Resource scheduling method and device

Abstract

The invention discloses a resource scheduling method and device. The method comprises the steps that cross-domain network data is collected, and the cross-domain network data at least comprises resource load indexes of a plurality of cloud nodes in a cross-domain network in a preset time period, network bandwidth indexes of the plurality of cloud nodes, a cross-domain rate function and security policy parameters; a global dynamic graph is constructed based on the cross-domain network data, and the global dynamic graph is used for representing the state of the cross-domain network; obtaining a plurality of sub-networks corresponding to each cloud node, and determining the global state of the cross-domain network based on a pre-constructed partial differential equation operator, the global state including the state of each cloud node in the plurality of sub-networks; determining a resource scheduling strategy among the plurality of cloud nodes according to a pre-constructed discrete resource flow operator and the global state of the cross-domain network; and executing a resource scheduling strategy among the plurality of cloud nodes, and allocating resources of the plurality of cloud nodes.

Description

Technical Field

[0001] The present application relates to the fields of artificial intelligence, cloud computing, and large model technology, and more specifically, to a resource scheduling method and device. Background Art

[0002] In hybrid or multi-cloud environments, domains often have different network bandwidth, service costs, and security policies. To achieve efficient cross-domain network data transmission and direct network collaboration, existing technical solutions generally combine distributed scheduling algorithms or resource orchestration systems to formulate network flow paths. However, these solutions still face various technical challenges in application, especially when considering factors such as cross-domain bandwidth, costs, and security compliance. There are problems such as the inability to simultaneously consider multi-dimensional hard constraints, the lack of detailed characterization of dynamic and time-varying environments, and the single-minded approach to handling security scenarios, lacking multi-expert sub-networks or multi-factor collaboration.

[0003] Existing technologies for cross-domain network data transmission and scheduling in multi-cloud or hybrid clouds generally suffer from the inability to simultaneously address bandwidth, cost, and security requirements in dynamic, time-varying environments, and to effectively and rapidly iterate globally balanced resource allocation solutions based on scenario-based policies. This is especially true when there's a conflict between security and cost, or when there are peak loads, requiring more flexible and sophisticated cloud-based routing and scheduling mechanisms to adapt to diverse business policies. Summary of the Invention

[0004] The embodiments of the present application provide a resource scheduling method and device to at least solve the technical problem that the resource scheduling accuracy is low due to the related technology determining the resource scheduling strategy based on a single dimensional indicator.

[0005] According to one aspect of an embodiment of the present application, a resource scheduling method is provided, characterized in that it includes: collecting cross-domain network data, wherein the cross-domain network data at least includes: resource load indicators of multiple cloud nodes in the cross-domain network within a preset time period, network bandwidth indicators of the multiple cloud nodes, cross-domain rate functions and security policy parameters; constructing a global dynamic graph based on the cross-domain network data, wherein the global dynamic graph is used to represent the state of the cross-domain network; obtaining multiple sub-networks corresponding to each cloud node, and determining the global state of the cross-domain network based on a pre-constructed partial differential equation operator, wherein the global state includes: the state of each cloud node under the multiple sub-networks; determining a resource scheduling strategy between the multiple cloud nodes according to a pre-constructed discrete resource flow operator and the global state of the cross-domain network; executing the resource scheduling strategy between the multiple cloud nodes, and allocating resources to the multiple cloud nodes.

[0006] Optionally, obtaining multiple sub-networks corresponding to each cloud node includes: respectively obtaining state functions of multiple sub-networks corresponding to each cloud node, wherein the multiple sub-networks include at least one of the following: a security sub-network, a cost sub-network, a load sub-network, and a delivery sub-network; the state functions of the multiple sub-networks include at least one of the following: a first state function for representing the state of the cloud node in the security sub-network, a second state function for representing the state of the cloud node in the cost sub-network, a third state function for representing the state of the cloud node in the load sub-network, and a fourth state function for representing the state of the cloud node in the delivery sub-network. The first state function is used to represent the security assessment value of the cloud node in the security subnetwork, the second state function is used to represent the cost pressure value of the cloud node in the cost subnetwork, the third state function is used to represent the load saturation of the cloud node in the load subnetwork, and the fourth state function is used to represent the data flow deliverability of the cloud node in the delivery subnetwork; obtaining the global state of the cross-domain network includes: grouping the state functions of the multiple subnetworks into a state function set; determining the partial differential equation operator according to the state function set, and determining the global state of the cross-domain network based on the partial differential equation operator.

[0007] Optionally, determining the partial differential equation operator based on the state function set includes: obtaining a first parameter of the cloud node, wherein the first parameter includes at least one of the following: a domain node set of the cloud node, a coupling coefficient between the cloud node and the neighborhood node, a penalty term of the business scenario corresponding to the cloud node, an adjustment coefficient for adjusting the weight of the penalty term, and an external incentive term for representing external incentives; respectively obtaining the state function set of the cloud node and the state function set of the neighborhood node; determining the partial differential equation operator based on the first parameter of the cloud node, the state function set of the cloud node, and the state function set of the neighborhood node.

[0008] Optionally, determining the global state of the cross-domain network based on the partial differential equation operator includes: obtaining the subnetwork corresponding to the cloud node and the business scenario corresponding to the cloud node; determining the constraints of the partial differential equation operator based on the subnetwork corresponding to the cloud node and the business scenario corresponding to the cloud node, and the constraints of the partial differential equation operator are used to correct the cloud node state output by the partial differential equation operator; iterating the initial state of the cloud node based on the partial differential equation operator and the constraints of the partial differential equation operator until the state of the cloud node converges to obtain the final state of the cloud node; and combining the final state of each cloud node to obtain the global state.

[0009] Optionally, the resource scheduling strategy between the multiple cloud nodes is determined based on a pre-constructed discrete resource flow operator and the global state of the cross-domain network, including: determining the cost pressure value and security assessment value of each edge in the cross-domain network based on the global state; obtaining the edge capacity and edge bandwidth of each edge in the cross-domain network, and determining the initial traffic of each edge based on the edge capacity, the edge bandwidth, the cost pressure value and the security assessment value of each edge to obtain an initial traffic distribution matrix; using the discrete resource flow operator to update the initial traffic distribution matrix until the traffic distribution matrix converges to obtain a final traffic distribution matrix; determining the resource scheduling strategy between the multiple cloud nodes based on the final traffic distribution matrix.

[0010] Optionally, the discrete resource flow operator is used to update the initial traffic allocation matrix until the traffic allocation matrix converges to obtain a final traffic allocation matrix, including: obtaining a second parameter of each edge in the cross-domain network, wherein the second parameter includes at least one of the following: a security assessment value of the edge, the available bandwidth of the edge, the maximum capacity of the edge, and the load difference between the two nodes connected by the edge; and constructing the discrete resource flow operator based on the second parameter.

[0011] Optionally, the discrete resource flow operator is used to update the initial traffic allocation matrix until the traffic allocation matrix converges to obtain a final traffic allocation matrix, including: obtaining the traffic transmission rates for different time periods; determining the total cost of traffic transmission based on the traffic transmission rates for different time periods; and determining that the traffic allocation matrix converges when the load saturation and safety assessment values ​​corresponding to the traffic allocation matrix meet preset requirements and the total cost of the traffic transmission no longer decreases.

[0012] Optionally, the method further includes: using each subnetwork in turn to predict each cloud node in the cross-domain network to obtain a prediction result for each subnetwork; obtaining the contribution weight of each subnetwork, and using the contribution weight of each subnetwork to weight the prediction result of each subnetwork to obtain a weighted result for each subnetwork; determining the weighted result of each subnetwork as the initial state value, using the partial differential equation operator to iterate the initial state value until convergence, and obtaining a final state value; determining an initial traffic distribution matrix based on the final state value, and using the discrete resource flow operator to update the initial traffic distribution matrix until the objective function value is minimized, and obtaining a final traffic distribution matrix.

[0013] Optionally, the method further includes: respectively obtaining the contribution of each subnetwork in the process of determining the cloud node status; determining the net contribution of each subnetwork in the process of determining the cloud node status based on the contribution of each subnetwork in the process of determining the cloud node status and a preset evaluation operator; and eliminating the subnetworks whose net contribution is lower than a preset threshold from the multiple subnetworks.

[0014] According to another aspect of an embodiment of the present application, a resource scheduling device is also provided, characterized in that it includes: an acquisition module for collecting cross-domain network data, wherein the cross-domain network data at least includes: resource load indicators of multiple cloud nodes in the cross-domain network within a preset time period, network bandwidth indicators of the multiple cloud nodes, cross-domain rate functions and security policy parameters; a construction module for constructing a global dynamic graph based on the cross-domain network data, wherein the global dynamic graph is used to represent the state of the cross-domain network; a determination module for obtaining multiple sub-networks corresponding to each cloud node, and determining the global state of the cross-domain network based on a pre-constructed partial differential equation operator, wherein the global state includes: the state of each cloud node under multiple sub-networks; a policy module for determining the resource scheduling strategy between the multiple cloud nodes based on the pre-constructed discrete resource flow operator and the global state of the cross-domain network; and an execution module for executing the resource scheduling strategy between the multiple cloud nodes and allocating resources to the multiple cloud nodes.

[0015] According to another aspect of the embodiment of the present application, a computer device is further provided, characterized in that it includes: a memory and a processor, wherein the memory is used to store program instructions; the processor is connected to the memory and is used to execute the above-mentioned resource scheduling method.

[0016] According to another aspect of the embodiments of the present application, a non-volatile storage medium is provided, which includes a stored computer program, wherein the device where the non-volatile storage medium is located executes the above-mentioned resource scheduling method by running the computer program.

[0017] According to another aspect of the embodiments of the present application, a computer program product is provided, including computer instructions, characterized in that when the computer instructions are executed by a processor, the above-mentioned resource scheduling method is implemented.

[0018] In an embodiment of the present application, cross-domain network data is collected, wherein the cross-domain network data includes at least: resource load indicators of multiple cloud nodes in the cross-domain network within a preset time period, network bandwidth indicators of the multiple cloud nodes, cross-domain rate functions and security policy parameters; a global dynamic graph is constructed based on the cross-domain network data, wherein the global dynamic graph is used to represent the state of the cross-domain network; multiple sub-networks corresponding to each cloud node are obtained, and the global state of the cross-domain network is determined based on a pre-constructed partial differential equation operator, wherein the global state includes: the state of each cloud node under the multiple sub-networks; a resource scheduling strategy between the multiple cloud nodes is determined based on a pre-constructed discrete resource flow operator and the global state of the cross-domain network; the resource scheduling strategy between the multiple cloud nodes is executed, and resources of the multiple cloud nodes are allocated. By determining the global state of the cross-domain network based on partial differential equation operators through cross-domain network data, and then determining the resource scheduling strategy, the purpose of taking into account multi-dimensional data in a dynamic time-varying environment in cross-domain network data transmission and scheduling of multi-cloud or hybrid clouds to achieve timely scheduling and resource flow optimization, thereby achieving the technical effect of improving the accuracy of resource scheduling, and thus solving the technical problem of low resource scheduling accuracy due to the fact that related technologies determine resource scheduling strategies based on single-dimensional indicators. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0020] Figure 1 is a hardware structure block diagram of a computer terminal for implementing a resource scheduling method according to an embodiment of the present application;

[0021] Figure 2 is a flow chart of a resource scheduling method according to an embodiment of the present application;

[0022] Figure 3 is a flowchart of another resource scheduling method according to an embodiment of the present application;

[0023] Figure 4 is a flowchart of another resource scheduling method according to an embodiment of the present application;

[0024] Figure 5 is a flow chart of determining the global state of the cross-domain network based on the partial differential equation operator according to an embodiment of the present application;

[0025] Figure 6 This is a flowchart of constructing a discrete resource flow operator DRFO according to an embodiment of the present application;

[0026] Figure 7 This is a structural diagram of a resource scheduling device according to an embodiment of the present application. DETAILED DESCRIPTION

[0027] In order to enable those skilled in the art to better understand the present invention, the following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts should fall within the scope of protection of this application.

[0028] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequential order. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in a sequence other than those illustrated or described herein. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0029] In order to better understand the embodiments of the present application, the technical terms involved in the embodiments of the present application are explained as follows:

[0030] Partial Differential Equation Operator (PDE Operator):

[0031] A technical mechanism that describes the collaborative evolution of multiple expert subnetworks (such as security, cost, load, and delivery subnetworks) by iteratively solving coupling equations between nodes on a graph or discrete space. By incorporating state differences, scenario-based boundary conditions, and external incentives from each subnetwork into a single coupling model, it achieves dynamic balancing and iterative calculation of multi-dimensional hard constraints (security, cost, load, and high availability).

[0032] DRFO operator (Discrete Resource Flow Operator):

[0033] One proposed approach is to develop a discrete optimization technology for cross-domain network edge (link) traffic scheduling. By incorporating multiple criteria, including security requirements, bandwidth capacity constraints, time-based costs, and load drivers, the technology iterates and updates traffic allocation for each link. While ensuring security compliance and cost control, it achieves fine-grained migration and optimization of cross-domain network data flows, ultimately delivering a globally optimal or suboptimal resource flow allocation solution.

[0034] The information collected in the embodiments of the present application is information and data authorized by the user or fully authorized by all parties, and the collection, storage, use, processing, transmission, provision, disclosure and application of the relevant data comply with the relevant laws, regulations and standards of the relevant regions, take necessary confidentiality measures, do not violate public order and good morals, and provide corresponding operation entrances for users to choose to authorize or reject the automated decision results; if the user chooses to reject, the expert decision-making process will be entered.

[0035] In order to solve the problems existing in the related art, the embodiment of the present application provides a resource scheduling method, which can be run on Figure 1 In the computer terminal shown, the computer terminal is explained below.

[0036] The resource scheduling method embodiment provided in the embodiment of the present application can be executed in a mobile terminal, a computer terminal or a similar computing device. Figure 1 FIG1 shows a hardware structure block diagram of a computer terminal for implementing a resource scheduling method. Figure 1 As shown, the computer terminal 10 may include one or more (illustrated by 102a, 102b, ..., 102n in the figure) processors (the processor may include but is not limited to a processing device such as a microprocessor MCU or a programmable logic device FPGA), a memory 104 for storing data, and a transmission module 106 for communication functions connected via a wired and / or wireless network. In addition, it may also include: a display, a keyboard, a cursor control device, an input / output interface (I / O interface), a universal serial bus (USB) port (which may be included as one of the ports of the I / O interface), a network interface, and a BUS bus. It will be understood by those skilled in the art that Figure 1 The structure shown is only for illustration and does not limit the structure of the above electronic device. Figure 1 More or fewer components than shown, or with Figure 1 Different configurations shown.

[0037] It should be noted that the one or more processors and / or other data processing circuitry described above can be referred to herein generically as "data processing circuitry." The data processing circuitry can be embodied as software, hardware, firmware, or any combination thereof, in whole or in part. Moreover, the data processing circuitry can be a single standalone processing module, or incorporated in whole or in part within any one of the other elements of the computer terminal 10. As referred to in the embodiments of the present application, the data processing circuitry serves as a processor to control, for example, the selection of the variable resistance terminal path in connection with the interface.

[0038] The memory 104 can be used to store software programs of application software and modules, such as program instructions / data storage means corresponding to the resource scheduling method in the embodiments of the present application. The processor executes various functional applications and data processing by running the software programs and modules stored in the memory 104, i.e. implements the resource scheduling method described above. The memory 104 can include a high-speed random access memory, and can further include a non-volatile memory, such as one or more magnetic storage devices, flash memories, or other non-volatile solid-state memories. In some examples, the memory 104 can further include a memory disposed remotely with respect to the processor, which can be connected to the computer terminal 10 through a network. Examples of the network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0039] The transmission module 106 is configured to receive or send data via a network. Specific examples of the network can include a wireless network provided by a communication provider of the computer terminal 10. In one example, the transmission module 106 includes a network interface controller (NIC) that can be connected to other network devices through a base station so as to be able to communicate with the Internet. In one example, the transmission module 106 can be a radio frequency (RF) module configured to communicate with the Internet in a wireless manner.

[0040] The display can be, for example, a touch screen type liquid crystal display (LCD) that enables a user to interact with the user interface of the computer terminal 10.

[0041] It should be noted that in some alternative embodiments, the above-mentioned Figure 1 The computer terminal shown can include hardware elements (including circuitry), software elements (including computer code stored on a computer readable medium), or a combination of both hardware and software elements. It should be noted that in some embodiments, the functions of the computer terminal described above can be performed by one or more of the other elements of the computer terminal. Figure 1 is merely one example of a particular implementation, and is intended to illustrate the types of components that can be present in the computer terminal described above.

[0042] In the above-mentioned operating environment, an embodiment of the present application provides an embodiment of a resource scheduling method. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0043] Figure 2 is a flow chart of a resource scheduling method according to an embodiment of the present application. Figure 2 As shown, the method includes the following steps:

[0044] Step S202: collecting cross-domain network data, wherein the cross-domain network data includes at least: resource load indicators of multiple cloud nodes in the cross-domain network within a preset period, network bandwidth indicators of the multiple cloud nodes, cross-domain rate functions, and security policy parameters;

[0045] Step S204: constructing a global dynamic graph based on the cross-domain network data, wherein the global dynamic graph is used to represent the state of the cross-domain network;

[0046] Step S206: Acquire multiple sub-networks corresponding to each cloud node, and determine the global state of the cross-domain network based on a pre-built partial differential equation operator, wherein the global state includes: the state of each cloud node in the multiple sub-networks;

[0047] In step S206 , the multiple sub-networks represent multiple sub-networks in the model, which are respectively used to determine the multi-dimensional status of each cloud node, such as: security assessment value, cost pressure value, load saturation, and data flow deliverability.

[0048] It should be noted that the security assessment value is an indicator that measures the current security status of a cloud node or cross-domain link. For example, node B is part of a public cloud. Due to multiple recent security vulnerability reports, its security assessment value may be set to a lower value. The cost pressure value represents the cost indicator required to perform specific operations (such as data transmission and storage) on a specific cloud node. For example, the cost pressure value of link AC may be 0.02 yuan / GB. Load saturation is an indicator used to describe the current processing power and resource usage of a cloud node. For example, the load saturation of node C is 85%, indicating that its resources are running close to full capacity. Data flow deliverability is an indicator used to represent the data transmission capability of a cloud node or cross-domain link.

[0049] Step S208: determining a resource scheduling strategy between the plurality of cloud nodes according to a pre-built discrete resource flow operator and the global state of the cross-domain network;

[0050] Step S210: executing a resource scheduling strategy among the plurality of cloud nodes, and allocating resources to the plurality of cloud nodes.

[0051] Through the above steps S202 to S210, cross-domain network data is collected, wherein the cross-domain network data at least includes: resource load indicators of multiple cloud nodes in the cross-domain network within a preset time period, network bandwidth indicators of the multiple cloud nodes, cross-domain rate functions and security policy parameters; a global dynamic graph is constructed based on the cross-domain network data, wherein the global dynamic graph is used to represent the state of the cross-domain network; multiple sub-networks corresponding to each cloud node are obtained, and the global state of the cross-domain network is determined based on a pre-constructed partial differential equation operator, wherein the global state includes: the state of each cloud node under the multiple sub-networks; a resource scheduling strategy between the multiple cloud nodes is determined according to the pre-constructed discrete resource flow operator and the global state of the cross-domain network; the resource scheduling strategy between the multiple cloud nodes is executed, and resources of the multiple cloud nodes are allocated. By using partial differential equation operators to determine the global state of the cross-domain network using cross-domain network data, and then determining the resource scheduling strategy, this approach achieves the goal of timely scheduling and resource flow optimization in a dynamic, time-varying environment, taking into account multi-dimensional data in cross-domain network data transmission and scheduling across multiple or hybrid clouds. This improves resource scheduling accuracy and addresses the low resource scheduling accuracy caused by related technologies that determine resource scheduling strategies based on single-dimensional indicators. This is explained in detail below.

[0052] Figure 3 Another resource scheduling method is shown, such as Figure 3 As shown, the process includes: Step 1: cross-domain network data aggregation and dynamic graph construction. Step 2: building a PDE operator expert collaboration mechanism on the graph. Step 3: building a discrete resource flow operator (DRFO). Step 4: merging a sparse expert hybrid model with dual operators. Step 5: executing scheduling policies and deploying data transmission paths.

[0053] In the technical solution provided in step S206 of the above-mentioned resource scheduling method, the corresponding subnetwork of the cloud node can be determined in the following manner: obtaining multiple subnetworks corresponding to each cloud node, including: respectively obtaining the state functions of the multiple subnetworks corresponding to each cloud node, wherein the multiple subnetworks include at least one of the following: a security subnetwork, a cost subnetwork, a load subnetwork, and a delivery subnetwork; the state functions of the multiple subnetworks include at least one of the following: a first state function for representing the state of the cloud node in the security subnetwork, a second state function for representing the state of the cloud node in the cost subnetwork, a third state function for representing the state of the cloud node in the load subnetwork, and a third state function for representing the state of the cloud node in the load subnetwork. The fourth state function of the state of the point in the delivery subnetwork, the first state function is used to represent the security assessment value of the cloud node in the security subnetwork, the second state function is used to represent the cost pressure value of the cloud node in the cost subnetwork, the third state function is used to represent the load saturation of the cloud node in the load subnetwork, and the fourth state function is used to represent the data flow deliverability of the cloud node in the delivery subnetwork; obtaining the global state of the cross-domain network includes: combining the state functions of the multiple subnetworks into a state function set; determining the partial differential equation operator according to the state function set, and determining the global state of the cross-domain network based on the partial differential equation operator. By fusing the various network state functions through the PDE operator, a comprehensive balance of security, cost, load and data transmission efficiency in the hybrid cloud environment is ensured, thereby achieving multi-dimensional global optimization of the cross-domain network.

[0054] like Figure 4As shown, the steps for cross-domain network data aggregation and dynamic graph construction are as follows: S11. Collect cross-domain resource load indicators and network bandwidth information: Obtain the resource load indicators Lj(t) and network bandwidth indicators Bjk(t) of the private cloud and multiple public clouds at time t∈T, where j represents the jth cloud node, j,k∈V, and V represents the cloud node set. The resource load indicators and network bandwidth information are sorted into time series to obtain multi-domain load sequences and cross-domain bandwidth sequences. These sequences are aligned according to a unified timestamp t, and a preliminary dataset Dt is generated. S12. Extract cross-domain network rates and security policy parameters: Further, based on cross-domain execution fees, network quota limits, and security level requirements, collect the cross-domain rate function Cjk(t) and security compliance factor (security policy parameter) Sjk(t). Cjk(t) represents the dynamic change function of the cross-domain network cost from cloud node j to cloud node k over time, and Sjk(t) represents the security compliance level of the cross-domain connection at time t. Discretize Cjk(t) and Sjk(t) according to the billing cycle and security policy cycle to obtain a rate sequence and security compliance sequence synchronized with time t. S13. Obtain business requirement scenario labels and establish constraint boundaries: Further analyze the security priority scenarios, security compliance thresholds, cost-sensitive thresholds, and high-availability service quality assurance factors in the business requirements corresponding to the cloud nodes. Define the scenario labels as the set Q = {q1, q2, …, qm}, where qm represents the specific constraint value of the business on security or cost. Map each element in Q to the constraint boundary of the corresponding node or cross-domain connection and register it in the dataset Dt to facilitate constraint judgment during subsequent discrete solution and dynamic evolution. S14. Construct a global dynamic graph and annotate dynamic attributes: Further, each cloud node j, k ∈ V is considered to be the vertex set V of the graph G. The available cross-domain network links between each cloud node are considered to be the edge set E of the graph G, denoted as ejk ∈ E, where j, k ∈ V and j ≠ k.

[0055] At time t∈T, the above attributes and business scenario constraints Q are uniformly incorporated into the dynamic graph Gt=(V,Et), so that the real-time status of the cross-domain network can be obtained at each t. To further enable subsequent steps to measure the comprehensive dynamic impact of edge ejk, this application proposes and constructs an attribute measurement formula, as shown below:

[0056]

[0057] Where W jk (t) represents the comprehensive dynamic weight of edge ejk in the time interval [t, t+Δt], τ represents the integral independent variable, and a, b, and c represent weight coefficients, which are used to balance the impact of rate, bandwidth, and security compliance. jk (τ) represents the network rate value, B jk (τ) represents the available bandwidth, Sjk (τ) represents the value of the safety compliance factor. 1-S jk (τ) represents the change in potential security risk within the time interval. By calculating W jk (t), in the subsequent cross-domain scheduling and path selection process, a more precise balance between cost, bandwidth and security factors can be achieved.

[0058] In this application embodiment, compared to existing static resource collection technologies that are unable to simultaneously balance the bandwidth, cost, and security compliance characteristics of cross-domain network data, this technical solution integrates multiple sources of rate and security information in the time dimension, and applies dynamic attribute annotations to cross-domain connections in different autonomous domains. By integrating multiple factors in the improved attribute measurement formula, bandwidth and security blind spots are incorporated into the comprehensive graph, overcoming the limitation of traditional solutions in which a single static method cannot reflect dynamic changes in a timely manner, thereby providing a relatively accurate and flexible foundation for dynamic scheduling and data transmission path optimization.

[0059] In some examples of the present application, the global state of the cross-domain network can be determined in the following manner: determining the global state of the cross-domain network based on the partial differential equation operator, including: obtaining the sub-network corresponding to the cloud node and the business scenario corresponding to the cloud node; determining the constraint conditions of the partial differential equation operator based on the sub-network corresponding to the cloud node and the business scenario corresponding to the cloud node, the constraint conditions of the partial differential equation operator being used to correct the cloud node state output by the partial differential equation operator; iterating the initial state of the cloud node according to the partial differential equation operator and the constraint conditions of the partial differential equation operator until the state of the cloud node converges to obtain the final state of the cloud node; combining the final states of each cloud node to obtain the global state. By applying the constraint conditions to correct the output of the PDE operator during the iterative process, the cloud node state reflects the dynamic environment changes while meeting the actual business needs, thereby achieving the optimization and convergence of the global state of the cross-domain network.

[0060] Wherein, determining the partial differential equation operator according to the state function set includes: obtaining a first parameter of the cloud node, wherein the first parameter includes at least one of the following: a domain node set of the cloud node, a coupling coefficient between the cloud node and the neighborhood node, a penalty term of the business scenario corresponding to the cloud node, an adjustment coefficient for adjusting the weight of the penalty term, and an external incentive term for representing external incentives; respectively obtaining the state function set of the cloud node and the state function set of the neighborhood node; determining the partial differential equation operator according to the first parameter of the cloud node, the state function set of the cloud node, and the state function set of the neighborhood node.

[0061] Specifically, Figure 5The flowchart of constructing the PDE operator expert collaboration mechanism on the graph is shown in Figure 5 As shown, it includes: S21, defining multiple expert subnetworks and their discrete state functions: Obtain the dynamic graph Gt = (V, Et) output by step S1 and the scenario constraint set Q = {q1, q2, …, qm}. Based on the scenario requirements, the expert subnetwork is divided into a security subnetwork, a cost subnetwork, a load subnetwork, and a delivery subnetwork. V is decomposed into several subdomains Vs, Vc, Vl, and Vd, corresponding to the four types of nodes of interest: security, cost, load, and delivery, respectively. For each subdomain, state functions Xis(t), Xic(t), Xil(t), and Xid(t) are defined, where i∈V; Xis(t) represents the security assessment value of the security subnetwork at node i at time t, Xic(t) represents the cost pressure value of the cost subnetwork at node i at time t, Xil(t) represents the load saturation of the load subnetwork at node i at time t, and Xid(t) represents the data flow deliverability of the delivery subnetwork at node i at time t. The above state function set is denoted as X(t) = [Xs(t), Xc(t), Xl(t), Xd(t)] to describe the initial discrete state of the multi-expert sub-network. S22. Construct hybrid cloud collaborative PDE and coupling terms: In the discrete space on the graph, for each node i∈V, a PDE discrete form based on local neighborhood interaction is established to describe the collaborative evolution process between the multi-expert sub-networks. To further characterize the iterative coupling relationship between the four sub-networks of security, cost, load and delivery, this application proposes and defines the following nonlinear PDE operators:

[0062]

[0063] Among them, α∈{s,c,l,d} represents the categories of the four sub-networks: security, cost, load, and delivery; Represents the state value of node i in sub-network α; represents the neighborhood set of node i; represents the coupling coefficient between node i and node j in the sub-network α framework; η α represents the adjustment coefficient for the local influence term; represents the penalty term generated at node i after mapping with the scenario label Q, for example, the security pressure value when the security subnetwork faces a high-security scenario label; Represents an external incentive term, which is used to inject the impact of cross-domain bandwidth, rate and security level changes on the state of the sub-network. It is established that based on this formula, the state difference between the node and the neighborhood and the local nonlinear term can be used to achieve the collaborative evolution between multiple expert sub-networks, overcoming the problem that the existing technology cannot strictly comply with the hard constraints of multiple scenarios through only a single perspective. S23. Construct scenario boundary conditions and complete PDE decoupling: Based on the security priority scenario and cost-sensitive scenario labels registered to the data set Dt in step S1, different boundary conditions (constraints) are imposed on each sub-network, and let Obtain mandatory restrictions when business constraints are met. Define piecewise functions to describe scenario boundaries:

[0064]

[0065] in, represents the state value of node i in sub-network α at the next moment, Q i represents the scene label set related to node i; Θ s and Θ c They represent the nonlinear boundary constraint mapping functions set for the safety domain and the cost domain respectively; δ α represents a constant state increment. By i Embedded in the scenario boundary conditions, ensure that each sub-network truly reflects the scenario requirements of security, cost, and high-availability service quality during the iteration process. It is established that based on this piecewise function, the corresponding constraint value can be forced to be activated at each node or edge, replacing the traditional attention or fixed threshold control method, so as to improve the adaptability to various scenario requirements. S24, iterative fusion and generation of global expert state After defining the multi-expert sub-network state equation and scenario boundary conditions), this application can start from the initial state X(0) of S21 and directly use explicit iteration and implicit iteration methods to synchronously update the state value of each sub-network. The update direction is determined by the four sub-networks of security, cost, load and delivery. Through the coupling operation of the difference between the node and the neighborhood state and the superposition of the scene label boundary constraints, the state of each sub-network is gradually corrected. The value of . When the joint state tends to be stable in numerical calculation after several iterations, the obtained X(t) = [Xs(t), Xc(t), Xl(t), Xd(t)] is regarded as the fusion state of global expert collaboration. In step S2 of this application plan, the core is to deeply combine the traditional PDE framework with multi-scenario and multi-expert sub-networks, and break through the inflexible limitations of existing technologies on multi-dimensional hard constraint boundary processing through collaborative nonlinear coupling and scenario-based boundary conditions. This solution is based on a multi-network iterative approximation process, which enables each sub-network to maintain state evolution while meeting specific safety and cost constraints. It can ensure effective supervision of each sub-domain scenario without adding additional attention gating structures, thereby improving the accuracy and efficiency of state fusion.

[0066] In some instances of the present application, the resource scheduling strategy between the multiple cloud nodes is determined based on a pre-constructed discrete resource flow operator and the global state of the cross-domain network, including: determining the cost pressure value and security assessment value of each edge in the cross-domain network based on the global state; obtaining the edge capacity and edge bandwidth of each edge in the cross-domain network, and determining the initial traffic of each edge based on the edge capacity, the edge bandwidth, the cost pressure value and the security assessment value of each edge to obtain an initial traffic allocation matrix; using the discrete resource flow operator to update the initial traffic allocation matrix until the traffic allocation matrix converges to obtain a final traffic allocation matrix; determining the resource scheduling strategy between the multiple cloud nodes based on the final traffic allocation matrix. By utilizing the discrete resource flow operator in combination with the global state of the cross-domain network, the initial traffic allocation matrix is ​​iteratively updated until convergence, thereby refining the optimal resource scheduling strategy between multiple cloud nodes, thereby achieving the optimization of data transmission paths and bandwidth utilization while meeting cost and security requirements.

[0067] Among them, the discrete resource flow operator is used to update the initial traffic allocation matrix until the traffic allocation matrix converges to obtain the final traffic allocation matrix, including: obtaining the traffic transmission rates of different time periods; determining the total cost of traffic transmission according to the traffic transmission rates of different time periods; and determining that the traffic allocation matrix converges when the load saturation and safety assessment value corresponding to the traffic allocation matrix meet preset requirements and the total cost of the traffic transmission no longer decreases.

[0068] Specifically, Figure 6 The flowchart of constructing the discrete resource flow operator DRFO is shown in FIG. Figure 6 As shown, it includes: S31, initializing flow variables based on the global expert state output by the PDE operator on the graph: obtaining the global expert fusion state X(t) = [Xs(t), Xc(t), Xl(t), Xd(t)] output by step S2, mapping the cost pressure and safety assessment information therein to the edge set Et of the graph Gt = (V, Et). Define a discrete flow variable F for each edge ejk∈Et jk (n), where n represents the number of iterations of the DRFO operator and Fjk(0) represents the initial flow distribution. The state value of each node in the corresponding sub-network is comprehensively considered with information such as edge capacity and bandwidth limit to determine the feasible initial flow range. The initial flow distribution matrix F(0) = [F jk (0)], used for iterative optimization in the discrete resource flow operator. S32, further establish the variable resource flow migration operator and impose bandwidth and capacity constraints: To achieve load-driven traffic migration in each iteration and force the traffic of security non-compliant edges to zero, the following coupled update equation is constructed. Let Sjk (t) represents the security level of edge ejk, q safe represents the scene safety threshold, B jk (t) represents the available bandwidth of the edge, Cap jk (t) represents the maximum capacity constraint of the edge:

[0069]

[0070] where Δ j k represents the flow gain coefficient driven by load difference, represents the load state difference between node j and node k; jk represents the bandwidth penalty coefficient; ε is a small amount to prevent the denominator from being zero; Cap jk (t) is the upper limit of the capacity of edge ejk, which is used to avoid overload during traffic migration. S33, embed the security level restriction operator and map the scenario-based security constraints: In each round of iteration, based on Xs(t), identify the edge ejk with insufficient security level. When S jk (t) safe When F jk (n+1) is assigned a value of 0 to implement forced blocking of the path for sensitive business data. If the security level meets the requirements, bandwidth-limited load difference traffic migration is performed according to the upper branch to ensure the effective allocation of cross-domain flows under the dual constraints of security sub-network and load sub-network. S34, integrate the time-based cost sub-operator and perform multi-criteria measurement: the cross-domain transmission cost C jk (t) Segmented as peak rate Normal rate and valley rates The total cost of traffic allocation on edge ejk is calculated at each iteration step n. The following formula is proposed to complete the segmentation constraint of the periodized cost:

[0071]

[0072] Where T high ,T mid ,T low For different time intervals, the rate of each time period triggers different discretized cost increments. In the iterative process, for the flow F of each edge jk (n) Calculate the corresponding costs in segments and incorporate them into the multi-criteria measurement together with the security constraints and load migration results to achieve a dynamic optimization of cross-domain resources. S35, execute DRFO operator iteration and determine the convergence condition: Further in each iteration, update all F according to the variable resource flow migration branch and security constraint branch of S32 jk ​(n). The total costs incurred by each edge are then compared based on the periodized cost function. If it is found that further adjusting the flow can reduce the cost, the next round of iteration is started; if the load and security scenarios meet the requirements and the cost no longer decreases significantly, the DRFO is determined to have converged. Output the final flow allocation matrix F * =F jk (n) * As a globally optimal and suboptimal discrete flow solution, it provides a decision-making basis for subsequent hybrid cloud scheduling and resource management. This solution overcomes the limitations of existing technologies that rely solely on fuzzy mechanisms such as minimum cost flow or attention mechanisms by incorporating security restrictions and periodized costs into discrete traffic iterations, enhances adaptability to security and dynamic rate scenarios, and ensures the reasonable flow of cross-domain resources under high load conditions. To put it more specifically, in the above steps, this application proposes a discrete resource flow operator DRFO that performs fine-grained redistribution of traffic by integrating multiple sub-network states and scenario labels within each iteration, avoiding the local optimal problem caused by ignoring multi-scenario security restrictions or unified rate modeling in existing technologies. This application further constructs a segmented cost function and a security restriction function and embeds load difference feedback, so that the resource flow optimization process can achieve elastic scheduling and cost savings of cross-domain resources while ensuring security, with higher flexibility and accuracy.

[0073] In some examples of the present application, each subnetwork is used in turn to predict each cloud node in the cross-domain network to obtain a prediction result for each subnetwork; the contribution weight of each subnetwork is obtained, and the prediction result of each subnetwork is weighted using the contribution weight of each subnetwork to obtain a weighted result for each subnetwork; the weighted result of each subnetwork is determined as the initial state value, and the initial state value is iterated using the partial differential equation operator until convergence to obtain a final state value; the initial traffic distribution matrix is ​​determined based on the final state value, and the discrete resource flow operator is used to update the initial traffic distribution matrix until the objective function value is minimized to obtain the final traffic distribution matrix. By integrating the predictions of multiple expert subnetworks and weighted fusion, the partial differential equation operator is used to iteratively solve the final state value, and then the initial traffic distribution matrix is ​​determined based on this state value. Finally, the discrete resource flow operator is used to update the matrix until the objective function value is minimized, thereby achieving cross-domain network resource scheduling optimization based on multi-dimensional prediction, ensuring the comprehensive efficiency of data transmission in terms of security, cost and load.

[0074] Specifically, the sparse expert hybrid large model and the dual operator fusion operation steps are as follows: S41, multi-expert initialization and scenario-based pre-reasoning: load the multi-expert large model structure, activate the security sub-network, cost sub-network, real-time load sub-network, traffic prediction sub-network and other sub-networks respectively, and let them perform separate reasoning in their respective associated data domains to obtain the preliminary judgment result Eα (n), where α represents the sub-network type and n represents the number of iterations. Construct the sub-network output vector E(n)=[E1(n),E2(n),...,E α (n)], which is then used as the initial value for the PDE operator. After completing the preliminary reasoning of multiple experts, in order to further characterize the relative contribution of each sub-network, the following aggregation operator is constructed to normalize and synthesize the outputs of multiple sub-networks:

[0075]

[0076] where φ α represents the function of weighted transformation of the output of sub-network α, Γ represents the total number of currently activated sub-networks, t0 and t1 represent the observation range of this round of reasoning in the time interval [t0, t1], W α Indicates the weight of sub-network α in this round of comprehensive judgment. After completing the aggregation, it provides the weighted initial value for the subsequent PDE operator iteration. S42, integrate PDE operator and realize global node state fusion: The sub-network output vector E(n) and its corresponding weight W obtained in step S41 are α Map to the node V of the graph G = (V, E) to form the initial value of the node state X′(0). Construct a PDE operator consistent with step S2 to perform spatiotemporal evolution of the interaction between nodes. Further, let the state function X corresponding to node j be j (t), β represents the sub-network number, and the discretized PDE evolution equation is based on the following multiple integral form:

[0077]

[0078] Where Nbr(j) represents the set of neighboring nodes of node j, Ω j (t′) represents the external expert-driven term of node j at time t′, Π jk Represents the coupling coefficient connecting the two nodes j and k. Through the above formula, the states of each node are superimposed and iterated within the discrete time step δ, and finally a globally consistent or converged node state distribution X′(τ) is obtained, where τ represents the PDE operator reaching the specified convergence time. S43. Discrete data flow decision based on DRFO operator: further obtain the global node state distribution X′(τ) output by step S42, combine the bandwidth, capacity, security level, and time-based rate information, execute the discrete resource flow operator DRFO defined in step S3, and generate a cross-domain network data transmission plan F(n). In each DRFO iteration, a time-based objective function is proposed for multi-criteria evaluation to guide the discrete optimization of traffic distribution and load scheduling:

[0079]

[0080] Among them F jkrepresents the discrete flow distribution along edge ejk, Cost jk (F jk ,t) represents the cross-domain time-based cost, Θ jk .S jk (t) / represents the constraint violation penalty under the safety level, Ψ jk represents the additional cost or benefit brought by the load difference, μ1, μ2, μ3 are weight coefficients. By repeatedly updating and minimizing J DRFO , under the conditions of security and cost coordination, the final cross-domain network data flow distribution matrix F is obtained * S44. Construct a sparse evaluation operator and further perform dynamic expert selection: To avoid maintaining the low-contribution or negative-contribution sub-networks in repeated iterations, a sparse evaluation operator G is proposed. α (n) measures the net benefit of sub-network α in the nth round of fusion. A dynamic shutdown strategy is implemented for sub-networks that have no positive contribution for a long period of time:

[0081]

[0082] where E′ α (t′) represents the actual contribution of sub-network α to the node state fusion at time t′, κ α Represents the contribution gain coefficient, v α represents the fixed cost of maintaining the subnetwork α. If G α (n+1) is lower than a certain safety threshold for a long time or shows negative growth, then this sub-network will be closed in subsequent iterations. Further, when a new scenario appears and a new safety rule appears, the relevant sub-network will be dynamically reactivated according to the scenario label, so that it can participate in the PDE and DRFO reasoning process again to ensure the adaptive scheduling capability for diverse application needs. S45, output the final scheduling strategy and transmission plan and perform execution monitoring: After completing multiple rounds of fusion iterations of steps S41-S44, monitor the change rate of global indicators such as comprehensive cost, delay, and safety compliance. If the new round of iteration no longer brings significant gains or exceeds the system preset reasoning time limit, stop the iteration and output the global scheduling and transmission results, including the task scheduling plan T * , data transmission path R * And the corresponding security policy table S * The final result T * 、R * 、S *This is then sent to the hybrid cloud execution environment to monitor bandwidth usage, costs, and security compliance in real time. The status information generated by this continuous monitoring is then fed back to step S41, forming a closed-loop update mechanism for a new round of sparse expert hybrid large model reasoning and dual-operator fusion. By integrating this sparse expert hybrid large model with dual-operator fusion, this application enables multi-level coordination of PDE and DRFO operators in cross-subnet scenarios, overcoming the resource waste inherent in existing technologies that rely on a single operator or operate with all expert networks enabled.

[0083] In some examples of this application, the contribution of each subnetwork in determining the cloud node state is obtained; the net contribution of each subnetwork in determining the cloud node state is determined based on the contribution of each subnetwork in determining the cloud node state and a preset evaluation operator; and subnetworks whose net contribution is below a preset threshold are eliminated from the multiple subnetworks. By quantifying the actual contribution of each subnetwork in determining the cloud node state and calculating its net contribution based on a preset evaluation operator, subnetworks whose contribution is below the threshold are eliminated, thereby optimizing the collaborative structure of multiple expert subnetworks and ensuring efficient and targeted resource scheduling decisions.

[0084] Specifically, the steps of scheduling strategy execution and data transmission path deployment are as follows: S51, allocate hybrid cloud instance and load container runtime environment: obtain the task scheduling solution T output in step S4 * In the hybrid cloud scheduling platform, a virtual machine instance or container instance is allocated to each autonomous domain or cloud node. The instance set corresponding to each node j is recorded as Vj. Vj contains several containers IDj,1, IDj,2, etc., ensuring that the T * After completing instance initialization, the required computing and storage resources are configured for the container startup process. Let Mj(k) represent the CPU and memory capacity vector allocated to the kth container on node j to meet the subsequent data processing and application load requirements. Based on this allocation process combined with the node status X output by S4 ′ (τ), avoid resource waste and ensure real-time availability under high load conditions. S52, deploy cross-domain network strategy and specify data flow attributes: the data transmission path R obtained in step S4 is * Map to the cross-domain network link E and follow R * The edge sequence specified in the example is used to assign priority levels Prie and encryption methods Ence to data flows. Prie is used to distinguish the importance of different business flows, and Ence is used to specify whether to enable high-strength encryption. For inter-domain links that require traffic speed limit and bandwidth reservation, based on node capacity and time-based cost constraints, a speed limit threshold Thresb is set in the hybrid cloud scheduling platform and a bandwidth reservation policy is enabled. Let Brie represent the reservable bandwidth quota on edge e. By comparing it with the F output of step S3, *The matrix is ​​matched to ensure cross-domain network data transmission under the global optimal or suboptimal flow allocation framework. Further, after the deployment is executed, the bandwidth usage, real-time cost, security level and other indicators of each autonomous domain j are continuously collected. Let Rj(t) represent the actual resource usage of node j at time t, and let It represents the predicted resource usage of node j by the large model during inference in step S4. Based on the difference between the two, a deviation formula is constructed to make predictions:

[0085]

[0086] where η j represents the bias weight coefficient, ε represents a small positive number to avoid the denominator being zero, ζ safe represents the safety threshold, ζ j (n+1) represents the cumulative deviation of node j during the n+1th round of monitoring, j (n+1) represents the indicator function that triggers the emergency strategy, j (n+1)=1, emergency switching is performed. j When (n+1)=1, that is, a serious deviation is detected in node j, such as a surge in public cloud prices or an overload of private cloud nodes, the emergency strategy is triggered to obtain candidate solutions from S4 for short-term switching, or to initiate an online small-scale re-inference request to the multi-expert large model to re-update the task scheduling and data flow allocation, thereby maintaining the security and feasibility of cross-domain scheduling in the new scenario. In the monitoring and emergency strategy links, this application integrates the deviation between the prediction and actual usage and constructs a trigger indicator function to overcome the limitations of the existing technology of lacking adaptive deviation monitoring and lacking a fast online correction mechanism. If a large deviation is found, a partial online re-evaluation is initiated in a short time to achieve a rapid response to dynamic scenarios at a low computational cost. This solution significantly improves the flexibility and reliability of resource scheduling and security management in a hybrid cloud environment by connecting the scheduling strategy with the dynamic update of the data transmission path.

[0087] Figure 7 A resource scheduling device is shown, the device comprising:

[0088] A collection module 70 is configured to collect cross-domain network data, wherein the cross-domain network data includes at least: resource load indicators of multiple cloud nodes in the cross-domain network within a preset time period, network bandwidth indicators of the multiple cloud nodes, cross-domain rate functions, and security policy parameters;

[0089] A construction module 72 is configured to construct a global dynamic graph based on the cross-domain network data, wherein the global dynamic graph is used to represent the state of the cross-domain network;

[0090] A determination module 74 is configured to obtain multiple sub-networks corresponding to each cloud node and determine a global state of the cross-domain network based on a pre-constructed partial differential equation operator, wherein the global state includes: a state of each cloud node in the multiple sub-networks;

[0091] a policy module 76 for determining a resource scheduling policy between the plurality of cloud nodes based on a pre-built discrete resource flow operator and a global state of the cross-domain network;

[0092] The execution module 78 is used to execute the resource scheduling strategy among the multiple cloud nodes and allocate resources to the multiple cloud nodes.

[0093] The method adopts the method of collecting cross-domain network data, wherein the cross-domain network data includes at least: resource load indicators of multiple cloud nodes in the cross-domain network within a preset time period, network bandwidth indicators of the multiple cloud nodes, cross-domain rate functions and security policy parameters; constructing a global dynamic graph based on the cross-domain network data, wherein the global dynamic graph is used to represent the state of the cross-domain network; obtaining multiple sub-networks corresponding to each cloud node, and determining the global state of the cross-domain network based on a pre-constructed partial differential equation operator, wherein the global state includes: the state of each cloud node under the multiple sub-networks; determining the resource scheduling strategy between the multiple cloud nodes according to the pre-constructed discrete resource flow operator and the global state of the cross-domain network; executing the resource scheduling strategy between the multiple cloud nodes, and allocating resources to the multiple cloud nodes. A global dynamic graph is constructed through cross-domain network data to represent the cross-domain network status, and the global status of the cross-domain network is determined based on partial differential equation operators, and the resource scheduling strategy is determined. This achieves the purpose of timely scheduling and resource flow optimization in a dynamic time-varying environment while taking into account multi-dimensional hard constraints in cross-domain network data transmission and scheduling in multi-cloud or hybrid clouds, thereby reducing uncertainty and resource waste in the scheduling process, and further solving the technical problem of low resource scheduling accuracy due to the fact that related technologies determine resource scheduling strategies based on single-dimensional indicators.

[0094] The policy module 76 includes an acquisition submodule, a determination submodule, and a prediction submodule, wherein the acquisition submodule is used to acquire multiple subnetworks corresponding to each cloud node, including: respectively acquiring state functions of multiple subnetworks corresponding to each cloud node, wherein the multiple subnetworks include at least one of the following: a security subnetwork, a cost subnetwork, a load subnetwork, and a delivery subnetwork; the state functions of the multiple subnetworks include at least one of the following: a first state function for representing the state of the cloud node in the security subnetwork, a second state function for representing the state of the cloud node in the cost subnetwork, a third state function for representing the state of the cloud node in the load subnetwork, and a third state function for representing the state of the cloud node in the The fourth state function of the state under the delivery subnetwork, the first state function is used to represent the security assessment value of the cloud node under the security subnetwork, the second state function is used to represent the cost pressure value of the cloud node under the cost subnetwork, the third state function is used to represent the load saturation of the cloud node under the load subnetwork, and the fourth state function is used to represent the data flow deliverability of the cloud node under the delivery subnetwork; obtaining the global state of the cross-domain network, including: grouping the state functions of the multiple subnetworks into a state function set; determining the partial differential equation operator according to the state function set, and determining the global state of the cross-domain network based on the partial differential equation operator.

[0095] The acquisition submodule includes a determination unit for determining the partial differential equation operator based on the state function set, including: obtaining a first parameter of the cloud node, wherein the first parameter includes at least one of the following: a domain node set of the cloud node, a coupling coefficient between the cloud node and the neighborhood node, a penalty term of the business scenario corresponding to the cloud node, an adjustment coefficient for adjusting the weight of the penalty term, and an external incentive term for representing external incentives; respectively obtaining the state function set of the cloud node and the state function set of the neighborhood node; and determining the partial differential equation operator based on the first parameter of the cloud node, the state function set of the cloud node, and the state function set of the neighborhood node.

[0096] The determination unit is also used to determine the global state of the cross-domain network based on the partial differential equation operator, including: obtaining the sub-network corresponding to the cloud node and the business scenario corresponding to the cloud node; determining the constraints of the partial differential equation operator according to the sub-network corresponding to the cloud node and the business scenario corresponding to the cloud node, and the constraints of the partial differential equation operator are used to correct the cloud node state output by the partial differential equation operator; iterating the initial state of the cloud node according to the partial differential equation operator and the constraints of the partial differential equation operator until the state of the cloud node converges to obtain the final state of the cloud node; combining the final state of each cloud node to obtain the global state.

[0097] The determining submodule is configured to determine the resource scheduling strategy among the plurality of cloud nodes according to the pre-constructed discrete resource flow algorithm and the global state of the cross-domain network, including: determining a cost pressure value and a security evaluation value of each edge in the cross-domain network according to the global state; obtaining an edge capacity and an edge bandwidth of each edge in the cross-domain network, and determining an initial flow of each edge according to the edge capacity, the edge bandwidth, the cost pressure value and the security evaluation value of each edge, to obtain an initial flow allocation matrix; updating the initial flow allocation matrix by using the discrete resource flow algorithm until the flow allocation matrix converges, to obtain a final flow allocation matrix; and determining the resource scheduling strategy among the plurality of cloud nodes according to the final flow allocation matrix.

[0098] The determining submodule includes an updating unit configured to update the initial flow allocation matrix by using the discrete resource flow algorithm until the flow allocation matrix converges, to obtain a final flow allocation matrix, including: obtaining a second parameter of each edge in the cross-domain network, where the second parameter includes at least one of the following: a security evaluation value of an edge, an available bandwidth of an edge, a maximum capacity of an edge, and a load difference between two nodes connected by an edge; and constructing the discrete resource flow algorithm according to the second parameter.

[0099] The updating unit is further configured to update the initial flow allocation matrix by using the discrete resource flow algorithm until the flow allocation matrix converges, to obtain a final flow allocation matrix, including: obtaining a flow transmission rate of different time periods; determining a total cost of flow transmission according to the flow transmission rate of different time periods; and determining that the flow allocation matrix converges in a case where a load saturation degree and a security evaluation value corresponding to the flow allocation matrix meet preset requirements and the total cost of flow transmission no longer decreases.

[0100] The prediction submodule is configured to sequentially use each sub-network to predict each cloud node in the cross-domain network, to obtain a prediction result of each sub-network; obtain a contribution weight of each sub-network, and weight the prediction result of each sub-network by using the contribution weight of each sub-network, to obtain a weighted result of each sub-network; determine the weighted result of each sub-network as an initial state value, and iteratively use the partial differential equation algorithm on the initial state value until convergence, to obtain a final state value; determine an initial flow allocation matrix according to the final state value, and update the initial flow allocation matrix by using the discrete resource flow algorithm until a target function value is minimum, to obtain a final flow allocation matrix.

[0101] The prediction submodule includes an elimination unit for respectively obtaining the contribution of each subnetwork in the process of determining the cloud node state; determining the net contribution of each subnetwork in the process of determining the cloud node state according to the contribution of each subnetwork in the process of determining the cloud node state and a preset evaluation operator; and eliminating the subnetworks whose net contribution is lower than a preset threshold from the multiple subnetworks.

[0102] It should be noted that Figure 7 The resource scheduling device shown is used to execute Figure 2 The resource scheduling method shown in the figure, therefore the relevant explanations in the above resource scheduling method are also applicable to the resource scheduling device, and will not be repeated here.

[0103] An embodiment of the present application further provides a computer device, comprising: a memory and a processor, wherein the memory is used to store program instructions; and the processor is connected to the memory and is used to execute the above-mentioned resource scheduling method.

[0104] An embodiment of the present application further provides a non-volatile storage medium, which includes a stored computer program, wherein the device where the non-volatile storage medium is located executes the above-mentioned resource scheduling method by running the computer program.

[0105] An embodiment of the present application further provides a computer program product, comprising computer instructions, which, when executed by a processor, implement the steps of the resource scheduling method in the present application.

[0106] The serial numbers of the above embodiments of the present application are for description only and do not represent the advantages or disadvantages of the embodiments.

[0107] In the above embodiments of the present application, the description of each embodiment has its own focus. For parts that are not described in detail in a certain embodiment, please refer to the relevant description of other embodiments.

[0108] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. Among them, the device embodiments described above are only exemplary. For example, the division of the units can be a logical function division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of units or modules, which can be electrical or other forms.

[0109] The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple units. Some or all of the units may be selected according to actual needs to achieve the purpose of the present embodiment.

[0110] In addition, the functional units in the various embodiments of the present application may be integrated into a single processing unit, or each unit may exist physically separately, or two or more units may be integrated into a single unit. The aforementioned integrated units may be implemented in the form of hardware or software functional units.

[0111] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for enabling a computer device (which can be a personal computer, a server or a network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: various media that can store program codes, such as a U disk, a read-only memory (ROM), a random access memory (RAM), a mobile hard disk, a magnetic disk or an optical disk.

[0112] The above is only a preferred embodiment of the present application. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present application. These improvements and modifications should also be regarded as the scope of protection of the present application.

Claims

1. A resource scheduling method, characterized in that: include: Collecting cross-domain network data, wherein the cross-domain network data includes at least: resource load indicators of multiple cloud nodes in the cross-domain network within a preset time period, network bandwidth indicators of the multiple cloud nodes, cross-domain rate functions, and security policy parameters; Building a global dynamic graph based on the cross-domain network data, wherein the global dynamic graph is used to represent the state of the cross-domain network; Acquire multiple sub-networks corresponding to each cloud node, and determine the global state of the cross-domain network based on a pre-constructed partial differential equation operator, wherein the global state includes: the state of each cloud node in the multiple sub-networks; Determining a resource scheduling strategy among the plurality of cloud nodes based on a pre-built discrete resource flow operator and a global state of the cross-domain network; Execute a resource scheduling strategy among the multiple cloud nodes and allocate resources to the multiple cloud nodes.

2. The method according to claim 1, characterized in that Obtaining multiple subnetworks corresponding to each cloud node, including: respectively obtaining state functions of the multiple subnetworks corresponding to each cloud node, wherein the multiple subnetworks include at least one of the following: a security subnetwork, a cost subnetwork, a load subnetwork, and a delivery subnetwork; the state functions of the multiple subnetworks include at least one of the following: a first state function for representing a state of the cloud node in the security subnetwork, a second state function for representing a state of the cloud node in the cost subnetwork, a third state function for representing a state of the cloud node in the load subnetwork, and a fourth state function for representing a state of the cloud node in the delivery subnetwork, the first state function being used to represent a security assessment value of the cloud node in the security subnetwork, the second state function being used to represent a cost pressure value of the cloud node in the cost subnetwork, the third state function being used to represent a load saturation of the cloud node in the load subnetwork, and the fourth state function being used to represent a data flow deliverability of the cloud node in the delivery subnetwork; Acquiring the global state of the cross-domain network, comprising: combining the state functions of the plurality of sub-networks into a state function set; The partial differential equation operator is determined according to the state function set, and the global state of the cross-domain network is determined based on the partial differential equation operator.

3. The method according to claim 2, characterized in that Determining the partial differential equation operator according to the state function set includes: Obtaining a first parameter of the cloud node, wherein the first parameter includes at least one of the following: a domain node set of the cloud node, a coupling coefficient between the cloud node and a neighboring node, a penalty term of a business scenario corresponding to the cloud node, an adjustment coefficient for adjusting a weight of the penalty term, and an external incentive term for representing an external incentive; Respectively obtaining a state function set of the cloud node and a state function set of the neighborhood node; The partial differential equation operator is determined according to the first parameter of the cloud node, the state function set of the cloud node, and the state function set of the neighborhood nodes.

4. The method according to claim 2, characterized in that Determining a global state of the cross-domain network based on the partial differential equation operator includes: Obtain the subnetwork corresponding to the cloud node and the business scenario corresponding to the cloud node; Determining a constraint condition of the partial differential equation operator according to the subnetwork corresponding to the cloud node and the business scenario corresponding to the cloud node, wherein the constraint condition of the partial differential equation operator is used to correct the cloud node state output by the partial differential equation operator; Iterating the initial state of the cloud node according to the partial differential equation operator and the constraint condition of the partial differential equation operator until the state of the cloud node converges to obtain the final state of the cloud node; The final state of each cloud node is combined to obtain the global state.

5. The method according to claim 1, wherein Determining a resource scheduling strategy between the plurality of cloud nodes according to a pre-built discrete resource flow operator and a global state of the cross-domain network includes: Determining a cost pressure value and a security assessment value of each edge in the cross-domain network according to the global state; Obtaining the edge capacity and edge bandwidth of each edge in the cross-domain network, and determining the initial flow of each edge according to the edge capacity, the edge bandwidth, the cost pressure value, and the security assessment value of each edge, to obtain an initial flow distribution matrix; Using the discrete resource flow operator to update the initial flow allocation matrix until the flow allocation matrix converges, to obtain a final flow allocation matrix; A resource scheduling strategy among the multiple cloud nodes is determined according to the final traffic allocation matrix.

6. The method according to claim 5, characterized in that The initial traffic allocation matrix is ​​updated using the discrete resource flow operator until the traffic allocation matrix converges to obtain a final traffic allocation matrix, including: Acquire a second parameter of each edge in the cross-domain network, wherein the second parameter includes at least one of the following: a security assessment value of the edge, an available bandwidth of the edge, a maximum capacity of the edge, and a load difference between two nodes connected by the edge; The discrete resource flow operator is constructed according to the second parameter.

7. The method according to claim 5, characterized in that The initial traffic allocation matrix is ​​updated using the discrete resource flow operator until the traffic allocation matrix converges to obtain a final traffic allocation matrix, including: Get traffic transmission rates at different time periods; Determine the total cost of traffic transmission based on the traffic transmission rates for the different time periods; When the load saturation and the safety assessment value corresponding to the traffic distribution matrix meet preset requirements and the total cost of the traffic transmission no longer decreases, it is determined that the traffic distribution matrix has converged.

8. The method according to claim 1, characterized in that The method further comprises: Using each sub-network in turn to predict each cloud node in the cross-domain network, and obtaining a prediction result for each sub-network; Obtain the contribution weight of each sub-network, and use the contribution weight of each sub-network to weight the prediction results of each sub-network to obtain the weighted results of each sub-network; Determining the weighted result of each sub-network as an initial state value, and iterating the initial state value using the partial differential equation operator until convergence to obtain a final state value; An initial flow distribution matrix is ​​determined according to the final state value, and the initial flow distribution matrix is ​​updated using the discrete resource flow operator until the objective function value is minimized, thereby obtaining a final flow distribution matrix.

9. The method according to claim 8, characterized in that The method further comprises: Obtain the contribution of each sub-network in determining the cloud node status; Determining a net contribution of each subnetwork in the process of determining the cloud node state according to the contribution of each subnetwork in the process of determining the cloud node state and a preset evaluation operator; The subnetworks whose net contribution is lower than a preset threshold are eliminated from the multiple subnetworks.

10. A resource scheduling device, characterized in that: include: A collection module, configured to collect cross-domain network data, wherein the cross-domain network data includes at least: resource load indicators of multiple cloud nodes in the cross-domain network within a preset time period, network bandwidth indicators of the multiple cloud nodes, cross-domain rate functions, and security policy parameters; A construction module, configured to construct a global dynamic graph based on the cross-domain network data, wherein the global dynamic graph is used to represent the state of the cross-domain network; A determination module is configured to obtain multiple sub-networks corresponding to each cloud node, and determine a global state of the cross-domain network based on a pre-built partial differential equation operator, wherein the global state includes: a state of each cloud node in the multiple sub-networks; A policy module, configured to determine a resource scheduling policy among the plurality of cloud nodes based on a pre-built discrete resource flow operator and a global state of the cross-domain network; The execution module is used to execute the resource scheduling strategy among the multiple cloud nodes and allocate resources to the multiple cloud nodes.

11. A computer device, characterized in that: include: A memory and a processor, wherein the memory is used to store program instructions; The processor is connected to the memory and is used to execute the resource scheduling method described in any one of claims 1 to 9.

12. A computer program product comprising computer instructions, characterized in that When the computer instructions are executed by a processor, the resource scheduling method according to any one of claims 1 to 9 is implemented.

Images