Regional computing power network infrastructure oriented to computing power scheduling and service

Abstract

The invention discloses a regional computing power network infrastructure oriented to computing power scheduling and service, which comprises a computing power node infrastructure layer, a computing power network sensing layer, a computing power description layer, a computing power resource layer, a computing power routing layer, a computing power service layer and a computing power network scheduling layer, the computing power network sensing layer is used for carrying out network access processing on a plurality of computing power resources from the computing power node infrastructure layer through a network so as to obtain regional computing power network resources, and sending the regional computing power network resources to the computing power description layer; and the computing power description layer is used for performing mapping processing on the regional computing power network resources from the computing power network sensing layer by using a computing power template so as to map the regional computing power network resources to a unified dimension and obtain abstract description of the regional computing power network resources. According to the invention, the technical problem that the existing computing power scheduling and comprehensive management technology is difficult to provide heterogeneous computing power service can be solved.

Description

Technical Field

[0001] The present invention belongs to the technical field of computing power networks and infrastructures, and more specifically, relates to a regional computing power network infrastructure for computing power scheduling and services. Background Art

[0002] The demand for computing power continues to expand in all walks of life, and the demand for inclusive, flexible, and efficient computing power services is bursting out. The increasing number of intelligent products and applications are also constantly putting forward higher requirements for computing power. The demand for the integration of multiple computing powers is everywhere, and the concept of regional computing power network has emerged. The regional computing power network infrastructure is the cornerstone of computing power scheduling and comprehensive management. By building a cross-regional network architecture, it realizes the coordination of computing resources, storage resources, and network resources between different regions, provides users with a unified platform interface, and dynamically submits, schedules, and monitors tasks in real time. It is a comprehensive application of multiple technologies such as network, computing power scheduling, and computing power management. It can better cope with fluctuations in regional computing needs and avoid waste or shortage of resources.

[0003] At present, the research on computing power scheduling and comprehensive management technology at home and abroad is still in its infancy, mainly providing users with unified cloud computing services in the form of internal networking. Google's (Google Distributed Cloud) has realized cross-domain resource sharing within Google; Amazon's Greengrass has introduced local computing, message sending and receiving, data management, synchronization and ML reasoning functions into edge devices; Huawei's large model development kit, which combines MindSpore and ModelArts, is mainly aimed at algorithm development, parallel computing, storage optimization, and breakpoint continuation training to achieve efficient development of large models. It mainly focuses on the research and scheduling of various fields or homogeneous resources, and has not yet formed a complete and unified scheduling and management solution for heterogeneous ubiquitous resources.

[0004] However, the existing computing power scheduling and comprehensive management technologies still have some technical problems that cannot be ignored: first, it is difficult to provide heterogeneous computing power services; second, it is unable to fully integrate and schedule a large amount of idle computing power resources in the society on demand; third, it has no standardized architecture for computing power scheduling and comprehensive management, and cannot achieve comprehensive management of computing power. Summary of the invention

[0005] In response to the above defects or improvement needs of the prior art, the present invention provides a regional computing power network infrastructure for computing power scheduling and services, with the aim of solving the technical problems that the existing computing power scheduling and comprehensive management technologies are difficult to provide heterogeneous computing power services, and the technical problems that a large number of idle computing power resources in the scheduling society cannot be fully integrated on demand, as well as the technical problems that there is no standardized architecture for computing power scheduling and comprehensive management, and the comprehensive management of computing power cannot be achieved.

[0006] To achieve the above-mentioned purpose, according to one aspect of the present invention, a regional computing power network infrastructure for computing power scheduling and service is provided, including a computing power node infrastructure layer, a computing power network perception layer, a computing power description layer, a computing power resource layer, a computing power routing layer, a computing power service layer, and a computing power network scheduling layer. The computing power node infrastructure layer is used to provide a variety of computing power resources and send all computing power resources to the computing power network perception layer;

[0007] The computing power network perception layer is used to perform network access processing on multiple computing power resources from the computing power node infrastructure layer through the network to obtain regional computing power network resources and send the regional computing power network resources to the computing power description layer.

[0008] The computing power description layer is used to map the regional computing power network resources from the computing power network perception layer using the computing power capability template, so as to map the regional computing power network resources to a unified dimension, obtain an abstract description of the regional computing power network resources, and send the abstract description of the regional computing power network resources to the computing power resource layer.

[0009] The computing power resource layer is used to perform service-oriented encapsulation processing on the abstract description of the regional computing power network resources from the computing power description layer to obtain computing power network resource information and send the computing power network resource information to the computing power routing layer.

[0010] The computing power service layer is used to obtain the business service level agreement (SLA) request from the user through the API gateway, parse the business SLA request to generate user resource demand information, and send the user resource demand information to the computing power routing layer.

[0011] The computing power routing layer includes the control plane and the forwarding plane;

[0012] The control plane is used to extract indicators from the computing network resource information from the computing resource layer and the user resource demand information from the computing service layer to obtain the resource status of the current computing network. Based on the resource status, the control plane uses a multi-objective optimization scheduling method and the hot, cold and warm characteristics of the application to obtain computing nodes that meet the resource requirements corresponding to the business SLA request from the current computing network. All the computing nodes obtained constitute the optimal resource computing node combination.

[0013] The forwarding plane is used to process the optimal resource computing node combination obtained by the control plane using load balancing and quality of service QoS routing algorithm to obtain the optimal scheduling solution corresponding to the business SLA request.

[0014] The computing power network scheduling layer is connected to the computing power network perception layer, the computing power resource layer, and the computing power routing layer respectively, and is used to perform resource monitoring, task monitoring, and billing management operations on the regional computing power network resources obtained by the computing power network perception layer, the computing power network resource information obtained by the computing power resource layer, and the optimal resource computing power node combination obtained by the computing power routing layer, so as to obtain resource monitoring results, task monitoring results, and billing results respectively.

[0015] Preferably, the computing power node infrastructure layer includes computing power nodes in multiple data centers;

[0016] Data centers include supercomputing centers, provincial government clouds, municipal and state clouds, and data centers above a certain scale.

[0017] Computing resources include one or any combination of CPU, GPU, FPGA, DPU, and DSP computing capabilities.

[0018] Preferably, the computing power capacity template is a rule that provides standard computing power measurements in computing power routing, computing power equipment management, computing power billing and other functions;

[0019] Service-oriented encapsulation processing provides resource models based on the computing needs of different applications and encapsulates them into computing application programming interfaces (APIs) to meet the diverse computing needs of various industries.

[0020] The computing network resource information is used to evaluate the computing resources currently available in the computing network, and includes processor performance, memory size, storage space, network and connection speed, etc.

[0021] Preferably, the process of parsing the business SLA request is to obtain the specific requirements of the task submitted by the user from the business SLA request, including the amount of calculation, data scale, and execution time.

[0022] Preferably, the computing network resource information from the computing resource layer and the user resource demand information from the computing service layer are subjected to an index extraction operation to obtain the resource status of the current computing network. Specifically, the available computing resources of the current computing network are first obtained through the computing network resource information, including processor type, memory capacity, network bandwidth, computing power, current load, task queue, delay and bandwidth, node location and node cost. Then, task performance indicators (including response time, processing time, bandwidth, delay, computing power), task priority and task execution status instructions are obtained from the user resource demand information. The available computing resources of the current computing network and the task performance indicators obtained together constitute the resource status of the current computing network.

[0023] The computing nodes that meet the resource requirements corresponding to the business SLA requests refer to the computing nodes whose type, capacity, performance, cost, bandwidth and latency all meet the business SLA requests.

[0024] Preferably, the resource monitoring process is specifically as follows: first, the computing network resource information obtained by the computing resource layer is monitored and summarized in real time to obtain the monitoring statistical results of the computing network as the resource monitoring results, including the number of jobs, job status, data location, network bandwidth, and computing / storage resource load, etc.

[0025] Preferably, the task monitoring process is specifically as follows: first, the optimal resource computing power node combination obtained from the computing power routing layer is classified according to the task ID of each task in the business SLA request to obtain the optimal resource computing power node combination corresponding to the task; then, based on the obtained optimal resource computing power node combination corresponding to the task, the computing power network resource information obtained from the computing power resource layer is monitored and summarized in real time to obtain the job status information corresponding to the task; finally, all the job status information corresponding to all the tasks obtained is summarized to obtain the overall real-time monitoring information as the task monitoring result, including recording key information such as the task running status, progress, and performance.

[0026] Preferably, the billing management process is specifically as follows: first, classify the information of the optimal resource computing power node combination from the computing power routing layer according to the task ID of each task in the business SLA request to obtain the optimal resource computing power node combination corresponding to the task; then, based on the obtained optimal resource computing power node combination corresponding to each task (that is, all jobs under the task), monitor and summarize the computing power network resource information obtained from the computing power resource layer in real time to obtain the resource usage information of all jobs of the task in the cluster; finally, map the obtained resource usage information of all jobs of the task in the cluster according to the computing power capacity template to obtain the cost result generated by the task using resources in the computing power network, and the cost results corresponding to all tasks constitute the billing result.

[0027] In summary, compared with the prior art, the above technical solutions conceived by the present invention can achieve the following beneficial effects:

[0028] 1. The present invention designs a regional computing network infrastructure for computing power scheduling and services, adopts a computing power node infrastructure layer, a computing power network perception layer, and a computing power description layer to manage underlying physical resources and virtual resources, and provides computing power scheduling and comprehensive management services upward through the computing power resource layer, turning heterogeneous resources into resource pools to provide a variety of computing power services, and carrying out universal computing power public services for ubiquitous users, so that heterogeneous computing power is routed to multi-field application demand parties in the region, solving the problem of difficulty in providing heterogeneous computing power services;

[0029] 2. The present invention adopts computing power control and computing power perception based on the computing power routing layer, and completes the mapping and negotiation between computing power applications and computing power networks to support "user needs" and "computing interconnection resources" through the cooperation of three functional modules: computing power resource layer, computing power routing layer and computing power service layer. It achieves the coordination and unification of user needs, network environment and computing power resources, provides users with the most suitable computing power services, realizes optimized configuration, and solves the problem of a large number of idle computing power resources that cannot be fully integrated and dispatched on demand;

[0030] 3. The present invention adopts a computing power operation structure based on the computing power network scheduling layer. By monitoring, measuring and managing the computing power network perception layer, computing power resource layer and computing power routing layer, it completes operational management functions such as resource monitoring, task monitoring and billing, solves the problem of no standardized architecture for computing power scheduling and comprehensive management, and promotes the construction of a regional computing power network service system. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 It is a schematic diagram of the regional computing power network infrastructure for computing power scheduling and services of the present invention;

[0032] Figure 2 It is a schematic diagram of the computing power node infrastructure layer in the present invention. DETAILED DESCRIPTION

[0033] In order to make the purpose, technical scheme and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention. In addition, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0034] like Figure 1 and Figure 2 As shown, the present invention provides a regional computing power network infrastructure for computing power scheduling and services, including a computing power node infrastructure layer, a computing power network perception layer, a computing power description layer, a computing power resource layer, a computing power routing layer, a computing power service layer, and a computing power network scheduling layer.

[0035] The computing node infrastructure layer includes all computing infrastructure in the regional computing network, which is used to provide a variety of computing resources and send all computing resources to the computing network perception layer;

[0036] Specifically, the computing power node infrastructure layer in the regional computing power network is as follows: Figure 2As shown, it includes computing nodes of multiple data centers (including supercomputing centers, provincial government cloud, municipal and state clouds, and data centers above a certain scale); computing resources include CPU (including single-core CPU and multi-core CPU), GPU, FPGA, DPU, DSP and other computing capabilities, one or any combination of them.

[0037] The computing power network perception layer (which is one of the two major infrastructure layers for realizing the computing power perception network) is used to process the network access of multiple computing power resources from the computing power node infrastructure layer through the 5G network (including 5G access network, 5G metropolitan area network, and 5G backbone network, etc.) (even if the computing power resources of different data centers that were originally isolated from each other due to geographical reasons can be connected) to obtain regional computing power network resources, and send the regional computing power network resources to the computing power description layer.

[0038] The computing power description layer is used to map the regional computing power network resources from the computing power network perception layer using the computing power capability template (which is a model function) to map the regional computing power network resources to a unified dimension, obtain an abstract description of the regional computing power network resources, and send the abstract description of the regional computing power network resources to the computing power resource layer.

[0039] Specifically, the computing power capacity template is a rule that provides standard computing power measurements in functions such as computing power routing, computing power equipment management, and computing power billing.

[0040] The computing power resource layer (which is one of the two major infrastructure layers of the computing power-aware network) is used to perform service-oriented encapsulation processing on the abstract description of the regional computing power network resources from the computing power description layer (that is, to provide resource models based on the computing needs of different applications and encapsulate them into a computing power application programming interface (Application Programming Interface, API for short) to meet the diverse computing needs of various industries) to obtain computing power network resource information and send the computing power network resource information to the computing power routing layer (for resource awareness).

[0041] Specifically, the computing power network resource information is an assessment of the computing resources currently available in the computing power network, including processor performance, memory size, storage space, network connection speed, etc.

[0042] The computing power service layer (which carries various services and applications of ubiquitous computing) is used to obtain business service level agreement (SLA) requests from users through the API gateway, and parse the business SLA requests to generate user resource demand information, and send the user resource demand information to the computing power routing layer (whose purpose is to achieve demand perception).

[0043] Specifically, a business SLA request is an agreement between a customer and a service provider that specifies the service quality standards, performance requirements, and various indicators that quantify service quality, including response time, error rate, etc.

[0044] Specifically, the process of parsing the business SLA request is to obtain the specific requirements of the task submitted by the user, including the amount of calculation, data scale, execution time, etc.

[0045] The computing power routing layer (which is one of the two core functional module layers for realizing the computing power perception function system, divided into the control plane and the forwarding plane), the control plane is used to extract indicators from the computing power network resource information from the computing power resource layer and the user resource demand information from the computing power service layer to obtain the resource status of the current computing power network, and use the multi-objective optimization scheduling method and the hot, cold and warm characteristics of the application according to the resource status to obtain the computing power nodes that meet the resource requirements corresponding to the business SLA request from the current computing power network. All the computing power nodes obtained constitute the optimal resource computing power node combination.

[0046] Specifically, the computing network resource information from the computing resource layer and the user resource demand information from the computing service layer are subjected to indicator extraction operations to obtain the resource status of the current computing network. Specifically, the available computing resources of the current computing network are obtained through the computing network resource information, including processor type, memory capacity, network bandwidth, computing power, current load, task queue, delay and bandwidth, node location and node cost. Then, task performance indicators (response time, processing time, bandwidth, latency, computing power), task priority and task execution status instructions are obtained from the user resource demand information. The available computing resources of the current computing network and the task performance indicators obtained together constitute the resource status of the current computing network.

[0047] The computing nodes that meet the resource requirements corresponding to the business SLA request refer to the computing nodes whose type, capacity, performance, cost, bandwidth and latency all meet the business SLA request.

[0048] The forwarding plane is used to process the optimal resource computing node combination obtained by the control plane using load balancing and Quality of Service (QoS) routing algorithms to obtain the optimal scheduling solution corresponding to the business SLA request (it is used to realize the mapping and negotiation of "user needs" and "computing interconnection resources" between computing power applications and computing power networks, as well as computing power network programmability and business automatic adaptation, to complete computing power control and computing power perception).

[0049] The computing power network scheduling layer (one of the two core functional module layers for realizing the computing power perception function system) is connected to the computing power network perception layer, the computing power resource layer, and the computing power routing layer respectively, and is used to successively perform resource monitoring, task monitoring, and billing management operations on the regional computing power network resources obtained by the computing power network perception layer, the computing power network resource information obtained by the computing power resource layer, and the optimal resource computing power node combination obtained by the computing power routing layer, so as to obtain resource monitoring results, task monitoring results, and billing results respectively (to realize computing power operation) 。

[0050] Specifically, the resource monitoring process is as follows: first, the computing network resource information obtained by the computing resource layer is monitored and summarized in real time to obtain the monitoring statistical results of the computing network as the resource monitoring results, including but not limited to: number of jobs, job status, data location, network bandwidth, computing / storage resource load, etc.

[0051] Specifically, the task monitoring process is as follows: first, the optimal resource computing power node combination obtained from the computing power routing layer is classified according to the task ID of each task in the business SLA request to obtain the optimal resource computing power node combination corresponding to the task; then, based on the obtained optimal resource computing power node combination corresponding to the task (that is, all jobs under the task), the computing power network resource information obtained from the computing power resource layer is monitored and summarized in real time to obtain the job status information corresponding to the task; finally, all the job status information corresponding to all the tasks is summarized to obtain the overall real-time monitoring information as the task monitoring result, including recording key information such as task running status, progress, and performance.

[0052] This process makes it easy for users to understand the task execution status at any time, realize real-time tracking and management of users' tasks, and ensure that tasks can be successfully completed according to user needs.

[0053] Specifically, the billing management process is as follows: first, the optimal resource computing power node combination from the computing power routing layer is classified according to the task ID of each task in the business SLA request to obtain the optimal resource computing power node combination corresponding to the task; then, based on the obtained optimal resource computing power node combination corresponding to each task (that is, all jobs under the task), the computing power network resource information obtained from the computing power resource layer is monitored and summarized in real time to obtain the resource usage information of all jobs of the task in the cluster; finally, the resource usage information of all jobs of the task in the cluster is mapped according to the computing power capacity template to obtain the cost result generated by the task using resources in the computing power network, and the cost results corresponding to all tasks constitute the billing result.

[0054] This process is called a dynamic metering and billing mechanism. By establishing a service-oriented system center billing coordination service, it ensures that users only pay for the resources they actually use, and also ensures that resource providers receive reasonable benefits.

[0055] It will be easily understood by those skilled in the art that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A regional computing network infrastructure for computing power scheduling and services, including a computing power node infrastructure layer, a computing power network perception layer, a computing power description layer, a computing power resource layer, a computing power routing layer, a computing power service layer, and a computing power network scheduling layer, characterized in that: The computing node infrastructure layer is used to provide a variety of computing resources and send all computing resources to the computing network perception layer; The computing power network perception layer is used to perform network access processing on multiple computing power resources from the computing power node infrastructure layer through the network to obtain regional computing power network resources and send the regional computing power network resources to the computing power description layer. The computing power description layer is used to map the regional computing power network resources from the computing power network perception layer using the computing power capability template, so as to map the regional computing power network resources to a unified dimension, obtain an abstract description of the regional computing power network resources, and send the abstract description of the regional computing power network resources to the computing power resource layer. The computing power resource layer is used to perform service-oriented encapsulation processing on the abstract description of the regional computing power network resources from the computing power description layer to obtain computing power network resource information and send the computing power network resource information to the computing power routing layer. The computing power service layer is used to obtain the business service level agreement (SLA) request from the user through the API gateway, parse the business SLA request to generate user resource demand information, and send the user resource demand information to the computing power routing layer. The computing power routing layer includes the control plane and the forwarding plane; The control plane is used to extract indicators from the computing network resource information from the computing resource layer and the user resource demand information from the computing service layer to obtain the resource status of the current computing network. Based on the resource status, the control plane uses a multi-objective optimization scheduling method and the hot, cold and warm characteristics of the application to obtain computing nodes that meet the resource requirements corresponding to the business SLA request from the current computing network. All the computing nodes obtained constitute the optimal resource computing node combination. The forwarding plane is used to process the optimal resource computing node combination obtained by the control plane using load balancing and quality of service QoS routing algorithm to obtain the optimal scheduling solution corresponding to the business SLA request. The computing power network scheduling layer is connected to the computing power network perception layer, the computing power resource layer, and the computing power routing layer respectively, and is used to perform resource monitoring, task monitoring, and billing management operations on the regional computing power network resources obtained by the computing power network perception layer, the computing power network resource information obtained by the computing power resource layer, and the optimal resource computing power node combination obtained by the computing power routing layer, so as to obtain resource monitoring results, task monitoring results, and billing results respectively.

2. The regional computing power network infrastructure for computing power scheduling and services according to claim 1, characterized in that: The computing node infrastructure layer includes computing nodes in multiple data centers; Data centers include supercomputing centers, provincial government clouds, municipal and state clouds, and data centers above a certain scale. Computing resources include one or any combination of CPU, GPU, FPGA, DPU, and DSP computing capabilities.

3. The regional computing power network infrastructure for computing power scheduling and service according to claim 1 or 2, characterized in that: The computing capacity template is a rule that provides standard computing power measurements in computing power routing, computing power equipment management, computing power billing and other functions; Service-oriented encapsulation processing provides resource models based on the computing needs of different applications and encapsulates them into computing application programming interfaces (APIs) to meet the diverse computing needs of various industries. The computing network resource information is used to evaluate the computing resources currently available in the computing network, and includes processor performance, memory size, storage space, network and connection speed, etc.

4. The regional computing network infrastructure for computing power scheduling and services according to any one of claims 1 to 3, characterized in that: The process of parsing the business SLA request is to obtain the specific requirements of the task submitted by the user from the business SLA request, including the amount of calculation, data scale, and execution time.

5. The regional computing power network infrastructure for computing power scheduling and services according to claim 4, characterized in that: The index extraction operation is performed on the computing network resource information from the computing resource layer and the user resource demand information from the computing service layer to obtain the resource status of the current computing network. Specifically, the available computing resources of the current computing network are first obtained through the computing network resource information, including processor type, memory capacity, network bandwidth, computing power, current load, task queue, delay and bandwidth, node location and node cost. Then, the task performance indicators (including response time, processing time, bandwidth, delay, computing power), task priority and task execution status instructions are obtained from the user resource demand information. The available computing resources of the current computing network and the task performance indicators obtained together constitute the resource status of the current computing network. The computing nodes that meet the resource requirements corresponding to the business SLA requests refer to the computing nodes whose type, capacity, performance, cost, bandwidth and latency all meet the business SLA requests.

6. The regional computing network infrastructure for computing power scheduling and services according to claim 5, characterized in that: The resource monitoring process is as follows: first, the computing network resource information obtained by the computing resource layer is monitored and summarized in real time to obtain the monitoring statistics of the computing network as the resource monitoring results, including the number of jobs, job status, data location, network bandwidth, and computing / storage resource load, etc.

7. The regional computing network infrastructure for computing power scheduling and services according to claim 6, characterized in that: The task monitoring process is as follows: first, the optimal resource computing power node combination obtained from the computing power routing layer is classified according to the task ID of each task in the business SLA request to obtain the optimal resource computing power node combination corresponding to the task; then, based on the obtained optimal resource computing power node combination corresponding to the task, the computing power network resource information obtained from the computing power resource layer is monitored and summarized in real time to obtain the job status information corresponding to the task; finally, all the job status information corresponding to all the tasks is summarized to obtain the overall real-time monitoring information as the task monitoring result, including key information such as task running status, progress, and performance.

8. The regional computing power network infrastructure for computing power scheduling and services according to claim 7, characterized in that: The billing management process is as follows: first, the optimal resource computing power node combination from the computing power routing layer is classified according to the task ID of each task in the business SLA request to obtain the optimal resource computing power node combination corresponding to the task; then, based on the obtained optimal resource computing power node combination corresponding to each task (that is, all jobs under the task), the computing power network resource information obtained from the computing power resource layer is monitored and summarized in real time to obtain the resource usage information of all jobs of the task in the cluster; finally, the resource usage information of all jobs of the task in the cluster is mapped according to the computing power capacity template to obtain the cost result generated by the task using resources in the computing power network, and the cost results corresponding to all tasks constitute the billing result.

Images