Feasibility constraint routing and period combined scheduling method, system, device and medium

Abstract

The application relates to a feasibility constraint routing and period joint scheduling method, system, device and medium. The method comprises the following steps: acquiring a to-be-scheduled service flow, a network topology, a link state and a period resource state in a current scheduling window, and constructing a current scheduling state; integrating candidate next-hop nodes and candidate period offsets, and constructing a candidate joint action space; performing feasibility verification on each candidate joint action in the candidate joint action space, and generating a feasibility mask; screening a movable action set according to the feasibility mask, and calculating an action space retention rate and an action space compression rate; selecting a joint scheduling action and updating the current scheduling state; and when the to-be-scheduled service flow reaches a target node, generating a joint scheduling result. By using the method, the scheduling search space can be compressed through multi-constraint pre-pruning, the full-dimension compliance of a scheduling scheme can be guaranteed, the scheduling process efficiency can be quantified synchronously, and the solving efficiency and the scheme feasibility of the routing and period joint scheduling can be effectively improved.

Description

Technical Field

[0001] This invention relates to the fields of industrial internet, time-sensitive networks, and deterministic networks, and in particular to a method, system, device, and medium for joint scheduling of feasibility-constrained routing and periodic scheduling. Background Technology

[0002] With the development of applications such as the Industrial Internet, intelligent manufacturing, and predictive maintenance, networks need to simultaneously carry multiple types of service flows, including status monitoring, anomaly alarms, maintenance control, and operation and maintenance management. These services have different requirements in terms of end-to-end latency, bandwidth utilization, reliability, and transmission stability. Deterministic networks provide bounded latency and low jitter transmission guarantees for critical services through mechanisms such as time synchronization, resource reservation, queue scheduling, and periodic forwarding. Among them, the cyclic designated queue forwarding mechanism divides the forwarding process into fixed-length scheduling periods, allowing data packets to be forwarded hop-by-hop in multi-hop links according to the specified period, thus providing the foundation for end-to-end deterministic transmission.

[0003] In traditional technologies, traffic routing and period offset configuration are typically handled as two independent processes: first, the transmission path is determined, and then the forwarding period for each hop is configured on the selected path. However, under the round-robin assigned queue forwarding mechanism, the transmission performance of traffic depends not only on the selected physical path but also on the period offset configuration and period resource occupancy status on each hop link. If routing is based solely on the shortest path or static link cost, traffic may concentrate on a few links, leading to rapid depletion of corresponding period resources. If period offsets are configured separately after the path is determined, problems such as path reachability but unavailable period resources, and local period feasibility but end-to-end latency exceeding limits can easily occur. Furthermore, existing methods often lack explicit feasibility screening mechanisms for connectivity, acyclic constraints, period capacity constraints, deadline reachability constraints, and period offset validity constraints when selecting candidate actions. This results in a large number of invalid candidate actions participating in the evaluation, reducing scheduling decision efficiency and network resource utilization. Summary of the Invention

[0004] Therefore, it is necessary to provide a method, system, device, and medium for joint scheduling of routing and periodic offsets that can achieve joint scheduling of routing and periodic offsets through feasibility constraints, in order to address the above-mentioned technical problems.

[0005] Firstly, this application provides a joint scheduling method for feasibility-constrained routing and periodic scheduling, including:

[0006] Obtain the pending service flow requests, network topology, link status, and periodic resource status within the current scheduling window, and construct the current scheduling status based on the pending service flow requests, network topology, link status, and periodic resource status;

[0007] Based on the network topology, obtain the candidate next-hop node and candidate period offset of the current node in the current scheduling state, and integrate the candidate next-hop node and candidate period offset to obtain the candidate joint action space;

[0008] Based on connectivity constraints, acyclic constraints, periodic capacity constraints, deadline reachability constraints, and periodic offset legality constraints, the feasibility of each candidate joint action in the candidate joint action space is verified, and a feasibility mask is generated.

[0009] Based on the feasibility mask, candidate joint actions that are not executable are eliminated to obtain a set of actionable actions. Based on the size of the candidate joint action space and the size of the set of actionable actions, the action space retention rate and action space compression rate are calculated.

[0010] Based on network topology, link status, and periodic resource status, determine joint scheduling actions from the set of actionable actions, and update the current scheduling status based on the joint scheduling actions;

[0011] When the service flow to be scheduled arrives at the target node in the current scheduling state, a joint scheduling result is generated based on the current scheduling state, action space retention rate, and action space compression rate.

[0012] Secondly, this application also provides a feasibility-constrained routing and periodic joint scheduling system, which includes:

[0013] The data acquisition module is used to acquire the pending service flow requests, network topology, link status, and periodic resource status within the current scheduling window, and to construct the current scheduling status based on the pending service flow requests, network topology, link status, and periodic resource status.

[0014] The action space construction module is used to obtain the candidate next-hop node and candidate period offset of the current node in the current scheduling state according to the network topology, and integrate the candidate next-hop node and candidate period offset to obtain the candidate joint action space.

[0015] The feasibility verification module is used to verify the feasibility of each candidate joint action in the candidate joint action space based on connectivity constraints, acyclic constraints, periodic capacity constraints, deadline reachability constraints, and periodic offset legality constraints, and generate a feasibility mask.

[0016] The action filtering module is used to eliminate unexecutable candidate joint actions based on the feasibility mask, obtain the set of actionable actions, and calculate the action space retention rate and action space compression rate based on the size of the candidate joint action space and the size of the set of actionable actions.

[0017] The action determination module is used to determine the joint scheduling action from the set of actionable actions based on the network topology, link status and periodic resource status, and update the current scheduling status based on the joint scheduling action;

[0018] The scheduling result generation module is used to generate a joint scheduling result based on the current scheduling state, action space retention rate, and action space compression rate when the service flow to be scheduled arrives at the target node in the current scheduling state.

[0019] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the above-described feasibility constraint routing and periodic joint scheduling method.

[0020] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the above-described feasibility-constrained routing and periodic joint scheduling method.

[0021] The aforementioned feasible constraints on routing and periodic joint scheduling methods, systems, devices, and media, by acquiring the scheduled service flow requests, network topology, link status, and periodic resource status within the current scheduling window, integrate and construct the current scheduling state, and extract the candidate next-hop node and candidate period offset of the current node in the current scheduling state based on the network topology, and integrate the candidate next-hop node and candidate period offset to obtain the candidate joint action space, can achieve the unified construction of the joint action space of routing path and periodic scheduling in two dimensions, complete the standardized quantitative representation of the full-dimensional scheduling state and the systematic sorting of joint scheduling candidate actions, and can solve the problems of traditional routing and periodic scheduling being separated, joint scheduling lacking a unified action carrying space, incomplete scheduling state coverage leading to omissions or redundancies in candidate actions, and subsequent scheduling decisions lacking complete state basis;

[0022] By performing feasibility checks on all candidate joint actions within the candidate joint action space based on connectivity constraints, acyclic constraints, periodic capacity constraints, deadline reachability constraints, and periodic offset legality constraints, a corresponding feasibility mask is generated. Then, all unexecutable candidate joint actions are eliminated based on the feasibility mask to obtain the set of actionable actions. At the same time, the action space retention rate and action space compression rate are calculated based on the size of the candidate space and the set of actionable actions. This technique enables precise multi-constraint pruning of the joint action space and quantitative evaluation of the action space compression efficiency. It reduces invalid scheduling attempts and filters invalid scheduling actions that violate constraint rules. This technique can solve the problems of traditional joint scheduling, which does not have a multi-constraint feasibility check step beforehand. The action space is mixed with a large number of unexecutable invalid actions, resulting in redundant search paths for scheduling decisions, low solution efficiency, and easy output of invalid scheduling schemes that violate constraint rules. At the same time, it is impossible to quantify the actual effect of constraint pruning.

[0023] By combining network topology, link status, and periodic resource status, the optimal joint scheduling action is determined from the set of feasible actions, and the current scheduling status is updated synchronously based on the joint scheduling action. This technical approach enables optimal decision-making of joint scheduling actions and dynamic iterative updates of the scheduling status under feasibility constraints. It ensures that each scheduling decision is consistent with the actual state of the current network and resources. This approach can solve the problems of insufficient decision-making accuracy of traditional scheduling methods within the constraint boundary, and the scheduling status update lagging behind the execution of scheduling actions, which leads to asynchronous scheduling process states, subsequent node decisions deviating from actual resource occupancy, and low executability of scheduling schemes.

[0024] By integrating current scheduling status data, action space retention rate, and action space compression rate when a pending service flow arrives at the target node in the current scheduling state, a joint scheduling result is generated. This technical approach enables the complete output of a joint scheduling scheme for the entire service flow path and the synchronous presentation of scheduling process performance indicators. It generates a scheduling result system that integrates scheduling execution schemes and process performance evaluation. This addresses the problems of traditional scheduling methods that only output simple path and periodic scheduling schemes, lack quantitative data on scheduling process performance, fail to intuitively reflect the actual value of feasibility constraint pruning, and have a single dimension in scheduling results that cannot provide sufficient evaluation support for network performance optimization. This approach can simultaneously consider path reachability and periodic resource availability, improving the completeness and reference value of joint scheduling results and providing data support for network scheduling performance optimization and iterative improvement of constraint mechanisms. Attached Figure Description

[0025] To more clearly illustrate the technical solutions in the embodiments or related technologies of this application, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0026] Figure 1 This is a flowchart of a feasibility-constrained routing and periodic joint scheduling method according to the present invention;

[0027] Figure 2 This is a flowchart illustrating the construction of a candidate joint action space in one optional embodiment of the present invention;

[0028] Figure 3 This is a structural diagram of a feasibility-constrained routing and periodic joint scheduling system according to the present invention. Detailed Implementation

[0029] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0030] In one embodiment, such as Figure 1 As shown, a feasibility-constrained routing and periodic joint scheduling method is provided. This embodiment illustrates the application of this method to a joint scheduling terminal. It is understood that this method can also be applied to a joint scheduling server, and further to a joint scheduling system including a joint scheduling terminal and a joint scheduling server, and is implemented through the interaction between the joint scheduling terminal and the joint scheduling server. In this embodiment, the method includes the following steps:

[0031] Step S101: Obtain the pending service flow requests, network topology, link status, and periodic resource status within the current scheduling window, and construct the current scheduling status based on the pending service flow requests, network topology, link status, and periodic resource status.

[0032] Optionally, the joint scheduling terminal can set a scheduling window period. Based on the scheduling window period, it obtains all pending service flow requests arriving within the current scheduling window from the network management node, and simultaneously obtains the latest network topology from the network topology management module, the link status of each directed link from the link monitoring module, and the remaining available capacity data of each directed link in each specified queue forwarding cycle from the periodic resource management module to obtain the periodic resource status. The joint scheduling terminal can also set status coding rules to integrate the pending service flow requests, network topology, link status, and periodic resource status to obtain the current scheduling status.

[0033] Specifically, the scheduled service flow request can be a deterministic transmission request initiated by the industrial field terminal system. The scheduled service flow request can include information such as source node, destination node, message length, bandwidth requirement, maximum allowable end-to-end delay, and service type identifier. The network topology can be a directed graph structure composed of switching nodes and directed links. The network topology can be used to characterize the physical connectivity between nodes. The link status can include information such as link connectivity identifier, fixed transmission delay, and currently occupied bandwidth. The periodic resource status can include the remaining allocable capacity of each directed link in each cyclic specified queue forwarding cycle. The periodic resource status can be used to characterize the resource availability in the time dimension.

[0034] Step S102: Based on the network topology, obtain the candidate next-hop node and candidate period offset of the current node in the current scheduling state, integrate the candidate next-hop node and candidate period offset to obtain the candidate joint action space.

[0035] Optionally, the joint scheduling terminal can extract the current node from the current scheduling state, and based on the adjacency relationship of the network topology, query all directed links starting from the current node, and take the endpoint of each directed link as a candidate next-hop node; the joint scheduling terminal can obtain the total number of forwarding cycles of the circular specified queue in the circular specified queue forwarding mechanism from the periodic resource state, obtain the maximum cycle offset, and generate an integer set from 1 to the maximum cycle offset to obtain a candidate cycle offset set; the joint scheduling terminal can use Cartesian product operation to combine the candidate next-hop nodes and candidate cycle offsets in pairs to obtain the candidate joint action space.

[0036] Specifically, the current node can be the network switching node where the service flow to be scheduled is located in the current scheduling state; the candidate next-hop node can be an adjacent node that is directly connected to the current node through a directed link and can be used as the next-hop forwarding target of the service flow to be scheduled; the candidate period offset can be the number of periods that the service flow to be scheduled needs to wait from being received by the current node until it is forwarded from the corresponding directed link; the candidate joint action space can be all combinations of all candidate next-hop nodes and all candidate period offsets.

[0037] Step S103: Based on connectivity constraints, acyclic constraints, periodic capacity constraints, deadline reachability constraints, and periodic offset legality constraints, perform feasibility verification on each candidate joint action in the candidate joint action space and generate a feasibility mask.

[0038] Optionally, the joint scheduling terminal can sequentially perform connectivity constraint verification, acyclic constraint verification, periodic capacity constraint verification, deadline reachability constraint verification, and periodic offset legality constraint verification on each candidate joint action in the candidate joint action space. The joint scheduling terminal can perform a logical AND operation on the verification results of connectivity constraint verification, acyclic constraint verification, periodic capacity constraint verification, deadline reachability constraint verification, and periodic offset legality constraint verification to obtain the feasibility mask of each candidate joint action.

[0039] Specifically, connectivity constraints can refer to the existence of a usable directed link between the candidate next-hop node and the current node; acyclic constraints can refer to the requirement that the candidate next-hop node must not appear in the node path already traversed by the scheduled service flow to avoid forming forwarding loops; periodic capacity constraints can refer to the requirement that the remaining resource capacity of the link corresponding to the target forwarding period must not be less than the resource requirements of the scheduled service flow; deadline reachability constraints can refer to the requirement that the cumulative delay after executing the current candidate action, plus the minimum delay of subsequent paths, must not exceed the maximum allowable end-to-end delay of the scheduled service flow; period offset legality constraints can refer to the requirement that the value of the candidate period offset must be within the preset minimum and maximum offset range and must not exceed the legal value boundary of periodic scheduling; the feasibility mask can be a set of identifiers consisting of 0 and 1, where a value of 1 indicates that the corresponding candidate joint action satisfies all constraints and can be executed, and a value of 0 indicates that the corresponding candidate joint action violates at least one constraint and cannot be executed.

[0040] Step S104: Based on the feasibility mask, eliminate candidate joint actions that are not executable to obtain a set of actionable actions. Then, based on the size of the candidate joint action space and the size of the set of actionable actions, calculate the action space retention rate and the action space compression rate.

[0041] Optionally, the joint scheduling terminal can remove all candidate joint actions with a mask value of 0 from the candidate joint action space based on the value of the feasibility mask, and retain all candidate joint actions with a mask value of 1 to obtain the set of actionable actions. The joint scheduling terminal can count the number of candidate joint actions in the candidate joint action space and the number of actionable actions in the set of actionable actions, and calculate the action space retention rate and action space compression rate by ratio calculation.

[0042] Specifically, the set of actionable actions can be a subset of actions in the candidate joint action space that satisfy all feasibility constraints; the action space retention rate can be the proportion of the number of actionable actions to the total number of candidate joint actions, and the action space retention rate can be used to characterize the sufficiency of the available scheduling space in the current state; the action space compression rate can be the proportion of the number of inactionable actions that are removed to the total number of candidate joint actions, and the action space compression rate can be used to characterize the pruning strength of feasibility constraints on the action space, and indirectly reflect the periodic resource pressure and deadline tension of the current network.

[0043] Step S105: Based on the network topology, link status, and periodic resource status, determine the joint scheduling action from the set of actionable actions, and update the current scheduling status according to the joint scheduling action.

[0044] Optionally, the joint scheduling terminal can input the network topology, link load status, and link periodic resource status into a graph convolutional network to aggregate neighborhood information and obtain a topology-aware feature representation. The joint scheduling terminal can then input the topology-aware feature representation, the current scheduling status, and the attribute information of each candidate joint action into deep reinforcement learning to evaluate action value and obtain the original action value corresponding to each candidate joint action. The joint scheduling terminal can reconstruct the original action value based on a feasibility mask, replacing the original action value of inactive actions with a low penalty value to obtain the reconstructed action value. Within the set of feasible actions, the joint scheduling terminal can select the candidate joint action with the highest reconstructed action value as the joint scheduling action. Based on the joint scheduling action, the joint scheduling terminal can update the current node, the set of visited nodes, the cumulative end-to-end latency, and simultaneously update the link periodic resource occupancy status.

[0045] Specifically, a graph convolutional network (GCNN) can be a neural network model used to process graph-structured data. A GCNN can aggregate link load and periodic resource information in the neighborhood of a node to generate a feature vector with topology awareness. A deep reinforcement learning model can be an action value evaluation model based on a deep Q-network. A deep reinforcement learning model can be used to evaluate the long-term scheduling benefits of different candidate joint actions in the current state. A topology-aware feature representation can be a high-dimensional feature vector output by a GCNN. A topology-aware feature representation can fuse global information such as the entire network topology, link load distribution, and periodic resource occupancy. The original action value can be a score output by a deep reinforcement learning model that represents the scheduling benefits of the corresponding candidate joint action. The reconstructed action value can be the action value after feasibility masking.

[0046] Step S106: When the service flow to be scheduled arrives at the target node in the current scheduling state, a joint scheduling result is generated based on the current scheduling state, action space retention rate, and action space compression rate.

[0047] Optionally, after completing the scheduling decision for each hop, the joint scheduling terminal can determine whether the updated current node is consistent with the target node; if the current node is consistent with the target node, it is determined that the scheduled service flow has been completed; the joint scheduling terminal can generate a complete routing path according to the order of the visited node set, generate a period offset sequence according to the period offset selected by each hop scheduling, and summarize the action space retention rate and action space compression rate data of each hop scheduling during the scheduling process to obtain the joint scheduling result.

[0048] Specifically, the target node can be the destination receiving node of the service flow to be scheduled; the routing path can be an ordered sequence of all switching nodes that the service flow passes through from the source node to the destination node; the period offset sequence can be an ordered sequence of period offset values ​​configured on each hop link of the service flow; and the joint scheduling result can be a complete scheduling output including path configuration, period configuration, process performance indicators and scheduling status.

[0049] The aforementioned feasibility-constrained routing and periodic joint scheduling method constructs the current scheduling state by acquiring the scheduled service flow, network topology, link status, and periodic resource status within the current scheduling window; integrates candidate next-hop nodes and candidate periodic offsets to construct a candidate joint action space; performs feasibility verification on each candidate joint action in the candidate joint action space to generate a feasibility mask; based on the feasibility mask, selects a set of feasible actions and calculates the action space retention rate and action space compression rate; selects a joint scheduling action and updates the current scheduling state; when the scheduled service flow arrives at the target node, a joint scheduling result is generated. This method can compress the scheduling search space through multi-constraint pre-pruning, ensuring full-dimensional compliance of the scheduling scheme, simultaneously quantifying the efficiency of the scheduling process, and effectively improving the solution efficiency and feasibility of routing and periodic joint scheduling.

[0050] In one embodiment, the service flow request to be scheduled may include a source node, a destination node, and a maximum allowed end-to-end latency; the current scheduling state is constructed based on the service flow request to be scheduled, network topology, link status, and periodic resource status, and may include:

[0051] Step S201: Determine the positions of the source node and the destination node in the network topology according to the network topology, set the source node as the current node, and set the destination node as the target node; initialize the set of visited nodes to include the source node, initialize the cumulative end-to-end delay to zero, and initialize the remaining time limit to the maximum allowed end-to-end delay.

[0052] Optionally, after receiving a request for a service flow to be scheduled, the joint scheduling terminal can parse the source node identifier and destination node identifier of the service flow, and match and locate the topological positions of the source node and destination node in the network topology; the joint scheduling terminal can set the source node as the current node and the destination node as the target node; the joint scheduling terminal can create a set of visited nodes and add the source node as the first element to the set of visited nodes; the joint scheduling terminal can set the initial value of the accumulated end-to-end delay to zero; the joint scheduling terminal can assign the maximum allowed end-to-end delay of the service flow to be scheduled to the remaining time limit.

[0053] Specifically, the source node can be the switching node connected to the sending system of the service flow to be scheduled, and the source node can be the starting point of the service flow transmission; the destination node can be the switching node connected to the receiving system of the service flow to be scheduled, and the destination node can be the ending point of the service flow transmission; the visited node set can be a set recording all the nodes that the service flow to be scheduled has passed through since it started from the source node, and the visited node set can be used to detect forwarding loops; the cumulative end-to-end delay can be the total delay generated by the service flow to be scheduled from the source node to the current node; the remaining time limit can be the remaining usable delay budget of the service flow to be scheduled, and the remaining time limit can be equal to the maximum allowed end-to-end delay minus the cumulative end-to-end delay.

[0054] Step S202: Obtain the load information of each link in the link status as the link load status; obtain the remaining available capacity of each link in the periodic resource status in different cyclic specified queue forwarding cycles as the link periodic resource occupancy status.

[0055] Optionally, the joint scheduling terminal can obtain the real-time bandwidth utilization rate, the number of carried service flows, and the fixed transmission delay of each directed link from the link monitoring module, and integrate them to obtain the link load status; the joint scheduling terminal can obtain the remaining allocable bandwidth capacity of each directed link in each specified queue forwarding cycle from the periodic resource management module, and obtain the link periodic resource utilization status.

[0056] Specifically, the link load status can be status information representing the spatial dimension resource occupancy of each link, and the link load status can be used to reflect the overall load pressure of the link; the link periodic resource occupancy status can be status information representing the resource remaining status of each link in each cyclic specified queue forwarding cycle time slot, and the link periodic resource occupancy status can be used to reflect the fine-grained resource availability in the time dimension; the cyclic specified queue forwarding cycle can be the smallest scheduling time unit under the cyclic specified queue forwarding mechanism.

[0057] Step S203: Integrate the current node, target node, set of visited nodes, cumulative end-to-end delay, remaining time limit, link load status, and link periodic resource occupancy status to construct the current scheduling status.

[0058] Optionally, the joint scheduling terminal can set state vector concatenation rules and concatenate the current node identifier, target node identifier, visited node set, cumulative end-to-end delay, and remaining time limit according to the state vector concatenation rules to obtain a dynamic state vector; the joint scheduling terminal can organize the link load status and link periodic resource occupancy status into a graph-structured feature matrix; the joint scheduling terminal can integrate the dynamic state vector and the feature matrix to obtain the current scheduling status.

[0059] Specifically, the current scheduling state can be a standardized representation of all relevant information at the time of the scheduling decision.

[0060] In one embodiment, such as Figure 2 As shown, based on the network topology, the candidate next-hop nodes and candidate period offsets of the current node in the current scheduling state are obtained. Integrating the candidate next-hop nodes and candidate period offsets yields the candidate joint action space, which may include:

[0061] Step S301: Represent the network topology as a directed graph, filter out the directed links starting from the current node from the directed graph, and take the endpoint of each directed link as a candidate next-hop node to obtain a set of candidate next-hop nodes.

[0062] Optionally, the joint scheduling terminal can use all the switching nodes in the network topology as vertices of the directed graph and the directed communication links between the switching nodes as edges of the directed graph to construct the directed graph corresponding to the network topology. The joint scheduling terminal can start from the current node, traverse all outgoing edges in the directed graph, and use the endpoints of each outgoing edge as candidate next-hop nodes to obtain a set of candidate next-hop nodes.

[0063] Step S302: Obtain the number of forwarding cycles of the specified queue from the periodic resource status, obtain the maximum period offset, and determine the candidate period offset set as an integer set from 1 to the maximum period offset.

[0064] Optionally, the joint scheduling terminal can obtain the total number of cycles of the specified queue forwarding mechanism from the configuration parameters of the cycle resource management module, and use this total number as the maximum cycle offset; the joint scheduling terminal can generate a continuous integer sequence starting from 1, with a step size of 1, and ending at the maximum cycle offset to obtain a candidate cycle offset set.

[0065] Step S303: Calculate the Cartesian product of the candidate next-hop node set and the candidate periodic offset set to obtain the candidate joint action space.

[0066] Alternatively, the formula for calculating the candidate joint action space representation can be:

[0067] ;

[0068] ;

[0069] ;

[0070] In the formula, Let be the candidate joint action space, where each candidate joint action is represented as: , As a candidate next-hop node For candidate period offset; For the set of candidate next-hop nodes, For the current node; It is the set of all directed links in a directed graph; For the candidate period offset set, This represents the maximum period offset. This indicates the Cartesian product operation.

[0071] In one embodiment, the feasibility of each candidate joint action in the candidate joint action space is verified based on connectivity constraints, acyclic constraints, periodic capacity constraints, deadline reachability constraints, and periodic offset legality constraints, and a feasibility mask is generated, which may include:

[0072] Step S401: Perform feasibility verification on each candidate joint action. When a candidate joint action simultaneously satisfies connectivity constraints, acyclic constraints, periodic capacity constraints, deadline reachability constraints, and periodic offset legality constraints, set the feasibility mask of the candidate joint action to 1.

[0073] Optionally, the joint scheduling terminal can sequentially perform connectivity constraint verification, acyclic constraint verification, periodic capacity constraint verification, deadline reachability constraint verification, and period offset legality constraint verification on each candidate joint action in the candidate joint action space. The joint scheduling terminal can first verify whether a valid directed link exists between the current node and the candidate next-hop node; if it exists, the connectivity constraint verification is passed. The joint scheduling terminal can verify whether the candidate next-hop node already exists in the set of visited nodes of the current service flow; if it does not exist, the acyclic constraint verification is passed. The joint scheduling terminal can calculate the target forwarding period based on the current receiving period and the candidate period offset, and query the remaining available resource capacity of the corresponding directed link in the target forwarding period. If the remaining available resource capacity... If the source capacity exceeds the resource requirements of the service flow to be scheduled, the periodic capacity constraint check is passed. The joint scheduling terminal can calculate the cumulative end-to-end delay after executing the candidate joint action, and combine it with the minimum remaining reachable delay estimate from the candidate next-hop node to the destination node to obtain the total delay. If the total delay does not exceed the maximum allowable end-to-end delay of the service flow to be scheduled, the deadline reachability constraint check is passed. The joint scheduling terminal can check whether the candidate period offset is within the legal value range. If it is, the period offset legality constraint check is passed. The joint scheduling terminal can set the feasibility mask of the candidate joint action that has passed the connectivity constraint check, acyclic constraint check, periodic capacity constraint check, deadline reachability constraint check, and period offset legality constraint check to 1.

[0074] Alternatively, the formula for calculating the target forwarding period can be:

[0075] ;

[0076] In the formula, For the target forwarding period corresponding to the candidate joint action, This represents the current reception period of the service flow to be scheduled at the current node. For candidate period offset, Specify the total number of queue forwarding cycles for the loop. This indicates the modulo operation.

[0077] Specifically, the target forwarding period can be the period number in which the scheduled service flow is actually forwarded on the current link. The target forwarding period can be used to query the remaining available capacity of the link in the corresponding cyclic specified queue forwarding period. The minimum remaining reachable delay can be the theoretical minimum delay from the candidate next-hop node to the destination node.

[0078] Step S402: When a candidate joint action does not satisfy any of the following constraints: connectivity constraint, acyclic constraint, periodic capacity constraint, deadline reachability constraint, and periodic offset legality constraint, the feasibility mask of the candidate joint action is set to 0.

[0079] Optionally, the joint scheduling terminal can determine whether a candidate joint action does not satisfy any of the following constraints: connectivity constraint, acyclic constraint, periodic capacity constraint, deadline reachability constraint, and periodic offset legality constraint. If so, the feasibility mask of the candidate joint action is set to 0.

[0080] In one embodiment, based on a feasibility mask, unexecutable candidate joint actions are eliminated to obtain a set of feasible actions. Then, based on the size of the candidate joint action space and the size of the set of feasible actions, the action space retention rate and action space compression rate are calculated, which may include:

[0081] Step S501: Remove candidate joint actions with a feasibility mask value of 0 from the candidate joint action space to obtain a set of feasible actions.

[0082] Optionally, the joint scheduling terminal can traverse every candidate joint action in the candidate joint action space, filter out candidate joint actions with a feasibility mask value of 0, retain and summarize all candidate joint actions with a feasibility mask value of 1, and obtain a set of actionable actions.

[0083] Alternatively, the expression for the set of actionable actions can be:

[0084] ;

[0085] In the formula, Set up as an actionable set; For candidate joint action space; For candidate joint action The feasibility mask.

[0086] Specifically, the set of actionable actions can be the only valid range for subsequent scheduling action selection.

[0087] Step S502: Based on the size of the candidate joint action space and the size of the action set, calculate the action space retention rate and the action space compression rate.

[0088] Alternatively, the formula for calculating the action space retention rate can be:

[0089] ;

[0090] The formula for calculating the motion space compression ratio can be:

[0091] ;

[0092] In the formula, The number of elements in the set of possible actions; The number of elements in the candidate joint action space; Action space retention rate, used to represent the proportion of actionable actions in candidate joint actions; The action space compression ratio is used to represent the proportion of non-actionable actions that are eliminated.

[0093] In one embodiment, determining a joint scheduling action from the set of possible actions based on network topology, link status, and periodic resource status, and updating the current scheduling state based on the joint scheduling action, may include:

[0094] Step S601: Input the network topology, link status and periodic resource status into the graph convolutional network to extract structured features and obtain topology-aware feature representation.

[0095] Optionally, the joint scheduling terminal can input network topology, link load status, and periodic resource status into a graph convolutional network, perform aggregation updates and global pooling through multiple graph convolutional layers, and output a topology-aware feature representation.

[0096] Step S602: Input the topology-aware feature representation, the current scheduling state, and the candidate joint actions into the deep reinforcement learning model to evaluate the action value and obtain the original action value of each candidate joint action.

[0097] Optionally, the joint scheduling terminal can concatenate and fuse topology-aware feature representations, current scheduling states, and candidate joint actions, and input them into a deep reinforcement learning model to evaluate the long-term scheduling benefits of each candidate joint action and output the original action value of each candidate joint action.

[0098] Step S603: Based on the feasibility mask, reconstruct the original action value to obtain the reconstructed action value.

[0099] Optionally, the formula for calculating the value of a reconstructed action can be:

[0100] ;

[0101] ;

[0102] ;

[0103] In the formula, To reconstruct the value of actions, For candidate joint action Feasibility mask, For candidate joint action The original value of the action This is an action value penalty used to adjust the probability of an unselectable action being chosen; This represents topology-aware features. The current scheduling state is as follows. This represents a deep reinforcement learning model. These are the parameters of the deep reinforcement learning model; This is a directed graph corresponding to the network topology. Link status, For periodic resource status, This represents a graph convolutional network.

[0104] Step S604: Within the set of possible actions, select the candidate joint action with the highest reconstruction action value as the joint scheduling action; and update the link cycle resource occupancy status in the current scheduling state according to the joint scheduling action.

[0105] Optionally, the joint scheduling terminal can compare the reconstruction action values ​​of all candidate joint actions and select the candidate joint action with the largest reconstruction action value as the joint scheduling action; the joint scheduling terminal can set the candidate next-hop node as the new current node and add the current node to the set of visited nodes; the joint scheduling terminal can update the cumulative end-to-end delay and remaining time limit according to the joint scheduling action; the joint scheduling terminal can deduct the remaining available capacity on the target forwarding cycle of the corresponding link according to the selected candidate cycle offset and update the link cycle resource occupancy status.

[0106] Optionally, the selection expression for the joint scheduling action can be:

[0107] ;

[0108] In the formula, For joint scheduling actions, To find the operator for the joint scheduling action that maximizes the function output.

[0109] In one embodiment, when a service flow to be scheduled arrives at a target node in the current scheduling state, a joint scheduling result is generated based on the current scheduling state, action space retention rate, and action space compression rate, which may include:

[0110] Step S701: When the current node in the current scheduling state is the target node, generate a routing path according to the order in which each node in the visited node set is added, and integrate the candidate period offsets selected by each hop to obtain a period offset sequence. Integrate the routing path, period offset sequence, action space retention rate and action space compression rate to obtain a joint scheduling result.

[0111] Optionally, after each hop scheduling decision is completed, the joint scheduling terminal can determine whether the current node and the target node are the same node; if the current node and the target node are the same, the end-to-end scheduling of the service flow to be scheduled is determined to be successful; the joint scheduling terminal can arrange and generate end-to-end routing paths according to the order in which nodes are added to the set of visited nodes; the joint scheduling terminal can arrange the selected candidate period offsets of each hop link according to the path order to generate a period offset sequence; the joint scheduling terminal can summarize the action space retention rate and action space compression rate data of each hop during the scheduling process, and integrate the routing path, period offset sequence, action space retention rate and action space compression rate to obtain the joint scheduling result.

[0112] Specifically, a routing path can be a path that fully represents the forwarding trajectory of a service flow in the spatial dimension; a periodic offset sequence can be used to represent the periodic configuration of a service flow to be scheduled in the time dimension; and action space retention rate and action space compression rate can be used to reflect the constraint strength and resource pressure of scheduling from the process dimension, providing data support for network optimization.

[0113] Step S702: Convert the joint scheduling result into routing and periodic scheduling instructions that can be executed by the underlying network devices, and perform resource scheduling according to the routing and periodic scheduling instructions.

[0114] Optionally, the joint scheduling terminal can convert the routing path into forwarding table entries of each switching node, specify the ingress port, egress port, and matching identifier of the service flow to be scheduled, and convert the periodic offset sequence into the periodic forwarding mapping table and gating control table of each switching node, specifying the forwarding periodic queue of the service flow to be scheduled at the corresponding output port, thereby obtaining the routing and periodic scheduling instructions; the joint scheduling terminal can send the routing and periodic scheduling instructions to all switching nodes, and schedule the switching nodes to perform deterministic forwarding of resources.

[0115] Specifically, routing and periodic scheduling instructions can be configuration instructions that the underlying switching nodes can directly recognize and execute. Routing and periodic scheduling instructions can be directly written into the device's forwarding chip and scheduling chip without additional conversion. Routing and periodic scheduling instructions can be used to ensure the rapid implementation of scheduling schemes.

[0116] Step S703: After the scheduled service flow transmission ends, release the occupied link cycle resources and update the link cycle resource occupancy status.

[0117] Optionally, the joint scheduling terminal can monitor the transmission lifecycle of the service flow to be scheduled. When the transmission task of the service flow is completed or the lifecycle expires, the resource release process is triggered. The joint scheduling terminal can release the resource occupation of each link on the routing path in turn during the forwarding cycle, add the occupied capacity back to the corresponding remaining available capacity, and update the link cycle resource occupation status synchronously.

[0118] Specifically, the dynamic release and update of link lifecycle resources can be used to ensure the recycling of lifecycle resources, avoid long-term occupation of resources by completed services, and improve the overall utilization efficiency of lifecycle resources across the entire network.

[0119] Step S704: When there are no actionable actions, perform exception handling and obtain the exception handling result.

[0120] Optionally, when the set of available actions is empty, the joint scheduling terminal can determine that the scheduled service flow cannot be forwarded in the current state and set an exception handling strategy. Combining the priority and latency tolerance of the scheduled service flow, exception handling is performed: for low-priority service flows, service flow rejection is directly triggered and a scheduling failure response is returned to the service initiator; for service flows that can tolerate a certain latency, delayed scheduling is triggered, and the service flow is placed in the next scheduling window for re-attempted scheduling; for service flows with multi-path transmission capabilities, backup path search is triggered, and other alternative paths are reselected for scheduling starting from the source node; for large-scale scheduling failures caused by network-wide resource shortages, resource reconfiguration is triggered, and the resource allocation ratio of each period is dynamically adjusted to alleviate global resource pressure.

[0121] Specifically, a multi-tiered anomaly handling mechanism can be used to balance the service quality requirements of different services with the overall scheduling stability of the network, ensuring the transmission needs of high-priority services in resource-constrained scenarios.

[0122] It should be understood that although the steps in the flowcharts of the above embodiments are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the above embodiments may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0123] Based on the same inventive concept, this application also provides a feasibility-constrained routing and periodic joint scheduling system for implementing the aforementioned feasibility-constrained routing and periodic joint scheduling method. The solution provided by this system is similar to the implementation scheme described in the above method. Therefore, the specific limitations of one or more feasibility-constrained routing and periodic joint scheduling system embodiments provided below can be found in the limitations of the feasibility-constrained routing and periodic joint scheduling method described above, and will not be repeated here.

[0124] In one exemplary embodiment, such as Figure 3 As shown, a feasibility-constrained routing and periodic joint scheduling system 400 is provided, including:

[0125] The data acquisition module 401 can be used to acquire the pending service flow requests, network topology, link status and periodic resource status within the current scheduling window, and construct the current scheduling status based on the pending service flow requests, network topology, link status and periodic resource status;

[0126] The action space construction module 402 can be used to obtain the candidate next-hop node and candidate period offset of the current node in the current scheduling state according to the network topology, and integrate the candidate next-hop node and candidate period offset to obtain the candidate joint action space.

[0127] The feasibility verification module 403 can be used to verify the feasibility of each candidate joint action in the candidate joint action space based on connectivity constraints, acyclic constraints, periodic capacity constraints, deadline reachability constraints, and periodic offset legality constraints, and generate a feasibility mask.

[0128] The action filtering module 404 can be used to eliminate unexecutable candidate joint actions based on the feasibility mask, obtain a set of actionable actions, and calculate the action space retention rate and action space compression rate based on the size of the candidate joint action space and the size of the set of actionable actions.

[0129] The action determination module 405 can be used to determine joint scheduling actions from the set of actionable actions based on network topology, link status and periodic resource status, and update the current scheduling status based on the joint scheduling actions;

[0130] The scheduling result generation module 406 can be used to generate a joint scheduling result based on the current scheduling state, action space retention rate, and action space compression rate when the service flow to be scheduled arrives at the target node in the current scheduling state.

[0131] In one embodiment, a computer device is provided, including a memory and a processor, the memory storing a computer program, the processor executing the computer program to implement the steps of the feasibility-constrained routing and periodic joint scheduling method as described above.

[0132] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps in the above method embodiments.

[0133] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The device embodiments described above are merely illustrative. The components described as separate parts 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 network units. Some or all of the modules can be selected to achieve the purpose of this disclosure according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0134] The above embodiments merely illustrate several implementation methods of the embodiments of this application, and their descriptions are relatively specific and detailed, but they should not be construed as limiting the scope of the patent application. It should be noted that those skilled in the art can make several modifications and improvements without departing from the concept of the embodiments of this application, and these all fall within the protection scope of the embodiments of this application.

Claims

1. A joint scheduling method combining feasibility-constrained routing and periodic scheduling, characterized in that, The method includes: Obtain the pending service flow requests, network topology, link status, and periodic resource status within the current scheduling window, and construct the current scheduling status based on the pending service flow requests, network topology, link status, and periodic resource status; Based on the network topology, the candidate next-hop node and candidate period offset of the current node in the current scheduling state are obtained, and the candidate next-hop node and candidate period offset are integrated to obtain the candidate joint action space; Based on connectivity constraints, acyclic constraints, periodic capacity constraints, deadline reachability constraints, and periodic offset legality constraints, the feasibility of each candidate joint action in the candidate joint action space is verified, and a feasibility mask is generated. Based on the feasibility mask, the candidate joint actions that are not executable are eliminated to obtain the set of actionable actions. Based on the size of the candidate joint action space and the size of the set of actionable actions, the action space retention rate and the action space compression rate are calculated. Based on the network topology, the link status, and the periodic resource status, a joint scheduling action is determined from the set of possible actions, and the current scheduling status is updated based on the joint scheduling action; When the service flow to be scheduled arrives at the target node in the current scheduling state, a joint scheduling result is generated based on the current scheduling state, the action space retention rate, and the action space compression rate.

2. The method according to claim 1, characterized in that, The service flow request to be scheduled includes a source node, a destination node, and a maximum allowed end-to-end latency; the process of constructing the current scheduling state based on the service flow request to be scheduled, the network topology, the link status, and the periodic resource status includes: The positions of the source node and the destination node in the network topology are determined according to the network topology. The source node is set as the current node and the destination node is set as the target node. The set of visited nodes is initialized to include the source node. The cumulative end-to-end latency is initialized to zero. The remaining time limit is initialized to the maximum allowed end-to-end latency. Obtain the load information of each link in the link status as the link load status; obtain the remaining available capacity of each link in the periodic resource status in different cyclic specified queue forwarding cycles as the link periodic resource occupancy status. The current scheduling state is constructed by integrating the current node, the target node, the set of visited nodes, the cumulative end-to-end latency, the remaining time limit, the link load status, and the link periodic resource occupancy status.

3. The method according to claim 2, characterized in that, The step of obtaining the candidate next-hop node and candidate period offset of the current node in the current scheduling state according to the network topology, and integrating the candidate next-hop node and the candidate period offset to obtain the candidate joint action space includes: The network topology is represented as a directed graph. Directed links starting from the current node are selected from the directed graph. The endpoints of each directed link are used as candidate next-hop nodes to obtain a set of candidate next-hop nodes. The number of forwarding cycles of the specified cyclic queue is obtained from the periodic resource status, the maximum cycle offset is obtained, and the candidate cycle offset set is determined as an integer set from 1 to the maximum cycle offset; The Cartesian product of the candidate next-hop node set and the candidate periodic offset set is calculated to obtain the candidate joint action space; wherein, the calculation formula for the candidate joint action space is: ； ； ； In the formula, Let the candidate joint action space be defined as follows, wherein each candidate joint action is represented as: , This is the candidate next-hop node. The candidate period offset; The set of candidate next-hop nodes. The current node; The set of all directed links in the directed graph; The candidate period offset set, The maximum period offset; This indicates the Cartesian product operation.

4. The method according to claim 1, characterized in that, The feasibility verification of each candidate joint action in the candidate joint action space is performed based on connectivity constraints, acyclic constraints, periodic capacity constraints, deadline reachability constraints, and periodic offset legality constraints, generating a feasibility mask, including: The feasibility of each candidate joint action is verified. When the candidate joint action simultaneously satisfies the connectivity constraint, the acyclic constraint, the periodic capacity constraint, the deadline reachability constraint, and the periodic offset legality constraint, the feasibility mask of the candidate joint action is set to 1. When the candidate joint action does not satisfy any of the connectivity constraint, the acyclic constraint, the periodic capacity constraint, the deadline reachability constraint, and the periodic offset legality constraint, the feasibility mask of the candidate joint action is set to 0.

5. The method according to claim 1, characterized in that, The step involves eliminating unexecutable candidate joint actions based on the feasibility mask to obtain a set of feasible actions. Then, based on the size of the candidate joint action space and the size of the set of feasible actions, the action space retention rate and action space compression rate are calculated, including: The candidate joint actions with a feasibility mask value of 0 are removed from the candidate joint action space to obtain a set of actionable actions; wherein, the set of actionable actions includes all candidate joint actions with a feasibility mask value of 1; Based on the size of the candidate joint action space and the size of the action set, the action space retention rate and the action space compression rate are calculated; wherein, the expression for the action set is: ； The formula for calculating the motion space retention rate is: ； The formula for calculating the motion space compression ratio is: ； In the formula, For the set of possible actions, The number of elements in the set of possible actions; For the candidate joint action space, The number of elements in the candidate joint action space; For candidate joint action The aforementioned feasibility mask; The action space retention rate is used to represent the proportion of actionable actions in the candidate joint actions; The action space compression ratio is used to represent the proportion of inactive actions that are eliminated.

6. The method according to claim 3, characterized in that, The step of determining a joint scheduling action from the set of possible actions based on the network topology, the link status, and the periodic resource status, and updating the current scheduling status based on the joint scheduling action, includes: The network topology, link status, and periodic resource status are input into a graph convolutional network for structured feature extraction to obtain a topology-aware feature representation. The topology-aware feature representation, the current scheduling state, and the candidate joint actions are input into a deep reinforcement learning model to evaluate the action value and obtain the original action value of each candidate joint action. Based on the feasibility mask, the original action value is reconstructed to obtain the reconstructed action value; wherein, the calculation formula for the reconstructed action value is: ； ； ； In the formula, The value of the reconstructed action. For the candidate joint action The aforementioned feasibility mask, For the candidate joint action The value of the original action, This is an action value penalty used to adjust the probability of an unselectable action being chosen; This is the topology-aware feature representation. The current scheduling state is... This refers to the deep reinforcement learning model. These are the parameters of the deep reinforcement learning model; The directed graph corresponding to the network topology. The link status is as described. This refers to the periodic resource status. This represents the graph convolutional network; Within the set of possible actions, the candidate joint action with the highest reconstruction action value is selected as the joint scheduling action; and the link cycle resource occupancy status in the current scheduling state is updated according to the joint scheduling action.

7. The method according to claim 2, characterized in that, When the service flow to be scheduled arrives at the target node in the current scheduling state, a joint scheduling result is generated based on the current scheduling state, the action space retention rate, and the action space compression rate, including: When the current node in the current scheduling state is the target node, a routing path is generated according to the order in which each node in the visited node set is added, and the candidate period offsets selected by each hop are integrated to obtain a period offset sequence. The routing path, the period offset sequence, the action space retention rate, and the action space compression rate are integrated to obtain the joint scheduling result. The joint scheduling results are converted into routing and periodic scheduling instructions that can be executed by the underlying network devices, and resource scheduling is performed according to the routing and periodic scheduling instructions; After the scheduled service flow transmission ends, the occupied link cycle resources are released and the link cycle resource occupancy status is updated. When no action is available, exception handling is performed to obtain the exception handling result; wherein, the exception handling includes triggering service flow rejection, delay scheduling, alternative path search and resource reconfiguration processing.

8. A joint scheduling system for feasibility-constrained routing and periodic scheduling, characterized in that, The system includes: The data acquisition module is used to acquire the pending service flow requests, network topology, link status and periodic resource status within the current scheduling window, and to construct the current scheduling status based on the pending service flow requests, the network topology, the link status and the periodic resource status; The action space construction module is used to obtain the candidate next-hop node and candidate period offset of the current node in the current scheduling state according to the network topology, and integrate the candidate next-hop node and the candidate period offset to obtain the candidate joint action space. The feasibility verification module is used to verify the feasibility of each candidate joint action in the candidate joint action space based on connectivity constraints, acyclic constraints, periodic capacity constraints, deadline reachability constraints, and periodic offset legality constraints, and generate a feasibility mask. The action filtering module is used to eliminate the candidate joint actions that are not executable according to the feasibility mask, to obtain a set of actionable actions, and to calculate the action space retention rate and action space compression rate according to the size of the candidate joint action space and the size of the set of actionable actions. An action determination module is used to determine a joint scheduling action from the set of possible actions based on the network topology, the link status, and the periodic resource status, and to update the current scheduling status based on the joint scheduling action. The scheduling result generation module is used to generate a joint scheduling result based on the current scheduling state, the action space retention rate, and the action space compression rate when the service flow to be scheduled arrives at the target node in the current scheduling state.

9. A computer device, comprising a memory and a processor, characterized in that, The memory stores a computer program, and when the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 7.

Images