Computing power resource scheduling method and network

Abstract

The invention provides a computing power resource scheduling method and a network, and the method is applied to a computing network controller, and comprises the steps: generating a service identifier in response to a computing power service request sent by a user, and controlling a computing power gateway to deploy a computing power cluster based on the service identifier; a computing power resource table of the computing power resources is obtained, the computing power resource table comprises computing power resource information of the computing power nodes and computing power information sent by the computing power gateway, in response to a DNS request for searching the computing power resources, candidate computing power nodes are determined based on the computing power resource table, path calculation is executed based on the candidate computing power nodes, and scheduling path information is generated; the scheduling path information is a routing forwarding path generated by the plurality of segment identifiers; and sending the scheduling path information to the computing power gateway so as to execute computing power resource scheduling through the computing power gateway. Through management and distribution of computing power resources, the resources can be reasonably distributed, the utilization rate of the computing power resources is balanced, network congestion and system delay are reduced, and the efficiency of distributed computing is improved.

Description

Technical Field

[0001] The present application relates to the field of communication technology, and in particular to a computing resource scheduling method and network. Background Art

[0002] Computing power scheduling aims to reasonably allocate distributed computing resources to achieve efficient task execution and resource utilization. By optimizing the allocation of computing power resources, the scheduling system can improve the overall performance of the system, reduce energy consumption, and meet the growing computing needs.

[0003] Some scheduling mechanisms rely on pre-set rules and parameters to work. These rules may include but are not limited to task priority evaluation, which is used as a guideline to allocate computing resources to tasks that need them. When receiving user requests, this scheduling method will allocate computing tasks to different servers or nodes based on pre-defined rules or static configurations, without necessarily monitoring the status information of each node in real time, such as CPU utilization and memory occupancy, for dynamic adjustment.

[0004] However, the above methods do not take into account the dynamic factors. Therefore, during peak hours or when the network is in poor condition, it is impossible to balance the utilization of computing resources, which can easily cause some nodes to be overloaded while other nodes are idle. In addition, due to the lack of effective management of network traffic and path optimization, it is also easy to cause network congestion and increase system delays, resulting in low distributed computing efficiency. Summary of the invention

[0005] The present application provides a computing resource scheduling method and network to solve the problem of low efficiency of distributed computing.

[0006] In a first aspect, the present application provides a computing resource scheduling method, which is applied to a computing network controller, comprising:

[0007] In response to a computing power service request sent by a user, a service identifier is generated, where the service identifier is used to identify the computing power service to be accessed;

[0008] Based on the service identifier, control the computing power gateway to deploy a computing power cluster;

[0009] Obtain a computing resource table of the computing resource, wherein the computing resource table includes computing resource information of the computing node and computing information sent by the computing gateway;

[0010] In response to a DNS request for searching for computing resources, determining a candidate computing node based on the computing resource table;

[0011] Based on the candidate computing power node, perform path calculation to generate scheduling path information, where the scheduling path information is a routing forwarding path generated by multiple segment identifiers;

[0012] The scheduling path information is sent to the computing power gateway to perform computing power resource scheduling through the computing power gateway.

[0013] In some feasible embodiments, the computing network controller is connected to a computing power cluster through the computing power gateway, the computing power cluster includes a computing power perception client module, and the computing power gateway also includes a routing module;

[0014] The step of obtaining a computing resource table of the computing resource comprises:

[0015] Sending an acquisition instruction to the computing power perception server module in the computing power gateway to control the computing power perception server module to receive the computing power resource information of the computing power node collected by the computing power perception client module, wherein the computing power resource information includes computing power, and the computing power is measured by the number of operations and the number of floating-point operations;

[0016] Receiving the computing power resource information and generating a computing power resource sub-table;

[0017] Receive the computing power information sent by the routing module, and generate a computing power resource table based on the computing power resource table.

[0018] In some feasible embodiments, the computing network controller includes a path computing server module, an algorithm module, and a traffic engineering database, and the computing power gateway includes a path computing client;

[0019] The performing path calculation based on the candidate computing power node to generate scheduling path information includes:

[0020] In response to a path calculation request, receiving the topology information and traffic information obtained by the routing module through the border gateway protocol, and saving them to the traffic engineering database, wherein the path calculation request is sent by the path calculation client, and the scheduling path information is information calculated by a constrained shortest path finite algorithm;

[0021] Based on the topology information and traffic information, the algorithm module is called to perform path calculation to generate scheduling path information.

[0022] In some feasible embodiments, the computing network controller includes an application layer, an abstraction layer, and a driver layer;

[0023] Receiving a computing power service request through the application layer, and converting the computing power service request into a logical operation of the computing power resource;

[0024] Based on the logic operation, calling the abstraction layer to provide an interface through the abstraction layer and converting the logic operation into an operation instruction;

[0025] The driver layer is called to execute the operation instruction through the driver layer.

[0026] In some feasible embodiments, the DNS request includes a first DNS request and a second DNS request;

[0027] The step of responding to the DNS request for searching computing resources and determining the candidate computing nodes based on the computing resources table includes:

[0028] In response to a first DNS request initiated by a user, controlling the computing power gateway to perform a DNS query based on the first DNS request, where the first DNS request is to obtain an IP address mapped to the service identifier;

[0029] If the DNS query fails to find a resource record for the domain name, the DNS proxy function of the computing power gateway is called to initiate a second DNS request through the DNS proxy function, where the second DNS request is to obtain the IP address mapped to the service identifier;

[0030] In response to the second DNS request, a candidate computing power node is determined based on the computing power resource table.

[0031] In some feasible embodiments, determining the candidate computing power node based on the computing power resource table includes:

[0032] Presetting a time value and a disconnection time parameter at the application layer, configuring a keep-alive time of the border gateway protocol by the time value, and configuring a disconnection time of the border gateway protocol by the disconnection time parameter;

[0033] Sending the keep-alive time and the disconnection time to the computing power gateway to update the parameters of the border gateway protocol, and adjusting the border gateway protocol session according to the updated parameters, wherein the border gateway protocol session is used to exchange routing information;

[0034] The candidate computing power resource nodes are determined through the routing information and the computing power resource table.

[0035] In some feasible embodiments, generating scheduling path information includes:

[0036] Obtain the number of computing power gateways;

[0037] If there are more than or equal to two computing power gateways, and the computing power gateways are located in different networks, create a logical link, where the logical link is used to characterize the connection relationship between the computing power gateways;

[0038] If the path calculation includes computing power gateways located in different networks, the logical link is added to the traffic engineering database to generate scheduling path information based on the calculation information, wherein the calculation information includes the logical link and topology information.

[0039] In some feasible embodiments, sending the scheduling path information to a computing power gateway so as to perform computing power resource scheduling through the computing power gateway includes:

[0040] Acquire the computing power resources of the candidate computing power node based on the computing power resource table;

[0041] Evaluating the computing power resources and the scheduling path information to generate scores for candidate computing power nodes;

[0042] Determine a target service node according to the score, wherein the target service node is a candidate computing power node having a score greater than or equal to a score threshold;

[0043] Generate a query result based on the target service node, and send the query result to the user, wherein the query result is the IP address of the target service node;

[0044] If the computing power gateway receives a data message sent by a user, the computing power gateway is controlled to encapsulate the data message into a target message, and forward the target message to the computing power resource node based on the scheduling path information.

[0045] In some feasible embodiments, generating scheduling path information includes:

[0046] Get node information;

[0047] Based on the node information, matching a segment identifier, the segment identifier being used to point to an Internet Protocol address of a node;

[0048] A node list is generated based on the segment identifier, the node list is used to determine the forwarding order of the data packet, and the node list is a routing forwarding path.

[0049] In a second aspect, the present application provides a computing resource scheduling network, including: a computing network controller, a computing gateway, and a computing cluster;

[0050] The computing network controller generates a service identifier in response to a computing service request sent by a user, wherein the service identifier is used for the computing service to be accessed; and deploys a computing cluster based on the service identifier;

[0051] The computing power cluster is used to collect a computing power resource table of the computing power resources, wherein the computing power resource table includes computing power resource information of the computing power nodes and computing power information sent by the computing power gateway;

[0052] The computing network controller determines a candidate computing power node based on the computing power resource table in response to a DNS request for searching computing power resources; and performs path calculation based on the candidate computing power node to generate scheduling path information, wherein the scheduling path information is a routing forwarding path generated by multiple segment identifiers;

[0053] The computing power gateway is used to perform computing power resource scheduling based on the scheduling path information.

[0054] It can be seen from the above technical scheme that the present application provides a computing power resource scheduling method and a network method applied to a computing network controller, including: generating a service identifier in response to a computing power service request sent by a user, wherein the service identifier is used to identify the computing power service to be accessed, and based on the service identifier, controlling the computing power gateway to deploy a computing power cluster; then obtaining the computing power resource table of the computing power resource, the computing power resource table includes the computing power resource information of the computing power node and the computing power information sent by the computing power gateway, responding to a DNS request for finding computing power resources, determining a candidate computing power node based on the computing power resource table, and based on the candidate computing power node, performing path calculation in multiple ways such as the shortest path algorithm or the constrained shortest path algorithm or the adaptive multi-constraint routing algorithm to generate scheduling path information, the scheduling path information is a routing forwarding path generated by multiple segment identifiers; sending the scheduling path information to the computing power gateway to perform computing power resource scheduling through the computing power gateway. By managing and allocating computing power resources, it is ensured that resources can be reasonably allocated, which not only balances the utilization of computing power resources, but also reduces network congestion and system delays, and improves the efficiency of distributed computing. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] In order to more clearly illustrate the technical solution of the present application, the drawings required for use in the embodiments are briefly introduced below. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0056] Figure 1 A schematic diagram of the functional layer structure of the computing network controller provided in an embodiment of the present application;

[0057] Figure 2 A schematic diagram of the process flow of a computing network controller processing a computing power service request provided in an embodiment of the present application;

[0058] Figure 3 A flowchart of a computing resource scheduling method provided in an embodiment of the present application;

[0059] Figure 4 A schematic diagram of a computing network resource scheduling network structure provided in an embodiment of the present application;

[0060] Figure 5A schematic diagram of the interaction between the computing network controller and the computing power gateway provided in an embodiment of the present application;

[0061] Figure 6 A schematic diagram of the network path calculation algorithm design provided in the embodiment of the present application;

[0062] Figure 7 A schematic diagram of the computing network controller and computing power resource architecture provided in an embodiment of the present application;

[0063] Figure 8 A schematic diagram of the interaction between the computing network controller and computing resources provided in an embodiment of the present application. DETAILED DESCRIPTION

[0064] The following embodiments are described in detail, and examples thereof are shown in the accompanying drawings. When the following description refers to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The implementations described in the following embodiments do not represent all implementations consistent with the present application. They are only examples of systems and methods consistent with some aspects of the present application as detailed in the claims.

[0065] Some embodiments of the present application provide a computing resource scheduling method, which is applied to a computing network controller. The computing network controller can implement unified scheduling management. By managing and allocating computing resources, it ensures that resources can be reasonably allocated, which not only balances the utilization of computing resources, but also reduces network congestion and system delays, thereby improving the efficiency of distributed computing.

[0066] A computing network controller is a device or system that controls and manages computing network (Compute and Network) resources. It is used to receive user requests and coordinate the scheduling of computing resources and network resources.

[0067] like Figure 1 As shown, in some embodiments, the computing network controller includes an application layer (Northbound app), an abstract layer (Device model SAL), a driver layer (Device driver), and a connection protocol layer. In the layered functions, the lower layer (driver layer) provides services for the upper layer (application layer), and the middle layer (abstract layer) can isolate the changes in the upper and lower layer services. The abstract layer provides a stable interface for operating devices to the application layer, shielding the complexity and changes of the lower driver layer. Similarly, when the business of the application layer changes due to the computing power gateway, the abstract layer and the driver layer can maintain appropriate stability.

[0068] In the heterogeneous computing gateway environment of chips with different crystal brands and architectures, the abstraction layer and driver layer support the access of heterogeneous devices, and achieve higher computing performance, energy efficiency and task optimization through intelligent scheduling. The computing network controller can better support horizontal and vertical expansion in the heterogeneous chip environment, and respond to changing computing needs by dynamically adjusting resource configuration. At the same time, processor resources are flexibly allocated according to specific application requirements to improve operating efficiency and adaptability.

[0069] like Figure 2 As shown in the figure, the computing network controller provides computing power scheduling and orchestration services. The application layer is the top-level functional component of the computing network controller, which is used to receive computing power service requests from users or administrators and provide business processing capabilities based on user needs. It then decomposes the computing power service requests into operations on the device and calls the interface provided by the abstract layer.

[0070] In some embodiments, a computing power service request is received through the application layer, and the computing power service request is converted into a logical operation of computing power resources; based on the logical operation, the abstract layer is called to provide an interface through the abstract layer, and the logical operation is converted into an operation instruction; the driver layer is called to execute the operation instruction through the driver layer.

[0071] The application layer converts the received computing service request into the logical operation of computing resources. The logical operation is an abstract representation of the computing service request, describing the operation steps and resource requirements required to complete the request. The abstract layer provides a set of interfaces to support communication and data exchange between the application layer and the driver layer. After the application layer converts the computing service request into a logical operation, it calls the interface provided by the abstract layer and passes the logical operation to the abstract layer.

[0072] The abstraction layer is used to convert logical operations into operation instructions, which are more specific instructions that describe how to perform specific operations at the hardware or software level. The abstraction layer is also used to handle the differences between different hardware and software platforms to ensure that operation instructions can be executed correctly in different environments.

[0073] The driver layer is the interface between the computing resource scheduling network and the hardware. It is used to control and manage the operation of hardware devices and handle data exchange related to the hardware. After the abstract layer converts the logical operation into an operation instruction, it calls the interface provided by the driver layer and passes the operation instruction to the driver layer. After receiving the operation instruction, the driver layer parses the instruction and performs the corresponding hardware operation. This may include configuring hardware devices, starting computing tasks, reading or writing data, etc. The driver layer is also used to monitor the status of hardware devices to ensure that they are working properly and provide fault recovery and error handling mechanisms when necessary.

[0074] In weak network environments such as 5G and the Internet, network instability may have an adverse impact on the efficiency and reliability of data transmission. In order to improve the performance of data transmission, the application layer uses technologies such as Multipath Transmission with Preferred Selection to achieve network acceleration, which helps to improve the efficiency and reliability of data transmission and reduce retransmissions and delays caused by network instability.

[0075] Multiple transmission and selective reception, also known as multi-path packet replication or multi-path transmission, is a packet loss prevention technology. The sending device replicates the data packet and sends the original packet and the replicated packet simultaneously through multiple links. If packet loss occurs on one of the links, the receiving device can restore the data through the replicated packet on another link, thereby avoiding retransmission and reducing latency.

[0076] The connection protocol layer implements specific network protocols for data transmission and communication, ensuring that the computing network controller can operate effectively under different network conditions. For example, NETCONF (Network Configuration Protocol), OpenFlow (Open Flow Protocol), SNMP (Simple Network Management Protocol), REST (Representational State Transfer), OVSDB (OpenvSwitch Database).

[0077] NETCONF provides a protocol for communication between network management and network devices. Through the NETCONF protocol, network management can issue, modify, and delete the configuration of remote devices. OpenFlow is a management and control protocol used in software-defined networks (SDNs) that allows controller software to communicate directly with network devices to control the forwarding and processing of data packets. SNMP is an application layer protocol used for network management, providing a standardized way to manage and monitor network devices such as routers, switches, and servers. The SNMP protocol allows a network management system (NMS) to collect information from network devices and remotely configure and modify device settings.

[0078] REST is a software architecture style, a design and development method for network applications, based on standard protocols and data formats such as HTTP, URI, and XML (or JSON, etc.) to implement access and operation of network resources. The RESTful architecture style emphasizes resources and resource operations, as well as stateless communication between clients and servers. OVSDB is a database engine used on Open vSwitch for interaction between the manager and ovsdbserver to manage and configure Open vSwitch. OVSDB provides a mechanism to store and retrieve configuration and status information related to Open vSwitch.

[0079] like Figure 3 As shown, the method comprises the following steps:

[0080] S100: Generate a service identifier in response to a computing power service request sent by a user.

[0081] A computing service request is a request sent by a user or administrator to a computing network controller through a network, requesting access to specific computing resources or services. For example, a user initiates a request through a computing cluster or interface on a user-side device, such as a PC (Personal Computer).

[0082] The service identifier is generated by the computing network controller to identify the computing service to be accessed, that is, the computing service corresponding to the computing service request. Different service identifiers can be generated based on the content or service type of the user request. The service identifier may include a unique number and service type information. The service type information may include the computing power requirement and priority of the computing service request.

[0083] The user sends a computing power service request to the computing network controller, and the computing network controller responds to the computing power service request, generates a service identifier S_NAME (domain name) and returns the service identifier to the user.

[0084] S200: Based on the service identifier, control the computing power gateway to deploy the computing power cluster.

[0085] like Figure 4 As shown, in the computing resource scheduling network, including the computing network controller and the computing power gateway, the computing power gateway and the computing network controller are connected through multiple protocols, including BGP (Border Gateway Protocol), PCEP (Path Computation Element Communication Protocol), BGP-LS (BGP Link-state, Border Gateway Extension Protocol), etc., to ensure reliability and flexible network connection capabilities.

[0086] The computing network controller is also connected to the computing power cluster through the computing power gateway. The computing power cluster is a specific application or service instance requested by the user, such as application 1: virtual machine (VM), application 2: container (POD), etc. The computing power gateway is composed of routing devices or industrial gateways, which can issue various configurations and deploy computing power applications based on user computing power requests. Figure 4 In the figure, the dotted line represents the control channel and the solid line represents the overlay network. The control channel is used to transmit control information, such as BGP-LS and PCEP (Path Computation Engine Protocol), while the overlay network is used for actual data transmission.

[0087] Computing resources are computing power resources in a distributed computing network, including CPU, memory, bandwidth, etc. Computing resources can come from cloud computing centers, edge nodes, user devices, etc. Computing resources include node resources, topology resources, and dynamic resource information. Node resources are the hardware resource information of a single node; topology resources are the network connection information between computing nodes; dynamic resource information is the real-time load of the node.

[0088] The computing power gateway selects appropriate computing power resources based on the service identifier S_NAME and deploys a computing power cluster. The computing power cluster may include multiple, for example, computing power cluster 1 and computing power cluster 2, each of which includes multiple nodes, for example, node1 and node2, each of which is a different application, for example, node1 is a VM and node2 is a POD.

[0089] S300: Obtain a computing resource table of computing resources.

[0090] In some embodiments, the computing power cluster includes a computing power perception client module (CAN-agent), and the computing power gateway includes a computing power perception server module (CAN-server). Through the collaboration of CAN-Agent and CAN-Server, it is possible to monitor and respond to the status changes of computing power nodes in real time to ensure efficient use of computing power resources.

[0091] The computing power perception client module is deployed on each computing power node to collect data such as the CPU, GPU, and interface traffic of the computing power node, and periodically report relevant resource information. The computing power perception client module uses computing power as a key indicator, and the units include OPS (Operations Per Second) and FLOPS (Floating point operations Per). OPS refers to the number of operations that the processor can perform per second, and FLOPS refers to the number of floating-point operations that can be performed per second. FLOPS focuses more on reflecting the numerical computing capabilities of the processor.

[0092] Computing power perception uses computing power as a key indicator, which means that when monitoring and evaluating the performance of computing power nodes, priority is given to the computing power of the processor, thereby providing an important reference for the rational allocation and scheduling of resources, but at the same time it must also be combined with other performance indicators to obtain a more comprehensive evaluation.

[0093] Computing gateways cover a wide range, including not only traditional CPUs, but also various special chips such as GPUs, FPGAs, ASICs, DPUs, and TPUs. Different types of processors have their own unique performance indicators and application scenarios. In practical applications, a multi-dimensional measurement system is needed to comprehensively and accurately describe their performance. For example, in addition to computing power, factors such as memory bandwidth, power consumption, and latency also need to be considered.

[0094] The computing power perception server module can summarize and analyze local computing power resources to form a local computing power resource table. It can also work with the routing module to notify local computing power resources, and then use algorithms to evaluate the optimal computing power resources through modeling.

[0095] In some embodiments, an acquisition instruction is sent to a computing power perception server module in a computing power gateway to control the computing power perception server module to receive computing power resource information of computing power nodes collected by the computing power perception client module, wherein the computing power resource information includes computing power, and the computing power is measured by the number of operations and the number of floating-point operations; receiving computing power resource information and generating a computing power resource sub-table; receiving computing power information sent by the routing module, and generating a computing power resource table based on the computing power resource sub-table.

[0096] The computing power resource table includes the computing power resource information of the computing power node and the computing power information sent by the computing power gateway. The computing power resource table is a data structure that stores the computing resources of each computing power node, such as detailed information on CPU, GPU, memory, storage, etc.

[0097] By sending acquisition instructions to the computing power perception server module in the computing power gateway, the acquisition instructions enable the computing power perception server module to receive the computing power resource information of the computing power node from the computing power perception client module. The computing power perception client module is responsible for collecting various computing power resource information of the computing power node, including but not limited to computing power. The computing power is measured by the number of operations and the number of floating-point operations. These two indicators are parameters for evaluating the performance of computing power nodes.

[0098] When the computing power perception server module receives computing power resource information, it will generate a computing power resource sub-table based on the computing power resource information. The sub-table is a data structure used to organize and store computing power resource information from different computing power nodes.

[0099] In addition to receiving information from the computing power perception client module, the computing power perception server module also receives computing power information from the routing module. The computing power information includes additional parameters such as the availability, latency, and bandwidth of the computing power nodes. Based on the computing power resource table and the computing power information sent by the routing module, the computing power perception server module will further generate a computing power resource table. The generation of the computing power resource table can fully understand and manage computing power resources, thereby optimizing computing power allocation and scheduling strategies. In addition, it also helps to achieve dynamic load balancing of computing power resources, avoiding the situation where some computing power nodes are overloaded while other nodes are idle.

[0100] S400: In response to a DNS (Domain Name System) request for searching computing resources, a candidate computing node is determined based on a computing resource table.

[0101] The DNS request is used to resolve the domain name into an IP address. In this embodiment, the DNS is a computing power gateway, the domain name is S_NAME, and the DNS request is issued by a user-side device or a computing power gateway to resolve the IP address corresponding to the service identifier, that is, to map the service identifier to a specific computing power node.

[0102] Among them, the DNS request includes the first DNS request and the second DNS request. The first DNS request is the initial DNS request initiated by the user. The user initiates a request to the computing power gateway through an application or interface to obtain the IP address mapped by the service identifier through the computing power gateway; the second DNS request is a DNS request initiated by the computing power gateway agent when the domain name resource record is not found in the first DNS request. It can be understood that the sending entity of the first DNS request and the second DNS request are different.

[0103] In some embodiments, first, in response to a first DNS request initiated by a user, the computing power gateway is controlled to perform a DNS query based on the first DNS request, and the first DNS request is to obtain the IP address mapped to the service identifier; if the DNS query does not find the resource record of the domain name, the DNS proxy function of the computing power gateway is called to initiate a second DNS request through the DNS proxy function, and the second DNS request is to obtain the IP address mapped to the service identifier; in response to the second DNS request, the candidate computing power node is determined based on the computing power resource table.

[0104] After initiating the first DNS request, the IP address mapped to the service identifier is obtained so that the user device can establish a network connection with the service. After receiving the first DNS request, the computing power gateway performs a DNS query and attempts to directly obtain the IP address. If the DNS query does not find the resource record of the domain name, the computing power gateway calls its DNS proxy function. The DNS proxy function allows the computing power gateway to act as an intermediate proxy and initiate an additional DNS query request, that is, the second DNS request, to try again to obtain the IP address mapped to the service identifier.

[0105] The purpose of the second DNS request is the same as that of the first DNS request, which is to obtain the IP address mapped to the service identifier, but the second DNS request may be executed through a different DNS server and using different query parameters or policies.

[0106] In response to the second DNS request, one or more IP addresses are obtained as the resolution result. These IP addresses may correspond to different computing nodes that can provide the required services. After obtaining these IP addresses, the candidate computing nodes are further screened and determined based on the computing resource table. The computing resource table contains computing resource information about each computing node, such as computing power, memory size, storage capacity, etc., and other possible parameters, such as geographic location, network latency, etc.

[0107] By combining the DNS query and proxy functions of the computing power gateway and the management of computing power resources, it is possible to process DNS resolution requests and dynamically select the optimal computing power node to provide services based on actual needs, which helps to improve the availability and performance of services and optimize the utilization of computing power resources. In particular, it can improve user experience and system efficiency in scenarios such as distributed systems, cloud computing, and edge computing.

[0108] Taking into account the instability of the Internet, the application layer sets the BGP hold time and disconnection time to ensure that the BGP session will not be disconnected prematurely when the network is unstable. At the same time, the routing information can be updated in time when needed, which can reduce the amount of information transmitted in the network, improve network efficiency, and reduce network shocks caused by frequent small changes.

[0109] In some embodiments, a time value and a link-off time parameter are preset at the application layer, a keep-alive time of the border gateway protocol is configured by the time value, and a link-off time of the border gateway protocol is configured by the link-off time parameter;

[0110] Send the keep-alive time and disconnection time to the computing power gateway to update the parameters of the border gateway protocol, and adjust the border gateway protocol session with the updated parameters. The border gateway protocol session is used to exchange routing information;

[0111] The candidate computing power resource nodes are determined through routing information and the computing power resource table, wherein the candidate computing power nodes are computing power nodes that meet the user's request requirements and are screened out according to the computing power resource table.

[0112] The disconnection time parameter is a pre-set time parameter used to configure the disconnection time of the Border Gateway Protocol to determine how long it takes for the BGP session to be disconnected after no keep-alive messages are received. The keep-alive time is the interval between messages sent regularly in the BGP session to confirm that the session is still active, to ensure that the BGP session remains active and prevent disconnection due to long periods of no communication.

[0113] Use the preset time value to configure the BGP keepalive time. If the keepalive message of the other party is not received within the keepalive time, some checks or actions may be triggered. Send the configured keepalive time and disconnection time parameters to the computing power gateway, which is a network device or software used to manage and allocate computing power resources. After receiving the parameters, the computing power gateway will update its BGP configuration to reflect the new keepalive time and disconnection time.

[0114] With the updated BGP parameters, the computing power gateway will adjust its BGP session, including the process of establishing, maintaining, and terminating BGP sessions with neighboring devices. BGP sessions are channels for exchanging routing information between BGP devices. By adjusting BGP sessions, it can ensure that the routing information in the network is up to date and that changes in the network topology can be perceived and responded to in a timely manner.

[0115] The computing power gateway determines candidate computing power resource nodes through the routing information exchanged in the BGP session and the computing power resource table. By finely configuring BGP parameters and using the routing information exchanged in the BGP session, effective management and allocation of computing power resources can be achieved, which helps ensure that computing power resources in the network can be used efficiently and reliably.

[0116] S500: Execute path calculation based on the candidate computing power nodes to generate scheduling path information.

[0117] like Figure 5 As shown, the network controller includes a path calculation server module (PCE module), an algorithm module and a traffic engineering database (Traffic Engineering Database, TED), wherein the path calculation server module is used to receive path calculation requests and perform related calculations, and the algorithm module is used to perform a specific algorithm module for path calculation, which is implemented through a programming language, for example, using a constrained shortest path finite algorithm, etc. The traffic engineering database is a database that stores network topology information and traffic information, which is implemented through a database management system, such as MySQL, PostgreSQL, etc.

[0118] The computing network controller also includes a graph module, which is used to represent and process the graph structure of network topology and resources. It can manage the relationship between all nodes in the network, such as computing resources, routing devices, etc., and supports various graph algorithms, such as shortest path calculation and traffic optimization. During the path calculation process, the graph module graphically represents the state of the network topology to make the path calculation more efficient. The PCE module can quickly find the best path that meets the constraints based on the structural information provided by the graph.

[0119] The computing power gateway includes a path calculation client (PCC module), which is used to send a path calculation request. In some embodiments, in response to the path calculation request, the topology information and traffic information obtained by the routing module are received through the border gateway protocol and saved to the traffic engineering database, wherein the path calculation request is sent by the path calculation client, and the scheduling path information is calculated by a variety of methods such as the shortest path first (SPF) or the constrained shortest path algorithm (CSPF) or the self-adaptive multi-constrained routing algorithm (SAMCRA). Based on the topology information and traffic information, the algorithm module is called to perform path calculation and generate scheduling path information. The scheduling path information includes network path information from the user to the computing power node, such as delay, bandwidth, etc.

[0120] Among them, SPF selects a starting node and marks the node as visited, then calculates the distance from the starting node to all adjacent nodes, adds these nodes to the priority queue, selects the node with the shortest distance from the priority queue, and marks it as visited. For all unvisited neighbor nodes of the newly visited node, calculates the distance from the starting node to these neighbor nodes, and updates the priority queue until all nodes have been visited.

[0121] CSPF first defines the constraints in the network, such as bandwidth, delay, etc., then selects a starting node and marks the node as visited, calculates the distance from the starting node to all adjacent nodes, and checks whether the constraints are met, selects the node that meets the constraints and has the shortest distance from the priority queue, and marks it as visited, calculates the distance from the starting node to all unvisited neighbor nodes of the newly visited node, and checks whether the constraints are met, and then updates the priority queue until all nodes have been visited or no path that meets the constraints can be found.

[0122] SAMCRA defines multiple constraints in the network, such as bandwidth, delay, cost, etc., then selects a starting node and marks the node as visited, calculates the distance from the starting node to all adjacent nodes, and checks whether the constraints are met, selects the node that meets the constraints and has the shortest distance from the priority queue, and marks it as visited, for all unvisited neighbor nodes of the newly visited node, calculates the distance from the starting node to these neighbor nodes, and checks whether the constraints are met, then updates the priority queue, and dynamically adjusts the path selection strategy according to the real-time status of the network (such as bandwidth changes, delay changes) until all nodes have been visited or no path that meets the constraints can be found.

[0123] PCC and PCE interact through the PCEP protocol to make path calculation requests and feedback results. When the computing power gateway needs to find the optimal computing power resource path, it first sends a path calculation request to the path calculation server module of the computing network controller through the path calculation client. The path calculation server module responds to the path calculation request and obtains the network topology information and traffic information from the routing module through the border gateway protocol. This information is saved in the traffic engineering database.

[0124] The routing module interacts with TED through the BGP-LS protocol to obtain network topology information. BGP-LS needs to use the internal gateway protocol IGP (OSPF or ISIS) within the domain to obtain the topology status. There are great challenges in running the IGP protocol in a three-layer network across operator links. Manually defining logical links through the computing network platform can solve the problem of topological relationships between computing power gateways across the three-layer underlay network.

[0125] In some embodiments, the number of computing power gateways is obtained. If the number of computing power gateways is greater than or equal to two and the computing power gateways are located in different networks, a logical link is created, wherein the logical link is used to characterize the connection relationship between the computing power gateways; if the path calculation includes computing power gateways located in different networks, the logical link is added to the traffic engineering database to generate scheduling path information based on the calculation information, wherein the calculation information includes the logical link and topology information.

[0126] The routing module collects link state information from other routers or devices through the BGP-LS protocol. This information is stored in TED and used for path calculation and resource scheduling. The network administrator manually defines logical links through the network management interface or command line interface. When defining logical links, the source node, target node and related attributes are specified, such as bandwidth limit, delay requirements, etc.

[0127] Obtaining the number of computing power gateways and their network locations can be achieved by interacting with the network operator's interface, configuration management database (CMDB) or network management system (NMS). Then determine whether there are two or more computing power gateways located in different networks. If there are computing power gateways across networks, add the defined logical link information to TED and use it together with the automatically discovered topology information. Based on the complete topology information in TED, that is, the computing information, path calculation is performed to calculate the optimal path to ensure that data packets can be transmitted efficiently.

[0128] It is understood that the logical link can be dynamically adjusted according to the network status to adapt to different service requirements. The logical link can also define redundant paths to improve the reliability and fault tolerance of the network.

[0129] Based on the topology information and traffic information stored in TED, the path calculation server module calls the Constrained Shortest Path First (CSPF) algorithm or other applicable algorithms in the algorithm module. The CSPF algorithm calculates the optimal scheduling path information based on given constraints, such as computing resource requirements, QoS requirements, etc. and the actual topology of the network, and then returns the calculated scheduling path information to the path calculation client of the computing power gateway. The computing power gateway can use this information to configure network devices to ensure that data traffic can be transmitted to the target computing resource node along the optimal path.

[0130] The suboptimal path can also be calculated and used as an alternative path. When the optimal path encounters a failure, it can be quickly switched to the alternative path. Through the joint collaboration of the computing network controller and the computing power gateway, effective support is provided for the allocation and optimization of computing power resources, which helps to improve the efficiency and reliability of data transmission and reduce network latency.

[0131] The PCE module can obtain an ordered list of nodes, each of which represents an intermediate or terminal computing gateway on the path. To implement the Segment Routing IPv6 (SRv6) function, the SegmentID (SID) of each node on the path needs to be concatenated in the order of the path to form an SRv6 path, that is, an SRv6 SID list. The list will be used to guide the transmission of data packets in the SRv6 network.

[0132] Segment Routing is a network architecture that simplifies network operations by using segments, and SRv6 is implemented in an IPv6 environment. SRv6 uses the flexibility of the IPv6 address space to implement traffic engineering, path control, and quality of service through SIDs. In SRv6, each network node has one or more segment identifiers (SIDs), which are represented by IPv6 addresses. The PCE module calculates a path and generates an ordered list of SIDs for each node on the path. The SID list indicates how data packets are transmitted in the SRv6 network.

[0133] Among them, the scheduling path information is a routing forwarding path generated by multiple segment identifiers. In some embodiments, the node information is first obtained, and then the segment identifier is matched based on the node information. The segment identifier is used to point to the Internet Protocol address of the node. A node list is generated based on the segment identifier. The node list is used to determine the forwarding order of the data packet. The node list is the routing forwarding path.

[0134] For each node in the node list, the PCE module can obtain the node's Internet Protocol address (IPv6 address) and possible other attributes, such as node name, location, etc. after parsing. This information can be obtained through configuration files, databases, or network interfaces. Based on the node's IPv6 address, the PCE module matches the corresponding segment identifier (SID). SID is a special IPv6 address or a compressed address that uniquely identifies a node in an SRv6 network. In the order of the path, the SID of each node is concatenated to form an SRv6 SID list.

[0135] After generating the SRv6 SID list, the PCE module can also perform verification and optimization to ensure the reachability and efficiency of the path. For example, the validity of each SID can be checked and the path can be optimized to reduce latency or improve bandwidth utilization. The PCE module outputs the generated SRv6 SID list to the path calculation client of the computing power gateway. The list can be used to configure the SRv6 router or switch to forward data packets along the specified path.

[0136] S600: Send scheduling path information to the computing power gateway to perform computing power resource scheduling through the computing power gateway.

[0137] In some embodiments, computing resources of the candidate computing nodes are obtained based on the computing resource table;

[0138] Evaluate computing resources and scheduling path information to generate scores for candidate computing nodes;

[0139] According to the score, the target service node is determined, and the target service node is the candidate computing power node with a score greater than or equal to the score threshold;

[0140] Generate query results based on the target service node and send the query results to the user. The query results are the IP addresses of the target service nodes.

[0141] If the computing power gateway receives a data message sent by a user, it controls the computing power gateway to encapsulate the data message into a target message and forwards the target message to the computing power resource node based on the scheduling path information.

[0142] By querying the computing power resource table, the computing power resource status of all candidate computing power nodes is obtained. Then, according to the type and scale of user requests, the computing power resources of the candidate computing power nodes are evaluated to determine whether they meet the processing requirements of the requests. Combining the evaluation results of the computing power resources and the scheduling path information, each candidate computing power node generates a score that reflects the suitability of the node for processing user requests.

[0143] By presetting the scoring threshold, it is used to screen the computing nodes that meet the requirements, and all candidate computing nodes with scores greater than or equal to the scoring threshold are used as target service nodes. These nodes are able to process user requests. Based on the determined target service node, the query result is generated, and the query result includes the IP address of the target service node so that the user or subsequent system can communicate with it. The query result is sent to the user, and the user can initiate a request to the target service node based on this information.

[0144] When users send data packets to the computing power gateway, the computing power gateway will receive these data packets and encapsulate the data packets into target packets as needed. The target packets are in the form of data packets and may include IPv6 headers, SRv6 extension headers, and payloads. The computing power gateway will forward the encapsulated target packets to the corresponding computing power resource nodes for processing based on the previously obtained scheduling path information.

[0145] By evaluating computing resources and scheduling path information, we can select the computing node that is most suitable for processing user requests, and achieve efficient forwarding of data messages through the computing gateway, which not only improves the processing efficiency of the system, but also optimizes the user's request processing experience.

[0146] The computing power gateway receives the data message and encapsulates it according to the FIB table. The outer source IP is the user-side computing power gateway address, the outer destination IP is the SRH next-hop address, the inner IP source address is the user PCIP, and the destination address is S_IP (optimal node). The computing power gateway receives subsequent messages from the user and directly encapsulates and forwards them according to the FIB table entries.

[0147] like Figure 3As shown, if two computing power gateways are included, computing power gateway 1 and computing power gateway 2 communicate through SRv6 encapsulation and decapsulation. Among them, computing power gateway 1 can be a user-side computing power gateway, and computing power gateway 2 can be a resource-side computing power gateway. The user-side computing power gateway is located in the user's local network or edge network, and is the interface between the user device and the core network. The resource-side computing power gateway is located in a data center or cloud environment, and is the interface between computing power resources (such as VM, POD) and the external network. Computing power gateway 1 forwards the encapsulated message to computing power gateway 2 according to the FIB table entry, and computing power gateway 2 continues to forward the message according to the FIB table entry until the message reaches the target computing power resource.

[0148] Based on the above-mentioned computing resource scheduling method, some embodiments of the present application also provide a computing resource scheduling network, including: a computing network controller, a computing gateway and a computing cluster;

[0149] The computing network controller generates a service identifier in response to a computing service request sent by a user, wherein the service identifier is used for the computing service to be accessed; and deploys a computing cluster based on the service identifier;

[0150] The computing power cluster is used to collect a computing power resource table of the computing power resources, wherein the computing power resource table includes computing power resource information of the computing power nodes and computing power information sent by the computing power gateway;

[0151] The computing network controller determines a candidate computing power node based on the computing power resource table in response to a DNS request for searching computing power resources; and performs path calculation based on the candidate computing power node to generate scheduling path information, wherein the scheduling path information is a routing forwarding path generated by multiple segment identifiers;

[0152] like Figure 6 As shown, the PCC module sends a path calculation request to the network controller, and the network controller forwards the request to the PCE module. The PCE module calculates the optimal path according to the network status and returns the result to the network controller. The network controller instructs the PCC to establish the corresponding LSP (Label Switched Path) based on the path information provided by the PCE. The PCC establishes the LSP according to the instruction, thereby completing the path setting.

[0153] Network path calculation refers to the process of finding the optimal path or the path that meets specific conditions from the starting point to the end point in the network or graph. In the current network controller, the following three path calculation algorithms are supported: shortest path algorithm, constrained shortest path algorithm and adaptive multi-constrained routing algorithm.

[0154] The computing network controller supports the PCEP protocol based on the DNS request algorithm, and has an independent PCE functional module and PCC path calculation client. The PCE based on the DNS request algorithm is deployed in the computing network controller, and calculates the best path that meets the constraints based on the network topology information; the PCC is deployed in the new data plane gateway. After the PCE based on the DNS request algorithm calculates the best path, it sends the best path to the PCC through the PCEP protocol, thereby providing path calculation services for the new data plane gateway. The PCE based on the DNS request algorithm collects information such as the topology, link status, and resource usage in the network, and uses built-in algorithms or external systems, such as the support of the SDN controller, to calculate the optimal path from the source to the destination.

[0155] When the PCE calculates the optimal path, it encapsulates relevant information, such as the SID list and path priority, into an SRv6 TE Policy and sends it to the forwarder in the network through the PCEP session. The forwarder receives and parses the SRv6 TE Policy and forwards the traffic based on the path information.

[0156] When a data packet enters the network, the forwarder forwards the data packet according to the path information in the SRv6 TE Policy. The data packet passes through a series of network nodes during the forwarding process, and each node forwards the data packet according to the instructions in the SRv6 TE Policy.

[0157] In summary, the computing network controller supports multiple protocols, such as GRPC, BGP-LS, and PCEP, which provide a basis for communication between the controller and network devices. Compared with the traditional manual configuration method, the computing network controller based on the DNS request algorithm can automatically calculate and send path information, greatly simplifying the complexity of network configuration and management. By deploying PCE-InitiatedSRv6TE Policy, refined management and optimization of network traffic can be achieved, improving the overall performance and reliability of the network.

[0158] like Figure 7 As shown, the computing network controller includes a scheduling module and a collection module. When the collection module is connected to the computing power gateway and the computing power cluster, it can communicate with the computing power gateway in the computing power cluster through the southbound interface to obtain the status information of the network equipment, including link status, routing information, etc.

[0159] The acquisition module service abstraction layer (SAL) can shield the differences in southbound protocols and provide consistency services for upper-layer functional modules. When connecting to a computing power gateway or computing power cluster, SAL can abstract the service interface related to the computing power gateway or computing power cluster for upper-layer application calls. The computing network controller acquisition module also provides an extensible northbound API, allowing developers to customize functions according to their needs.

[0160] The computing network controller acquisition module can also monitor the network status in real time through communication with the computing power gateway / computing power cluster. When a link or node failure is detected, the computing network controller scheduling module can quickly trigger the load balancing mechanism and switch the traffic to the backup path to achieve rapid recovery.

[0161] The collection module of the computing network controller can participate in the process of route reconvergence, recalculate the route according to the change of network status, and switch the traffic back to the optimal path after the reconvergence is completed. The design principle of the computing network controller to connect to the computing power gateway or computing power cluster is mainly based on the centralized control, southbound interface support, service abstraction layer, scalable northbound API, and fault detection and recovery of the SDN controller. These features together constitute the basic framework and implementation method of connecting to computing power resources.

[0162] like Figure 8 As shown in the figure, the computing network controller sends computing power collection tasks to each node (Node) through the gRPC (Google Remote Procedure Call) protocol. The computing power collection module of each node regularly collects and reports the current computing power information, including the usage of resources such as CPU, memory (MEM), GPU, etc. The collection module of the computing network controller receives and processes this computing power information and generates a detailed computing power information table. The table lists the resource status of each node, including cluster ID, node, CPU, MEM, GPU, etc. The scheduling module of the computing network controller performs resource scheduling based on the data in the computing power information table to ensure the effective use and reasonable allocation of resources.

[0163] The computing power gateway is used to perform computing power resource scheduling based on the scheduling path information. The effects of the above embodiment when running can refer to the effects of the above method embodiment, which will not be repeated here.

[0164] Similar parts between the embodiments provided in this application can be referenced to each other. The specific implementation methods provided above are only a few examples under the general concept of this application and do not constitute a limitation on the protection scope of this application. For those skilled in the art, any other implementation methods expanded based on the scheme of this application without creative work belong to the protection scope of this application.

Claims

1. A computing resource scheduling method, characterized in that: Applied to computing network controllers, including: In response to a computing power service request sent by a user, a service identifier is generated, where the service identifier is used to identify the computing power service to be accessed; Based on the service identifier, control the computing power gateway to deploy a computing power cluster; Obtain a computing resource table of the computing resource, wherein the computing resource table includes computing resource information of the computing node and computing information sent by the computing gateway; In response to a DNS request for searching for computing resources, determining a candidate computing node based on the computing resource table; Based on the candidate computing power node, perform path calculation to generate scheduling path information, where the scheduling path information is a routing forwarding path generated by multiple segment identifiers; The scheduling path information is sent to the computing power gateway to perform computing power resource scheduling through the computing power gateway.

2. The computing resource scheduling method according to claim 1, characterized in that: The computing network controller is connected to a computing power cluster through the computing power gateway, the computing power cluster includes a computing power perception client module, and the computing power gateway also includes a routing module; The step of obtaining a computing resource table of the computing resource comprises: Sending an acquisition instruction to the computing power perception server module in the computing power gateway to control the computing power perception server module to receive the computing power resource information of the computing power node collected by the computing power perception client module, wherein the computing power resource information includes computing power, and the computing power is measured by the number of operations and the number of floating-point operations; Receiving the computing power resource information and generating a computing power resource sub-table; Receive the computing power information sent by the routing module, and generate a computing power resource table based on the computing power resource table.

3. The computing resource scheduling method according to claim 2, characterized in that: The computing network controller includes a path computing server module, an algorithm module and a traffic engineering database, and the computing power gateway includes a path computing client; The performing path calculation based on the candidate computing power node to generate scheduling path information includes: In response to a path calculation request, receiving the topology information and traffic information obtained by the routing module through the border gateway protocol, and saving them to the traffic engineering database, wherein the path calculation request is sent by the path calculation client, and the scheduling path information is information calculated by a constrained shortest path finite algorithm; Based on the topology information and traffic information, the algorithm module is called to perform path calculation to generate scheduling path information.

4. The computing resource scheduling method according to claim 1, characterized in that: The computing network controller includes an application layer, an abstraction layer and a driver layer; Receiving a computing power service request through the application layer, and converting the computing power service request into a logical operation of the computing power resource; Based on the logic operation, calling the abstraction layer to provide an interface through the abstraction layer and converting the logic operation into an operation instruction; The driver layer is called to execute the operation instruction through the driver layer.

5. The computing resource scheduling method according to claim 1, characterized in that: The DNS request includes a first DNS request and a second DNS request; The step of responding to the DNS request for searching computing resources and determining the candidate computing nodes based on the computing resources table includes: In response to a first DNS request initiated by a user, controlling the computing power gateway to perform a DNS query based on the first DNS request, where the first DNS request is to obtain an IP address mapped to the service identifier; If the DNS query fails to find a resource record for the domain name, the DNS proxy function of the computing power gateway is called to initiate a second DNS request through the DNS proxy function, where the second DNS request is to obtain the IP address mapped to the service identifier; In response to the second DNS request, a candidate computing power node is determined based on the computing power resource table.

6. The computing resource scheduling method according to claim 4, characterized in that: The determining the candidate computing power node based on the computing power resource table includes: Presetting a time value and a disconnection time parameter at the application layer, configuring a keep-alive time of the border gateway protocol by the time value, and configuring a disconnection time of the border gateway protocol by the disconnection time parameter; Sending the keep-alive time and the disconnection time to the computing power gateway to update the parameters of the border gateway protocol, and adjusting the border gateway protocol session according to the updated parameters, wherein the border gateway protocol session is used to exchange routing information; The candidate computing power resource nodes are determined through the routing information and the computing power resource table.

7. The computing resource scheduling method according to claim 3, characterized in that: The generating of the scheduling path information comprises: Obtain the number of computing power gateways; If there are more than or equal to two computing power gateways, and the computing power gateways are located in different networks, create a logical link, where the logical link is used to characterize the connection relationship between the computing power gateways; If the path calculation includes computing power gateways located in different networks, the logical link is added to the traffic engineering database to generate scheduling path information based on the calculation information, wherein the calculation information includes the logical link and topology information.

8. The computing resource scheduling method according to claim 1, characterized in that: The sending the scheduling path information to the computing power gateway to perform computing power resource scheduling through the computing power gateway includes: Acquire the computing power resources of the candidate computing power node based on the computing power resource table; Evaluating the computing power resources and the scheduling path information to generate scores for candidate computing power nodes; Determine a target service node according to the score, wherein the target service node is a candidate computing power node having a score greater than or equal to a score threshold; Generate a query result based on the target service node, and send the query result to the user, wherein the query result is the IP address of the target service node; If the computing power gateway receives a data message sent by a user, the computing power gateway is controlled to encapsulate the data message into a target message, and forward the target message to the computing power resource node based on the scheduling path information.

9. The computing resource scheduling method according to claim 1, characterized in that: The generating of the scheduling path information comprises: Get node information; Based on the node information, matching a segment identifier, the segment identifier being used to point to an Internet Protocol address of a node; A node list is generated based on the segment identifier, the node list is used to determine the forwarding order of the data packet, and the node list is a routing forwarding path.

10. A computing resource scheduling network, characterized in that: include: Computing network controller, computing power gateway and computing power cluster; The computing network controller generates a service identifier in response to a computing service request sent by a user, where the service identifier is used for the computing service to be accessed; and deploying a computing power cluster based on the service identifier; The computing power cluster is used to collect a computing power resource table of the computing power resources, wherein the computing power resource table includes computing power resource information of the computing power nodes and computing power information sent by the computing power gateway; The computing network controller determines a candidate computing power node based on the computing power resource table in response to a DNS request for searching for computing power resources; and performing path calculation based on the candidate computing power node to generate scheduling path information, wherein the scheduling path information is a routing forwarding path generated by multiple segment identifiers; The computing power gateway is used to perform computing power resource scheduling based on the scheduling path information.

Images