Distributed computing power resource scheduling method and system

Abstract

The present invention relates to the technical field of computing power resource scheduling, and in particular to a distributed computing power resource scheduling method and a system. The method comprises: step 1) querying a locally maintained mapping relationship between resource states of computing power resource nodes and forwarding nodes for a computing power resource service identifier in a computing power resource request message, so as to determine a group of candidate computing power resource nodes; step 2) on the basis of a resource demand preference field in the computing power resource request message, determining a corresponding computing power matching policy and, on the basis of the corresponding computing power matching policy and a computing power resource demand field and network resource demand field in the computing power resource request message, selecting from the group of candidate computing power resource nodes a computing power resource node meeting the resource demand of a user; and step 3) modifying the destination address of the computing power resource request message into the next hop forwarding address of the message corresponding to the selected computing power resource node. The present invention flexibly schedules computing power resources, and increases the overall utilization rate and scheduling success rate of computing power resources.

Description

A distributed computing resource scheduling method and system

[0001] Related applications

[0002] This application claims priority to Chinese patent application number 202311297837.2, entitled “A Distributed Computing Resource Scheduling Method and System,” filed on October 9, 2023, which is hereby incorporated by reference. Technical Field

[0003] The present invention relates to the technical field of computing power resource scheduling, and in particular to a distributed computing power resource scheduling method and system. Background Art

[0004] The development of the digital economy has driven the generation of massive amounts of data. Processing this massive data requires powerful computing power and extensive network connectivity across the cloud, edge, and end. Computing-network convergence has become a significant trend. Against this backdrop, a new networking paradigm has been proposed: the "computing network," which integrates computing resource scheduling capabilities into the network itself. This further deepens the integration of computing power and the network, aiming to interconnect distributed computing nodes, coordinate computing resources, network resources, and user needs, and achieve integrated scheduling of computing and network resources.

[0005] Various network service providers, represented by operators, have proposed their own computing power network architectures, and have put forward their own ideas for the coordination of computing power and network resources, and have scheduled computing power resources in a centralized or distributed manner.

[0006] However, with the rapid increase in distributed cloud, network, edge, and end computing nodes and the diverse computing power demands of users, the existing centralized computing resource scheduling mechanism in computing networks faces scalability limitations and is unsuitable for large-scale network deployments. This is especially true in scenarios where some computing nodes are subject to sudden high-load tasks or computing nodes frequently join and leave, causing highly dynamic changes in computing resource status. A single centralized scheduling platform faces a significant communication and computing scheduling burden, which can easily lead to single points of failure and processing scheduling performance bottlenecks. Distributed scheduling, on the other hand, offers advantages in scalability and reliability.

[0007] Existing distributed resource scheduling frameworks for computing networks, such as CFN (computing first network) and CAN (Computing Aware Network), are mostly based on a flat architecture, fully notifying and scheduling resource status information across all nodes with scheduling functions (resource scheduling nodes). In the face of such highly dynamic scenarios, it is difficult to quickly synchronize computing resource status information. Delayed resource status information can easily lead to inefficient resource utilization and scheduling failures. Moreover, the existing distributed resource scheduling framework is based on the ordinary TCP / IP (Transmission Control Protocol / Internet Protocol) network for scheduling. Its end-to-end routing method lacks flexibility and programmability. Especially in highly dynamic scenarios where the resource status of some computing nodes fluctuates violently, the computing demand message cannot dynamically adjust the scheduling purpose according to the real-time computing resource status changes during the routing process, which can also easily lead to inefficient resource utilization and scheduling failure. In addition, most of the resource scheduling frameworks in the existing computing network adopt a single, static computing resource scheduling strategy, such as optimizing node resource utilization or balance rate and minimizing resource usage cost. It does not have a special design for scheduling based on user resource preferences and cannot meet the diverse needs of users.

[0008] Summary of the Invention

[0009] The purpose of the present invention is to overcome the problem that in the existing computing power resource scheduling method, the computing power resource status information is difficult to synchronize quickly in the scenario of high dynamic changes in the computing power resource status, resulting in inefficient resource utilization and scheduling failure, and the lack of computing power resource scheduling design based on user resource preferences, resulting in the inability to meet diverse user needs, thereby providing a distributed computing power resource scheduling method.

[0010] To solve the above technical problems, the technical solution of the present invention provides a distributed computing resource scheduling method, which is implemented based on distributed computing routing nodes, computing resource access nodes, and computing resource nodes, and includes the following steps:

[0011] Step 1) The computing power routing node queries the computing power resource service identifier in the computing power resource request message in the mapping relationship between the computing power resource node resource status and the forwarding node maintained locally to determine a group of candidate computing power resource nodes;

[0012] Step 2) The computing power routing node determines a corresponding computing power matching strategy based on the resource requirement preference field in the computing power resource request message; based on the corresponding computing power matching strategy and the computing power resource requirement field and network resource requirement field in the computing power resource request message, selects a computing power resource node from the group of candidate computing power resource nodes that meets the user's resource requirements;

[0013] Step 3) The computing power routing node modifies the destination address of the computing power resource request message to the next-hop forwarding address of the computing power resource request message corresponding to the selected computing power resource node, and responds to the modified computing power resource request message to the user through the computing power resource access node. The user accesses the selected computing power resource node to call the relevant computing power resources.

[0014] As an improvement to the above method, in step 1), the computing power resource service identifier in the computing power resource request message is used to identify the type of the computing power resource node, wherein one computing power resource service identifier corresponds to a group of computing power resource nodes of the same type; the types of computing power resource nodes include: stand-alone computing power nodes, edge cluster computing power nodes, cloud data center computing power nodes and supercomputing center computing power nodes.

[0015] As an improvement to the above method, the specific components of the mapping relationship in step 1) include: computing power resource service identifier, computing power resource node resource status and the next hop forwarding address of the computing power resource request message corresponding to the computing power resource node;

[0016] The mapping relationship is generated in a distributed manner by classifying each computing resource node in units of computing resource service identifiers through the computing resource routing node based on the computing resource notification message received from the computing resource access node and the locally perceived network status.

[0017] As an improvement to the above method, step 1) specifically includes: the computing power routing node selects a group of candidate computing power resource nodes by comparing the computing power resource service identifier in the computing power resource request message with the computing power resource service identifier in the mapping relationship.

[0018] As an improvement to the above method, step 2) specifically includes:

[0019] The computing power routing node determines a corresponding computing power matching strategy based on the resource demand preference field in the computing power resource request message;

[0020] The computing power routing node compares the computing power resource requirement field and the network resource requirement field in the computing power resource request message with the resource status of the computing power resource node in the mapping relationship based on the corresponding computing power matching strategy, and selects a computing power resource node that meets the user's resource requirements.

[0021] As an improvement to the above method, in step 2),

[0022] The resource requirement preference field in the computing power resource request message is used to identify the user's preference when selecting resources, including: load balancing priority, CPU priority, memory priority, storage priority, price priority, and distance priority;

[0023] The computing resource requirement field in the computing resource request message is used to identify the computing resource requirement, including: CPU requirement, memory requirement and storage requirement; the method of identifying the computing resource requirement includes: using a level identifier;

[0024] The network resource requirement field in the computing power resource request message is used to identify the network resource requirement, including: delay requirement and bandwidth requirement; the method of identifying the network resource requirement includes: using level identification.

[0025] As an improvement to the above method, the computing power matching strategy in step 2) includes: load balancing priority matching, CPU priority matching, memory priority matching, storage priority matching, price priority matching and distance priority matching.

[0026] To achieve another object of the present invention, the present invention further provides a distributed computing resource scheduling system, the system comprising: distributed computing routing nodes, computing resource access nodes and computing resource nodes, wherein:

[0027] The computing power routing node is used to query the computing power resource service identifier in the computing power resource request message in the mapping relationship between the computing power resource node resource status and the forwarding node maintained locally to determine a group of candidate computing power resource nodes; is used to determine the corresponding computing power matching strategy based on the resource requirement preference field in the computing power resource request message; is used to select a computing power resource node that meets the user's resource requirements from the group of candidate computing power resource nodes based on the corresponding computing power matching strategy and the computing power resource requirement field and the network resource requirement field in the computing power resource request message; is used to modify the destination address of the computing power resource request message to the next hop forwarding address of the computing power resource request message corresponding to the selected computing power resource node, and is also used to respond to the modified computing power resource request message to the user through the computing power resource access node;

[0028] The user accesses the selected computing resource node to call the relevant computing resources.

[0029] As an improvement to the above system, the computing power routing node is also used to classify each computing power resource node based on the computing power resource notification message received from the computing power resource access node and the locally perceived network status, and to generate the mapping relationship in a distributed manner.

[0030] Compared with the existing technology, the distributed computing resource scheduling method provided by the present invention has the following advantages:

[0031] 1. The present invention proposes a distributed computing resource scheduling method, which improves scalability and reliability compared with centralized scheduling.

[0032] 2. The present invention utilizes the distributed scheduling capabilities of computing power routing nodes that integrate forwarding, storage, and computing, so that during the routing process of computing power resource request messages, the scheduling purpose can be adjusted in a timely manner according to real-time computing power resource status information, thereby realizing flexible scheduling of computing power resources and improving the utilization rate, balance rate, and success rate of resource scheduling of computing power resources.

[0033] 3. The present invention designs a variety of computing power matching strategies based on the resource demand preferences of various users; in the process of distributed routing scheduling, each computing power routing node implements intelligent decision-making and forwarding based on the preference field in the resource request message header, thereby meeting the diverse needs of users. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] FIG1 is a schematic diagram of a distributed computing resource scheduling method according to an embodiment of the present invention;

[0035] FIG2 is a flow chart of a distributed computing resource scheduling method provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0036] In order to make the objectives, technical solutions and advantages of the present invention more clear, the technical solutions of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0037] Example 1

[0038] As shown in FIG1 , this embodiment provides a distributed computing power resource scheduling method, which is implemented based on widely distributed computing power routing nodes, computing power resource access nodes, and computing power resource nodes.

[0039] As shown in FIG2 , the method provided in this embodiment includes the following steps:

[0040] The computing power routing node receives the computing power resource request message, reads the computing power resource service identifier, computing power resource requirement, and network resource requirement fields in the message, queries the mapping relationship between the locally maintained resource status and the message forwarding node, and determines a set of candidate forwarding nodes;

[0041] Based on the resource demand preferences carried in the message, the corresponding computing power matching strategy is determined, and a forwarding node that meets the user's resource requirements is selected based on the strategy;

[0042] Modify the destination address of the computing resource request message to the network address of the forwarding node and forward the message.

[0043] Preferably, in addition to basic routing and forwarding capabilities, the computing power routing node also has functions including but not limited to: reading, filling in and modifying relevant fields in the message header, and the modifiable fields include but are not limited to: destination address; based on the received computing power resource notification message, distributed generation and maintenance of the mapping relationship between resource status and message forwarding nodes; the implementation architecture of the computing power routing node includes but is not limited to: implementation based on programmable switches and locally deployed controllers.

[0044] Preferably, the functions of the computing resource access node include but are not limited to: sensing, monitoring and managing the resource status of the computing resource node, generating a computing resource service identifier, generating a computing resource notification message based on the computing resource service identifier and computing resource status of the resources managed by this node, and notifying the computing resource routing node.

[0045] Preferably, the types of computing resource nodes include but are not limited to: stand-alone computing nodes, edge cluster computing nodes, cloud data center computing nodes, and supercomputing center computing nodes.

[0046] Preferably, the computing power resource service identifier in the computing power resource request message header is used to represent a group of computing power resources with the same scale. Computing power resources of different scales use different computing power resource service identifiers. The method of dividing the computing power resource scale includes but is not limited to: single-machine computing power, edge cluster computing power, cloud data center computing power, and supercomputing center computing power.

[0047] Preferably, the computing power resource requirement field in the computing power resource request message header is used to identify specific computing power resource requirements, including but not limited to: CPU requirements, memory requirements, storage requirements; methods for identifying requirements include but are not limited to: using level identification.

[0048] Preferably, the network resource field in the computing power resource request message header is used to identify specific network resource requirements, including but not limited to: latency requirements, bandwidth requirements; methods for identifying requirements include but are not limited to: using level identification.

[0049] Preferably, the resource requirement preference field in the computing power resource request message header is used to identify the user's specific preferences when selecting resources, including but not limited to: load balancing priority, CPU priority, memory priority, storage priority, price priority, and distance priority.

[0050] Preferably, the specific components of the locally maintained mapping relationship between resource status and message forwarding nodes include but are not limited to: computing power resource service identifier, computing power resource node resource status, and the next-hop forwarding address of the message corresponding to the node; each computing power resource service identifier corresponds to a group of computing power resource nodes of this identifier type and their status and address information. Based on the resource notification messages it receives and the locally perceived network status, the computing power routing node classifies each computing power resource node by computing power resource service identifier and generates a mapping relationship. The computing power routing node compares the computing power resource service identifier in the resource request message with the identification field in the mapping relationship to find a group of candidate computing power resource nodes; based on the computing power matching strategy corresponding to the demand preference in the message, it compares the resource demand in the message with the node resource status in the mapping relationship to find a message next-hop forwarding node address that meets the resource demand. The mapping relationship implementation method includes but is not limited to: implementation based on the flow table method.

[0051] Preferably, the computing power matching strategy is used to select a forwarding node from a group of candidate forwarding nodes. Optional strategies include but are not limited to: load balancing priority matching, CPU priority matching, memory priority matching, storage priority matching, price priority matching, and distance priority matching.

[0052] The computing power routing node in this embodiment is implemented based on a programmable switch and a locally deployed controller. In addition to basic routing and forwarding functions, it also has the following functions:

[0053] (1) Read, fill in and modify relevant fields of the passing message;

[0054] (2) Receive and process the status information of the computing resource nodes reported by the computing resource access nodes;

[0055] (3) Notify the status information of the local computing resource node to other computing routing nodes, and receive notification messages from the computing resource nodes of other computing routing nodes;

[0056] (4) Perception and network status information between other computing power routing nodes;

[0057] (5) Based on the notification messages received from the computing resource nodes and the perceived network status information, a resource scheduling status information table is formed; based on the resource scheduling status information table, a mapping relationship between resource status and message forwarding nodes is generated;

[0058] The resource scheduling status information table contains the relevant status and address information of each computing power routing node, including: computing power resource service identifier, forwarding address information corresponding to the computing power resource node, number of available CPU cores, available memory size, available disk size, CPU utilization, memory utilization, disk utilization, communication delay and bandwidth between computing power routing nodes.

[0059] The computing resource access node in this embodiment has the following functions:

[0060] (1) Sense, monitor, and manage the resource status of computing resource nodes and generate computing resource service identifiers, and report the resource status information and computing resource service identifiers of computing resource nodes to computing resource routing nodes;

[0061] (2) Provide user access and computing resource access functions;

[0062] The computing resource node types in this embodiment include: stand-alone computing nodes, edge cluster computing nodes, cloud data center computing nodes, and supercomputing center computing nodes, each with a different computing resource service identifier. Stand-alone computing nodes only provide computing resources and have no organizational structure. Edge cluster computing nodes, cloud data center computing nodes, and supercomputing center computing nodes have native orchestration capabilities and do not need to worry about the details of cluster internal scheduling and management.

[0063] The computing resource request message in this embodiment is implemented based on a self-designed protocol stack. The computing resource requirement field and the network resource requirement field in the message header use levels to identify the specific resource requirements. The computing resource requirement field includes: CPU requirement level, memory requirement level, and storage requirement level, each occupying 4 bits and divided into 16 levels; the network resource requirement field includes: latency requirement level and bandwidth requirement level, each occupying 2 bits and divided into 4 levels. The correspondence between the level and the specific requirements is as follows:

[0064] The resource requirement preference field in the message header occupies a total of 8 bits. The specific options include: load balancing priority, CPU priority, memory priority, storage priority, price priority, and distance priority. The corresponding relationship between the fields is as follows:

[0065] In the present embodiment, the mapping relationship between the resource status maintained locally by the computing power routing node and the message forwarding node is implemented by the computing power routing flow table. The computing power routing flow table consists of a matching domain and an instruction block. The matching domain content includes: computing power resource service identification and resource demand preference; the instruction block content includes: computing power resource node resource status and the next hop forwarding address of the message corresponding to the computing power resource node. The executable instructions include: resource demand level comparison jump instructions, message field modification instructions, and flow table operation instructions. In other words, in the computing power routing flow table, different combinations of computing power resource service identification and resource demand preference correspond to a group of candidate computing power resource nodes and their related status and address information.

[0066] The computing power routing node classifies each computing power resource node based on the computing power resource service identifier based on the resource scheduling status information table it maintains; based on the computing power resource node evaluation function corresponding to each resource demand preference, it calculates and sorts each computing power resource node to form different computing power resource node sequences under different combinations of computing power resource service identifiers and resource demand preferences, thereby generating a computing power routing flow table.

[0067] The evaluation indicators of the computing resource node evaluation function include: load balancing index, CPU size, memory size, storage size, price and distance.

[0068] In this embodiment, the computing resource node evaluation function based on the load balancing indicator is specifically implemented as follows: BalancedResourseScore=(1-σ(CoreUsage, MemUsage, StoreUsage))*10 NodeScore=(LeastRequestedScore+BalancedResourseScore) / 2

[0069] Among them, LeastRequestedScore represents the resource load score, BalancedResourseScore represents the resource balance score, CoreUsage represents CPU utilization, MemUsage represents memory utilization, and StoreUsage represents disk utilization; NodeScore represents the node score. The higher the node score, the better the load balancing indicator of the computing resource node.

[0070] In this embodiment, the computing power routing node matches the computing power resource service identifier and resource requirement preference combination in the computing power resource request message with the computing power resource service identifier and resource requirement preference combination in the matching domain of the computing power routing flow table to find a set of candidate computing power resource nodes. In the instruction block of the computing power routing flow table, based on the computing power matching strategy corresponding to the resource requirement preference, the resource requirement level comparison jump instruction is executed, sequentially comparing the resource requirement in the message with the resource status of the computing power resource nodes in the computing power routing flow table, finding a next-hop forwarding node that meets the resource requirement, and modifying the destination field of the message for forwarding. The computing power matching strategy corresponds one-to-one with the resource requirement preference and the computing power resource node evaluation function, and is used to select a forwarding node from a set of candidate forwarding nodes. Optional strategies include: load balancing priority matching, CPU priority matching, memory priority matching, storage priority matching, price priority matching, and distance priority matching.

[0071] As shown in FIG2 , the specific steps of the distributed computing resource scheduling method provided in this embodiment are summarized as follows:

[0072] (1) The computing power routing node maintains the resource status and network status information of each computing power resource node in its domain.

[0073] (2) The computing power routing node calculates and generates a mapping relationship between the resource status and the message forwarding node based on the resource status and network status information of the computing power resource node.

[0074] (3) The user sends a computing resource request message to the computing routing node.

[0075] (4) The computing power routing node receives the computing power resource request message, reads the computing power resource service identifier, computing power resource requirement and network resource requirement fields in the computing power resource request message, queries the mapping relationship between the locally maintained resource status and the message forwarding node, and determines a group of candidate forwarding nodes.

[0076] (5) The computing power routing node determines the corresponding computing power matching strategy based on the resource demand preference carried in the computing power resource request message, and selects a forwarding node that meets the user's resource demand based on the computing power matching strategy.

[0077] (6) The computing power routing node modifies the destination address of the computing power resource request message as the network address of the forwarding node and forwards the message.

[0078] (7) Repeat steps 4 to 6 until the computing resource request message is forwarded to a computing resource access node that meets the user's resource requirements.

[0079] (8) The computing resource access node receives the computing resource request message and returns a response message to the user.

[0080] (9) The user accesses the computing resource node based on the relevant information contained in the response message and calls the relevant computing resources.

[0081] Example 2

[0082] This embodiment provides a distributed computing resource scheduling system, as shown in FIG1 , the system includes: computing routing nodes, computing resource access nodes, and computing resource nodes that are distributed;

[0083] As shown in Figure 2, the computing power routing node is used to query the computing power resource service identifier in the computing power resource request message in the mapping relationship between the resource status of the computing power resource node and the forwarding node maintained locally to determine a group of candidate computing power resource nodes; it is used to determine the corresponding computing power matching strategy based on the resource requirement preference field in the computing power resource request message; it is used to select a computing power resource node that meets the user's resource requirements from the group of candidate computing power resource nodes based on the corresponding computing power matching strategy and the computing power resource requirement field and the network resource requirement field in the computing power resource request message; it is used to modify the destination address of the computing power resource request message to the next hop forwarding address of the selected computing power resource request message, and is also used to respond to the modified computing power resource request message to the user through the computing power resource access node; the user accesses the selected computing power resource node to call the relevant computing power resources.

[0084] Preferably, the computing power routing node is also used to classify each computing power resource node based on the computing power resource notification message received from the computing power resource access node and the locally perceived network status, and to generate the mapping relationship in a distributed manner.

[0085] Finally, it should be noted that the above embodiments are intended only to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the embodiments, it should be understood by those skilled in the art that modifications or equivalent substitutions to the technical solutions of the present invention do not depart from the spirit and scope of the technical solutions of the present invention and are intended to be encompassed by the claims of the present invention.

Claims

1. A distributed computing resource scheduling method, based on distributed computing routing nodes, computing resource access nodes and computing resource nodes, comprises the following steps: Step 1) The computing power routing node queries the computing power resource service identifier in the computing power resource request message in the mapping relationship between the computing power resource node resource status and the forwarding node maintained locally to determine a group of candidate computing power resource nodes; Step 2) The computing power routing node determines the corresponding computing power matching strategy based on the resource requirement preference field in the computing power resource request message; based on the corresponding computing power matching strategy and the computing power resource requirement field and the network resource requirement field in the computing power resource request message, selects a computing power resource node that meets the user's resource requirements from the group of candidate computing power resource nodes; Step 3) The computing power routing node modifies the destination address of the computing power resource request message to the next hop forwarding address of the computing power resource request message corresponding to the selected computing power resource node, and responds to the user through the computing power resource access node with the modified computing power resource request message. The user accesses the selected computing power resource node to call the relevant computing power resources.

2. The distributed computing resource scheduling method according to claim 1, characterized in that: In the step 1), The computing power resource service identifier in the computing power resource request message is used to identify the type of the computing power resource node, wherein one computing power resource service identifier corresponds to a group of computing power resource nodes of the same type; the types of computing power resource nodes include: stand-alone computing power nodes, edge cluster computing power nodes, cloud data center computing power nodes and supercomputing center computing power nodes.

3. The distributed computing resource scheduling method according to claim 1, characterized in that: The specific components of the mapping relationship in step 1) include: a computing resource service identifier, a computing resource node resource status, and a next-hop forwarding address of a computing resource request message corresponding to the computing resource node; The mapping relationship is generated in a distributed manner by classifying each computing resource node in units of computing resource service identifiers through the computing resource routing node based on the computing resource notification messages received from the computing resource access nodes and the locally perceived network status.

4. The distributed computing resource scheduling method according to claim 3 is characterized in that: The step 1) specifically includes: the computing power routing node selects a group of candidate computing power resource nodes by comparing the computing power resource service identifier in the computing power resource request message with the computing power resource service identifier in the mapping relationship.

5. The distributed computing resource scheduling method according to claim 4, characterized in that: The step 2) specifically includes: The computing power routing node determines a corresponding computing power matching strategy based on a resource demand preference field in the computing power resource request message; The computing power routing node compares the computing power resource requirement field and the network resource requirement field in the computing power resource request message with the resource status of the computing power resource node in the mapping relationship based on the corresponding computing power matching strategy, and selects a computing power resource node that meets the user's resource requirements.

6. The distributed computing resource scheduling method according to claim 1 or 5, characterized in that: In the step 2), The resource requirement preference field in the computing resource request message is used to identify the user's preference when selecting resources, including: load balancing priority, CPU priority, memory priority, storage priority, price priority and distance priority; The computing resource requirement field in the computing resource request message is used to identify the computing resource requirement, including: CPU requirement, memory requirement and storage requirement; the method of identifying the computing resource requirement includes: using a level identifier; The network resource requirement field in the computing power resource request message is used to identify the network resource requirement, including: latency requirement and bandwidth requirement; the method of identifying the network resource requirement includes: using a level identifier.

7. The distributed computing resource scheduling method according to claim 1 or 5, characterized in that: The computing power matching strategies in step 2) include: load balancing priority matching, CPU priority matching, memory priority matching, storage priority matching, price priority matching and distance priority matching.

8. A distributed computing resource scheduling system, characterized in that: The system includes: distributed computing power routing nodes, computing power resource access nodes and computing power resource nodes, wherein: The computing power routing node is used to query the computing power resource service identifier in the computing power resource request message in the mapping relationship between the computing power resource node resource status and the forwarding node maintained locally to determine a group of candidate computing power resource nodes; to determine the corresponding computing power matching strategy based on the resource requirement preference field in the computing power resource request message; to select a computing power resource node that meets the user's resource requirements from the group of candidate computing power resource nodes based on the corresponding computing power matching strategy and the computing power resource requirement field and the network resource requirement field in the computing power resource request message; to modify the destination address of the computing power resource request message to the next hop forwarding address of the computing power resource request message corresponding to the selected computing power resource node, and also to forward the computing power resource request message to the next hop forwarding address of the computing power resource request message corresponding to the selected computing power resource node through the computing power routing node. The computing resource access node responds to the user with the modified computing resource request message; The user accesses the selected computing resource node to call the relevant computing resources.

9. The distributed computing resource scheduling system according to claim 8, characterized in that: The computing power routing node is also used to classify each computing power resource node in units of computing power resource service identifiers based on the computing power resource notification messages received from the computing power resource access nodes and the locally perceived network status, and to generate the mapping relationship in a distributed manner.

Images