Network resource slicing-based computing power network control method, system and device and medium

Abstract

The embodiment of the invention provides a computing power network control method and system based on network resource slicing. The method comprises the following steps: constructing a global resource map based on heterogeneous resource information and network connectivity; determining a plurality of logic slices and a resource table corresponding to each logic slice based on the global resource map and a slice configuration strategy; obtaining task information of a to-be-processed task, and matching a target slice for the to-be-processed task based on the task information and a task allocation rule; in response to the condition that the target slice is not matched, updating the plurality of logic slices based on a resource compression strategy, and re-matching the target slice based on the updated plurality of logic slices; and in response to the matched target slice, processing the to-be-processed task based on the target slice, and updating the resource table corresponding to the global resource map based on the target slice.

Description

Technical Field

[0001] This specification relates to the field of intelligent monitoring, and in particular to a computing power network control method, system, device, and medium based on network resource slicing. Background Technology

[0002] With the rapid development of the digital economy, computing networks, as a new type of information infrastructure that deeply integrates computing and network resources, are becoming a crucial cornerstone supporting the diverse computing needs of modern society. In computing networks, how to efficiently manage and schedule resources to meet dynamically changing user demands and fully utilize network resources is one of the core issues in current computing network development. Network resource slicing technology, as an emerging resource management method, is gradually being introduced into computing networks. This technology logically isolates physical network resources into multiple independent slices, each of which can dynamically allocate and expand resources on demand to meet specific business needs. Network resource slicing technology possesses the characteristics of logical isolation, dynamic allocation, on-demand expansion, and fine-grained control, providing important support for achieving efficient management and flexible scheduling of computing networks. Although network resource slicing technology has shown significant advantages in resource management and dynamic adjustment, how to apply it to the dynamic reconstruction of computing network resources to further solve the challenges of dynamic demand and multi-dimensional resource collaborative optimization remains a direction that urgently needs exploration.

[0003] Therefore, this paper aims to provide a computing network control method, system, device, and medium based on network resource slicing. Leveraging the technological advantages of network resource slicing, the computing network can quickly, flexibly, and with low disruption reconfiguration of allocated resources based on real-time conditions, and achieve unified management of computing power, algorithms, data, and heterogeneous resources. By applying network resource slicing technology to computing network resource scheduling, fine-grained resource management can be achieved, thereby fully utilizing all resources. Summary of the Invention

[0004] This specification provides one or more embodiments of a computing power network control method based on network resource slicing, comprising: constructing a global resource graph based on heterogeneous resource information and network connectivity; determining multiple logical slices and a resource table corresponding to each logical slice based on the global resource graph and a slice configuration strategy; obtaining task information of a task to be processed, and matching a target slice for the task to be processed based on the task information and task allocation rules; in response to not matching the target slice, updating the multiple logical slices based on a resource compression strategy, and re-matching the target slice based on the updated multiple logical slices; in response to matching the target slice, processing the task to be processed based on the target slice, and updating the resource table corresponding to the target slice in the global resource graph.

[0005] Furthermore, the method further includes: acquiring the load information of each logical slice in real time; in response to the existence of an abnormal slice whose load information meets a preset load condition, determining a compressed slice based on the resource compression strategy; distributing the computing resources of the compressed slice to the abnormal slice, updating the resource tables corresponding to the compressed slice and the abnormal slice, and updating the global resource graph.

[0006] Furthermore, the method further includes: in response to the completion of processing of the task to be processed, releasing the logical slice corresponding to the task to be processed, and updating the corresponding resource table and the global resource graph.

[0007] Furthermore, the method further includes: obtaining the current number of slices; in response to the current number of slices being greater than a preset number threshold, determining the slices to be merged based on a slice merging strategy, and merging the released logical slices into the slices to be merged.

[0008] Furthermore, the method further includes: in response to the failure to match the target slice of the task to be processed after updating the plurality of logical slices based on the resource compression strategy; determining a task to be replaced based on the task attributes of the task to be processed and the task attributes of the currently processed task; in response to the matching of the task to be replaced, using the logical slice corresponding to the task to be replaced as the target slice of the task to be processed; and in response to the failure to match the task to be replaced, indicating resource allocation failure and returning the estimated waiting time.

[0009] Furthermore, the method further includes: determining the waiting time based on the task attributes of the task to be processed and the task attributes of the currently being processed, using a prediction model, wherein the prediction model is a machine learning model.

[0010] Furthermore, the global resource graph includes at least one of computing nodes, network nodes, storage nodes, algorithm nodes, and data nodes; the edges of the global resource graph are configured to connect two nodes that satisfy preset connection conditions; the preset connection conditions include the existence of a physical or logical network connection relationship between the two nodes; and / or, the existence of a logical affiliation or dependency relationship between the two nodes.

[0011] Furthermore, the nodes of the global resource graph include inherent attributes and proprietary attributes. The inherent attributes include at least one of node ID, resource type, node location, and resource configuration. The proprietary attributes of the computing nodes include at least one of computing power, current utilization, availability status, and allocated quota. The proprietary attributes of the network nodes include at least one of device type, number of ports, and status. The proprietary attributes of the storage nodes include at least one of storage capacity, used space, and read / write bandwidth. The proprietary attributes of the algorithm nodes include at least one of algorithm identifier and version, input / output format, average processing latency, maximum concurrency, and current load. The proprietary attributes of the data nodes include at least one of dataset size, data format, access permissions, access popularity, and replica location.

[0012] This specification provides one or more embodiments of a computing power network control system based on network resource slicing, comprising: a construction module configured to construct a global resource graph based on heterogeneous resource information and network connectivity; a determination module configured to determine multiple logical slices and a resource table corresponding to each logical slice based on the global resource graph and a slice configuration strategy; a matching module configured to acquire task information of a task to be processed, and match a target slice for the task to be processed based on the task information and task allocation rules; an update module configured to update the multiple logical slices based on a resource compression strategy in response to not matching the target slice, and re-match the target slice based on the updated multiple logical slices; and an allocation module configured to process the task to be processed based on the target slice in response to matching the target slice, and update the resource table corresponding to the target slice in the global resource graph.

[0013] This specification provides one or more embodiments of a computing power network control device based on network resource slicing, including a processor, the processor being used to execute a computing power network control method based on network resource slicing.

[0014] This specification provides one or more embodiments of a storage medium that stores computer instructions. When a computer reads the computer instructions from the storage medium, the computer executes a computing power network control method based on network resource slicing. Attached Figure Description

[0015] Figure 1 This is a schematic diagram of the system structure of a computing power network control system based on network resource slicing, as shown in some embodiments of this specification.

[0016] Figure 2 This is an exemplary flowchart of a computing power network control method based on network resource slicing, as shown in some embodiments of this specification.

[0017] Figure 3 These are exemplary schematic diagrams of prediction models shown according to some embodiments of this specification;

[0018] Figure 4 This is an exemplary flowchart of a computing power network control method based on network resource slicing in one embodiment;

[0019] Figure 5 This is a logical schematic diagram of a computing power network control method based on network resource slicing in one embodiment. Detailed Implementation

[0020] To more clearly illustrate the technical solutions of the embodiments in this specification, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are merely some examples or embodiments of this specification. For those skilled in the art, these drawings can be applied to other similar scenarios without creative effort. Unless obvious from the context or otherwise specified, the same reference numerals in the drawings represent the same structures or operations.

[0021] It should be understood that the terms “system,” “device,” “unit,” and / or “module” used herein are one way to distinguish different components, elements, parts, sections, or assemblies at different levels. However, if other terms can achieve the same purpose, they may be replaced by other expressions.

[0022] As indicated in this specification and claims, unless the context clearly indicates otherwise, the words "a," "an," "an," and / or "the" do not specifically refer to the singular and may also include the plural. Generally speaking, the terms "comprising" and "including" only indicate the inclusion of expressly identified steps and elements, which do not constitute an exclusive list, and the method or apparatus may also include other steps or elements.

[0023] Flowcharts are used in this specification to illustrate the operations performed by the system according to embodiments of this specification. It should be understood that the preceding or following operations are not necessarily performed in exact order. Instead, the steps can be processed in reverse order or simultaneously. Furthermore, other operations can be added to these processes, or one or more steps can be removed from them.

[0024] Figure 1 This is a schematic diagram of the system structure of a computing power network control system 100 based on network resource slicing, as shown in some embodiments of this specification.

[0025] In some embodiments, the computing power network control system 100 based on network resource slicing may include a construction module 110, a determination module 120, a matching module 130, an update module 140, and an allocation module 150.

[0026] The building module is configured to construct a global resource graph based on heterogeneous resource information and network connectivity.

[0027] The determination module is configured to determine multiple logical slices and the resource table corresponding to each logical slice based on the global resource map and slice configuration strategy.

[0028] The matching module is configured to acquire task information of the task to be processed, and match target slices for the task to be processed based on the task information and task allocation rules.

[0029] The update module is configured to, in response to the failure to match the target slice, update the plurality of logical slices based on a resource compression strategy, and re-match the target slice based on the updated plurality of logical slices.

[0030] The allocation module is configured to, in response to a match with the target slice, process the task to be processed based on the target slice, and update the resource table of the global resource graph based on the target slice.

[0031] In some embodiments, the computing power network control system 100 may further include a first processing module, which is configured to acquire load information of each logical slice in real time; in response to the existence of an abnormal slice whose load information meets a preset load condition, determine a compressed slice based on the resource compression strategy; distribute the computing resources of the compressed slice to the abnormal slice, update the resource table corresponding to the compressed slice and the abnormal slice, and update the global resource map.

[0032] In some embodiments, the computing power network control system 100 may further include a second processing module, which is configured to release the logical slice corresponding to the task to be processed and update the corresponding resource table and the global resource map in response to the completion of the processing of the task to be processed.

[0033] In some embodiments, the computing power network control system 100 may further include a third processing module, which is configured to obtain the current number of slices, and in response to the current number of slices being greater than a preset number threshold, determine the slices to be merged based on a slice merging strategy, and merge the released logical slices into the slices to be merged.

[0034] In some embodiments, the computing power network control system 100 may further include a fourth processing module, which is configured to: in response to the following situation where, after updating the plurality of logical slices based on the resource compression strategy, no target slice for the task to be processed is still matched: determine a task to be replaced based on the task attributes of the task to be processed and the task attributes of the currently processed task; in response to the matching of the task to be replaced, use the logical slice corresponding to the task to be replaced as the target slice for the task to be processed; and in response to the failure to match the task to be replaced, indicate resource allocation failure and return the estimated waiting time.

[0035] In some embodiments, all or some of the modules included in the computing power network control system 100 may be integrated into the processor. For further description of the above modules, see [link to documentation]. Figures 2 to 4 The corresponding content.

[0036] It should be noted that the above description of the computing power network control system and its modules based on network resource slicing is for convenience only and should not be construed as limiting this specification to the scope of the embodiments described. It is understood that those skilled in the art, after understanding the principles of the system, may arbitrarily combine the various modules or construct subsystems connected to other modules without departing from these principles. In some embodiments, Figure 1 The construction module 110, determination module 120, matching module 130, update module 140, and allocation module 150 disclosed herein can be different modules within a single system, or a single module can implement the functions of two or more of the aforementioned modules. For example, the modules can share a single storage module, or each module can have its own separate storage module. Such variations are all within the scope of protection of this specification.

[0037] Figure 2 This is an exemplary flowchart illustrating a computing power network control method based on network resource slicing, according to some embodiments of this specification. Figure 2 As shown, process 200 includes the following steps. In some embodiments, process 200 may be executed by a processor.

[0038] Step 210: Construct a global resource graph based on heterogeneous resource information and network connectivity.

[0039] Heterogeneous resource information refers to information describing the attributes and status of computing, storage, network, algorithm, or data resources. For example, heterogeneous resource information includes the resource type, computing power, storage capacity, network bandwidth, current utilization, geographical or network location, and availability status.

[0040] Network connectivity refers to the property that enables any two or more resource nodes in a computing network to establish an effective data communication path based on the underlying physical or virtual network infrastructure.

[0041] A global resource graph is a data model that uses a graph structure to visualize or logically represent the resources and their relationships in a resource catalog. For example, a node in a global resource graph can represent a server (computing resource), and node attributes include its CPU utilization; an edge can represent a network link between servers, and edge attributes include the current latency and available bandwidth of that link.

[0042] The global resource graph includes at least one of computing nodes, network nodes, storage nodes, algorithm nodes, and data nodes.

[0043] Compute nodes can correspond to physical or virtual entities that provide computing power, such as physical servers, GPU clusters, edge computing devices, virtual machines (VMs), and container groups (Pods).

[0044] Network nodes can correspond to connection and forwarding devices in the network, such as routers, switches, gateways, load balancers, and Network Functions Virtualization (NFV) instances.

[0045] Storage nodes can correspond to entities that provide persistent data storage, such as storage arrays, distributed storage service nodes (like Ceph OSD), network attached storage (NAS), and database instances.

[0046] Algorithm nodes can be encapsulated and provide logical entities with specific algorithm capabilities, such as service instances that deploy machine learning models, microservices that provide image recognition APIs, and specific data processing function instances.

[0047] Data nodes can correspond to a specific set of data or source that is accessible in the system, such as a specific database, a file dataset (such as a directory in HDFS), or a real-time data stream (such as a Kafka Topic).

[0048] In some embodiments, the nodes of the global resource graph include inherent attributes and proprietary attributes. The inherent attributes include at least one of the following: node ID (such as a globally unique identifier), resource type (such as "CPU server", "NVSwitch", "object storage", "ResNet50 service", "training dataset V1.0"), node location (such as physical location (data center, rack) or network location (IP address, network segment)), and resource configuration (such as the total capacity limit of resources (such as the total number of CPU cores, total storage capacity in GB, maximum number of concurrent calls of the algorithm)).

[0049] The proprietary attributes of the computing node include at least one of the following: computing power (such as floating-point operations per second (FLOPS), processor clock speed), current utilization (such as CPU / GPU utilization percentage), availability status (such as whether it is healthy or can be scheduled for new tasks), and allocated quota (such as the amount of resources already occupied by logical slices).

[0050] The proprietary attributes of the network node include at least one of the following: device type (such as core switch, access switch), number of ports, and status.

[0051] The proprietary attributes of the storage node include at least one of storage capacity (such as total capacity), used space, and read / write bandwidth.

[0052] The proprietary attributes of the algorithm node include at least one of the following: algorithm identifier and version, input / output format, average processing latency, maximum concurrency, and current load (such as the number of requests being processed).

[0053] The proprietary attributes of the data node include at least one of the following: dataset size, data format, access permissions, access popularity, and replica location.

[0054] In some embodiments, the edges of the global resource graph are configured to connect two nodes that satisfy preset connection conditions; the preset connection conditions include the existence of a physical or logical network connection relationship between the two nodes; and / or, the existence of a logical affiliation or dependency relationship between the two nodes.

[0055] For example, if there is a direct, communicative physical link (such as a straight-through network cable) or logical link (such as an overlay network connection established through tunneling) between two nodes, an edge is established between them. This is typically based on the configuration information of network devices (such as MAC tables, ARP tables, SDN controller views) or the results of network discovery protocols (such as LLDP). As an example, if the network card of server Node-A is connected to the Eth0 / 1 port of switch Switch-1, then there is an edge between Node-A and Switch-1. Similarly, if two data centers are interconnected via a leased line, then there is an edge between the nodes representing the two core routers (or gateways).

[0056] For example, if there are dependencies such as deployment, mounting, or invocation between resources, edges can be established to represent these logical relationships, aiding in coordinated scheduling or fault impact analysis. As an example, an "algorithm node" (such as a model service) deployed on a "compute node" (virtual machine) could have a "deployed on" edge between them. Similarly, if a "compute node" (application server) frequently accesses a "data node" (database), there could be a "high-frequency access" edge between them.

[0057] In some embodiments, the attributes of an edge are primarily used to describe the performance and quality of a network link. These are key parameters for network resource scheduling (especially for meeting task bandwidth and latency requirements), including: bandwidth, latency, utilization, packet loss rate, and cost.

[0058] Bandwidth refers to the theoretical or maximum available data transmission rate of a link (e.g., 10 Gbps). Latency refers to the one-way or round-trip transmission delay of data packets from the source node to the destination node (e.g., 5 ms). Utilization refers to the percentage of current link bandwidth being used. Packet loss rate is an optional attribute, the probability of data packets being lost on the link. Cost / Cost is an optional attribute, the expense or internal cost metric incurred from using the link.

[0059] In some embodiments, the processor can acquire heterogeneous resource information and network connectivity based on monitoring probes or agents, and then construct a global resource map.

[0060] Step 220: Based on the global resource map and the slice configuration strategy, determine multiple logical slices and the resource table corresponding to each logical slice.

[0061] A logical slice is a resource collection unit that is formed by logically dividing physical resources, containing one or more resource types, and can be managed independently. For example, a logical resource slice can be allocated 10 CPU cores, 1 GPU, 100GB of memory, a copy of a specific path planning algorithm, and 500Mbps of dedicated network bandwidth.

[0062] The slice configuration strategy can include the initial number of slices n, the initial allocation ratio of each resource type, etc. For example, the slice configuration strategy can include pre-dividing the resources into n logical slices according to the default configuration, and evenly allocating computing power, algorithm data resources and network resources to each slice.

[0063] A resource table is a data structure used to record and track the details and usage status of resources contained within a logical slice. For example, a slice's resource table would list all virtual machine instance IDs, allocated storage volume IDs, occupied network ports and bandwidth quotas currently owned by that slice, and mark whether each resource instance is being used by a task.

[0064] The processor can logically allocate resources from the global resource pool according to the slicing configuration strategy, forming n initial logical slices. "Logically" does not actually involve physical migration of the physical resources; they are merely marked with ownership at the management level. Simultaneously, the processor can create an independent resource table for each slice, recording the specific list of resources allocated to that slice and their current usage status. Furthermore, it can update the status of these resources in the global resource graph to "allocated to a slice" to achieve synchronization.

[0065] Through the above-mentioned slicing, this solution can organize the "scattered" physical resources into flexible units (slices) that can be independently managed and scheduled, providing operational objects for "on-demand allocation".

[0066] Step 230: Obtain the task information of the task to be processed, and match the target slice for the task to be processed based on the task information and task allocation rules.

[0067] Task information for pending tasks can be obtained based on the task requirements submitted by the user, and the task information may include the task priority.

[0068] A task requirement refers to a description of resource needs submitted by a user or system to accomplish a specific computational goal. For example, a task requirement may include the required floating-point computing power, the specified version of the image recognition algorithm, the name of the training dataset to be accessed, the minimum required network bandwidth, and an indication of the urgency of the task.

[0069] Priority refers to an identifier used to distinguish the importance or order of priority of different tasks or logical resource slices during resource allocation and adjustment. For example, priority can be a numerical level (such as 1-10), with higher values ​​indicating greater importance and higher protection in resource competition.

[0070] The task allocation rules include that a logical slice whose current idle amount is greater than or equal to the task requirement and whose resource type matches can be used as the target slice.

[0071] Step 240: In response to the failure to match the target slice, update the plurality of logical slices based on the resource compression strategy, and rematch the target slice based on the updated plurality of logical slices.

[0072] Resource compression strategies can be based on the principle of "priority weighting". For example, the higher the priority of a slice, the smaller the proportion of resources that are compressed (contributed), or even zero (guaranteed by the highest priority); while for slices with low load and low priority, calculations are performed, and in their resource tables, without affecting the normal operation of their existing tasks, a portion of idle or compressible resources are logically "allocated". These resources are logically removed from the resource table of the source slice and logically added to the resource table of new tasks or slices with insufficient resources.

[0073] The processor can update the multiple logical slices based on the task information of the task to be processed and the task information of all currently being processed tasks, and rematch the target slice based on the updated multiple logical slices, so as to use the compressed resources to combine a "virtual slice" that meets the requirements of the new task, or to supplement the slice with insufficient resources.

[0074] Step 250: In response to matching the target slice, process the task to be processed based on the target slice, and update the resource table corresponding to the target slice in the global resource graph.

[0075] In some embodiments, the processor may assign a target slice to a task to be processed and update the slice resource table and global resource graph (marking resources as "in use"), and the task begins execution.

[0076] In some embodiments, process 200 may further include: acquiring load information for each logical slice in real time; in response to the existence of an abnormal slice whose load information meets preset load conditions, determining a compressed slice based on the resource compression strategy; distributing the computing resources of the compressed slice to the abnormal slice, updating the resource tables corresponding to the compressed slice and the abnormal slice, and updating the global resource graph.

[0077] In some embodiments, the processor can obtain the actual load of resources within each slice in real time (such as high CPU utilization or insufficient bandwidth) based on deployed monitoring probes.

[0078] In some embodiments, when a slice's resources are detected to be nearing or already insufficient, the processor can treat this as a "restructuring event." Dynamic restructuring is performed: the resource compression strategy from step 240 is reused, but the triggering conditions and targets are different. In this case, the compression source is another slice with relatively light load or lower priority. The compressed resources are dynamically transferred to the resource-deficient slice.

[0079] In some embodiments, the above steps can transform the system from one-time allocation to full lifecycle elastic management, directly addressing the core problem that the allocation mechanism cannot quickly respond to dynamic changes.

[0080] In some embodiments, the method further includes: in response to the completion of processing of the task to be processed, releasing the logical slice corresponding to the task to be processed, and updating the corresponding resource table and the global resource graph.

[0081] For example, after a task is completed, the processor can mark all resources occupied by that task in the slice's resource table as "released", restore the slice's resource table to its initial idle state (but the resources still belong to the slice), and mark these resources as idle and available in the global resource graph.

[0082] Furthermore, the method further includes: obtaining the current number of slices; in response to the current number of slices being greater than a preset number threshold, determining the slices to be merged based on a slice merging strategy, and merging the released logical slices into the slices to be merged.

[0083] In some embodiments, the preset number threshold may be the maximum number of slices preset by the processor. If there are released slices, the processor first determines the number of existing slices. If there are too many idle slices, causing the total number to approach or potentially exceed the preset number threshold, the processor can select the released slice or any idle slice, merge its resource logic into another idle or lightly loaded slice, and then delete the empty slice to ensure that the system maintains an efficient and streamlined slice set and avoids excessive management overhead.

[0084] By performing the above operations, the health and efficiency of the system architecture can be maintained during long-term operation, preventing the logical slices from expanding indefinitely. This is an important optimization step to ensure the sustainable operation of the solution.

[0085] Furthermore, the method further includes: in response to the failure to match the target slice of the task to be processed after updating the plurality of logical slices based on the resource compression strategy; determining a task to be replaced based on the task attributes of the task to be processed and the task attributes of the currently processed task; in response to the matching of the task to be replaced, using the logical slice corresponding to the task to be replaced as the target slice of the task to be processed; and in response to the failure to match the task to be replaced, indicating resource allocation failure and returning the estimated waiting time.

[0086] Furthermore, the method further includes: determining the waiting time based on the task attributes of the task to be processed and the task attributes of the currently being processed, using a prediction model, wherein the prediction model is a machine learning model. For more information on the prediction model, see [link to relevant documentation]. Figure 3 The corresponding content.

[0087] It should be noted that the above description of process 200 is for illustrative purposes only and does not limit the scope of this specification. Those skilled in the art can make various modifications and changes to process 200 under the guidance of this specification. However, these modifications and changes remain within the scope of this specification.

[0088] Figure 3 These are exemplary schematic diagrams of prediction models shown according to some embodiments of this specification. Figure 3 As shown, prediction model 330 is a machine learning model.

[0089] In some embodiments, the processor may determine the waiting time 340 based on the task attributes 310 of the task to be processed and the task attributes 320 of the task currently being processed, using a prediction model 330. The waiting time 340 refers to the time that the task to be processed needs to wait for to begin execution.

[0090] Prediction model 330 can be a machine learning model, such as a neural network model.

[0091] like Figure 3 As shown, the input to the prediction model 330 can include the task attribute 310 of the task to be processed and the task attribute 320 of the task currently being processed, and the output of the prediction model 330 can include the waiting time 340. More information about task attributes can be found in [link to relevant documentation]. Figure 2 The relevant description is provided. In some embodiments, the prediction model may also output the confidence level corresponding to the waiting time.

[0092] In some embodiments, the prediction model can be trained using multiple first training samples with a first label. For example, multiple first training samples with a first label can be input into an initial prediction model. A loss function is constructed using the first labels and the output of the initial prediction model. The parameters of the initial prediction model are iteratively updated based on the loss function using gradient descent or other methods. When preset conditions are met, a trained prediction model is obtained. These preset conditions may include loss function convergence, the number of iterations reaching a threshold, etc.

[0093] In some embodiments, the first training sample may include at least the task attributes of the sample task and the task attributes of the task in the sample processing.

[0094] The first label can be the sample task corresponding to the first training sample, and the waiting time from when it is sent to the system to when it is actually executed, in the case of the first training sample. The first label can be determined based on system records or manual review, and can be obtained through manual annotation.

[0095] In the embodiments of this specification, the processor uses a prediction model to combine the task attributes of the task to be processed and the task attributes of the currently being processed to evaluate the waiting time, which can obtain a highly reliable time prediction and improve the accuracy of the final waiting time.

[0096] In the embodiments of this specification, the waiting time is determined by a prediction model. The self-learning ability of the machine learning model can be used to find patterns from a large amount of simulated data, obtain the relationship between the task attributes of the task to be processed and the task attributes and waiting time of the currently processed task, and improve the accuracy of the output waiting time.

[0097] Figure 4 This is an exemplary flowchart of a computing power network control method based on network resource slicing in one embodiment.

[0098] like Figure 4 The computing power network control method based on network resource slicing may include the following steps:

[0099] S1: Unified management of heterogeneous resources and network resources of computing power network, and construction of resource graph to achieve real-time monitoring and dynamic update synchronization.

[0100] Step S1 may specifically include the following steps:

[0101] S11: Collects basic information on computing power, algorithms, data (including heterogeneous computing resources such as CPUs and GPUs) and network resources (such as bandwidth, link status, and latency) to build a unified resource catalog.

[0102] S12: Construct a real-time resource map using resource catalog information, marking the utilization rate, location, and availability status of each resource.

[0103] S13: Monitor resources in the resource graph to ensure their real-time performance in large-scale systems.

[0104] S2: Use network resource slicing technology to divide physical resources into multiple logical slices.

[0105] In some embodiments, step S2 may include the following steps:

[0106] S21: Based on the default configuration, the resources are pre-divided into n logical slices, and computing power, algorithm data resources and network resources are evenly allocated to each slice.

[0107] S22: Initialize the resource table in each logical slice, waiting to be assigned to user requirements.

[0108] S23: Monitor the status of each slice, including resource allocation and resource utilization.

[0109] S3: Based on the task requirements submitted by the user, parse the three required resource elements and network resources, and allocate logical fragments to the requirements.

[0110] In some embodiments, step S3 may include the following steps:

[0111] S31: When a user's request arrives, the system first analyzes the user's request and parses it into a request list and priority. For example, computationally intensive requests require significantly more computing power than algorithms and data.

[0112] S32: Based on the parsed list of requirements and priorities, assign a logical slice to the requirement, and the slice will also acquire priority.

[0113] S33: If a slice cannot meet the required amount of resources, the system will perform a resource compression operation on all slices. The principle of compression is to allocate idle resources to the unassigned requirements and slices without affecting the execution of tasks in the slice, thereby realizing resource reconstruction.

[0114] S34: If the resources of a slice are excessive for the demand, the system will retain some resource redundancy in the slice based on the current global situation, and then reclaim the excess resources and allocate them to other slices with scarce or idle resources to wait for subsequent demand.

[0115] S35: Resource compression operations take priority into account and are weighted according to priority. The higher the priority, the less it is compressed, and the highest priority slice may not be compressed or reduced.

[0116] S4: Monitor user needs in real time, dynamically adjust the resource configuration of slices in response to changes, and reconstruct computing network resources.

[0117] In some embodiments, step S4 may include the following steps:

[0118] S41: Monitor the load changes of resources in the slice in real time. When the slice resources are insufficient, perform resource compression operation on slices with low load and low priority to release more resources for high priority tasks. Reconstruct the resource composition in the slice, and add (for high priority slices) or delete (for low priority slices) some resources on the basis of retaining the original composition.

[0119] S42: After resource reconstruction, maintain the resource tables in each slice and the global resource graph to ensure that they are always synchronized.

[0120] S43: By using real-time synchronized resource tables and global resource graphs, it quickly finds idle resources to prepare for the next compression.

[0121] S5: After the task is completed, reclaim the resources in the logical slice.

[0122] In some embodiments, step S5 may include the following steps:

[0123] S51: When the task is completed, release the resources in the slice, that is, initialize the resource table and synchronize the global resource graph.

[0124] S52: Based on the current number of slices, decide whether to retain the current slice or merge the resources into other slices. The principle is to keep the total number of slices below the maximum value (max) and wait for subsequent demand.

[0125] Figure 5This is a logical diagram illustrating a computing network control method based on network resource slicing in one embodiment. Taking the following scenario as an example: A computing network platform needs to provide computing power and network support to multiple users. Submitted tasks include high-performance tasks (such as scientific computing), low-latency real-time tasks (such as autonomous driving data analysis), and high-bandwidth-demand tasks (such as high-definition video processing). The platform's resources are widely distributed and highly heterogeneous, including various CPUs, GPUs, and other computing power resources, various algorithm resources, various data resources, and network resources (such as bandwidth and link latency). The dynamic changes in user task requirements and resource competition make it difficult for traditional scheduling methods to meet the requirements of real-time performance and efficiency.

[0126] like Figure 5 As shown, the computing power network control method based on network resource slicing may include the following operations:

[0127] Step S101: First, all resources on the platform need to be registered and managed in a unified manner.

[0128] Collect computing power, algorithms, and data resources, and define a unified resource representation vector as follows:

[0129]

[0130] Where T represents the type of resource. M represents computing power, U represents storage size, U represents current utilization, and A represents availability (1 for available, 0 for unavailable). The network resources are then represented in the form of a topology diagram. Where N represents the set of network nodes and L represents the set of links, and the bandwidth B, latency D, and utilization of each link are labeled. This leads to the construction of a resource catalog and a global resource graph.

[0131] Step S102: First, divide the resources into n logical slices according to the default configuration. Each slice is allocated computing power, algorithm data, and network resources. Maintain a resource table to record the resource utilization rate of the slice. The load status L provides a reference for subsequent resource adjustments.

[0132] Step S103: Parse the user-submitted task requirement T and extract key task parameters:

[0133] ,in These represent the required computing power, algorithm, data, bandwidth, and priority information, respectively.

[0134] Based on the parsed task list, tasks are assigned to a suitable slice. If no suitable slice is available, a resource compression mechanism is triggered. This mechanism reserves sufficient redundant resources for online tasks based on priority and system availability, and contributes any excess resources. The resource compression process ensures that higher-priority slices are less affected.

[0135] Step S104: Monitor the working status of the slices. When a slice has insufficient resources, the system will compress the resources. Similarly, the higher the priority, the less resources are compressed. The compressed resources are then given to the slices with insufficient resources, completing the overall resource configuration reconstruction and improving the system's resource utilization.

[0136] After dynamic reconstruction, the slice resource table and global resource graph are updated to ensure that the resource status is synchronized in real time.

[0137] Step S105: When the user task is completed, the resources in the slice are released, i.e., remarked as available, and the resource table is initialized. A decision is made based on the current number of slices to decide whether to merge the released slice resources into another slice. The principle is to ensure that the number of slices in the system remains within a reasonable range below the maximum value.

[0138] In summary, this solution has the following significant effects:

[0139] Achieving rapid, flexible, and low-disruption dynamic reconfiguration of network resources: Leveraging the logical isolation characteristics of network resource slicing technology, resources are divided into independent logical units, each of which can be dynamically reconfigured as needed. When resources are insufficient, resource coordination mechanisms between slices are utilized to quickly release and reconfigure resources through compression. The dynamic reconfiguration process has low disruptive impact on online tasks, significantly improving the system's adaptability to sudden demands.

[0140] It supports fine-grained resource management, improving resource utilization and reducing system operating costs: Through the dynamic allocation and on-demand expansion characteristics of network resource slicing technology, fine-grained management of computing power, algorithms, data, and network resources is achieved. Resource utilization is monitored in real time within each slice, and resources are allocated weighted by task priority. After task completion, resource distribution is optimized through resource reclamation and slice merging, reducing resource waste and minimizing waiting time for subsequent operations.

[0141] It provides a unified resource management framework that can fully tap into the system's potential and improve overall scheduling efficiency: through a resource catalog and a global resource graph, it integrates computing power, algorithms, data, and network resources into a unified management framework. It manages heterogeneous resources and achieves real-time synchronization of resource status.

[0142] This approach breaks down the isolation of resource management, enabling collaborative optimization between slices and improving overall system performance. By maintaining slice states through a global resource graph, dynamic resource transfer and collaborative optimization between slices are achieved. Combined with real-time monitoring and feedback mechanisms, resource reallocation can be performed globally, optimizing overall system performance.

[0143] The basic concepts have been described above. Obviously, for those skilled in the art, the detailed disclosure above is merely illustrative and does not constitute a limitation of this specification. Although not explicitly stated herein, those skilled in the art may make various modifications, improvements, and corrections to this specification. Such modifications, improvements, and corrections are suggested in this specification and therefore remain within the spirit and scope of the exemplary embodiments described herein.

[0144] Furthermore, this specification uses specific terms to describe embodiments thereof. For example, "an embodiment," "one embodiment," and / or "some embodiments" refer to a particular feature, structure, or characteristic associated with at least one embodiment of this specification. Therefore, it should be emphasized and noted that references to "an embodiment," "one embodiment," or "an alternative embodiment" in different locations throughout this specification do not necessarily refer to the same embodiment. Moreover, certain features, structures, or characteristics in one or more embodiments of this specification can be appropriately combined.

[0145] Furthermore, unless expressly stated in the claims, the order of processing elements and sequences, the use of numbers and letters, or other names described in this specification are not intended to limit the order of the processes and methods described herein. Although various examples have been discussed in the foregoing disclosure of some embodiments of the invention that are currently considered useful, it should be understood that such details are for illustrative purposes only, and the appended claims are not limited to the disclosed embodiments; rather, the claims are intended to cover all modifications and equivalent combinations that conform to the spirit and scope of the embodiments described herein. For example, while the system components described above can be implemented using hardware devices, they can also be implemented solely using software solutions, such as installing the described system on existing servers or mobile devices.

[0146] Similarly, it should be noted that, in order to simplify the description disclosed herein and thus aid in the understanding of one or more embodiments of the invention, the foregoing description of embodiments in this specification may sometimes combine multiple features into a single embodiment, drawing, or description thereof. However, this method of disclosure does not imply that the subject matter of this specification requires more features than those mentioned in the claims. In fact, the embodiments contain fewer features than all the features of a single embodiment disclosed above.

[0147] In some embodiments, numbers describing the quantity of components and attributes are used. It should be understood that such numbers used in the description of embodiments are modified in some examples with the terms "approximately," "approximately," or "generally." Unless otherwise stated, "approximately," "approximately," or "generally" indicates that the numbers are allowed to vary by ±20%. Accordingly, in some embodiments, the numerical parameters used in the specification and claims are approximate values, which may be changed depending on the characteristics required by individual embodiments. In some embodiments, numerical parameters should take into account specified significant digits and employ a general method of digit reservation. Although the numerical ranges and parameters used to confirm their breadth of range in some embodiments of this specification are approximate values, in specific embodiments, such values ​​are set as precisely as feasible.

[0148] For each patent, patent application, patent application publication, and other material such as articles, books, specifications, publications, and documents referenced in this specification, the entire contents of which are incorporated herein by reference. This excludes historical application documents that are inconsistent with or conflict with the content of this specification, as well as documents that limit the broadest scope of the claims in this specification (currently or subsequently appended to this specification). It should be noted that in the event of any inconsistency or conflict between the descriptions, definitions, and / or terminology used in the supplementary materials to this specification and the content of this specification, the descriptions, definitions, and / or terminology used in this specification shall prevail.

[0149] Finally, it should be understood that the embodiments described in this specification are merely illustrative of the principles of the embodiments described herein. Other variations may also fall within the scope of this specification. Therefore, alternative configurations of the embodiments described herein are intended to be illustrative rather than limiting, and are considered consistent with the teachings of this specification. Accordingly, the embodiments described herein are not limited to those explicitly introduced and described herein.

Claims

1. A method for controlling a computing power network based on network resource slicing, characterized in that, The method comprises: constructing a global resource graph based on heterogeneous resource information and network connectivity; determining a plurality of logical slices and a resource table corresponding to each logical slice based on the global resource graph and slice configuration strategy; obtaining task information of a to-be-processed task, and matching a target slice for the to-be-processed task based on the task information and task allocation rule; in response to not matching the target slice, updating the plurality of logical slices based on a resource compression strategy, and re-matching the target slice based on the updated plurality of logical slices; in response to matching the target slice, processing the to-be-processed task based on the target slice, and updating the global resource graph based on the resource table corresponding to the target slice.

2. The method of claim 1, wherein, The method further comprises: obtaining load information of each logical slice in real time; in response to there being an abnormal slice whose load information meets a preset load condition, determining a compressed slice based on the resource compression strategy; downloading the computing resources of the compressed slice to the abnormal slice, updating the resource table corresponding to the compressed slice and the abnormal slice, and updating the global resource graph.

3. The method of claim 1, wherein, The method further comprises: in response to the to-be-processed task being completed, releasing the logical slice corresponding to the to-be-processed task, and updating the corresponding resource table and global resource graph.

4. The method of claim 3, wherein, The method further comprises: obtaining the current number of slices, and in response to the current number of slices being greater than a preset number threshold, determining a to-be-merged slice based on a slice merging strategy, and merging the released logical slice into the to-be-merged slice.

5. The method of claim 1, wherein, The method further comprises: in response to not matching the target slice of the to-be-processed task after updating the plurality of logical slices based on the resource compression strategy: determining a to-be-replaced task based on the task attribute of the to-be-processed task and the task attribute of the task being processed; in response to matching the to-be-replaced task, taking the logical slice corresponding to the to-be-replaced task as the target slice of the to-be-processed task; in response to not matching the to-be-replaced task, prompting resource allocation failure and returning an expected waiting time.

6. The method of claim 1, wherein, The global resource graph comprises at least one of a computing node, a network node, a storage node, an algorithm node, and a data node; The edges of the global resource graph are configured to connect two nodes that meet a preset connection condition; the preset connection condition comprises a physical or logical network connection relationship between the two nodes; and / or a logical belonging or dependency relationship between the two nodes.

7. The method of claim 6, wherein, The nodes of the global resource graph contain inherent attributes and special attributes; the inherent attributes comprise at least one of a node ID, a resource type, a node location, and a resource configuration; The special attributes of the computing node comprise at least one of computing power, current utilization rate, available state, and allocated quota; The special attributes of the network node comprise at least one of device type, port number, and state; The special attributes of the storage node comprise at least one of storage capacity, used space, and read-write bandwidth. The special attributes of the algorithm nodes include at least one of algorithm identification and version, input / output format, average processing delay, maximum concurrency, and current load; The special attributes of the data nodes include at least one of data set size, data format, access permission, access heat, and replica location. 8.A computing power network control system based on network resource slicing, characterized in that, Comprise: A construction module configured to construct a global resource graph based on heterogeneous resource information and network connectivity; A determination module configured to determine a plurality of logical slices and a resource table corresponding to each of the logical slices based on the global resource graph and a slice configuration strategy; A matching module configured to obtain task information of a to-be-processed task, and match a target slice for the to-be-processed task based on the task information and a task allocation rule; An update module configured to, in response to not matching the target slice, update the plurality of logical slices based on a resource compression strategy, and re-match the target slice based on the updated plurality of logical slices; An allocation module configured to, in response to matching the target slice, process the to-be-processed task based on the target slice, and update the global resource graph based on the resource table corresponding to the target slice. 9.A computing power network control device based on network resource slicing, comprising a processor, characterized in that, The processor is configured to execute the network resource slice-based computing power network control method in any of claims 1-7.

10. A computer-readable storage medium storing computer instructions, wherein, When a computer reads computer instructions in a storage medium, the computer executes the network resource slice-based computing power network control method in any of claims 1-7.

Images