A low-latency computing power network optimization method based on software-defined network

Abstract

The application discloses a low-latency computing power network optimization method based on a software-defined network, relates to the technical field of software-defined computing power network optimization, and comprises the following steps: receiving and analyzing a low-latency computing power service request, generating a computing power service scheduling descriptor in combination with a preset service arrangement decision time limit; perceiving network state information used for service communication within a service arrangement domain range configured in a service arrangement domain, screening and judging the network state information, and generating a service reachable path set; and based on a candidate service instance scheme set, performing joint scheduling decision of a computing power service node and a communication path under the constraint of the service arrangement decision time limit, and generating a low-latency computing power network service deployment result. Through determination of the service arrangement domain range and the service state perception granularity, the application reduces the influence of irrelevant state information on scheduling decision, improves the accuracy of computing power and network collaborative scheduling, and guarantees stable deployment of the low-latency computing power network service.

Description

Technical Field

[0001] This invention relates to the field of software-defined computing network optimization technology, and in particular to a low-latency computing network optimization method based on software-defined networks. Background Technology

[0002] With the rapid development of cloud computing, edge computing, and artificial intelligence applications, computing power networks for multi-service scenarios are gradually becoming an important form of network and computing convergence. In existing technologies, low-latency computing power services are usually achieved through a collaborative approach of network resource scheduling and computing power resource allocation. Among these, software-defined networking (SDN) is widely used for network scheduling and path control of computing power services due to its characteristics of separation of control and forwarding, centralized programmability, and global view. Conventional methods are generally based on a software-defined network controller, which parses service requests and completes network path selection, and then combines computing power resource catalogs and registration information to achieve computing power service discovery and deployment.

[0003] However, in low-latency computing service scenarios, existing methods usually only focus on the accessibility and availability of network and computing resources, without clearly constraining the time consumption of service orchestration decisions, resulting in insufficient correlation between scheduling timeliness and end-to-end latency targets; the scope of service orchestration and the granularity of network status awareness are mostly statically configured, making it difficult to adjust with changes in service requirements, and easily introducing redundant status information and candidate computing nodes. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides a low-latency computing power network optimization method based on software-defined networks to solve the problems of uncontrollable timeliness of service orchestration decisions and difficulty in dynamically adjusting the service orchestration scope and state awareness granularity in low-latency computing power service scenarios.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: This invention provides a method for optimizing low-latency computing power networks based on software-defined networks (SDNs). The method includes: receiving and parsing low-latency computing power service requests, and generating a computing power service scheduling descriptor by combining a preset service orchestration decision time limit; using the computing power service scheduling descriptor to determine the service orchestration architecture of the SDN controller, and determining the service orchestration domain range and service status awareness granularity that can participate in service discovery and scheduling, generating a service orchestration domain configuration; within the service orchestration domain range defined by the service orchestration domain configuration, sensing network status information used for service communication, and filtering and constraining the network status information to generate a set of service reachable paths; initiating computing power service discovery within the network service range defined by the service reachable path set, obtaining computing power resource information used to carry low-latency computing power services, and performing reverse filtering of the computing power resource information to bind computing power service nodes that meet the network service range with corresponding low-latency communication paths, generating a candidate service instance scheme set; based on the candidate service instance scheme set, performing joint scheduling decisions of computing power service nodes and communication paths under the constraints of the service orchestration decision time limit, generating a low-latency computing power network service deployment result.

[0007] As a preferred embodiment of the low-latency computing power network optimization method based on software-defined networking described in this invention, the steps for generating the computing power service scheduling descriptor are as follows: Receive and parse low-latency computing power service requests, extract computing power resource requirements and end-to-end latency requirements, and generate service requirement specifications. The service requirement specifications are integrated with the preset service orchestration decision time limit to generate a computing power service scheduling descriptor.

[0008] As a preferred embodiment of the low-latency computing power network optimization method based on software-defined networking described in this invention, the steps for determining the service orchestration architecture of the software-defined network controller using a computing power service scheduling descriptor are as follows: The service orchestration decision time limit in the computing power service scheduling descriptor is used as the time constraint parameter for service scheduling decision to generate scheduling delay constraints. Based on scheduling latency constraints, service scheduling processing latency parameters of the software-defined network controller are obtained under centralized service orchestration architecture, hierarchical service orchestration architecture and edge autonomous service orchestration architecture. Based on the service scheduling processing latency parameters, calculate the estimated decision latency required for each service orchestration architecture to complete the service scheduling decision, and generate the service orchestration architecture latency assessment results.

[0009] As a preferred embodiment of the low-latency computing power network optimization method based on software-defined networking described in this invention, the steps of determining the service orchestration domain range and service state awareness granularity that can participate in service discovery and scheduling, and generating service orchestration domain configurations, are as follows: By using scheduling latency constraints, the latency evaluation results of service orchestration architectures are constrained and filtered to obtain a set of feasible service orchestration architectures. Based on the set of feasible service orchestration architectures, the target service orchestration architecture, service orchestration level and coverage are determined according to the principle of minimizing expected decision latency, and the service orchestration domain boundary is generated. Based on the service orchestration domain boundary, configure the service state awareness granularity that matches the service orchestration level to generate the service orchestration domain configuration.

[0010] As a preferred embodiment of the low-latency computing power network optimization method based on software-defined networking described in this invention, the steps for sensing network status information used for service communication are as follows: Based on the service orchestration domain configuration, network link status, node load and available bandwidth information are perceived within the service orchestration domain to generate the original network service status set; Based on the service orchestration decision time limit, the original network service state set is filtered for timeliness, and dynamic indicators of links and nodes that constitute end-to-end latency are extracted to generate a latency-critical service state set.

[0011] As a preferred embodiment of the low-latency computing power network optimization method based on software-defined networking described in this invention, the steps of filtering and constraining network state information to generate a set of service reachable paths are as follows: Based on the state set of latency-critical services, the total end-to-end path latency is calculated, and latency constraint filtering is performed to generate a preliminary service reachable path set. Perform network reachability and capacity checks on the initial set of service reachable paths to generate a new set of service reachable paths.

[0012] As a preferred embodiment of the low-latency computing power network optimization method based on software-defined networking described in this invention, the steps for obtaining computing power resource information for supporting low-latency computing power services are as follows: Based on the network service range defined by the set of service reachable paths, determine the logical identifiers of all potential computing power service nodes within the network service range, and generate a candidate computing power node list; Query the predefined computing power service registry center to obtain real-time computing power resource information for each computing power service node in the candidate computing power node list.

[0013] As a preferred embodiment of the low-latency computing power network optimization method based on software-defined networks described in this invention, the steps for generating the candidate service instance scheme set are as follows: Based on the computing resource requirements in the computing service scheduling descriptor, capacity and performance compliance of computing resource information are determined, and a set of available computing service nodes is obtained. The available computing power service node set is matched and bound with the service reachable path set, and a communication path that meets the end-to-end latency constraint is associated with each computing power service node to construct a candidate service instance scheme set.

[0014] As a preferred embodiment of the low-latency computing power network optimization method based on software-defined networking described in this invention, the steps for performing joint scheduling decisions of computing power service nodes and communication paths to generate low-latency computing power network service deployment results are as follows: Extract the logical node identifier of computing power service, the associated communication path identifier, and the end-to-end latency parameters of each service instance scheme from the candidate service instance scheme set to generate a service instance parameter set; Based on the computing power service scheduling descriptor, the service instance parameter set is filtered for scheduling decision latency feasibility to generate a schedulable service instance set; Based on the schedulable service instance set, the service instance schemes are calculated and sorted according to the end-to-end latency satisfaction, generating the deployment result of low-latency computing power network services.

[0015] As a preferred embodiment of the low-latency computing power network optimization method based on software-defined networks described in this invention, the steps for calculating and sorting service instance schemes according to the degree of end-to-end latency satisfaction are as follows: Based on the set of schedulable service instances, the degree of satisfaction between the end-to-end latency of each service instance scheme and the end-to-end latency requirement in the computing power service scheduling descriptor is calculated, and service instance latency quality assessment results are generated. Based on the service instance latency quality assessment results, the schedulable service instance set is sorted, the target service instance scheme that meets the end-to-end latency requirements is selected, and the low-latency computing power network service deployment results are generated.

[0016] The beneficial effects of this invention are as follows: By receiving and parsing low-latency computing power service requests, and simultaneously generating a computing power service scheduling descriptor in conjunction with service orchestration decision time limits, the selection of computing power service nodes and communication path decisions are completed within time constraints, enhancing the controllability of low-latency computing power service deployment; by utilizing the computing power service scheduling descriptor to determine the service orchestration architecture of the software-defined network controller and to determine the service orchestration domain range and service status awareness granularity, the impact of irrelevant status information on scheduling decisions is reduced, improving the accuracy of computing power and network collaborative scheduling and ensuring the stable deployment of low-latency computing power network services. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart of a low-latency computing power network optimization method based on software-defined networks.

[0019] Figure 2 A flowchart generated for configuring the service orchestration domain.

[0020] Figure 3 A flowchart generated for the set of reachable paths to the service.

[0021] Figure 4 The flowchart shows the candidate service instance scheme and joint scheduling. Detailed Implementation

[0022] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0023] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0024] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0025] Reference Figures 1-4 This is one embodiment of the present invention, which provides a low-latency computing power network optimization method based on software-defined networks, including the following steps: S1. Receive and parse low-latency computing power service requests, and generate a computing power service scheduling descriptor by combining the preset service orchestration decision time limit.

[0026] S1.1 Receive and parse low-latency computing power service requests, extract computing power resource requirements and end-to-end latency requirements, and generate service requirement specifications.

[0027] Furthermore, the system receives low-latency computing power service requests and parses them according to their field structure, reading and extracting the computing power resource requirement field and the end-to-end latency requirement field item by item; it extracts the computing power type, computing power scale, and computing power performance indicators from the computing power resource requirement field as computing power resource requirement parameters, and extracts the maximum allowable end-to-end latency parameter from the end-to-end latency requirement field; and it integrates the extracted parameters to generate a service requirement specification.

[0028] S1.2 Integrate the service requirement specifications with the preset service orchestration decision time limit to generate a computing power service scheduling descriptor.

[0029] Furthermore, the preset service orchestration decision time limit is associated and bound with the service requirement specification as an independent time constraint parameter, so that the computing resource requirement parameter and the maximum allowable end-to-end latency parameter are uniformly organized with the service orchestration decision time limit under the same data structure, generating a computing service scheduling descriptor.

[0030] It should be noted that the process for setting the service orchestration decision time limit is as follows: The end-to-end latency requirement in the service requirement specification is used as the total latency budget. The end-to-end latency requirement is decomposed, and a time share is reserved for computing power service scheduling decisions. The time share is adjusted based on the scheduling processing time of the control plane at the current service orchestration level. At the same time, the effective time of the state information that the scheduling decision can rely on is matched and verified based on the network state awareness update cycle to determine the service orchestration decision time limit. An exemplary value range is 10 to 200 milliseconds. If it is less than 10 milliseconds, the service orchestration decision will not be completed within the specified time, resulting in a scheduling interruption. If it is more than 200 milliseconds, the scheduling decision will take too long and consume too much end-to-end latency budget.

[0031] The expression for obtaining the service orchestration decision timeframe is: ; in, It is the timeframe for service orchestration decisions; This represents the share of time allocated to the service orchestration decision phase within the maximum permissible end-to-end latency, used to characterize the proportion of time that service orchestration decisions account for in the overall end-to-end latency budget. It is the maximum permissible end-to-end latency parameter in the service requirement specification; It is the effective time of state information that scheduling decisions can rely on; It is the network state awareness update cycle; It is the network load factor, which reflects the current level of activity of the entire network.

[0032] It should be noted that the network load factor is obtained by sensing network link status, node load and available bandwidth information within the network scope configured in the service orchestration domain to generate an original network service status set, and extracting key dynamic indicators (such as link utilization and node CPU load) for normalization and weighted aggregation calculation. The exemplary value range is 0 to 5.

[0033] S2. Use the computing power service scheduling descriptor to determine the service orchestration architecture of the software-defined network controller, and determine the service orchestration domain range and service status awareness granularity that can participate in service discovery and scheduling, and generate service orchestration domain configuration.

[0034] S2.1. Use the service orchestration decision time limit in the computing power service scheduling descriptor as the time constraint parameter for service scheduling decision to generate scheduling delay constraints.

[0035] Furthermore, time semantic parsing is performed on the service orchestration decision time limit in the computing power service scheduling descriptor to clarify the maximum allowable scheduling decision duration corresponding to the service orchestration decision time limit, and map it to a time threshold used to constrain the execution process of the service scheduling decision. At the same time, the time threshold is recorded as a unified time constraint parameter for service scheduling decisions, forming a scheduling delay constraint used to limit the execution time of service scheduling decisions.

[0036] S2.2. Based on scheduling delay constraints, obtain the service scheduling processing delay parameters of the software-defined network controller under centralized service orchestration architecture, hierarchical service orchestration architecture, and edge autonomous service orchestration architecture.

[0037] Furthermore, under scheduling latency constraints, for centralized service orchestration architecture, hierarchical service orchestration architecture, and edge autonomous service orchestration architecture, the scheduling processing steps involved in the corresponding service scheduling decision are determined according to the service orchestration hierarchy and the distribution relationship of control nodes. The processing time of decision calculation, state aggregation, and decision distribution processes involved in each scheduling processing step is statistically analyzed and accumulated to obtain the service scheduling processing latency parameters of the software-defined network controller under centralized service orchestration architecture, hierarchical service orchestration architecture, and edge autonomous service orchestration architecture.

[0038] S2.3. Based on the service scheduling processing latency parameters, calculate the expected decision latency required for each service orchestration architecture to complete the service scheduling decision, and generate the service orchestration architecture latency evaluation results.

[0039] Furthermore, the service scheduling processing latency parameters corresponding to the centralized service orchestration architecture, the hierarchical service orchestration architecture, and the edge autonomous service orchestration architecture are read. The processing time of the scheduling processing links obtained from statistics under each service orchestration architecture is summarized and calculated to obtain the total processing time required to complete a service scheduling decision. The total processing time corresponding to each service orchestration architecture is determined as the corresponding expected decision latency. The expected decision latency of each service orchestration architecture is uniformly sorted to generate the service orchestration architecture latency evaluation result.

[0040] S2.4. Use scheduling latency constraints to constrain and filter the service orchestration architecture latency evaluation results to obtain a set of feasible service orchestration architectures.

[0041] Furthermore, the expected decision latency corresponding to each service orchestration architecture in the service orchestration architecture latency assessment results is compared with the scheduling latency constraint to determine whether the expected decision latency meets the requirement of not exceeding the scheduling latency constraint; the service orchestration architectures that meet the scheduling latency constraint are summarized to obtain a set of feasible service orchestration architectures.

[0042] S2.5. Based on the set of feasible service orchestration architectures, determine the target service orchestration architecture, service orchestration level and coverage according to the principle of minimizing expected decision latency, and generate the service orchestration domain boundary.

[0043] Furthermore, the expected decision latency of each service orchestration architecture within the set of feasible service orchestration architectures is sorted, and the target service orchestration architecture is determined according to the principle of minimizing the expected decision latency. After determining the target service orchestration architecture, the network area range participating in service discovery and service scheduling is clarified based on the service orchestration level and control node coverage corresponding to the target service orchestration architecture, and the service orchestration domain boundary is generated.

[0044] S2.6 Based on the service orchestration domain boundary, configure the service state awareness granularity that matches the service orchestration level, and generate the service orchestration domain configuration.

[0045] Furthermore, the network area range, service orchestration level, and control node coverage range determined in the service orchestration domain boundary are read. Based on the hierarchical requirements of the service orchestration level for service status perception, the service status perception granularity level required for the corresponding level is determined. Within the network area, the service status perception granularity level is configured to correspond with the network link status, node load status, and available bandwidth status, clarifying the status types and perception precision that need to be perceived under different service orchestration levels, and generating the service orchestration domain configuration.

[0046] S3. Within the service orchestration domain configured and limited by the service orchestration domain, perceive the network status information used for service communication, filter and constrain the network status information, and generate a set of service reachable paths.

[0047] S3.1 Based on the service orchestration domain configuration, perceive network link status, node load and available bandwidth information within the service orchestration domain, and generate the original network service status set.

[0048] Furthermore, the network area range and service status awareness granularity level determined in the service orchestration domain configuration are read. Within the network area range, network link status information is collected one by one for network links participating in service communication according to the service status awareness granularity level, node load information is collected one by one for network nodes participating in service communication, and available bandwidth information corresponding to the network links is collected synchronously. The collected information is associated and summarized according to network identifiers to form an original network service status set covering all sensing objects within the service orchestration domain.

[0049] S3.2. Based on the service orchestration decision time limit, the original network service state set is filtered for timeliness, and dynamic indicators of links and nodes that constitute end-to-end latency are extracted to generate a latency-critical service state set.

[0050] Furthermore, by utilizing the effective time range corresponding to the service orchestration decision time limit, the timestamps of each network link status information, node load information, and available bandwidth information in the original network service status set are compared one by one to filter out network service status records whose collection time is within the effective time range. The filtered network service status records are parsed, and the link transmission delay index is calculated based on the link transmission rate and link occupancy recorded in the network link status information. The node queuing delay index is calculated based on the processing load level and queuing status recorded in the node load information. Bandwidth-related indicators used to characterize the link bandwidth occupancy level are extracted from the available bandwidth information and integrated to generate a latency-critical service status set.

[0051] S3.3. Based on the latency-critical service state set, calculate the total end-to-end path latency, perform latency constraint filtering, and generate a preliminary service reachable path set.

[0052] Furthermore, within the network area defined by the service orchestration domain boundary, based on the link transmission delay index, node queuing delay index, and bandwidth-related index recorded in the latency-critical service status set, candidate end-to-end communication paths are enumerated and matched segment by segment. The corresponding link transmission delay index and node queuing delay index are accumulated and calculated according to the path order to obtain the total end-to-end path delay of the candidate end-to-end communication paths. The total end-to-end path delay is compared with the maximum allowable end-to-end delay parameter one by one to select candidate end-to-end communication paths whose total end-to-end path delay does not exceed the maximum allowable end-to-end delay parameter, and a preliminary service reachable path set is generated.

[0053] S3.4 Perform network reachability and carrying capacity judgment on the preliminary service reachable path set to generate a service reachable path set.

[0054] Furthermore, for each candidate end-to-end communication path in the initial service reachable path set, the network links and corresponding network nodes are checked segment by segment along the path to ensure they are available and reachable. If there are no link interruptions or unreachable nodes in the path, the candidate end-to-end communication path is deemed to meet the network reachability requirements. Given that the network reachability requirements are met, the carrying capacity of the candidate end-to-end communication path is further assessed. The available bandwidth of each network link in the path is checked segment by segment to ensure it is not lower than the corresponding communication bandwidth requirement in the service demand specification. Simultaneously, the node load of each network node in the path is checked to ensure it is within the allowable load range for service communication. When all network links and network nodes in the path meet the above carrying conditions, the candidate end-to-end communication path is deemed to have carrying capacity. Candidate end-to-end communication paths that simultaneously meet both the network reachability and carrying capacity criteria are retained and aggregated to generate a service reachable path set.

[0055] It should be noted that communication bandwidth requirement is a constraint parameter used to limit the minimum network transmission capacity required for low-latency computing services during communication. It is obtained by parsing the computing resource requirements contained in the low-latency computing service request to obtain the data exchange scale that the computing service needs to complete per unit time during the computing execution process. At the same time, combined with the end-to-end latency requirement, the transmission completion time of the data exchange scale under the end-to-end latency constraint is constrained and calculated to determine the minimum communication bandwidth requirement required to complete data transmission within the end-to-end latency requirement. The load range is formed during the network status awareness phase. By collecting the node load information within the service orchestration domain configuration limit, the current processing load level of the network node is obtained. At the same time, combined with the maximum processing capacity that the network node can carry, the remaining available capacity between the current processing load level and the maximum processing capacity is evaluated. When the remaining available capacity can support the processing consumption required for the communication of new computing services, it is determined that the network node is within the load range that allows it to carry service communication.

[0056] S4. Initiate computing power service discovery within the network service range defined by the service reachable path set, obtain computing power resource information for carrying low-latency computing power services, and perform reverse filtering on the computing power resource information to bind computing power service nodes that meet the network service range with the corresponding low-latency communication paths, generating a candidate service instance scheme set.

[0057] S4.1 Based on the network service range defined by the set of service reachable paths, determine the logical identifiers of all potential computing power service nodes within the network service range and generate a candidate computing power node list.

[0058] Furthermore, the network node identifier corresponding to the endpoint of each end-to-end communication path in the service reachable path set is extracted. Combined with the network area range determined in the service orchestration domain boundary, the network node identifiers are range-matched to filter out network node identifiers within the network service range. The filtered network node identifiers are then verified against the preset computing power service node registration information to determine whether the network node identifier is associated with a computing power resource capability description. Network node identifiers with computing power resource capability descriptions are retained, and the corresponding computing power service node logical identifiers are read. Finally, the logical identifiers of computing power service nodes within the network service range that have computing power resource capabilities are summarized to generate a candidate computing power node list.

[0059] It should be noted that the computing power service node registration information is a structured set of information used to describe whether a network node has computing power service capabilities and corresponding computing power attributes. It is used to support node identification and capability determination in the process of computing power service discovery and scheduling. The computing power service node registration information is set when the computing power service node accesses the network and provides computing power services to the outside world. It is formed by declaring and registering the computing power capabilities of the computing power service node, including the logical identifier of the computing power service node, the type of computing power resources, the available computing power scale, computing power performance indicators, computing power service interface identifier, and computing power service availability status.

[0060] S4.2 Query the predefined computing power service registration center to obtain the real-time computing power resource information of each computing power service node in the candidate computing power node list.

[0061] Furthermore, using the logical identifier of the computing service node in the candidate computing power node list as the query index, the corresponding computing service node registration entry is located in the predefined computing power service registration center, the current computing power resource information associated with the logical identifier of the computing power service node is obtained, and the current computing power resource information is collected node by node to form a computing power resource information set.

[0062] It should be noted that the computing power service registration center is a unified information management entity used to centrally manage and provide external information on the real-time computing power resource status of computing power service nodes. The computing power service registration center is generated during the computing power network deployment phase by agreeing on a unified registration structure and query interface to centrally access and manage computing power service nodes that have completed the registration. It is used to reflect the computing power capabilities that computing power service nodes can provide externally at the current moment and to provide a basis for subsequent computing power resource screening and joint scheduling.

[0063] S4.3. Based on the computing resource requirements in the computing service scheduling descriptor, determine the capacity and performance compliance of the computing resource information and obtain the set of available computing service nodes.

[0064] Furthermore, the computing resource information corresponding to each computing service node in the candidate computing power node list is read one by one to determine whether the computing resource type is consistent with the computing resource type required in the computing power service scheduling descriptor. If the type is consistent, and the current available computing power scale is greater than or equal to the computing power scale required in the computing power service scheduling descriptor, the computing power capacity is determined to be sufficient. If the capacity is sufficient, it is determined whether the computing power performance operation status meets the computing power performance indicators required in the computing power service scheduling descriptor. Computing service nodes with consistent computing resource type, sufficient computing power capacity, and sufficient computing power performance indicators are retained and aggregated to generate an available computing power service node set.

[0065] S4.4 Match and bind the set of available computing power service nodes with the set of service reachable paths, associate each computing power service node with a communication path that satisfies the end-to-end latency constraint, and construct a set of candidate service instance schemes.

[0066] Furthermore, the logical identifiers of the computing power service nodes in the set of available computing power service nodes are read one by one. The end-to-end communication paths with the corresponding network node identifiers as the path endpoints in the set of service reachable paths are associated with the logical identifiers of the computing power service nodes and encapsulated to generate a set of candidate service instance schemes.

[0067] S5. Based on the candidate service instance scheme set, under the constraint of the service orchestration decision time limit, perform joint scheduling decision of computing power service nodes and communication paths to generate low-latency computing power network service deployment results.

[0068] S5.1 Extract the logical node identifier of computing power service, the associated communication path identifier, and the end-to-end latency parameters of each service instance scheme from the candidate service instance scheme set to generate a service instance parameter set.

[0069] Furthermore, each service instance scheme included in the candidate service instance scheme set is analyzed item by item, and the corresponding computing power service logical node identifier, associated communication path identifier, and end-to-end latency parameter are extracted from the service instance scheme. These parameters are then collected and organized according to the service instance dimension to generate a service instance parameter set.

[0070] S5.2. Based on the computing power service scheduling descriptor, perform scheduling decision latency completion screening on the service instance parameter set to generate a schedulable service instance set.

[0071] Furthermore, based on the service orchestration decision time limit recorded in the computing power service scheduling descriptor, it is determined whether the scheduling decision time required for the corresponding service instance scheme to complete the selection of computing power service nodes and communication paths is within the allowable range of the service orchestration decision time limit. Based on the end-to-end latency requirements, it is determined whether the end-to-end latency parameters corresponding to the service instance scheme meet the end-to-end latency requirements. Service instance schemes that simultaneously meet the scheduling decision time constraints and end-to-end latency requirement constraints are retained and aggregated to generate a set of schedulable service instances.

[0072] It should be noted that the scheduling decision time refers to the actual processing time required to complete the selection of computing service nodes and the determination of communication paths during the service orchestration decision-making process. It is determined by statistically summarizing the processing time of each scheduling processing link under the service orchestration architecture.

[0073] S5.3 Based on the schedulable service instance set, calculate the degree of satisfaction between the end-to-end latency of each service instance scheme and the end-to-end latency requirement in the computing power service scheduling descriptor, and generate service instance latency quality assessment results.

[0074] Furthermore, for each service instance scheme within the schedulable service instance set, the end-to-end latency parameters corresponding to the service instance scheme are compared with the end-to-end latency requirements specified in the computing power service scheduling descriptor. By calculating the proportion of the end-to-end latency parameters relative to the end-to-end latency requirements, the degree to which the service instance scheme meets the end-to-end latency requirements is quantified. The degree of satisfaction is then uniformly combined with the computing power service logical node identifier and associated communication path identifier corresponding to the service instance scheme to generate a service instance latency quality assessment result that reflects the latency quality of each service instance scheme.

[0075] S5.4. Based on the service instance latency quality assessment results, sort the set of schedulable service instances, select the target service instance scheme that meets the end-to-end latency requirements, and generate the low-latency computing power network service deployment results.

[0076] Furthermore, the service instance schemes recorded in the service instance latency quality assessment results are sorted according to their degree of satisfaction, with the service instance schemes with higher satisfaction ranking higher. In the sorting results, the end-to-end latency parameters corresponding to the service instance schemes are checked from front to back to see if they meet the end-to-end latency requirements, and the first service instance scheme that meets the end-to-end latency requirements is selected as the target service instance scheme. The computing power service logical node identifier and associated communication path identifier contained in the target service instance scheme are encapsulated to generate the low-latency computing power network service deployment result.

[0077] In summary, this invention enhances the controllability of low-latency computing service deployment by: receiving and parsing low-latency computing service requests, and simultaneously generating a computing service scheduling descriptor based on service orchestration decision time constraints, thereby enabling the selection of computing service nodes and communication path decisions to be completed within time constraints; and by utilizing the computing service scheduling descriptor to determine the service orchestration architecture of the software-defined network controller and define the service orchestration domain range and service status awareness granularity, reducing the impact of irrelevant status information on scheduling decisions, improving the accuracy of computing power and network collaborative scheduling, and ensuring the stable deployment of low-latency computing network services.

[0078] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for optimizing low-latency computing networks based on software-defined networks, characterized in that: include, Receive and parse low-latency computing power service requests, and generate computing power service scheduling descriptors by combining them with preset service orchestration decision time limits; The service orchestration architecture of the software-defined network controller is determined by using the computing power service scheduling descriptor, and the scope of service orchestration domains and the service status awareness granularity that can participate in service discovery and scheduling are determined, and service orchestration domain configuration is generated. Within the service orchestration domain configured and limited by the service orchestration domain, network status information used for service communication is perceived, and the network status information is filtered and constrained to generate a set of service reachable paths; Within the network service range defined by the set of service reachable paths, a computing power service discovery is initiated to obtain computing power resource information for carrying low-latency computing power services. The computing power resource information is then filtered in reverse to bind computing power service nodes that meet the network service range with the corresponding low-latency communication paths, generating a set of candidate service instance schemes. Based on the candidate service instance scheme set, under the constraint of service orchestration decision time limit, a joint scheduling decision of computing power service nodes and communication paths is made to generate low-latency computing power network service deployment results.

2. The low-latency computing power network optimization method based on software-defined networks as described in claim 1, characterized in that: The steps for generating the computing power service scheduling descriptor are as follows: Receive and parse low-latency computing power service requests, extract computing power resource requirements and end-to-end latency requirements, and generate service requirement specifications. The service requirement specifications are integrated with the preset service orchestration decision time limit to generate a computing power service scheduling descriptor.

3. The low-latency computing power network optimization method based on software-defined networks as described in claim 1, characterized in that: The steps for determining the service orchestration architecture of the software-defined network controller using the computing power service scheduling descriptor are as follows: The service orchestration decision time limit in the computing power service scheduling descriptor is used as the time constraint parameter for service scheduling decision to generate scheduling delay constraints. Based on scheduling latency constraints, service scheduling processing latency parameters of the software-defined network controller are obtained under centralized service orchestration architecture, hierarchical service orchestration architecture and edge autonomous service orchestration architecture. Based on the service scheduling processing latency parameters, calculate the estimated decision latency required for each service orchestration architecture to complete the service scheduling decision, and generate the service orchestration architecture latency assessment results.

4. The low-latency computing power network optimization method based on software-defined networks as described in claim 1, characterized in that: The steps for determining the service orchestration domain scope and service state awareness granularity that can participate in service discovery and scheduling, and generating service orchestration domain configurations, are as follows: By using scheduling latency constraints, the latency evaluation results of service orchestration architectures are constrained and filtered to obtain a set of feasible service orchestration architectures. Based on the set of feasible service orchestration architectures, the target service orchestration architecture, service orchestration level and coverage are determined according to the principle of minimizing expected decision latency, and the service orchestration domain boundary is generated. Based on the service orchestration domain boundary, configure the service state awareness granularity that matches the service orchestration level to generate the service orchestration domain configuration.

5. The low-latency computing power network optimization method based on software-defined networks as described in claim 1, characterized in that: The steps for sensing network status information used for service communication are as follows: Based on the service orchestration domain configuration, network link status, node load and available bandwidth information are perceived within the service orchestration domain to generate the original network service status set; Based on the service orchestration decision time limit, the original network service state set is filtered for timeliness, and dynamic indicators of links and nodes that constitute end-to-end latency are extracted to generate a latency-critical service state set.

6. The low-latency computing power network optimization method based on software-defined networks as described in claim 1, characterized in that: The steps for filtering and constraining network state information to generate a set of service reachable paths are as follows: Based on the state set of latency-critical services, the total end-to-end path latency is calculated, and latency constraint filtering is performed to generate a preliminary service reachable path set. Perform network reachability and capacity checks on the initial set of service reachable paths to generate a new set of service reachable paths.

7. The low-latency computing power network optimization method based on software-defined networks as described in claim 1, characterized in that: The steps for obtaining computing resource information used to support low-latency computing services are as follows: Based on the network service range defined by the set of service reachable paths, determine the logical identifiers of all potential computing power service nodes within the network service range, and generate a candidate computing power node list; Query the predefined computing power service registry center to obtain real-time computing power resource information for each computing power service node in the candidate computing power node list.

8. The low-latency computing power network optimization method based on software-defined networks as described in claim 1, characterized in that: The steps for generating the candidate service instance scheme set are as follows: Based on the computing resource requirements in the computing service scheduling descriptor, capacity and performance compliance of computing resource information are determined, and a set of available computing service nodes is obtained. The available computing power service node set is matched and bound with the service reachable path set, and a communication path that meets the end-to-end latency constraint is associated with each computing power service node to construct a candidate service instance scheme set.

9. The low-latency computing power network optimization method based on software-defined networks as described in claim 1, characterized in that: The steps for jointly scheduling computing power service nodes and communication paths to generate low-latency computing power network service deployment results are as follows: Extract the logical node identifier of computing power service, the associated communication path identifier, and the end-to-end latency parameters of each service instance scheme from the candidate service instance scheme set to generate a service instance parameter set; Based on the computing power service scheduling descriptor, the service instance parameter set is filtered for scheduling decision latency feasibility to generate a schedulable service instance set; Based on the schedulable service instance set, the service instance schemes are calculated and sorted according to the end-to-end latency satisfaction, generating the deployment result of low-latency computing power network services.

10. The low-latency computing power network optimization method based on software-defined networks as described in claim 9, characterized in that: The steps for calculating and ranking service instance schemes based on end-to-end latency satisfaction are as follows: Based on the set of schedulable service instances, the degree of satisfaction between the end-to-end latency of each service instance scheme and the end-to-end latency requirement in the computing power service scheduling descriptor is calculated, and service instance latency quality assessment results are generated. Based on the service instance latency quality assessment results, the schedulable service instance set is sorted, the target service instance scheme that meets the end-to-end latency requirements is selected, and the low-latency computing power network service deployment results are generated.

Images