock can be avoided in the switch clusters and between the switch clusters, and therefore, the reliability of the routing method can be improved.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings that are needed in the embodiments of the present application will be briefly described below, it should be understood that the following drawings only illustrate some embodiments of the present application and should not be considered as limiting the scope, and other related drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.
Fig. 1 is a schematic diagram of a conventional computer network system in the form of a building.
Fig. 2 is a schematic topology diagram of a computer network system according to an embodiment of the present application.
Fig. 3 is a schematic diagram of another topology of a computer network system according to an embodiment of the present application.
Fig. 4 is a schematic diagram of a topology of a switch cluster.
Fig. 5 is a schematic diagram of a topology of another switch cluster.
Fig. 6 is a schematic diagram of another topology of a computer network system according to an embodiment of the present application.
Fig. 7 is a schematic diagram of an installation manner of a plurality of switch clusters in a computer system according to an embodiment of the present application.
Fig. 8 is a schematic diagram of another installation of a plurality of switch clusters in the computer system shown in fig. 7.
Fig. 9 is a flowchart of steps of a routing method according to an embodiment of the present application.
Detailed Description
For the purposes, technical solutions and advantages of the embodiments of the present application, the technical solutions of the embodiments of the present application will be described below with reference to the drawings in the embodiments of the present application. Furthermore, it should be noted that: like reference numerals and letters denote like items in the following figures, and thus once an item is defined in one figure, no further definition or explanation thereof is necessary in the following figures.
Existing computer network systems 100 are typically built entirely in the form of fat tree structures, such as the full Fattree solution shown in fig. 1, and then the computer network system 100 needs to incorporate a second cluster of switches 120 for transit in addition to the plurality of first clusters of switches 110 for connecting computing nodes.
Referring to fig. 1, for each first switch cluster 110 in the plurality of first switch clusters 110, a top-mounted topology network is built by a multi-layer switch arranged in the first switch cluster 110 in a fat tree structure, and a plurality of switches included in a top-mounted switch arranged in the first switch cluster 110 are connected with a plurality of switches included in the second switch cluster 120 in a one-to-one correspondence manner. It should be understood that, for ensuring simplicity of the drawing, only a plurality of switches included in a top-level switch set inside one first switch cluster 110 are shown in fig. 1 to be connected to a plurality of switches included in a second switch cluster 120 in a one-to-one correspondence manner, but connection manners of the top-level switches set inside other first switches 110 and the second switch cluster 120 may be inferred from this.
Because of the addition of the second switch cluster 120, a large number of long-wire connections are involved in the deployment of the computer network system 100, which is costly to build. Taking 200G of computer network system 1000 as an example, it is embodied in:
1. the two exchanger nodes are connected through copper cables, so that the cost is the lowest (about hundreds of RMB), but the length of the exchanger nodes can only be limited to be within 2M, and the length can only be used for interconnection between exchanger nodes in one cabinet;
2. when the connection length exceeds 2M, only optical fibers can be used for interconnection, and the cost of 10M optical fibers is about several thousands of rmbs, so that the cost of the computer network system 100 is greatly increased when a large amount of optical fibers are used;
3. the cost of the fiber may also increase with increasing length, for example, 50M fiber costs more than 2 times the cost of less than 20M fiber.
In summary, the cost of building an existing computer network system 100 is high, and particularly, after the computer network system 100 is scaled up to tens of thousands of nodes, the cost increases dramatically.
To solve the above-described problems, embodiments of the present application provide a computer network system 200. Referring to fig. 2, a computer network system 200 provided in an embodiment of the present application includes a plurality of switch clusters 210.
A multi-layer switch is disposed inside each switch cluster 210 Of the plurality Of switch clusters 210, and the multi-layer switch builds a Top Of the Tree (TOR) topology network in the form Of a fat tree structure, a bottom layer switch 211 disposed inside each switch cluster 210 Of the plurality Of switch clusters 210 is used for connecting computing nodes, and a Top layer switch 212 disposed inside each switch cluster 210 Of the plurality Of switch clusters 210 is respectively connected with a Top layer switch 212 disposed inside adjacent other switch clusters 210, so as to build a ring (Torus) topology network, so as to form the computer network system 200 including the plurality Of switch clusters 210.
It should be noted that, in the embodiment of the present application, the inside of each switch cluster 210 of the plurality of switch clusters 210 is provided with a multi-layer switch, that is, at least two layers of switches are provided (fig. 2 shows a case where two layers of switches are provided inside of each switch cluster 210 of the plurality of switch clusters 210).
In addition, in the embodiment of the present application, for each switch cluster 210 in the plurality of switch clusters 210, other switch clusters 210 adjacent to the switch cluster 210 may refer to other switch clusters 210 adjacent to the switch cluster 210 in all dimension directions in the computer network system 200.
For example, the computer network system 200 shown in fig. 2 is a two-dimensional system, including an X-dimension direction and a Y-dimension direction, and then the top-level switches 212 disposed inside each of the switch clusters 210 in the plurality of switch clusters 210 are respectively connected to the top-level switches 212 disposed inside other switch clusters 210 adjacent to the first switch cluster in the X-dimension direction, and are respectively connected to the top-level switches 212 disposed inside other switch clusters 210 adjacent to the first switch cluster in the Y-dimension direction.
Referring to fig. 3, for another example, the computer network system 200 is a three-dimensional system including an X-dimension direction, a Y-dimension direction, and a Z-dimension direction, then a top-level switch 212 (not shown in fig. 3) disposed inside each switch cluster 210 of the plurality of switch clusters 210 is connected to a top-level switch 212 disposed inside other switch clusters 210 adjacent to the X-dimension direction, and is connected to a top-level switch 212 disposed inside other switch clusters 210 adjacent to the Y-dimension direction, and is connected to a top-level switch 212 disposed inside other switch clusters 210 adjacent to the Z-dimension direction.
Compared to the computer network system 100 fully constructed in the fat tree structure in the prior art, for example, the computer network system 100 shown in fig. 1, the computer network system 200 provided in the embodiment of the present application omits the second switch clusters 120, that is, omits a large number of long wires used for connecting the plurality of first switch clusters 110 with the second switch clusters 120 respectively, so that the computer network system 200 provided in the embodiment of the present application does not involve a large number of long wire connections in the deployment process, and has lower construction cost compared to the computer network system 100.
Further, in the embodiment of the present application, in the plurality of switch clusters 210, the bottom layer switch 211 set inside each switch cluster 210 includes M first switches, and the top layer switch 212 set inside the switch cluster 210 includes N second switches, where M is greater than or equal to 1, N is greater than or equal to 1, and is an integer, and regarding the fat tree structure, the building principle may be defined as that for each first switch of the M first switches, each second switch of the first switch and the N second switches is connected through at least one uplink. Based on the method, in the subsequent routing process, the first switch can directly communicate with each second switch in the N second switches without compensation by a routing algorithm, so that the development difficulty of the routing algorithm is reduced, and the routing efficiency is also improved.
In the following, a manner of assigning switch ports inside each switch cluster 210 of the plurality of switch clusters 210 will be described with reference to fig. 2.
For each switch cluster 210 of the plurality of switch clusters 210, it is assumed that two layers of switches are provided inside the switch cluster 210, wherein the bottom layer switch 211 includes 8 first switches, the top layer switch 212 includes 6 second switches, and the first and second switches are melanon HDR 200G switches (melanon HDR 200G switches include 40 communication ports, and each communication port is 200G). In practical implementation, each first switch included in the bottom layer switch 211 is connected to 3 ports of each second switch included in the top layer switch 212 through 3 ports in a one-to-one correspondence manner, so that each first switch in the bottom layer switch 211 has 3×6=18 ports connected to the top layer switch 212, and the remaining 40-18=22 communication ports can be connected to the computing node, so that a total of 22×8=176 communication ports can be connected to the computing node in the 8 first switches included in the bottom layer switch 211.
In actual implementation, however, the bottom layer switch 211 includes only 20 communication ports for each first switch to connect to the computing node, so a total of 20×8=160 communication ports can connect to the computing node out of the 8 first switches included in the bottom layer switch 211.
In addition, since the top-level switch 212 includes 6 second switches, each of which has 3 communication ports connected to the bottom-level switch 211, there remain 40-3x8=16 communication ports for connecting the top-level switch 212 disposed inside the other switch cluster 210, and then the top-level switch 212 includes 16 x6=96 communication ports for connecting the top-level switch 212 disposed inside the other switch cluster 210 in total among the 6 second switches.
In addition, it should be noted that, in the embodiment of the present application, if two layers of switches are provided in each switch cluster 210 of the plurality of switch clusters 210, in the case that the number of switch ports is limited and a TOR topology network needs to be built in the switch cluster 210 in the form of a fat tree structure, the expansion requirement of the switch nodes cannot be met, and specifically, the description is made with reference to the switch cluster 310 shown in fig. 4. It is assumed that, in the switch cluster 310, each switch included in the bottom layer switch 311 has 4 ports, where two ports are used to connect to the computing nodes, and two ports are used to connect to the top layer switch 312, and the number of ports is limited, so that the requirement of expanding the switch nodes cannot be met.
To solve the above problem, in the embodiment of the present application, at least three layers of switches may be disposed in the internal essence of each switch cluster 210 in the plurality of switch clusters 210, and the at least three layers of switches build a TOR topology network in the form of a fat tree structure, which is described below in connection with the switch cluster 320 shown in fig. 5.
In the switch cluster 320 shown in fig. 5, each switch included in the bottom layer switch 321 has 4 ports, where two ports are used for connecting computing nodes, two ports are used for connecting the middle layer switch 323, and then the middle layer switch 323 is connected to the top layer switch 322, so that, in the case that the number of ports is limited, and the TOR topology network needs to be built in the switch cluster 210 in the form of a fat tree structure, the requirement of expanding the switch nodes can still be met.
Further, in this embodiment, as an optional implementation manner, N second switches disposed inside each switch cluster 210 of the switch clusters 210 may be respectively connected with N second switches disposed inside other adjacent switch clusters 210 in a one-to-one correspondence manner, so as to form a Torus topology network, as shown in fig. 2, and this connection manner may also be referred to as a full connection manner. Based on the full connection mode, the compensation by a routing algorithm is not needed in the subsequent routing process, so that the development difficulty of the routing algorithm is reduced, and the routing efficiency is improved.
Of course, in actual implementation, the N second switches set in each switch cluster 210 of the plurality of switch clusters 210 may be connected to the N second switches set in the adjacent other switch clusters 210 in a one-to-one correspondence manner according to the setting positions of the switches, so as to form a Torus topology network.
Taking the switch cluster 210 (a) included in the computer network system 200 shown in fig. 2 as an example, among 6 second switches disposed inside the switch cluster 210 (a), the second switch A1 is connected to the second switch B1 disposed inside the adjacent switch cluster 210 (B), and is connected to the second switch D1 disposed inside the adjacent switch cluster 210 (D).
Because the N second switches in each switch cluster 210 of the plurality of switch clusters 210 are connected with the N second switches in the adjacent other switch clusters 210 in a one-to-one correspondence manner according to the setting positions of the switches, the connection operation is more convenient, and the construction efficiency of the computer network 200 can be further improved.
For the connection manner between the switch clusters 210, in this embodiment of the present application, as two alternative embodiments, for each switch cluster 210 in the multiple switch clusters 210, the N second switches disposed inside the switch cluster 210 include multiple groups of second switch sets, and for each group of second switch sets in the multiple groups of second switch sets, the multiple second switches included in the second switch sets are respectively connected with the multiple second switches included in the N second switches disposed inside the other switch clusters 210 adjacent to the target dimension in a one-to-one correspondence manner, so as to form a Torus topology network, where the target dimension includes at least one dimension direction of the Torus topology network, as shown in fig. 6, the connection manner may also be referred to as a half-connection manner.
Taking the switch cluster 210 (a) included in the computer network system 200 shown in fig. 6 as an example, 6 second switches are set on the top layer switch 212 in the switch cluster 210 (a), and the 6 second switches include two groups of second switch groups, each group of second switch groups includes 3 second switches, where the 3 second switches included in the first group of second switch groups are respectively a second switch A1, a second switch A2, and a second switch A3, the second switches A1, a second switch A2, and a second switch A3 are connected with 3 second switches (second switch B1, second switch B2, and second switch B3) included in 6 second switches set inside the switch cluster 210 (B) adjacent in the X dimension direction, the 3 second switches included in the second group of second switches are respectively a second switch A4, a second switch A5, and a second switch A6, and the second switch A4, a second switch A5, and a second switch A6 are respectively connected with 3 second switches (second switch B4, D) included in the second switch cluster 210 adjacent in the Y dimension direction, and a second switch D (D) set inside the switch cluster 6).
Based on the half-connection mode, the long-line connection can be further reduced, and compared with the existing computer network system, the construction cost of the computer network system can be further reduced.
Likewise, in this embodiment of the present application, for each group of second switch groups in the plurality of groups of second switch groups, a plurality of second switches included in the second switch group are respectively connected with a plurality of second switches included in N second switches that are set inside other switch clusters 210 that are adjacent in the target dimension according to the setting positions of the switches, so as to form a Torus topology network, so that the connection operation is more convenient, and the efficiency of building a computer network can be further improved, and specifically, see fig. 6, which is not described in detail in this embodiment of the present application.
Further, referring to fig. 7, regarding the installation manner of the plurality of switch clusters 210 in the computer system 200, in this embodiment, as a first alternative implementation manner, each switch cluster 210 in the plurality of switch clusters 210 is installed inside the same first cabinet 220. In the manner of installation shown in fig. 7, a large number of long wire connections are less involved within each first cabinet 220, with lower construction costs relative to the existing computer network system 100.
Further, based on the installation manner shown in fig. 7, the multi-layer switches arranged inside each switch cluster 210 of the plurality of switch clusters 210 may be connected by copper cables, and only the top-layer switch 212 arranged inside each switch cluster 210 of the plurality of switch clusters 210 is connected with the top-layer switch 212 arranged inside the adjacent other switch clusters 210 by optical fibers. Because copper cabling costs are lower and fiber usage is less, the cost of building the computer network system 200 can be further reduced.
Referring to fig. 8 (the computer network system shown in fig. 8 may be considered as an installation modification of the computer network system shown in fig. 7), for the installation mode of the plurality of switch clusters 210 in the computer system 200, in this embodiment of the present application, as a second alternative implementation manner, the plurality of second switches with matched positions may be installed in the same second cabinet 230 in the plurality of switch clusters 210, and for each switch cluster 210 in the plurality of switch clusters 210, the non-top-layer switch set in the switch cluster 210 is installed in the same third cabinet 240.
For example, the second switch A1 in the switch cluster 210 (a), the second switch B1 in the switch cluster 210 (B), the second switch C1 in the switch cluster 210 (C), the second switch D1 in the switch cluster 210 (D), the second switch E1 in the switch cluster 210 (E), the second switch F1 in the switch cluster 210 (F), the second switch G1 in the switch cluster 210 (G), the second switch H1 in the switch cluster 210 (H), and the second switch I1 in the switch cluster 210 (I) are installed inside the same second cabinet 230, and the non-top-level switches (the first switch A1, the first switch A2, the first switch A3, the first switch A4, the first switch A5, the first switch A6, the first switch A7, and the first switch A8) provided inside the switch cluster 210 (a) are installed inside the same third cabinet 240. It should be understood that, for ensuring simplicity of the drawing, only the connection lines of the non-top-level switches and the top-level switches, which are disposed inside the switch cluster 210 (a), are shown in fig. 8, but the connection lines of the non-top-level switches and the top-level switches, which are disposed inside the other switch clusters 210, can be inferred therefrom.
According to the installation manner shown in fig. 8, a short-line connection may be adopted inside each second cabinet 230 to reduce the construction cost of the computer network system 200, while for each switch cluster 210 of the plurality of switch clusters 210, a non-top-layer switch disposed inside the second cabinet adopts a long-line connection, so that the distance limitation problem is not required to be considered, and therefore, the application scenario is more common.
Further, based on the installation manner shown in fig. 8, the plurality of second switches with matched positions in the plurality of switch clusters 210 are connected by copper cables, and the plurality of switches arranged inside each switch cluster 210 in the plurality of switch clusters 210 are connected by optical fibers. Because copper cables are less costly and the amount of copper cables used is reduced relative to existing computer network systems, the computer network system 100 provided by embodiments of the present application has lower construction costs.
The embodiment of the application also provides a routing method, which comprises step S100, step S200 and step S300, so as to be applied to the computer network system 200. It should be noted that, the routing method provided in the embodiment of the present application is not limited by the sequence shown in fig. 9 and the following, and the specific flow and steps of the routing method are described below with reference to fig. 9.
Step S100, for each switch cluster included in the computer network system, any switch arranged in the switch cluster judges a target port pointed by a target address of the message to be sent after receiving the message to be sent.
It should be noted that, before implementing the routing method provided in the embodiment of the present application, a plurality of switch clusters included in the computer network system may be traversed by a subnet controller node (not shown in fig. 2 and fig. 4 to 9) included in the computer network system, so as to obtain the full-network topology characterization information for characterizing the topology structure of the computer network system, and send the full-network topology characterization information to each switch in the plurality of switch clusters. In this embodiment of the present application, the topology characterization information of the whole network may include dimensions (location characterization information) of each switch cluster, the switch cluster of any switch in each switch cluster, a switch layer where any switch in each switch cluster is located, and a port set, a port connection manner, etc. of any switch in each switch cluster.
When executing step S100, for each switch cluster included in the computer network system, after receiving the message to be sent, any switch set in the switch cluster extracts the destination address from the message to be sent, and determines, according to the topology characterization information of the whole network, the destination port pointed by the destination address, so as to determine the switch in which the destination port pointed by the destination address is located.
Step S200, if the target port pointed by the target address is located on other switches set in the switch cluster, the message to be sent is sent to the target port through the intra-group routing algorithm.
In the embodiment of the present application, the intra-group routing algorithm may be, but is not limited to, the Updown algorithm.
Step S300, if the target port pointed by the target address is located on a switch set in other switch clusters in the computer network system, the message to be sent is sent to the target port through an inter-group routing algorithm.
In the embodiment of the application, the inter-group routing algorithm may be, but is not limited to, a dimension sequence routing algorithm (Dimension Order Routing, DOR).
Obviously, in the routing method provided by the embodiment of the application, the intra-group routing algorithm is adopted in the switch clusters, and the inter-group routing algorithm is adopted between the switch clusters, so that the risk of deadlock can be avoided in the switch clusters and between the switch clusters, and the reliability of the routing method can be improved.
In summary, the computer network system 200 provided in the embodiment of the present application includes a plurality of switch clusters 210, wherein a multi-layer switch is disposed inside each switch cluster 210 in the plurality of switch clusters 210, and the multi-layer switch builds a TOR topology network in a fat tree structure, a bottom layer switch 211 disposed inside each switch cluster 210 in the plurality of switch clusters 210 is used for connecting with a computing node, and a top layer switch 212 disposed inside each switch cluster 210 in the plurality of switch clusters 210 is respectively connected with a top layer switch 212 disposed inside other adjacent switch clusters 210, so as to build a Torus topology network, thereby forming the computer network system 200 including the plurality of switch clusters 210. Compared to the computer network system 100 fully constructed in the fat tree structure in the prior art, for example, the computer network system 100 shown in fig. 1, the computer network system 200 provided in the embodiment of the present application omits the second switch clusters 120, that is, omits a large number of long wires used for connecting the plurality of first switch clusters 110 with the second switch clusters 120 respectively, so that the computer network system 200 provided in the embodiment of the present application does not involve a large number of long wire connections in the deployment process, and has lower construction cost compared to the computer network system 100.
In the routing method provided by the embodiment of the application, the intra-group routing algorithm is adopted in the switch clusters, and the inter-group routing algorithm is adopted between the switch clusters, so that the risk of deadlock can be avoided in the switch clusters and between the switch clusters, and therefore, the reliability of the routing method can be improved.
In the description of the present application, it should be noted that, unless explicitly specified and limited otherwise, the terms "connected", "arranged" and "configured" should be interpreted broadly, and may be, for example, mechanically fixed, detachably or integrally connected, electrically connected, communicatively connected, wherein the communicatively connected may be further wired or wireless, and further may be directly connected, or indirectly connected through an intermediate medium, or may be in communication with the inside of two elements, and it will be understood by those skilled in the art that the specific meaning of the terms in this application may be understood according to the specific circumstances. Furthermore, the terms "first," "second," "third," and the like are used merely to distinguish between descriptions and should not be construed as indicating or implying relative importance.
The foregoing is only a few examples of the present application and is not intended to limit the present application, and various modifications and variations may be made to the present application by those skilled in the art. Any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the present application should be included in the protection scope of the present application.
Claims (9)
1. A computer network system comprising a plurality of switch clusters;
a multi-layer switch is arranged in each switch cluster in the plurality of switch clusters, and the multi-layer switch builds a top-of-rack topology network in a fat tree structure;
the top-level switch arranged in each switch cluster in the plurality of switch clusters is respectively connected with the top-level switch arranged in other adjacent switch clusters, and a ring topology network is built to form a computer network system comprising the plurality of switch clusters;
in the switch clusters, a top-level switch arranged in each switch cluster comprises N second switches, wherein N is more than or equal to 1 and is an integer;
for each switch cluster in the switch clusters, the N second switches arranged inside the switch clusters comprise a plurality of groups of second switch units, and for each group of second switch units in the plurality of groups of second switch units, the plurality of second switches included in the second switch units are respectively connected with the plurality of second switches included in the N second switches arranged inside other switch clusters adjacent to the target dimension in a one-to-one correspondence manner so as to form the ring topology network, wherein the target dimension comprises at least one dimension direction of the ring topology network.
2. The computer network system according to claim 1, wherein the bottom layer switch arranged inside each switch cluster in the plurality of switch clusters comprises M first switches, M is greater than or equal to 1 and is an integer;
for each first switch in the M first switches, the first switch is connected with each second switch in the N second switches through at least one uplink line so as to form the top-of-rack topology network.
3. The computer network system according to claim 1 or 2, wherein at least three layers of switches are arranged inside each of the plurality of switch clusters, and wherein the at least three layers of switches build a top-of-rack topology network in the form of a fat tree structure.
4. The computer network system according to claim 1, wherein the N second switches disposed inside each of the plurality of switch clusters are respectively connected in one-to-one correspondence with the N second switches disposed inside adjacent other switch clusters, so as to form the ring topology network.
5. The computer network system according to claim 4, wherein the N second switches arranged inside each switch cluster in the plurality of switch clusters are connected in one-to-one correspondence with the N second switches arranged inside the adjacent other switch clusters according to the arrangement positions of the switches, respectively, so as to form the ring topology network.
6. The computer network system according to claim 1, wherein for each of the plurality of sets of second switch groups, the plurality of second switches included in the second switch group are connected in one-to-one correspondence with the plurality of second switches included in N second switches disposed inside other switch groups adjacent in the target dimension, respectively, according to the disposed positions of the switches, so as to form the ring topology network.
7. The computer network system of claim 1, wherein each of the plurality of switch clusters is mounted within a same first cabinet.
8. The computer network system of claim 1, wherein a plurality of second switches of the plurality of switch clusters are located within a same second cabinet;
for each of the plurality of switch clusters, non-top-level switches disposed within the switch cluster are mounted within a same third enclosure.
9. A routing method applied to the computer network system of any one of claims 1 to 8, the routing method comprising:
for each switch cluster included in the computer network system, judging a target port pointed by a destination address of a message to be sent after any switch arranged in the switch cluster receives the message to be sent;
if the target port pointed by the target address is positioned on other switches arranged in the switch cluster, sending a message to be sent to the target port through an intra-group routing algorithm;
and if the target port pointed by the target address is positioned on a switch arranged in other switch clusters in the system, sending the message to be sent to the target port through an inter-group routing algorithm.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111043324.XA CN113810286B (en) | 2021-09-07 | 2021-09-07 | Computer network system and routing method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111043324.XA CN113810286B (en) | 2021-09-07 | 2021-09-07 | Computer network system and routing method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113810286A CN113810286A (en) | 2021-12-17 |
| CN113810286B true CN113810286B (en) | 2023-05-02 |
Family
ID=78940645
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111043324.XA Active CN113810286B (en) | 2021-09-07 | 2021-09-07 | Computer network system and routing method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113810286B (en) |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104767681A (en) * | 2015-04-14 | 2015-07-08 | 清华大学 | Data center network routing method capable of tolerating error connection |
| CN112202620A (en) * | 2020-10-12 | 2021-01-08 | 曙光信息产业(北京)有限公司 | Network topology structure generation method and system |
Family Cites Families (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US8358597B2 (en) * | 2009-10-01 | 2013-01-22 | Hei Tao Fung | Method for building scalable Ethernet switch network and huge Ethernet switch |
| US8750122B1 (en) * | 2012-03-22 | 2014-06-10 | Avaya, Inc. | Method and apparatus for layer 2 loop prevention in a multi-node switch cluster |
| US9106578B2 (en) * | 2012-05-31 | 2015-08-11 | Hewlett-Packard Development Company, L.P. | Core network architecture |
| US8842575B2 (en) * | 2012-06-21 | 2014-09-23 | Alcatel Lucent | Method and apparatus for providing a non-overlapping ring-mesh network topology |
| US10084718B1 (en) * | 2013-03-15 | 2018-09-25 | Google Llc | Bi-Connected hierarchical data center network based on multi-ported network interface controllers (NICs) |
| US9426060B2 (en) * | 2013-08-07 | 2016-08-23 | International Business Machines Corporation | Software defined network (SDN) switch clusters having layer-3 distributed router functionality |
| US9307018B2 (en) * | 2013-09-11 | 2016-04-05 | International Business Machines Corporation | Workload deployment with real-time consideration of global network congestion |
| CN108040020A (en) * | 2017-12-21 | 2018-05-15 | 郑州云海信息技术有限公司 | A kind of fiber optic network switching port virtual management system and method |
| US10742513B2 (en) * | 2018-02-05 | 2020-08-11 | David I-Keong Wong | Network interconnect as a switch |
| CN108429679B (en) * | 2018-04-23 | 2020-12-22 | 曙光信息产业(北京)有限公司 | Topological structure of extended interconnection network and routing method thereof |
| CN109302466B (en) * | 2018-09-18 | 2021-10-26 | 华为技术有限公司 | Data processing method, related device and computer storage medium |
-
2021
- 2021-09-07 CN CN202111043324.XA patent/CN113810286B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104767681A (en) * | 2015-04-14 | 2015-07-08 | 清华大学 | Data center network routing method capable of tolerating error connection |
| CN112202620A (en) * | 2020-10-12 | 2021-01-08 | 曙光信息产业(北京)有限公司 | Network topology structure generation method and system |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113810286A (en) | 2021-12-17 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Chen et al. | The features, hardware, and architectures of data center networks: A survey | |
| CN105706404B (en) | Method and apparatus for managing direct interconnect switch wiring and growth for computer networks | |
| CN110474848B (en) | System and method for supporting efficient load balancing in a high performance computing environment | |
| US10250958B2 (en) | Optical network-on-chip, optical router, and signal transmission method | |
| EP2549388A1 (en) | Computer system | |
| CN102780628B (en) | On-chip interconnection network routing method oriented to multi-core microprocessor | |
| US11537443B2 (en) | Reconfigurable computing pods using optical networks | |
| US9871749B2 (en) | Switch assembly | |
| JP6436394B2 (en) | Computing system framework having a network switching structure incorporating integrated storage, processing, and network switches, and methods of creating and using them | |
| CN116506348A (en) | Communication network system, message forwarding method and message forwarding device | |
| CN113810286B (en) | Computer network system and routing method | |
| CN106209294B (en) | A kind of full light interconnection network system of data center and communication means of high extension | |
| CN112217652A (en) | Network topology device and method based on central communication mode | |
| JP6737747B2 (en) | MPP network, MPP network construction method, MPP network design apparatus, and MPP network design method | |
| CN106161169A (en) | A kind of multi-host network exchange system | |
| CN214101389U (en) | Interconnection system and interconnection module | |
| CN112579510B (en) | Chip cluster | |
| CN110825689B (en) | Electronic chip and implementation method thereof | |
| Inoguchi et al. | Shifted recursive torus interconnection for high performance computing | |
| CN109992060B (en) | Stacked multi-path server system and server | |
| CN105207823B (en) | A kind of network topology structure | |
| CN102090027B (en) | Hierarchical passive networks | |
| CN220325833U (en) | 5G network equipment based on link protection of D-RAN access | |
| CN108768897B (en) | Port expansion device and stacking system | |
| JP4158803B2 (en) | Data transfer network and operation method of data transfer network |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |