Communication system, resource management method and device

Abstract

The embodiment of the invention provides a communication system and a resource management method and device.The communication system determines a service application to which each service flow belongs according to a pre-constructed application networking relationship, and determines bandwidth parameters of the service applications according to bandwidth information of the service flows belonging to the same service application, and adjusting a management and control strategy of the service application according to the bandwidth parameter of the service application and the performance parameter of the computing node running the service application. Real-time monitoring of service applications and computing nodes in a communication system is achieved, the network bandwidth congestion condition of the service applications and the busy degree of the computing nodes are sensed, and therefore the operation strategy of the service applications can be adjusted in a targeted mode, the AI service data center such as an intelligent computing center which is high in burst flow and needs to be frequently changed and adjusted can be effectively dealt with, and the service performance of the AI service data center is improved. And the network operation and maintenance efficiency and quality are improved, so that the resource utilization rate and the processing capability of the data center can be utilized to the maximum extent.

Description

Technical Field

[0001] One or more embodiments of the present specification relate to the field of communication technology, and in particular, to a communication system, a resource management method and an apparatus. Background Art

[0002] A data center (DC) is a facility that centrally stores, processes and distributes data. A data center interconnect (DCI) network is a network facility that connects multiple data centers. The DCI network is designed to achieve high-speed, reliable and secure communication between data centers.

[0003] In recent years, AI (Artificial Intelligence) technology has developed rapidly, and various AI large-scale model applications and software have emerged one after another. The network traffic of data centers has increased rapidly, causing tremendous pressure on data center network equipment. The characteristics of AI business applications are high burst traffic and very frequent changes and adjustments. Traditional network operation and maintenance strategies cannot effectively cope with them, resulting in frequent network congestion. Summary of the invention

[0004] In order to improve the operation and maintenance efficiency and quality of a data center and ensure stable operation of business applications, the embodiments of this specification provide a communication system, a resource management method and an apparatus.

[0005] In a first aspect, an embodiment of the present specification provides a communication system, the communication system comprising a data center, a data center network, and a management and operation platform, the data center is deployed with multiple business applications, each business application runs on a computing node of the data center, and the management and operation platform comprises an intelligent analysis platform and an intelligent management and control platform;

[0006] The intelligent analysis platform is used to obtain bandwidth information of each service flow in the data center network and performance parameters of each computing node included in the data center, and determine the service application to which each service flow belongs based on the pre-built application networking relationship, and determine the bandwidth parameters of the service application according to the bandwidth information of the service flows belonging to the same service application, wherein the application networking relationship includes the service flow transmission path of each service application;

[0007] The intelligent management and control platform is used to adjust the management and control strategy for running each business application according to the bandwidth parameter of the business application and the performance parameter of the computing node running the business application, wherein the management and control strategy includes: allocating network bandwidth to the business application when the bandwidth parameter reaches the set bandwidth, and adjusting the operating load of the computing node when the performance parameter reaches the performance threshold.

[0008] In some embodiments, the data center network includes a cloud cluster switch, and the cloud cluster switch includes multiple network nodes. Flow information of each business flow in the cloud cluster switch is collected through each network node. The flow information includes the traffic identifier of the business flow and the bandwidth information. The intelligent analysis platform receives the flow information of each business flow sent by the cloud cluster switch.

[0009] In some implementations, the multiple network nodes of the cloud cluster switch include an NCP node and an NCC node, the NCP node is used to collect the flow identifier of the service flow sent by the computing node, and establish a first flow table based on the flow identifier of the service flow, and send the first flow table to the NCC node;

[0010] The NCC node is used to count the traffic of each service flow according to the received first flow table to obtain bandwidth information of each service flow.

[0011] In some implementations, the NCP node is specifically used to, when the storage space is sufficient, establish a first flow table according to the flow identifier of the received service flow, and send the first flow table to the NCC node; when the storage space is insufficient, establish the first flow table according to the flow identifier of the received service flow of the preset message type, and send the first flow table to the NCC node;

[0012] The NCC node is specifically used to establish a second flow table according to the flow identifier of the service flow in the first flow table when there is sufficient storage space; and to establish a second flow table according to the flow identifier of the service flow of the preset message type in the first flow table when there is insufficient storage space, and to count the flow of each service flow in the second flow table to obtain the broadband information of each service flow.

[0013] In some embodiments, the intelligent analysis platform is specifically used to:

[0014] Obtaining the path port number corresponding to each business flow, and matching the path port number with the transport layer port number corresponding to each business application in the application networking relationship, to determine the business application to which each business flow belongs;

[0015] The bandwidth information of the service flows belonging to the same service application is summed to obtain the bandwidth parameter corresponding to each service application.

[0016] In some embodiments, the management and operation platform further includes a computing management platform, which is used to collect performance parameters of each computing node and send the performance parameters of each computing node to the intelligent analysis platform;

[0017] Wherein, in the case where the computing node is a GPU computing node, the performance parameters of the GPU computing node include floating-point computing performance and GPU memory bandwidth.

[0018] In some implementations, the intelligent analysis platform is further used to obtain packet loss information of each service flow in the data center network, and determine the packet loss rate of the service application based on the packet loss information of the service flows belonging to the same service application;

[0019] The intelligent management and control platform is also used to adjust the management and control strategy for running the business application according to the bandwidth parameters of the business application, the packet loss rate and the performance parameters of the computing node running the business application, wherein the management and control strategy includes: when the packet loss rate reaches a preset threshold, adjusting the message forwarding strategy of the business application.

[0020] In some embodiments, the management and operation platform also includes an intelligent deployment platform, which is used to obtain all business applications deployed on the data center and the transport layer port number of each business application, wherein the transport layer port number includes the path port number of each business flow included in the business application. The business application is used as a node and the transmission path of the business flow is used as an edge. The application networking relationship is generated according to the path port number of the business flow of each business application.

[0021] In a second aspect, an embodiment of the present specification provides a resource management method, which is applied to a communication system, wherein the communication system includes a data center and a data center network, wherein the data center is deployed with multiple business applications, and the business applications are run on computing nodes of the data center, wherein the method includes:

[0022] Obtaining bandwidth information of each service flow in the data center network and performance parameters of each computing node included in the data center;

[0023] Based on the pre-built application networking relationship, determine the service application to which each service flow belongs, and determine the bandwidth parameter of the service application according to the bandwidth information of the service flow belonging to the same service application, wherein the application networking relationship includes the service flow transmission path of each service application;

[0024] For each business application, a management and control strategy for running the business application is adjusted according to the bandwidth parameter of the business application and the performance parameter of the computing node running the business application, wherein the management and control strategy includes: allocating network bandwidth to the business application when the bandwidth parameter reaches a set bandwidth, and adjusting the operating load of the computing node when the performance parameter reaches a performance threshold.

[0025] In some implementations, the data center network includes a cloud cluster switch, the cloud cluster switch includes a plurality of network nodes, and obtaining bandwidth information of each service flow in the data center network includes:

[0026] The flow information of each service flow in the cloud cluster switch is collected through each network node, and the flow information includes a flow identifier of the service flow and the bandwidth information.

[0027] In some implementations, the multiple network nodes of the cloud cluster switch include an NCP node and an NCC node, and collecting flow information of each service flow in the cloud cluster switch through each network node includes:

[0028] The NCP node collects the flow identifier of the service flow sent by the computing node, establishes a first flow table based on the flow identifier of the service flow, and sends the first flow table to the NCC node;

[0029] The NCC node counts the traffic of each service flow according to the received first flow table to obtain bandwidth information of each service flow.

[0030] In some implementations, the NCP node collects the flow identifier of the service flow sent by the computing node, establishes a first flow table based on the flow identifier of the service flow, and sends the first flow table to the NCC node, including:

[0031] When the storage space is sufficient, the NCP node establishes a first flow table according to the flow identifier of the received service flow, and sends the first flow table to the NCC node; when the storage space is insufficient, the NCP node establishes a first flow table according to the flow identifier of the received service flow of the preset message type, and sends the first flow table to the NCC node;

[0032] The NCC node counts the traffic of each service flow according to the received first flow table, and obtains the bandwidth information of each service flow, including:

[0033] When the storage space is sufficient, the NCC node establishes a second flow table according to the flow identifier of the service flow in the first flow table. When the storage space is insufficient, the NCC node establishes a second flow table according to the flow identifier of the service flow of the preset message type in the first flow table, counts the flow of each service flow in the second flow table, and obtains the broadband information of each service flow.

[0034] In some implementations, determining the service application to which each service flow belongs based on the pre-built application networking relationship, and determining the bandwidth parameter of the service application according to the bandwidth information of the service flows belonging to the same service application, includes:

[0035] Obtaining the path port number corresponding to each business flow, and matching the path port number with the transport layer port number corresponding to each business application in the application networking relationship, to determine the business application to which each business flow belongs;

[0036] The bandwidth information of the service flows belonging to the same service application is summed to obtain the bandwidth parameter corresponding to each service application.

[0037] In some implementations, the data center includes a first computing node. When the first computing node is a GPU computing node, the process of obtaining the performance parameter of the GPU computing node includes:

[0038] The floating-point computing performance and the GPU memory bandwidth of the GPU computing node are obtained, wherein the performance parameters of the GPU computing node include the floating-point computing performance and the GPU memory bandwidth.

[0039] In some implementations, adjusting the management and control strategy for running the business application according to the bandwidth parameter of the business application and the performance parameter of the computing node running the business application includes:

[0040] Determining a packet loss rate of a service application according to packet loss information of a service flow belonging to the same service application;

[0041] According to the bandwidth parameter of the business application, the packet loss rate and the performance parameter of the computing node running the business application, the control strategy for running the business application is adjusted, wherein the control strategy includes: when the packet loss rate reaches a preset threshold, adjusting the message forwarding strategy of the business application.

[0042] In some implementations, the process of pre-building the application networking relationship includes:

[0043] Obtaining all business applications deployed on the data center and the transport layer port number of each business application, wherein the transport layer port number includes the path port number of each business flow included in the business application;

[0044] The application networking relationship is generated by taking the business application as a node and the transmission path of the business flow as an edge, and according to the path port number of the business flow of each business application.

[0045] In a third aspect, an embodiment of the present specification provides a resource management device, which is applied to a communication system, wherein the communication system includes a data center and a data center network, wherein the data center is deployed with multiple business applications, and the business applications run on computing nodes of the data center, wherein the device includes:

[0046] An information acquisition module, used to acquire bandwidth information of each service flow in the data center network and performance parameters of each computing node included in the data center;

[0047] A bandwidth determination module, used to determine the service application to which each service flow belongs based on a pre-built application networking relationship, and determine the bandwidth parameters of the service application according to the bandwidth information of the service flows belonging to the same service application, wherein the application networking relationship includes the service flow transmission path of each service application;

[0048] A policy management module is used to adjust the management and control policy for running each business application according to the bandwidth parameter of the business application and the performance parameter of the computing node running the business application, wherein the management and control policy includes: allocating network bandwidth to the business application when the bandwidth parameter reaches the set bandwidth, and adjusting the running load of the computing node when the performance parameter reaches the performance threshold.

[0049] In some embodiments, the data center network includes a cloud cluster switch, and the cloud cluster switch includes multiple network nodes. The information acquisition module is specifically used to collect flow information of each business flow in the cloud cluster switch through each network node, and the flow information includes the traffic identifier of the business flow and the bandwidth information.

[0050] In some implementations, the plurality of network nodes of the cloud cluster switch include NCP nodes and NCC nodes, and the information acquisition module is specifically configured to:

[0051] The NCP node collects the flow identifier of the service flow sent by the computing node, establishes a first flow table based on the flow identifier of the service flow, and sends the first flow table to the NCC node;

[0052] The NCC node counts the traffic of each service flow according to the received first flow table to obtain bandwidth information of each service flow.

[0053] In some implementations, the information acquisition module is specifically used to:

[0054] When the storage space of the NCP node is sufficient, a first flow table is established according to the flow identifier of the received service flow, and the first flow table is sent to the NCC node; when the storage space is insufficient, a first flow table is established according to the flow identifier of the received service flow of the preset message type, and the first flow table is sent to the NCC node;

[0055] When the NCC node has sufficient storage space, a second flow table is established according to the traffic identifier of the service flow in the first flow table. When the storage space is insufficient, a second flow table is established according to the traffic identifier of the service flow of the preset message type in the first flow table, and the traffic of each service flow in the second flow table is counted to obtain the broadband information of each service flow.

[0056] In some implementations, the bandwidth determination module is specifically configured to:

[0057] Obtaining the path port number corresponding to each business flow, and matching the path port number with the transport layer port number corresponding to each business application in the application networking relationship, to determine the business application to which each business flow belongs;

[0058] The bandwidth information of the service flows belonging to the same service application is summed to obtain the bandwidth parameter corresponding to each service application.

[0059] In some embodiments, the data center includes a first computing node. When the first computing node is a GPU computing node, the information acquisition module is specifically used to obtain the floating-point computing performance and GPU memory bandwidth of the GPU computing node. The performance parameters of the GPU computing node include the floating-point computing performance and the GPU memory bandwidth.

[0060] In some implementations, the policy control module is specifically used to:

[0061] Determining a packet loss rate of a service application according to packet loss information of a service flow belonging to the same service application;

[0062] According to the bandwidth parameter of the business application, the packet loss rate and the performance parameter of the computing node running the business application, the control strategy for running the business application is adjusted, wherein the control strategy includes: when the packet loss rate reaches a preset threshold, adjusting the message forwarding strategy of the business application.

[0063] In some embodiments, the apparatus further comprises a relationship building module, wherein the relationship building module is configured to:

[0064] Obtaining all business applications deployed on the data center and the transport layer port number of each business application, wherein the transport layer port number includes the path port number of each business flow included in the business application;

[0065] The application networking relationship is generated by taking the business application as a node and the transmission path of the business flow as an edge, and according to the path port number of the business flow of each business application.

[0066] The communication system of the implementation mode of this specification determines the business application to which each business flow belongs based on the pre-built application networking relationship, determines the bandwidth parameters of the business application based on the bandwidth information of the business flow belonging to the same business application, and adjusts the management and control strategy of the business application based on the bandwidth parameters of the business application and the performance parameters of the computing node running the business application. It realizes real-time monitoring of business applications and computing nodes in the communication system, perceives the network bandwidth congestion of business applications and the busyness of computing nodes, so as to adjust the operation strategy of business applications in a targeted manner, and can effectively deal with AI business data centers such as intelligent computing centers with high burst traffic and frequent changes and adjustments, improve network operation and maintenance efficiency and quality, and maximize the resource utilization and processing capacity of data centers. BRIEF DESCRIPTION OF THE DRAWINGS

[0067] Figure 1 It is a logical architecture diagram of a communication system in an exemplary embodiment of this specification.

[0068] Figure 2 is a system architecture diagram of a communication system in an exemplary embodiment of this specification.

[0069] Figure 3 It is a schematic diagram of application networking relationships in an exemplary implementation of this specification.

[0070] Figure 4 is a flowchart of a resource management method in an exemplary embodiment of this specification.

[0071] Figure 5 is a structural block diagram of a resource management device in an exemplary embodiment of this specification. DETAILED DESCRIPTION

[0072] The technical solution of this specification will be described clearly and completely below in conjunction with the accompanying drawings. Obviously, the described implementation is a part of the implementation of this specification, not all of the implementation. Based on the implementation in this specification, all other implementations obtained by ordinary technicians in this field without creative work are within the scope of protection of this specification. In addition, the technical features involved in the different implementations of this specification described below can be combined with each other as long as they do not conflict with each other.

[0073] A data center (DC) is a facility that centrally stores, processes, and distributes data. It usually includes servers, storage devices, etc., and is used to provide various computing and storage services. A data center interconnect (DCI) network is a network facility that connects multiple data centers. The DCI network is designed to achieve high-speed, reliable, and secure communication between data centers to meet the high bandwidth and low latency requirements for data and applications. Data center networks include network transmission technology, fiber optic connections, routers, switches, and other equipment.

[0074] In recent years, AI (Artificial Intelligence) technology has developed rapidly, and various AI large-scale model applications and software have emerged in an endless stream. The computing capacity and network traffic of data centers have increased rapidly, which has put tremendous pressure on data center network equipment. Traditional network operation and maintenance management can no longer meet the requirements, causing network congestion. For example, the cloud service business provided by traditional data centers is relatively fixed, and the traffic of each business is relatively clear and controllable. Traffic conflicts generally occur during the expected business peak period, so network operation and maintenance can ensure the normal operation of the network by adjusting and changing the network according to the business peak. However, the characteristics of AI business applications are extremely high real-time requirements, and according to different user inputs, the computing capacity and communication volume of business applications have geometric differences, resulting in high business burst traffic and very frequent changes and adjustments. Traditional network operation and maintenance strategies cannot effectively cope with it, causing network congestion.

[0075] An intelligent computing center is a data center that focuses on intelligent data analysis, computing and processing. It has more powerful computing capabilities to meet complex needs such as high-performance computing and AI computing. At present, the business of an intelligent computing center generally includes two parts: one is the AI ​​business services provided to users, which complete real-time computing based on user requests, and the amount of computing and communication varies exponentially depending on user needs; the other is the learning and training of the AI ​​business itself, which often requires even greater computing and communication.

[0076] Figure 1 The logical architecture of the intelligent computing center system is shown below. Figure 1 Taking the intelligent computing center system shown as an example, the logical architecture of the communication system in this specification is explained.

[0077] like Figure 1 As shown in the figure, the bottom layer of the communication system is the hardware layer, which is also the data center. The hardware layer includes hardware that provides computing and storage capabilities, such as servers that provide traditional computing, servers that provide AI computing, and storage platforms. The hardware layer also includes computer rooms that provide power and cooling for the hardware, which are all hardware infrastructures of standardized modules.

[0078] The upper layer of the hardware layer is the network layer, which is also the data center network. The network layer refers to the DCI network facilities that provide communication for the intelligent computing center. For example, the network layer adopts a network system based on cloud cluster switches. The complex DCI network is deployed and managed through cloud cluster switches. To the outside, the entire network looks like one or several devices, greatly simplifying the complexity of the network.

[0079] The upper layer of the network layer is the management and operation platform, which is used to provide operation and management of various parts of the intelligent computing center. The management and operation platform of the traditional intelligent computing center includes the computing management platform, the network management platform and the storage management platform. The computing management platform generally uses virtualization to create multiple virtual machine instances on a physical server to achieve flexible allocation and management of computing resources. The network management platform refers to a platform used to manage the DIC network, such as the aforementioned DCI network based on cloud cluster switches. The storage management platform provides the intelligent computing center with high-performance, high-reliability and high-scalability storage solutions.

[0080] The top layer of the communication system is the business operation platform, which provides various business applications. The business applications of the intelligent computing center can include traditional cloud service businesses as well as AI application businesses. Business applications run on hardware-layer devices.

[0081] In the implementation manner of this specification, in addition to the above-mentioned computing management platform, network management platform and storage management platform, the management and operation platform is further deployed with an intelligent management and control platform, an intelligent deployment platform and an intelligent analysis platform. The intelligent management and control / deployment / analysis platform is used to implement the resource management method described below in this specification, to achieve real-time traffic monitoring, intelligent analysis and network policy adjustment for all business applications of the intelligent computing center, to determine the congestion level of business applications by sensing the traffic of business applications, and to intelligently allocate business resources to business applications, to ensure the normal development of business and maximize the utilization of data center capabilities.

[0082] In some implementations, this specification provides a resource management method, which can be applied to a communication system, such as Figure 1 The intelligent computing center shown in the figure may also be other data centers, and this specification does not limit this. Figure 1 Taking the intelligent computing center shown as an example, the resource management method implemented in this specification can be executed by the management and operation platform, and the real-time traffic collection, intelligent analysis, resource adjustment, etc. of the business applications of the intelligent computing center can be realized through the method of this specification.

[0083] Figure 2 The network architecture diagram of the communication system in some embodiments of this specification is shown below. Figure 2 Provide explanation.

[0084] See also Figure 2 As shown, the network layer of the communication system includes one or more cloud cluster switches, such as Figure 2 Two cloud cluster switches are shown, namely, cloud cluster switch 1 and cloud cluster switch 2. Taking any cloud cluster switch as an example, the cloud cluster switch includes multiple network nodes, such as a network cloud controller (NCC, Network Cloud Controller) node, a network cloud fabric (NCF, Network Cloud Fabric) node, and a network cloud packet forwarder (NCP, Network Cloud Packet-Forwarder) node, and the number of each network node can be multiple. Each of the above network nodes can be specifically an independent physical device, such as a switch, a white box switch, etc. Multiple network nodes are physically connected through a network management transmission (MGT, a three-layer switch, supporting common routing protocols, multicast protocols, etc.) network, and are aggregated into a cloud cluster switch, which appears to the outside as a network device.

[0085] In the cloud cluster switch, the connection between the NCC node, NCF node, NCP node, and MGT is represented as a physical connection. A cluster management network is established between each node through MGT. The cluster management network can use the LIPC protocol. The LIPC protocol is in the same position as the TCP / IP protocol stack, and is encapsulated through the Socket layer and provides a unified external interface to the business application. In some embodiments, the functions provided by the LIPC protocol to each node in the cloud cluster switch include but are not limited to: addressing function transparent to the protocol stack; topology management function; location transparent unicast; location opaque unicast; connection-based transport layer reliable unicast; multicast group management function; location transparent multicast; connection-based transport layer reliable multicast; non-connection-based unicast datagram transmission, etc.

[0086] Combination Figure 2 As shown, in the implementation mode of this specification, traffic collection in a cloud cluster switch is implemented based on chip capabilities, for example Figure 2As shown, a flow collector 400 can be set in each cloud cluster switch. The function of the flow collector 400 is to collect the flow information of each business flow in the cloud cluster switch, and send the collected flow information of the business flow to the intelligent analysis platform in the form of a flow table. The intelligent analysis platform analyzes the operation status of each business application based on the collected flow information. The operation status of the business application can include two aspects: one is whether the operating bandwidth of the business application reaches the upper limit of the allocated bandwidth, and the other is whether the performance of the server device running the business application reaches the upper limit. The intelligent analysis platform can send the analysis results to the intelligent management and control platform. The intelligent management and control platform adjusts different management and control strategies for the business application based on different analysis results to ensure that each business application can run smoothly. Specifically, the flow collector 400 can be deployed in each physical device of the cloud cluster switch, such as in the NCP node and the NCC node, so as to collect the corresponding flow information for each business flow reached.

[0087] The asset management method of the embodiment of this specification can be applied to a communication system, which includes a data center and a data center network. A data center can include one or more computing nodes, and a computing node can refer to one physical server or multiple physical servers. Figure 2 Taking the intelligent computing center system shown as an example, the data center is the hardware layer, which may include multiple computing nodes, such as cloud server computing nodes, GPU (graphics processing unit, graphics processor) computing nodes, etc.

[0088] The data center is deployed with multiple business applications, which run on computing nodes. A computing node can run one business application or multiple business applications, and a business application can also run on multiple computing nodes. Business applications can include traditional cloud service businesses and AI businesses, etc., which are not limited in this specification.

[0089] The data center network is the aforementioned network layer, for example Figure 2 In the example, the data center network refers to the cloud cluster switch. The business applications running on the data center forward messages through the data center network to achieve business communication between data centers. Business flow refers to the data flow of messages sent and received by business applications. The transmission path of a business flow includes the egress port and ingress port of the traffic, which defines the source and destination of the business flow. A business application may include one business flow path or multiple business flow paths.

[0090] It can be understood that in the data center network, what is collected by the aforementioned traffic collector 400 is the flow information of each business flow. In order to realize the correspondence between the business flow and the business application, in the implementation mode of this specification, the intelligent deployment platform can be used to pre-build the application networking relationship, and the application networking relationship can be used to establish an association between the business application and the collected business flow, so as to facilitate the subsequent intelligent analysis platform to analyze the operation status of each business application. The process of building the application networking relationship is explained below.

[0091] Combination Figure 1 As shown, first, the intelligent deployment platform can obtain all business applications deployed on the entire communication system from the business operation platform, that is, the relevant information of each business application. The relevant information may include, for example, the computing node resources, storage resources, message identifiers, priorities, and other information occupied by the business application. The message identifier may include the IP address, Vxlan identifier, and transport layer port number of the business application. The IP address is used to identify the business application address, the Vxlan identifier is used to identify and distinguish the identifiers of different virtual networks, and the transport layer port number is the logical address used by the transport layer protocol (such as TCP, UDP) to distinguish different business applications. In the implementation of this specification, the transport layer port number may include the path port number of each business flow included in the business application, such as the source port number and destination port number of each business flow.

[0092] For example, in an example, the business application information collected by the intelligent deployment platform is shown in Table 1 below:

[0093] Table 1

[0094]

[0095]

[0096] For example, taking the first row in Table 1 as an example, the business application is "cloud business 1", the computing node resource it occupies is "virtual machine (VM) 1", the storage resource it occupies is "node 1", and the corresponding message identifier includes "IP address 1, Vxlan identifier 1, transport layer port number 1", and the priority parameter is "100". In the example in Table 1, the priority indicates the running priority of the business application, and the lower the priority parameter, the higher the corresponding priority.

[0097] According to the above, the relevant information of the business application obtained above includes the path port number of each business flow in the business application. The path port number can identify the transmission path of a certain business flow. Therefore, in the implementation manner of this specification, the business application can be used as a node and the transmission path of the business flow as an edge to construct an application networking relationship used to represent the transmission path of each business application.

[0098] For example Figure 3A schematic diagram showing the application networking relationship in some implementation modes of this specification is shown. Figure 3 In the example application networking relationship, each node represents a business application, and each edge represents a transmission path of a business flow. Figure 3 The application networking relationship shown can help you understand the service application to which each service flow belongs and the transmission path of the service flow.

[0099] In some implementations, the intelligent deployment platform can perform network deployment for each business application and design corresponding network forwarding strategies based on the above-mentioned application networking relationship, such as allocating appropriate network bandwidth for the business application, to achieve network deployment of the business application. For the network deployment process of the business application, those skilled in the art can understand and fully implement it in combination with relevant technologies, and this specification will not elaborate on it.

[0100] like Figure 4 As shown, in some implementations, the resource management method of the example in this specification includes:

[0101] S410: Obtain bandwidth information of each service flow in the data center network and performance parameters of each computing node included in the data center.

[0102] As can be seen from the foregoing, in the implementation of this specification, the monitoring of the operation of business applications mainly includes two aspects: one is whether the operation bandwidth of the business application has reached the upper limit, and the other is whether the performance of the computing node running the business application has reached the bottleneck. Therefore, in some implementations, during the operation of the business application, the bandwidth of the business application and the performance of the computing node can be detected respectively.

[0103] Specifically, combined Figure 2 As shown, each service flow in the network can be collected by a traffic collector 400 deployed in the data center network, bandwidth information of the service flow can be obtained by statistics, and the bandwidth information of the service flow can be sent to the intelligent analysis platform for analysis and processing.

[0104] exist Figure 2 In the example, the traffic collector 400 may be a processing chip deployed in an NCP node and an NCC node and specifically used for traffic information collection. During the process of network forwarding messages, the processing chip may collect flow information of each service flow in the network.

[0105] In the implementation manner of this specification, the flow information of the business flow may include the traffic identifier of the business flow. The traffic identifier may include different information depending on the message type. For example, taking TCP / UDP / Vxlan messages as an example, the traffic identifier may include five-tuple information, namely: source IP, destination IP, source port number, destination port number and protocol number; for example, taking RoCEv2 messages as an example, the traffic identifier may include: source IP, destination IP, destination QP (Destination Queue Pair, used to identify a RoCEv2 flow).

[0106] In some implementations, the flow information of the service flow may also include: flow start time, flow end time, message input port / output port, TCP SN, etc. The flow start time and flow end time can be used to identify the duration of a service flow. For TCP messages, the time when the SYN message is received is the flow start time, and the time when the FIN message is received is the flow end time; for UDP / RoCEv2 messages, the time when the first message is received is the flow start time, and the time when the last message is received is the flow end time; for Vxlan messages, what is collected is the Vxlan inner message, and the meaning of the field can be determined by the type of the inner message. The input port / output port of the message is the path port number described in this specification, which is used to record the transmission path of the service flow and realize the visibility of the transmission path. TCP SN represents the next TCP message sequence number, which is used to detect the packet loss of TCP messages. Of course, those skilled in the art can understand that the collected flow information of the service flow can also include more or fewer information types, and this implementation is not limited to this.

[0107] In the following, data center network is referred to as Figure 2 Taking the cloud cluster switch shown in the figure as an example, the process of collecting bandwidth information of business flows is described in detail.

[0108] Combination Figure 2 As shown, the NCP node can receive the business flow sent by the computing node, so that the traffic collector 400 in the NCP node can collect the traffic identifier of each business flow. The following business flow message type takes TCP message as an example, and the traffic identifier is a five-tuple information, which includes: source IP, destination IP, source port, destination port, and message protocol.

[0109] For NCP nodes, the forwarding of service traffic is performed by the forwarding chip of the NCP node. In the implementation mode of this specification, when the NCP node collects service flows, it is necessary to establish a flow table according to the memory storage space of the forwarding chip. The specific process is as follows:

[0110] When a service flow arrives at the NCP node, the current memory storage space of the NCP node needs to be judged. If the storage space is sufficient, it means that the current storage space can accommodate all service flow information, so that a flow table can be established for the received service flow. If the storage space is insufficient, it means that the current storage space is tight and cannot accommodate all service flow information, so a flow table can be established only for messages of a preset message type. The preset message type is, for example, a SYN / FIN message, which indicates a start and end message. In the case of insufficient storage space, a flow table can be established only for SYN / FIN messages.

[0111] In the implementation manner of this specification, the process of determining whether the memory storage space of the NCP node is sufficient may be: a critical threshold is set in advance for the memory of the NCP node, and if the memory usage of the NCP node reaches the critical threshold, it is determined that the storage space is insufficient, otherwise it is determined that the storage space is sufficient. The specific value of the critical threshold can be selected according to demand, and this specification does not limit this.

[0112] In the implementation manner of this specification, the flow table established by the NCP node is defined as the first flow table. After the NCP node creates the first flow table according to the aforementioned process, the first flow table can be sent to the NCC node. In some implementation manners, when the storage space of the NCP node is sufficient, the first flow table can be sent to the NCC node after waiting for the first flow table to age or reach the reporting period. When the storage space of the NCP node is insufficient, the first flow table can be sent to the NCC node immediately after the first flow table is established based on the SYN / FIN message.

[0113] For example, in an example, the first flow table established by the NCP node may be as shown in Table 2 below:

[0114] Table 2

[0115] Source IP Source Port Destination IP Destination Port protocol 1.1.1.1 8756 3.3.3.1 80 TCP 1.1.1.1 20024 3.3.3.1 80 TCP 1.1.1.1 21124 3.3.3.1 80 TCP 2.2.2.1 9265 4.4.4.1 21 TCP 2.2.2.1 15623 4.4.4.1 21 TCP 2.2.2.1 23542 4.4.4.1 21 TCP

[0116] In the example of Table 2, each row represents a service flow. The first flow table includes the traffic identifier of the service flow, that is, the five-tuple information of the service flow, which are source IP, source port, destination IP, destination port and message protocol.

[0117] Combination Figure 2As shown, after the NCC node receives the first flow table sent by the NCP node, the flow table creation process for the service flow is similar to that of the NCP node. Specifically, the NCC node determines the current memory storage space. If the storage space is sufficient, it means that the current storage space can accommodate all service flow information, so that the flow table can be established for all service flows in the received first flow table. If the storage space is insufficient, it means that the current storage space is tight and cannot accommodate all service flow information, so the flow table can be established only for the messages of the preset message type. The preset message type is, for example, SYN / FIN message. SYN / FIN message indicates the start and end message. In the case of insufficient storage space, the flow table can be established only for the SYN / FIN message.

[0118] In the implementation manner of this specification, the process of determining whether the memory storage space of the NCC node is sufficient may be: presetting a critical threshold for the memory of the NCC node, and if the memory usage of the NCC node reaches the critical threshold, determining that the storage space is insufficient, and vice versa, determining that the storage space is sufficient. The specific value of the critical threshold can be selected according to demand, and this specification does not limit this.

[0119] In the implementation manner of this specification, the flow table established by the NCC node is defined as the second flow table. After the NCC node creates the second flow table according to the aforementioned process, the second flow table can be sent to the analyzer module in the NCC node. In some implementation manners, when the storage space of the NCC node is sufficient, the second flow table can be sent to the analyzer module after waiting for the second flow table to age or reach the reporting period. When the storage space of the NCC node is insufficient, the second flow table can be sent to the analyzer module of the NCC node immediately after the second flow table is established based on the SYN / FIN message.

[0120] The analyzer module is a module in the NCC node for traffic statistics. By using the analyzer module to count the traffic of each service flow in the second flow table, the bandwidth information corresponding to each service flow can be obtained. In some embodiments of the present specification, the bandwidth information of the service flow can be represented by the amount of message data and the number of data packets transmitted per unit time, and the unit time can be, for example, 1 second. For example, in one example, the statistical results obtained by the analyzer module of the NCC node for performing traffic statistics on the service flow of the second flow table can be shown in the following Table 3:

[0121] Table 3

[0122] Source IP Source Port Destination IP Destination Port protocol Bandwidth Information 1.1.1.1 8756 3.3.3.1 80 TCP 88KB / 150Pkts 1.1.1.1 20024 3.3.3.1 80 TCP 188KB / 250Pkts 1.1.1.1 21124 3.3.3.1 80 TCP 488KB / 360Pkts 2.2.2.1 9265 4.4.4.1 21 TCP 1MB / 1100Pkts 2.2.2.1 15623 4.4.4.1 21 TCP 3MB / 3200Pkts 2.2.2.1 23542 4.4.4.1 21 TCP 2MB / 2018Pkts

[0123] In the example of Table 3, each row represents a service flow, and the bandwidth information represents the bandwidth information of the service flow. For example, taking the bandwidth information "88KB / 150Pkts" of the service flow in the first row as an example, it means that the amount of message data transmitted per unit time of this service flow is 88KB, and the number of data packets transmitted per unit time is 150.

[0124] Through the above process, the flow information of each business flow in the data center network can be collected. The flow information includes traffic identification and bandwidth information. Next, the NCC node can send the flow information of each business flow in the second flow table to the aforementioned intelligent analysis platform, and the intelligent analysis platform will analyze and process the business flow data.

[0125] At the same time, the computing management platform can obtain the performance parameters of each computing node in the data center, and then send the performance parameters of each computing node to the intelligent analysis platform. Figure 1 As shown, the computing management platform refers to a functional platform used to manage the computing resources of the data center. When the data center is running, the computing management platform can monitor the performance parameters of each computing node in real time, so that the computing management platform can obtain the performance parameters of each computing node and send the performance parameters of each computing node to the intelligent analysis platform.

[0126] It can be understood that the performance of a computing node refers to the ability of the computing node to process data and calculate, and the performance parameters can reflect the busyness of the current business operation of the computing node. For example, in an example scenario, multiple business applications are running in parallel on a computing node, and multiple business applications are in the peak business stage at the same time. The computing resources required by the computing node increase, and the corresponding performance parameters also increase.

[0127] The performance parameters of computing nodes may vary according to the type of computing node. For example, taking a CPU computing node as an example, a CPU computing node mainly consumes CPU computing power. Therefore, for a CPU computing node, its corresponding performance parameter may be CPU utilization, that is, the higher the CPU utilization, the closer the current performance of the CPU computing node is to the performance bottleneck or upper limit. For another example, taking a GPU computing node as an example, a CPU computing node can effectively handle AI applications, AI model training and other businesses that require a large amount of concurrent computing. For a CPU computing node, its two most important performance indicators are floating-point computing performance and GPU memory bandwidth. Therefore, its corresponding performance parameters may include floating-point computing performance and GPU memory bandwidth.

[0128] Of course, those skilled in the art will appreciate that computing nodes may include other types, and corresponding performance parameters may include any other parameters that can characterize the capabilities of computing nodes, which will not be elaborated in this specification.

[0129] S420: Based on the pre-built application networking relationship, determine the service application to which each service flow belongs, and determine the bandwidth parameter of the service application according to the bandwidth information of the service flows belonging to the same service application.

[0130] As can be seen from the above, the intelligent analysis platform can receive the flow information of each business flow sent on the data center network on the one hand, and the performance parameters of each computing node sent by the computing management platform on the other hand. The intelligent analysis platform can perform data analysis, perceive the operation status of business applications and computing nodes, and determine whether the business application is congested, whether the computing node is busy, etc.

[0131] Combined with the example in Table 3 above, what is sent to the intelligent analysis platform on the data center network is the flow information of each business flow. Different business flows may belong to the same business application. The intelligent analysis platform needs to process the business flows, determine the business flows belonging to the same business application, and determine the bandwidth parameters of the business application based on the business flows belonging to the same business application.

[0132] In the implementation manner of this specification, the intelligent analysis platform can utilize the aforementioned pre-built application networking relationship, which includes all business applications and the transport layer port numbers corresponding to the business applications. Based on the matching of the transport layer port number and the traffic identifier of the business flow, the business application to which the business flow belongs can be identified.

[0133] Specifically, the path port number of each business flow can be matched with the transport layer port number of each business application. The path port number of the business flow can include, for example, the source port and destination port shown in Table 3. By matching the path port number of each business flow with the transport layer port number of each business application, the business application to which each business flow belongs can be determined. Then, for each business application, the bandwidth information of all business flows included in the business application can be summed to obtain the bandwidth parameter corresponding to the business application. For example, in an example, as shown in Table 3, assuming that the first three rows of business flows in Table 3 belong to the same business application A, the bandwidth information of the first three rows of business flows can be summed to obtain the bandwidth parameter corresponding to business application A.

[0134] Through the above process, the intelligent analysis platform can process the business flow in combination with the application networking relationship to obtain the bandwidth parameters of each business application.

[0135] S430: For each business application, adjust the management and control strategy for running the business application according to the bandwidth parameters of the business application and the performance parameters of the computing node running the business application.

[0136] Take a certain business application A as an example, assuming that business application A runs on computing node a. According to the aforementioned method process of this specification, the intelligent analysis platform can determine the bandwidth parameters of business application A and the performance parameters of computing node a running business application A.

[0137] Then, the intelligent analysis platform can send the bandwidth parameters and performance parameters corresponding to business application A to the intelligent management and control platform, and the intelligent management and control platform will adjust the management and control strategy for business application A according to the bandwidth of the business application and the busyness of the computing node.

[0138] It can be understood that when the intelligent deployment platform pre-deploys business applications, it can allocate bandwidth for business application A according to the business situation of business application A, and can pre-set a set bandwidth according to the allocated bandwidth. The set bandwidth indicates that the bandwidth of business application A reaches the critical value of the upper limit. The set bandwidth can be equal to the allocated bandwidth of business application A or less than the allocated bandwidth.

[0139] In some implementations, the bandwidth parameter of business application A can be compared with the set bandwidth. If the bandwidth parameter is greater than or equal to the set bandwidth, it means that the current network transmission of business application A has reached the maximum processing capacity and is prone to congestion. At this time, the intelligent management and control platform needs to allocate network bandwidth to business application A to increase the allocated bandwidth of business application A. For example, in one example, the intelligent management and control platform can use a cloud cluster to expand bandwidth, such as adding a new network device to the original network device to form a cloud cluster, and adding the port of the new member device to the existing aggregation group to increase the network bandwidth.

[0140] In the implementation mode of this specification, a corresponding performance threshold can be set in advance for each computing node. The performance threshold indicates that the current computing performance of the computing node has reached the critical value of the busy state. If the performance parameter of the computing node is greater than or equal to the performance threshold, it means that the current computing node is already in a busy state.

[0141] For example, in the above example, for business application A, the performance parameters of computing node a running business application A can be compared with the performance threshold. If the performance parameters of computing node a are greater than or equal to the performance threshold, it means that the current computing node a is in a busy state. At this time, the intelligent management and control platform needs to reduce the operating load of computing node a so that computing node a can run in a normal state. For example, in one example, the intelligent management and control platform can adjust the number of business applications running on computing node a, reduce the number of business applications running on computing node a through load balancing, thereby reducing the operating load of computing node a, so that computing node a can run in a normal state.

[0142] In the above example, if the bandwidth parameter of business application A is greater than or equal to the set bandwidth, and the performance parameter of computing node a is greater than or equal to the performance threshold, the network bandwidth can be increased and the computing node load can be reduced at the same time according to the above-mentioned management and control strategy. Those skilled in the art can understand this and this specification will not elaborate on it.

[0143] In some implementations, the flow information of the service flow collected in the data center network includes packet loss information. For example, in the above example, when the traffic collector 400 collects the service flow, the TCP SN information of the service flow can be collected, and the TCP SN information indicates the next TCP message sequence number, so that the TCP SN information can be used as the packet loss information of the service flow.

[0144] The data center network sends the collected flow information of the business flow to the intelligent analysis platform. The intelligent analysis platform determines the bandwidth parameters of the business application based on the bandwidth information of the business flow belonging to the same business application. At the same time, the intelligent analysis platform can also determine the packet loss rate of the business application based on the packet loss information of the business flow belonging to the same business application. For example, the packet loss rate corresponding to the business application can be determined based on the TCP SN information of each business flow included in the business application.

[0145] In the example of this specification, the intelligent analysis platform can send the bandwidth parameters and packet loss information corresponding to the business application, as well as the performance parameters of the computing nodes running the business application to the intelligent management and control platform. The process of the intelligent management and control platform adjusting the management and control strategy of the business application according to the bandwidth parameters and the performance parameters of the computing nodes is the same as the above, and will not be repeated here.

[0146] The intelligent management and control platform can pre-set preset thresholds corresponding to business applications, and the preset thresholds represent the critical value of the packet loss rate of business applications. If the packet loss rate of a business application is greater than or equal to the preset threshold, it means that the business application is running unstably and there is a problem with the message forwarding policy corresponding to the business application, so the message forwarding policy of the business application can be adjusted. Ways to adjust the message forwarding policy include but are not limited to: bandwidth guarantee, priority queue, QoS (Quality of Service), network speed limit, etc., and this manual does not restrict this. After adjusting the message forwarding policy of the business application, the intelligent management platform can send the message forwarding policy to each network device to ensure sufficient network services for the business application.

[0147] In some implementations, the flow information of the business traffic may further include information such as delay and jitter, and the intelligent analysis platform determines the running status of the business application based on the information such as delay and jitter, and adjusts the corresponding control strategy. Based on the above implementations, those skilled in the art can understand and fully implement this, and this specification will not elaborate on it.

[0148] In some implementations, after the intelligent deployment platform pre-builds the application networking relationship, the application networking relationship can be displayed on the management and operation platform. For example, the display interface can be as follows: Figure 3 In this embodiment, the communication system can use the aforementioned resource management method to monitor in real time whether each transmission path is congested. For example, when the bandwidth parameter of a certain service application reaches the set bandwidth, it means that the transmission path of the service application is congested. At this time, Figure 3 In the application networking relationship shown, the transmission path of the service flow included in the service application is identified to prompt the management personnel to have a network congestion problem.

[0149] For example, in a normal case, the transmission path can be as follows Figure 3 As shown, the transmission path is blue. When a transmission path is congested, the transmission path can be displayed in red in the application networking relationship, or it can be further combined with the flashing state, etc. The management personnel can observe which business application transmission path is congested through the application networking relationship.

[0150] In some implementations, an intelligent management and control platform can be used to count the congestion of business applications and the busyness of computing nodes. If business applications in the communication system are frequently congested, and / or computing nodes are frequently busy, it means that the processing capacity of the entire communication system has reached its upper limit. At this time, it is necessary to consider expanding the capacity of the entire communication system or adjusting the operating strategy, such as abandoning some low-priority business applications to ensure the service quality of high-priority business applications.

[0151] From the above, it can be seen that in the implementation mode of this specification, through intelligent operation and maintenance management strategies, using the collaboration of intelligent control platforms, intelligent analysis platforms and intelligent deployment platforms, real-time monitoring of business applications and computing nodes in the communication system can be achieved, and the network bandwidth congestion of business applications and the busyness of computing nodes can be perceived, so that the operation strategy of business applications can be adjusted in a targeted manner, and it can effectively deal with AI business data centers such as intelligent computing centers that have high burst traffic and require frequent changes and adjustments, improve network operation and maintenance efficiency and quality, and maximize the resource utilization and processing capacity of the data center.

[0152] In some implementations, this specification provides a resource management device, which is applied to a communication system, wherein the communication system includes a data center and a data center network, wherein the data center is deployed with multiple business applications, and the business applications are run on computing nodes of the data center, such as Figure 5 As shown, the device comprises:

[0153] An information acquisition module 10 is used to acquire bandwidth information of each service flow in the data center network and performance parameters of each computing node included in the data center;

[0154] The bandwidth determination module 20 is used to determine the service application to which each service flow belongs based on the pre-built application networking relationship, and determine the bandwidth parameters of the service application according to the bandwidth information of the service flows belonging to the same service application, wherein the application networking relationship includes the service flow transmission path of each service application;

[0155] The policy control module 30 is used to adjust the control policy for running the business application for each business application according to the bandwidth parameter of the business application and the performance parameter of the computing node running the business application, wherein the control policy includes: allocating network bandwidth to the business application when the bandwidth parameter reaches the set bandwidth, and adjusting the operating load of the computing node when the performance parameter reaches the performance threshold.

[0156] In some embodiments, the data center network includes a cloud cluster switch, and the cloud cluster switch includes multiple network nodes. The information acquisition module 10 is specifically used to collect flow information of each business flow in the cloud cluster switch through each network node, and the flow information includes the traffic identifier of the business flow and the bandwidth information.

[0157] In some implementations, the plurality of network nodes of the cloud cluster switch include NCP nodes and NCC nodes, and the information acquisition module 10 is specifically used to:

[0158] The NCP node collects the flow identifier of the service flow sent by the computing node, establishes a first flow table based on the flow identifier of the service flow, and sends the first flow table to the NCC node;

[0159] The NCC node counts the traffic of each service flow according to the received first flow table to obtain bandwidth information of each service flow.

[0160] In some implementations, the information acquisition module 10 is specifically used to:

[0161] When the storage space of the NCP node is sufficient, a first flow table is established according to the flow identifier of the received service flow, and the first flow table is sent to the NCC node; when the storage space is insufficient, a first flow table is established according to the flow identifier of the received service flow of the preset message type, and the first flow table is sent to the NCC node;

[0162] When the NCC node has sufficient storage space, a second flow table is established according to the traffic identifier of the service flow in the first flow table. When the storage space is insufficient, a second flow table is established according to the traffic identifier of the service flow of the preset message type in the first flow table, and the traffic of each service flow in the second flow table is counted to obtain the broadband information of each service flow.

[0163] In some implementations, the bandwidth determination module 20 is specifically configured to:

[0164] Obtaining the path port number corresponding to each business flow, and matching the path port number with the transport layer port number corresponding to each business application in the application networking relationship, to determine the business application to which each business flow belongs;

[0165] The bandwidth information of the service flows belonging to the same service application is summed to obtain the bandwidth parameter corresponding to each service application.

[0166] In some embodiments, the data center includes a first computing node. When the first computing node is a GPU computing node, the information acquisition module is specifically used to obtain the floating-point computing performance and GPU memory bandwidth of the GPU computing node. The performance parameters of the GPU computing node include the floating-point computing performance and the GPU memory bandwidth.

[0167] In some implementations, the policy control module 30 is specifically used to:

[0168] Determining a packet loss rate of a service application according to packet loss information of a service flow belonging to the same service application;

[0169] According to the bandwidth parameter of the business application, the packet loss rate and the performance parameter of the computing node running the business application, the control strategy for running the business application is adjusted, wherein the control strategy includes: when the packet loss rate reaches a preset threshold, adjusting the message forwarding strategy of the business application.

[0170] In some embodiments, the apparatus further comprises a relationship building module, wherein the relationship building module is configured to:

[0171] Obtaining all business applications deployed on the data center and the transport layer port number of each business application, wherein the transport layer port number includes the path port number of each business flow included in the business application;

[0172] The application networking relationship is generated by taking the business application as a node and the transmission path of the business flow as an edge, and according to the path port number of the business flow of each business application.

[0173] Obviously, the above embodiments are merely examples for the purpose of clear explanation, and are not intended to limit the embodiments. For those skilled in the art, other different forms of changes or modifications can be made based on the above description. It is not necessary and impossible to list all the embodiments here. The obvious changes or modifications derived therefrom are still within the scope of protection of the present invention.

Claims

1. A communication system, It is characterized in that The communication system includes a data center, a data center network, and a management and operation platform. The data center is deployed with multiple business applications, each of which runs on a computing node in the data center. The management and operation platform includes an intelligent analysis platform and an intelligent management and control platform. The intelligent analysis platform is used to obtain bandwidth information of each service flow in the data center network and performance parameters of each computing node included in the data center, and determine the service application to which each service flow belongs based on the pre-built application networking relationship, and determine the bandwidth parameters of the service application according to the bandwidth information of the service flows belonging to the same service application, wherein the application networking relationship includes the service flow transmission path of each service application; The intelligent management and control platform is used to adjust the management and control strategy for running each business application according to the bandwidth parameter of the business application and the performance parameter of the computing node running the business application, wherein the management and control strategy includes: allocating network bandwidth to the business application when the bandwidth parameter reaches the set bandwidth, and adjusting the operating load of the computing node when the performance parameter reaches the performance threshold.

2. The communication system according to claim 1, It is characterized in that The data center network includes a cloud cluster switch, and the cloud cluster switch includes multiple network nodes. The flow information of each business flow in the cloud cluster switch is collected through each network node. The flow information includes the traffic identifier of the business flow and the bandwidth information. The intelligent analysis platform receives the flow information of each business flow sent by the cloud cluster switch.

3. The communication system according to claim 2, It is characterized in that The multiple network nodes of the cloud cluster switch include NCP nodes and NCC nodes. The NCP node is used to collect the flow identifier of the service flow sent by the computing node, establish a first flow table based on the flow identifier of the service flow, and send the first flow table to the NCC node; The NCC node is used to count the traffic of each service flow according to the received first flow table to obtain bandwidth information of each service flow.

4. The communication system according to claim 3, It is characterized in that The NCP node is specifically used to, when the storage space is sufficient, establish a first flow table according to the flow identifier of the received service flow, and send the first flow table to the NCC node; when the storage space is insufficient, establish the first flow table according to the flow identifier of the received service flow of the preset message type, and send the first flow table to the NCC node; The NCC node is specifically used to establish a second flow table according to the flow identifier of the service flow in the first flow table when there is sufficient storage space; and to establish a second flow table according to the flow identifier of the service flow of the preset message type in the first flow table when there is insufficient storage space, and to count the flow of each service flow in the second flow table to obtain the broadband information of each service flow.

5. The communication system according to claim 1, It is characterized in that The intelligent analysis platform is specifically used for: Obtaining the path port number corresponding to each business flow, and matching the path port number with the transport layer port number corresponding to each business application in the application networking relationship, to determine the business application to which each business flow belongs; The bandwidth information of the service flows belonging to the same service application is summed to obtain the bandwidth parameter corresponding to each service application.

6. The communication system according to claim 1, It is characterized in that The management and operation platform also includes a computing management platform, which is used to collect performance parameters of each computing node and send the performance parameters of each computing node to the intelligent analysis platform; Wherein, in the case where the computing node is a GPU computing node, the performance parameters of the GPU computing node include floating-point computing performance and GPU memory bandwidth.

7. The communication system according to any one of claims 1 to 6, It is characterized in that The intelligent analysis platform is also used to obtain packet loss information of each service flow in the data center network, and determine the packet loss rate of the service application according to the packet loss information of the service flows belonging to the same service application; The intelligent management and control platform is also used to adjust the management and control strategy for running the business application according to the bandwidth parameters of the business application, the packet loss rate and the performance parameters of the computing node running the business application, wherein the management and control strategy includes: when the packet loss rate reaches a preset threshold, adjusting the message forwarding strategy of the business application.

8. The communication system according to any one of claims 1 to 6, It is characterized in that The management and operation platform also includes an intelligent deployment platform, which is used to obtain all business applications deployed on the data center and the transport layer port number of each business application, wherein the transport layer port number includes the path port number of each business flow included in the business application, and the business application is used as a node and the transmission path of the business flow is used as an edge. The application networking relationship is generated according to the path port number of the business flow of each business application.

9. A resource management method, It is characterized in that Applied to a communication system, the communication system includes a data center and a data center network, the data center is deployed with multiple business applications, and the business applications run on computing nodes in the data center, the method includes: Obtaining bandwidth information of each service flow in the data center network and performance parameters of each computing node included in the data center; Based on the pre-built application networking relationship, determine the service application to which each service flow belongs, and determine the bandwidth parameter of the service application according to the bandwidth information of the service flow belonging to the same service application, wherein the application networking relationship includes the service flow transmission path of each service application; For each business application, a management and control strategy for running the business application is adjusted according to the bandwidth parameter of the business application and the performance parameter of the computing node running the business application, wherein the management and control strategy includes: allocating network bandwidth to the business application when the bandwidth parameter reaches a set bandwidth, and adjusting the operating load of the computing node when the performance parameter reaches a performance threshold.

10. The method according to claim 9, It is characterized in that The process of pre-building the application networking relationship includes: Obtaining all business applications deployed on the data center and the transport layer port number of each business application, wherein the transport layer port number includes the path port number of each business flow included in the business application; The application networking relationship is generated by taking the business application as a node and the transmission path of the business flow as an edge, and according to the path port number of the business flow of each business application.

11. The method according to claim 9 or 10, It is characterized in that The determining of the service application to which each service flow belongs based on the pre-built application networking relationship, and determining the bandwidth parameter of the service application according to the bandwidth information of the service flows belonging to the same service application, includes: Obtaining the path port number corresponding to each business flow, and matching the path port number with the transport layer port number corresponding to each business application in the application networking relationship, to determine the business application to which each business flow belongs; The bandwidth information of the service flows belonging to the same service application is summed to obtain the bandwidth parameter corresponding to each service application.

12. A resource management device, It is characterized in that Applied to a communication system, the communication system includes a data center and a data center network, the data center is deployed with multiple business applications, the business applications run on computing nodes of the data center, and the device includes: An information acquisition module, used to obtain bandwidth information of each service flow in the data center network and performance parameters of each computing node included in the data center; A bandwidth determination module, used to determine the service application to which each service flow belongs based on the pre-built application networking relationship, and determine the bandwidth parameters of the service application according to the bandwidth information of the service flows belonging to the same service application, wherein the application networking relationship includes the service flow transmission path of each service application; A policy management and control module is used to adjust the management and control policy for running the business application for each business application according to the bandwidth parameter of the business application and the performance parameter of the computing node running the business application, wherein the management and control policy includes: allocating network bandwidth to the business application when the bandwidth parameter reaches the set bandwidth, and adjusting the running load of the computing node when the performance parameter reaches the performance threshold.

13. The device according to claim 12, It is characterized in that It also includes a relationship building module, wherein the relationship building module is used to: Obtaining all business applications deployed on the data center and the transport layer port number of each business application, wherein the transport layer port number includes the path port number of each business flow included in the business application; The application networking relationship is generated by taking the business application as a node and the transmission path of the business flow as an edge, and according to the path port number of the business flow of each business application.

14. The device according to claim 12 or 13, It is characterized in that The bandwidth determination module is specifically used for: Obtaining the path port number corresponding to each business flow, and matching the path port number with the transport layer port number corresponding to each business application in the application networking relationship, to determine the business application to which each business flow belongs; The bandwidth information of the service flows belonging to the same service application is summed to obtain the bandwidth parameter corresponding to each service application.

Images