Network scheduling method and apparatus, computer device, storage medium and program product

Abstract

Embodiments of the present application provide a network scheduling method and device, computer equipment, storage medium and program product, relate to the technical field of network scheduling, and can realize topology perception of distributed training, so that the traffic transmission path can be adjusted in time, thereby the communication bandwidth utilization in network communication can be improved, the network communication quality can be improved, and the efficiency of distributed training can be further improved. The method comprises the following steps: obtaining network demand information of a target tenant server and topology information of a network architecture to which the target tenant server belongs; determining a target communication path of the target tenant server according to the network demand information and the topology information; determining graphics card configuration information corresponding to a target cloud host server according to the target communication path; and connecting the target tenant server to the target cloud host server according to the graphics card configuration information.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of network scheduling, and in particular to a network scheduling method and device, computer equipment, a storage medium and a program product. BACKGROUND

[0002] In an era of popular large model applications, as the complexity of deep learning models increases, the number of model parameters also increases rapidly. To provide more accurate predictions, these models often need to be trained on a large amount of data, which requires powerful computing power to support the training of these large model applications. However, due to manufacturing process limitations, the development speed of artificial intelligence chips lags far behind the computing power required by the model. In order to solve the computing power challenge faced by large models, people use distributed training, that is, use multiple machines and calculations to cooperate in training, thereby improving computing power and training speed.

[0003] However, in a distributed training task, each graphics card calculation needs to frequently synchronize the parameters and gradients of the model training, thereby ensuring the consistency and accuracy of the model. Therefore, as the size of the distributed training system increases, the communication overhead accounts for a higher and higher proportion of the entire training overhead, thereby becoming a performance bottleneck of large-scale distributed training.

[0004] However, in traditional network communication, the underlying collection communication library cannot directly perceive the actual physical networking topology in the network architecture, thereby making it difficult to adjust the traffic transmission path in a timely manner, further easily causing a reduction in communication bandwidth utilization, and thereby affecting the efficiency of distributed training. SUMMARY

[0005] Therefore, it is necessary to provide a network scheduling method, device, computer equipment, storage medium and program product, which can realize topology perception of distributed training, thereby adjusting the traffic transmission path in a timely manner, thereby improving the communication bandwidth utilization in network communication, improving the network communication quality, and further improving the efficiency of distributed training.

[0006] In a first aspect, a network scheduling method is provided, comprising:

[0007] obtaining network requirement information of a target tenant server and topology information of a network architecture to which the target tenant server belongs;

[0008] determining a target communication path of the target tenant server according to the network requirement information and the topology information;

[0009] determining graphics card configuration information corresponding to a target cloud host server according to the target communication path;

[0010] According to the graphics card configuration information, the target tenant server is accessed to the target cloud host server.

[0011] In some embodiments, the network requirement information includes at least one of a number of graphics cards required by the target tenant server and a target communication bandwidth required by the target tenant server.

[0012] In some embodiments, the determining, according to the target communication path, of the graphics card configuration information corresponding to the target cloud host server comprises:

[0013] According to the target communication path, the number of target cloud host servers is determined;

[0014] According to the number of target cloud host servers, the graphics card configuration information corresponding to the target cloud host server is determined;

[0015] The graphics card configuration information includes single-machine graphics card configuration information and multi-machine graphics card configuration information. The single-machine graphics card configuration information includes a target configuration graphics card in one target cloud host server that is communicatively connected to the target tenant server. The multi-machine graphics card configuration information includes target configuration graphics cards in at least two target cloud host servers that are communicatively connected to the target tenant server, target connection graphics cards for communicatively connecting different target cloud host servers, and a connection relationship between the target connection graphics cards.

[0016] In some embodiments, the accessing, according to the graphics card configuration information, of the target tenant server to the target cloud host server comprises:

[0017] In the case where the target cloud host server includes one, the target tenant server is accessed to the target cloud host server through the target configuration graphics card according to the single-machine graphics card configuration information.

[0018] In the case where the target cloud host server includes at least two, the target tenant server is accessed to different target cloud host servers through the target configuration graphics cards according to the multi-machine graphics card configuration information, and at least two target cloud host servers are communicatively connected according to the connection relationship through the target connection graphics cards.

[0019] In some embodiments, the determining, according to the network requirement information and the topology information, of the target communication path of the target tenant server comprises:

[0020] According to the network requirement information and the topology information, at least one initial communication path is determined.

[0021] The switch hop count and network communication bandwidth of the initial communication path are obtained.

[0022] determining the target communication path according to at least one of the switch hop count and the network communication bandwidth.

[0023] In some embodiments, before the step of obtaining the network demand information of the target tenant server and the topology information of the network architecture to which the target tenant server belongs, the method further comprises:

[0024] obtaining original topology information of the network architecture of the target tenant server;

[0025] monitoring a running state of the network architecture;

[0026] correcting the original topology information according to the running state to obtain the topology information of the network architecture to which the target tenant server belongs;

[0027] The network scheduling method further comprises:

[0028] generating alarm information according to the network link position where the abnormality occurs and the abnormal state in the case that the running state of the network architecture is abnormal.

[0029] A second aspect of the embodiment of the application provides a network scheduling device, comprising:

[0030] an information obtaining module configured to obtain network demand information of a target tenant server and topology information of a network architecture to which the target tenant server belongs;

[0031] a first determining module configured to determine a target communication path of the target tenant server according to the network demand information and the topology information;

[0032] a second determining module configured to determine graphics card configuration information corresponding to a target cloud host server according to the target communication path;

[0033] an access control module configured to access the target tenant server to the target cloud host server according to the graphics card configuration information.

[0034] A third aspect of the embodiment of the application provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the following steps when executing the computer program:

[0035] obtaining network demand information of a target tenant server and topology information of a network architecture to which the target tenant server belongs;

[0036] determining a target communication path of the target tenant server according to the network demand information and the topology information;

[0037] According to the target communication path, determine graphic card configuration information corresponding to a target cloud host server;

[0038] According to the graphic card configuration information, access the target tenant server to the target cloud host server.

[0039] A fourth aspect of the embodiment of the present application provides a computer readable storage medium, which stores a computer program, and the computer program is executed by a processor to implement the following steps:

[0040] Obtain network requirement information of a target tenant server and topology information of a network architecture to which the target tenant server belongs;

[0041] According to the network requirement information and the topology information, determine a target communication path of the target tenant server;

[0042] According to the target communication path, determine graphic card configuration information corresponding to a target cloud host server;

[0043] According to the graphic card configuration information, access the target tenant server to the target cloud host server.

[0044] A fifth aspect of the embodiment of the present application provides a computer program product, which comprises a computer program, and the computer program is executed by a processor to implement the following steps:

[0045] Obtain network requirement information of a target tenant server and topology information of a network architecture to which the target tenant server belongs;

[0046] According to the network requirement information and the topology information, determine a target communication path of the target tenant server;

[0047] According to the target communication path, determine graphic card configuration information corresponding to a target cloud host server;

[0048] According to the graphic card configuration information, access the target tenant server to the target cloud host server.

[0049] The network scheduling method, device, computer device, storage medium and program product provided by the embodiments of the present application can further determine the target communication path of the target tenant server accessing the communication network according to the network demand information and the topology information of the network architecture, determine the target cloud host server in the target communication path that is in communication connection with the target tenant server by first determining the target communication path of the target tenant server, and further determine the graphics card configuration information in each target cloud host server, implement decomposition of the target communication path, so that the target cloud host server can directly optimize the access quality of the target tenant server according to the graphics card configuration information, and further implement topology perception of the target host server on the network architecture. Therefore, the network scheduling method provided by the embodiments of the present application can implement topology perception of distributed training, so that the traffic transmission path can be adjusted in time, and then the communication bandwidth utilization in network communication can be improved, the network communication quality can be improved, and further the efficiency of distributed training can be improved. BRIEF DESCRIPTION OF DRAWINGS

[0050] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the related art, the drawings needed to be used in the embodiments or the related art description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0051] Figure 1 A schematic flow chart of a network scheduling method provided by the embodiments of the present application is shown in the figure.

[0052] Figure 2 A schematic structural diagram of a target communication path of a network scheduling method provided by the embodiments of the present application is shown in the figure.

[0053] Figure 3 A schematic structural diagram of a target communication path of a network scheduling method provided by the embodiments of the present application is shown in the figure.

[0054] Figure 4 A schematic structural diagram of a target configuration graphics card of a network scheduling method provided by the embodiments of the present application is shown in the figure.

[0055] Figure 5 A target cloud host server connection schematic diagram of a network scheduling method provided by the embodiments of the present application is shown in the figure.

[0056] Figure 6 A target cloud host server connection schematic diagram of another network scheduling method provided by the embodiments of the present application is shown in the figure.

[0057] Figure 7 A target cloud host server connection diagram of another network scheduling method provided by an embodiment of the present application;

[0058] Figure 8 A target cloud host server connection diagram of another network scheduling method provided by an embodiment of the present application;

[0059] Figure 9 A schematic structural diagram of a network scheduling device provided by an embodiment of the present application;

[0060] Figure 10 A schematic structural diagram of a computer device provided by an embodiment of the present application;

[0061] Figure 11 A schematic structural diagram of a computer readable storage medium provided by an embodiment of the present application;

[0062] Figure 12 A schematic structural diagram of a computer program product provided by an embodiment of the present application. DETAILED DESCRIPTION

[0063] In order to make the purpose, technical solutions and advantages of the present application clearer, the present application will be further described in detail below in combination with the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and do not limit the present application.

[0064] In the era of popular large model applications, as the complexity of deep learning models increases, the number of model parameters also grows rapidly. In addition, in order to provide more accurate predictions, these models need to be trained on a large amount of data, which requires powerful computing power to support the training of these large model applications. Taking OpenAI's GPT (Generative Pre-trained Transformer) as an example, the GPT-3 model has 175 billion parameters, which is 116 times the number of GPT-2 parameters. The training of GPT-3 uses one thousand Nvidia A100 graphics cards and takes several weeks. The iteration of GPT-4 has increased the number of parameters by 10 times.

[0065] However, the development speed of artificial intelligence chips lags far behind the computing power required by models due to manufacturing processes. In order to solve the computing power challenge faced by large models, people use distributed training, that is, use multiple machines and GPGPUs (General-Purpose on Graphics Computing Unit) to cooperate in training, so as to improve computing power and training speed. Among them, the commonly used distributed training frameworks are PyTorch and TensorFlow, and the bottom layer realizes the communication operation between multiple computing GPGPUs in a single or multiple nodes by calling a collection communication library such as NCCL of Nvidia. In the distributed training task, each graphics card computing GPGPU needs to frequently synchronize the parameters and gradients of the model training, so as to ensure the consistency and accuracy of the model, and therefore the communication overhead accounts for a higher and higher proportion in the entire training overhead as the scale of the distributed training system increases, thereby becoming the performance bottleneck of large-scale distributed training. In the intelligent computing scene, cloud computing as a provider of computing power, too much communication consumption will also cause the blocking of computing power and the waste of computing resources. The underlying collection communication library cannot perceive the actual physical networking topology, and the networking topology with too many hops will greatly reduce the network bandwidth and the training efficiency. The distributed training method taking NCCL collection communication library as an example has the following shortcomings in the intelligent computing scene: first, NCCL itself does not perceive the physical networking topology, that is, in the intelligent computing networking, the communication scene across the core switch may occur, affecting the utilization rate of network bandwidth. Secondly, NCCL uses NBT (Double Binary Tree) structure in multiple scene, but only maintains the local topology information of the graphics card (such as the parent node and child node of the GPGPU on the binary tree) in the node, and only has node granularity topology information between nodes, lacking GPGPU granularity global topology information, and cannot perceive the actual networking information and link.

[0066] Therefore, it is currently a technical problem to be solved to provide a network scheduling method that can realize topology perception of network distributed training.

[0067] As shown in Figure 1 The first aspect of the embodiments of the present application provides a network scheduling method, comprising:

[0068] Step S110, obtaining network demand information of a target tenant server and topology information of a network architecture to which the target tenant server belongs.

[0069] Exemplarily, the topology information of the network architecture to which the target tenant server belongs can be determined according to initial deployment of the network architecture. The topology information can be physical networking information of the network architecture, and specifically can include positions, connection relationships of access switches and core switches in the network architecture, physical positions of each cloud host server, and connection relationships between the cloud host server and the core switch.

[0070] In step S120, a target communication path of the target tenant server is determined according to the network requirement information and the topology information.

[0071] Exemplarily, based on the network requirement information and the topology information, a combination of all cloud host servers meeting the requirements of the target tenant server can be determined, and further according to the combination of the cloud host servers, an access mode of a switch between the target tenant server and the cloud host server is determined, so that the target communication path can be determined.

[0072] In step S130, graphic card configuration information corresponding to the target cloud host server is determined according to the target communication path.

[0073] As shown in Figure 2 and Figure 3 , it can be seen that, Figure 3 The complexity and path length of the target communication path provided are obviously smaller than Figure 2 the initial communication path provided.

[0074] Exemplarily, according to the target communication path, the connection conditions of each graphic card in the target cloud host server can be determined. After the target communication path is determined, the target cloud host server in the target communication path can be assigned a corresponding NodeRank serial number, so as to facilitate identification and scheduling in the subsequent distributed training process.

[0075] In step S140, the target tenant server is accessed to the target cloud host server according to the graphic card configuration information.

[0076] The network scheduling method, device, computer device, storage medium and program product provided by the embodiments of the present application can further determine the target communication path of the target tenant server accessing the communication network according to the network requirement information and the topology information of the network architecture, determine the target cloud host server in the target communication path that is in communication connection with the target tenant server by first determining the target communication path of the target tenant server, further determine the graphics card configuration information in each target cloud host server, realize decomposition of the target communication path, and thus the graphics card configuration information can be distributed to the corresponding target cloud host server, so that the target cloud host server can directly optimize the access quality of the target tenant server according to the graphics card configuration information, and further realize topology perception of the target host server on the network architecture. Therefore, the network scheduling method provided by the embodiments of the present application can realize topology perception of distributed training, so that the traffic transmission path can be adjusted in time, and thus the communication bandwidth utilization in network communication can be improved, the network communication quality can be improved, and further the efficiency of distributed training can be improved.

[0077] In some possible embodiments, the network requirement information includes at least one of a required number of graphics cards of the target tenant server and a required target communication bandwidth of the target tenant server.

[0078] For example, at least one cloud host server with a number of graphics cards greater than or equal to the required number of graphics cards of the target tenant server can be determined according to the required number of graphics cards of the target tenant server.

[0079] For example, at least one cloud host server with a graphics card bandwidth greater than or equal to the required target communication bandwidth of the target tenant server can be determined according to the required target communication bandwidth of the target tenant server.

[0080] The network scheduling method provided by the embodiments of the present application can determine the target communication path through the required number of graphics cards and the required target communication bandwidth of the target tenant server, further improve the network access quality of the target tenant server, improve the network communication quality, guarantee the efficiency and stability of network communication, and further improve the efficiency of distributed training.

[0081] In some possible implementation manners, according to the target communication path, the GPU configuration information corresponding to the target cloud host server is determined, including: according to the target communication path, the number of target cloud host servers is determined; according to the number of target cloud host servers, the GPU configuration information corresponding to the target cloud host server is determined; wherein the GPU configuration information includes single-machine GPU configuration information and multi-machine GPU configuration information, the single-machine GPU configuration information includes a target configuration GPU in one target cloud host server that is in communication connection with the target tenant server, and the multi-machine GPU configuration information includes target configuration GPUs in at least two target cloud host servers that are in communication connection with the target tenant server, target connection GPUs for communication connection between different target cloud host servers, and a connection relationship between the target connection GPUs.

[0082] Exemplarily, the target transmission path can be transmitted in the form of a parameter or a configuration file, and can be transmitted to an MPI framework and a PyTorch framework and the like. For a parallel MPI framework, the serial number of the target cloud host server can be bound to the corresponding node IP based on a rankfile rule as a configuration file, and the order of the host nodes can be controlled by running a distributed command mpirun on a Master node. For a training framework TensorFlow, the number of nodes nnodes of the entire distributed training task is defined, the information of the Master node is filled into master_addr and master_port parameters, based on the node order planned in advance, the corresponding node_rank is distributed to each node, and the training script of the node is executed. Further, the Local Rank and the Rank serial number of each GPU can be allocated according to the number of GPUs in each computing node and the node order, which respectively represent the serial number of the GPU in the computing node and the global serial number of the GPU in the entire distributed training task.

[0083] The network scheduling method provided by the embodiments of the present application can determine the GPU connection mode in each target cloud host server through the single-machine GPU configuration information and the multi-machine GPU configuration information, and determine the connection mode between different target cloud host servers when the target cloud host server includes multiple target cloud host servers, so that the topology awareness of the distributed training can be realized, the traffic transmission path can be adjusted in time, and the network communication quality can be further improved, and the efficiency of the distributed training can be improved.

[0084] In some feasible implementations, connecting the target tenant server to the target cloud host server based on the graphics card configuration information includes: when there is only one target cloud host server, connecting the target tenant server to the target cloud host server through the target configured graphics card based on the single-machine graphics card configuration information; when there are at least two target cloud host servers, connecting at least two target cloud host servers to each other through the target connected graphics card based on the connection relationship according to the multi-machine graphics card configuration information, and connecting the target tenant server to different target cloud host servers through the target configured graphics card.

[0085] like Figure 4 As shown, when there are four target cloud server servers, the target configuration graphics cards within each target cloud server are interconnected internally. For example... Figure 5 As shown, when there are four target cloud server servers, they are connected to each other using the first connection method. Figure 6 As shown, in Figure 5 In the first connection method shown, all graphics cards within each target cloud host server establish a communication connection. For example... Figure 7 As shown, when there are four target cloud server servers, they are connected to each other using the second connection method. Figure 8 As shown, in Figure 7 In the second connection method shown, all graphics cards within each target cloud host server establish a communication connection.

[0086] For example, a collection communication library can be used to establish a PCIe system topology on each target cloud host server based on the target cloud host server and its corresponding number of target graphics cards provided by the upper layer. A rank relationship is bound to each graphics card, and the optimal path and bandwidth between GPGPU nodes are found through topology searching. The graphics card topology within a single node is also maintained. In a multi-machine scenario, the collection communication library establishes a binary tree structure based on the number of host nodes, further connecting the graphics cards responsible for inter-host communication in each node, thus forming a multi-machine, multi-GPU topology.

[0087] The network scheduling method provided in this application embodiment can clarify the graphics card connection method in each target cloud host server by using single-machine graphics card configuration information and multi-machine graphics card configuration information. At the same time, when there are multiple target cloud host servers, it can determine the connection method between different target cloud host servers, thereby improving the convenience of network access, realizing topology awareness of the underlying communication library, and further realizing topology awareness of distributed training, so as to adjust the traffic transmission path in a timely manner, further improving network communication quality and increasing the efficiency of distributed training.

[0088] In some possible implementation manners, the target communication path of the target tenant server is determined according to the network demand information and the topology information, including: determining at least one initial communication path according to the network demand information and the topology information; obtaining switch hop counts and network communication bandwidths of the initial communication paths; and determining the target communication path according to at least one of the switch hop counts and the network communication bandwidths.

[0089] For example, in the process of determining the target communication path, the initial communication path that does not involve the core switch can be determined as the target communication path. It should be noted that, in the spine-leaf network architecture, if the network communication can be completed between the access layer switches, the switch hop counts of the communication path will be the smallest, so as to improve the communication speed and further improve the efficiency of the network communication. However, for the case that the network communication must pass through the core switch, for example, in the case that the number of the target cloud host servers determined according to the number of the graphics cards required by the target tenant server exceeds the number of the cloud host servers that can be connected by one access switch, or in the case that the distance between different graphics cards connected by the target tenant server is far and a large amount of cross-node communication is required, the target communication path usually has to pass through the core switch.

[0090] For example, in the process of determining the target communication path, the initial communication path that does not involve the core switch can be determined as the target communication path. It should be noted that, in the spine-leaf network architecture, if the network communication can be completed between the access layer switches, the switch hop counts of the communication path will be the smallest, so as to improve the communication speed and further improve the efficiency of the network communication. However, for the case that the network communication must pass through the core switch, for example, in the case that the number of the target cloud host servers determined according to the number of the graphics cards required by the target tenant server exceeds the number of the cloud host servers that can be connected by one access switch, or in the case that the distance between different graphics cards connected by the target tenant server is far and a large amount of cross-node communication is required, the target communication path usually has to pass through the core switch.

[0091] It should be noted that, in the case of determining the target communication path through the topology structure, there is a possibility that the target cloud host server fails. The failed cloud host server can cause the network environment accessed by the target tenant server to be poor, or even cause the target tenant server to be unable to access the network environment. Therefore, for the node that can fail, the node can be excluded from the topology information.

[0092] The network scheduling method provided by the embodiment of the application can further shorten the network transmission time of the target tenant server, improve the efficiency and stability of the network communication, and improve the network bandwidth accessed by the target tenant server, so as to guarantee the access quality of the target tenant server.

[0093] In some possible implementation manners, before the steps of acquiring the network requirement information of the target tenant server and the topology information of the network architecture to which the target tenant server belongs, the method further includes: acquiring original topology information of the network architecture to which the target tenant server belongs; monitoring a running state of the network architecture; and correcting the original topology information according to the running state to obtain the topology information of the network architecture to which the target tenant server belongs.

[0094] Exemplarily, taking NCCL as an example, the graphics card tree structure in a single node (presented in a linked list structure) and the node tree structure information between multiple nodes (presented in a double binary tree structure) are integrated to construct a graphics card tree structure between nodes, each tree node representing a graphics card node for communication between the host node and other host nodes, and containing all graphics card topology information in the node. The actual networking information is further persisted in a YAML form for feedback, and the implementation of the optimized networking is further verified.

[0095] The network scheduling method provided in the embodiments of the present application can facilitate correction of the topology information by reconstructing a global topology networking of a graphics card granularity, thereby further improving the accuracy of network scheduling and further improving the network access quality of the target tenant server.

[0096] In some possible implementation manners, the network scheduling method further includes: in a case where the running state of the network architecture is abnormal, generating alarm information according to the abnormal network link position and the abnormal state.

[0097] Exemplarily, the corrected actual networking information can be pushed to a Pushgateway component, data is collected and stored through a Prometheus ecosystem, and monitoring of the graphics card granularity on the entire communication link is implemented, thereby providing data support for fault diagnosis and rapid recovery.

[0098] The network scheduling method provided in the embodiments of the present application can facilitate network access maintenance by monitoring the running state of the network architecture and generating alarm information in a timely manner according to network abnormalities, can improve the stability and robustness of the distributed training task, and further can improve the security of network access.

[0099] It should be understood that although each step in the flowchart involved in the above-described embodiments is shown in sequence according to the arrow, the steps are not necessarily executed in the order indicated by the arrow. Unless otherwise specified herein, the execution of the steps is not strictly limited in sequence, and the steps can be executed in other orders. Moreover, at least some of the steps in the flowchart involved in the above-described embodiments can include multiple steps or multiple stages, which are not necessarily executed at the same time but can be executed at different times, and the execution order of the steps or stages is not necessarily sequential but can be alternately or alternately executed with at least part of other steps or stages.

[0100] Based on the same inventive concept, the embodiments of the present application also provide a memory management device for implementing the above-mentioned memory management method. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme described in the above method, so the specific limitations in one or more memory management device embodiments provided below can refer to the limitations of the memory management method described above, which will not be repeated here.

[0101] As shown in Figure 9 The second aspect of the embodiments of the present application provides a network scheduling device, which comprises:

[0102] An information acquisition module 100 is configured to acquire network demand information of a target tenant server and topology information of a network architecture to which the target tenant server belongs.

[0103] A first determination module 200 is configured to determine a target communication path of the target tenant server according to the network demand information and the topology information.

[0104] A second determination module 300 is configured to determine graphics card configuration information corresponding to a target cloud host server according to the target communication path.

[0105] An access control module 400 is configured to access the target tenant server to the target cloud host server according to the graphics card configuration information.

[0106] Each module in the above memory management device can be realized by software, hardware and their combinations in whole or in part. The above modules can be embedded in or independent of the processor in the computer device in hardware form, or stored in the memory in the computer device in software form, so as to be called and executed by the processor to perform the operations corresponding to the above modules.

[0107] As shown in Figure 10As shown, in a third aspect of this application, a computer device is provided. This computer device may be a device integrating multiple target applications, and the specific implementation process of a memory monitoring module in one of the target applications may be executed by the computer device. The computer device may be a server, and its internal structure diagram may be as shown below. Figure 10 As shown, this computer device includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operating system and computer programs stored in the non-volatile storage media. The database stores memory data. The I / O interfaces are used for exchanging information between the processor and external devices. The communication interface is used for communicating with external terminals via a network. When the computer program is executed by the processor, it implements a memory management method.

[0108] Those skilled in the art will understand that the structure shown in the figures is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figures, or may combine certain components, or may have different component arrangements.

[0109] like Figure 11 As shown, in a fourth aspect of this application, a computer-readable storage medium is provided, on which a computer program is stored, wherein the computer program, when executed by a processor, performs the following steps:

[0110] Obtain the network requirements information of the target tenant server and the topology information of the network architecture to which the target tenant server belongs;

[0111] Based on network demand information and topology information, determine the target communication path of the target tenant server;

[0112] Based on the target communication path, determine the graphics card configuration information corresponding to the target cloud host server;

[0113] Based on the graphics card configuration information, connect the target tenant server to the target cloud host server.

[0114] like Figure 12 As shown, in a fifth aspect of this application, a computer program product is provided, including a computer program that, when executed by a processor, performs the following steps:

[0115] obtain network requirement information of the target tenant server and topology information of a network architecture to which the target tenant server belongs;

[0116] determine a target communication path of the target tenant server according to the network requirement information and the topology information;

[0117] determine graphics card configuration information corresponding to the target cloud host server according to the target communication path;

[0118] access the target tenant server to the target cloud host server according to the graphics card configuration information.

[0119] It should be noted that the information (including but not limited to application information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in the present application are all information and data fully authorized by all parties, and the collection, use and processing of related data need to comply with relevant regulations.

[0120] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer readable storage medium, and when the computer program is executed, the processes of the above-mentioned embodiments of the methods can be included. Any reference to memory, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical storage, high-density embedded non-volatile memory, resistive memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration but not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc. The database involved in the embodiments provided in the present application can include at least one of a relational database and a non-relational database. The non-relational database can include a distributed database based on a block chain, etc., without being limited thereto. The processor involved in the embodiments provided in the present application can be a general-purpose processor, a central processing unit, a graphics processing unit, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., without being limited thereto.

[0121] Any combination of the technical features of the above embodiments can be made. In order to make the description simple, all possible combinations of the technical features in the above embodiments are not described, however, as long as the combination of the technical features does not exist, it should be considered as the scope of the present application.

[0122] The above embodiments only express several implementation manners of the present application, and the description is more specific and detailed, but it should not be understood as a limitation on the scope of the patent of the present application. It should be pointed out that for ordinary skilled in the art, without departing from the concept of the present application, a number of modifications and improvements can be made, which are all within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the appended claims.

Claims

1. A network scheduling method, characterized by, The method comprises the following steps: obtaining network requirement information of a target tenant server and topology information of a network architecture to which the target tenant server belongs; determining a target communication path of the target tenant server according to the network requirement information and the topology information; determining a number of target cloud host servers according to the target communication path; determining graphics card configuration information corresponding to the target cloud host servers according to the number of the target cloud host servers; wherein the graphics card configuration information comprises single-machine graphics card configuration information and multi-machine graphics card configuration information, the single-machine graphics card configuration information comprises a target configuration graphics card in one of the target cloud host servers and connected in communication with the target tenant server, and the multi-machine graphics card configuration information comprises at least two target configuration graphics cards in different target cloud host servers, a target connection graphics card for communication connection between different target cloud host servers, and a connection relationship between the target connection graphics cards; accessing the target tenant server to the target cloud host servers according to the graphics card configuration information.

2. The network scheduling method of claim 1, wherein, The network requirement information comprises at least one of a required number of graphics cards of the target tenant server and a required target communication bandwidth of the target tenant server.

3. The network scheduling method of claim 1, wherein, The accessing the target tenant server to the target cloud host servers according to the graphics card configuration information comprises: in the case that the target cloud host servers comprise one, accessing the target tenant server to the target cloud host server through the target configuration graphics card according to the single-machine graphics card configuration information; in the case that the target cloud host servers comprise at least two, making at least two target cloud host servers in communication connection through the target connection graphics card according to the connection relationship and accessing the target tenant server to different target cloud host servers through the target configuration graphics card according to the multi-machine graphics card configuration information.

4. The network scheduling method of claim 1, wherein, The determining the target communication path of the target tenant server according to the network requirement information and the topology information comprises: determining at least one initial communication path according to the network requirement information and the topology information; obtaining switch hop counts and network communication bandwidths of the initial communication paths; determining the target communication path according to at least one of the switch hop counts and the network communication bandwidths.

5. The network scheduling method of claim 1, wherein, Before the steps of obtaining the network requirement information of the target tenant server and the topology information of the network architecture to which the target tenant server belongs, the method further comprises the following steps: obtaining original topology information of the network architecture to which the target tenant server belongs; monitoring a running state of the network architecture; correcting the original topology information according to the running state to obtain the topology information of the network architecture to which the target tenant server belongs. The network scheduling method further comprises the following steps: in the case that the running state of the network architecture is abnormal, generating alarm information according to a network link position at which the abnormality occurs and an abnormal state.

6. A network scheduling apparatus characterized by comprising: The method comprises the following steps: an information obtaining module, configured to obtain network requirement information of a target tenant server and topology information of a network architecture to which the target tenant server belongs; A first determining module is configured to determine a target communication path of the target tenant server according to the network requirement information and the topology information. A second determining module is configured to determine a number of target cloud host servers according to the target communication path, and determine corresponding graphics card configuration information of the target cloud host servers according to the number of target cloud host servers, wherein the graphics card configuration information comprises single-machine graphics card configuration information and multi-machine graphics card configuration information, the single-machine graphics card configuration information comprises a target configuration graphics card in one of the target cloud host servers and connected with the target tenant server, and the multi-machine graphics card configuration information comprises at least two target configuration graphics cards in the target cloud host servers and connected with the target tenant server, a target connection graphics card for communication connection between different target cloud host servers, and a connection relationship between the target connection graphics cards. An access control module is configured to access the target tenant server to the target cloud host servers according to the graphics card configuration information. 7.A computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the computer device is configured to perform the method according to any one of claims 1-6 when the computer program is executed by the processor. The processor executes the computer program to implement the steps of the network scheduling method in any one of claims 1 to 5.

8. A computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by the processor to implement the steps of the network scheduling method in any one of claims 1 to 5.

9. A computer program product comprising a computer program, characterized in that, The computer program is executed by the processor to implement the steps of the network scheduling method in any one of claims 1 to 5.

Images