A digital twin-based IDC (Internet Data Center) bandwidth resource allocation system

Abstract

This invention discloses a bandwidth resource allocation system for IDC (Internet Data Center) data centers based on digital twins, comprising: a digital twin modeling module, which collects network topology information, link resource status information, and service flow status information to construct a digital twin link graph; a path generation module, which generates a set of candidate transmission paths for service flows based on the digital twin link graph; a link disturbance calculation module, which performs virtual bandwidth disturbances on the links and calculates the link disturbance propagation amount; an association graph construction module, which constructs a service flow association graph and generates a link disturbance impact matrix; an improved SBM (Streaming Bus Module) processing module, which generates traffic blocks and calculates traffic block-level disturbance response values; a bandwidth allocation calculation module, which generates a bandwidth allocation matrix; and a bandwidth execution and feedback module, which executes bandwidth allocation and updates the digital twin link graph. This invention improves bandwidth allocation accuracy and global optimization capabilities.

Description

Technical Field

[0001] This invention relates to the field of computer networks and digital twin technology, and in particular to an IDC (Internet Data Center) bandwidth resource allocation system based on digital twins. Background Technology

[0002] With the rapid development of cloud computing, big data, and artificial intelligence, the scale of business traffic in Internet Data Centers (IDCs) continues to grow, and network link load exhibits highly dynamic characteristics. The rationality of bandwidth resource allocation directly affects the quality of service and resource utilization efficiency. Regarding the bandwidth resource allocation problem in IDCs, existing technologies mainly rely on traffic scheduling methods based on static policies or bandwidth allocation mechanisms based on simple load balancing. However, these methods generally suffer from the following problems in real-world complex network environments: Traditional bandwidth allocation methods often schedule service flows based on fixed thresholds or preset priorities, lacking the ability to dynamically perceive real-time changes in network status and struggling to adapt to the impact of link load fluctuations and sudden increases in service flows. While some methods introduce path selection mechanisms, they primarily rely on the shortest or maximum bandwidth path for decision-making, ignoring the coupling relationships between links and the impact of bandwidth disturbances on the propagation of global service flows. This leads to locally optimal resource allocation results but limited overall performance. Furthermore, existing methods typically employ simple graph structures or independent flow models when modeling competition relationships between service flows, failing to effectively depict the complex interactions between service flows through shared links and limiting the ability to express network congestion propagation mechanisms. In terms of network status modeling, most existing technologies do not incorporate digital twin technology for high-precision mapping of data center rooms, lacking a unified data-driven modeling environment, resulting in deviations between bandwidth allocation strategies and actual operating conditions. Simultaneously, existing graph model-based methods often employ traditional random block models or simple clustering methods, lacking the ability to model link disturbance propagation and multi-scale competition relationships, making it difficult to achieve unified processing of structured grouping and fine-grained scheduling of service flows.

[0003] Therefore, how to provide an IDC data center bandwidth resource allocation system based on digital twins is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0004] One objective of this invention is to propose a bandwidth resource allocation system for IDC (Internet Data Center) data centers based on digital twins. This invention constructs a digital twin link graph and introduces link disturbance propagation modeling, combined with an improved multi-scale random block model, to achieve a structured expression of the competitive relationship between service flows. It generates traffic blocks by fusing local and diffused competition relationships, and calculates bandwidth request intensity based on service flow status and traffic block disturbance response to generate a bandwidth allocation matrix. Finally, it combines execution feedback to form a closed-loop update mechanism, achieving dynamic adaptive and globally optimized allocation of IDC bandwidth resources.

[0005] According to an embodiment of the present invention, an IDC (Internet Data Center) bandwidth resource allocation system based on digital twins includes: The digital twin modeling module is used to collect network topology information, link resource status information, and service flow status information of the IDC data center to construct a digital twin link diagram of the IDC data center. The path generation module is used to generate a set of candidate transmission paths for each service flow based on the digital twin link diagram of the IDC data center. The link disturbance calculation module is used to perform virtual bandwidth disturbance on the links in the candidate transmission path set, update the service flow completion time corresponding to each candidate transmission path, and generate link disturbance propagation volume. The association graph construction module is used to construct a service flow association graph based on the shared link relationship between service flows and the propagation amount of the link disturbance, and to generate a link disturbance impact matrix based on the association edge weights of the service flow association graph. An improved SBM processing module is used to generate local competition scale and diffusion competition scale based on the link disturbance impact matrix, fuse the local competition scale and diffusion competition scale to obtain a multi-scale disturbance impact matrix, and input the multi-scale disturbance impact matrix into the improved SBM model to generate traffic blocks and calculate traffic block-level disturbance response values. The bandwidth allocation calculation module is used to generate the service flow bandwidth request strength based on the service flow status information and the traffic block-level disturbance response value, and to generate a bandwidth allocation matrix based on the service flow bandwidth request strength. The bandwidth execution and feedback module is used to send the bandwidth allocation matrix to network devices to perform bandwidth allocation, and collect the execution results to update the digital twin link diagram of the IDC data center.

[0006] Optionally, the digital twin modeling module includes: Collect server nodes, switching nodes, link connections, and port connections in the IDC data center, establish connection mapping relationships between nodes and links, and form network topology data; Collect the bandwidth usage and maximum bandwidth capacity of each link, subtract the bandwidth usage from the maximum bandwidth capacity to calculate the remaining bandwidth of the link, and record the amount of data to be transmitted in the port queue as the port queue length to form link resource status data. Collect the source and destination nodes, current bandwidth occupied by the service flow, remaining data volume, and completion time limit of each service flow. Divide the remaining data volume by the current bandwidth to calculate the basic transmission time of the service flow. Subtract the basic transmission time from the completion time limit of the service flow to calculate the time limit margin of the service flow. Based on the basic transmission time and the time limit margin of the service flow, form the service flow status data. Based on the network topology data, establish a set of nodes and a set of links, assign link resource status data to the corresponding links, map service flow status data to the corresponding nodes and link paths, and construct a digital twin link graph that includes node connection relationships, remaining link bandwidth, port queue length, and service flow distribution status.

[0007] Optionally, the path generation module includes: In the digital twin link diagram of the IDC data center, the source node of each business flow is used as the starting node and the destination node is used as the ending node. An initial set of reachable paths is generated based on the node connection relationship. Each path consists of a node sequence and a corresponding link sequence. For each path in the initial reachable path set, calculate the path length, which is the sum of the lengths of each link in the path. Calculate the average and standard deviation of all path lengths, and retain paths whose path lengths are less than the sum of the average and standard deviations to form candidate path set one. For each path in the candidate path set one, calculate the remaining bandwidth of the links. The path bandwidth is taken as the minimum value of the remaining bandwidth of each link in the path. Calculate the average value of the bandwidth of all paths and retain the paths whose bandwidth is greater than the average value to form the candidate path set two. Queuing delay is calculated for each path in the candidate path set 2. The path queuing delay is the sum of the ratios of the queue length of the corresponding port of each link in the path to the bandwidth of the corresponding link. The minimum and average queuing delays of all paths are calculated. Paths whose queuing delays are less than the sum of the minimum and average values ​​are retained to form a candidate transmission path set.

[0008] Optionally, the link disturbance calculation module includes: For each path in the candidate transmission path set, extract the remaining bandwidth, port queue length, and remaining data volume and currently occupied bandwidth of each link in the path, calculate the path bandwidth, which is the minimum of the remaining bandwidth of each link in the path, and calculate the initial completion time of the corresponding service flow based on the path bandwidth. The initial completion time is obtained by summing the queuing delay and transmission delay of each link in the path. The queuing delay is the ratio of the port queue length to the link bandwidth, and the transmission delay is the ratio of the remaining data volume to the path bandwidth. Perform bandwidth perturbation operation on each link in the path, and reduce the remaining bandwidth of the link by a certain proportion of the current remaining bandwidth of the link to obtain the perturbed remaining bandwidth of the link. The path bandwidth is recalculated based on the remaining bandwidth of the link after the disturbance, and the completion time of the corresponding service flow is calculated based on the updated path bandwidth to obtain the disturbance completion time. The difference between the disturbance completion time and the initial completion time is calculated as a ratio to the bandwidth disturbance amount to obtain the disturbance impact value of the path on the corresponding link; The disturbance impact values ​​corresponding to all paths passing through the same link are summed to obtain the link disturbance propagation amount.

[0009] Optionally, the association graph construction module includes: Using each service flow as a graph node, based on the link sequence of the corresponding path of the service flow in the candidate transmission path set, identify the shared link in any two service flow paths, establish the connection relationship between service flow nodes, and form the node set and edge set of the service flow association graph. For any pair of business flow nodes that are connected, extract the link disturbance propagation amount of each link in the corresponding shared link set, and sum up the link disturbance propagation amounts of each link in the shared link set to obtain the associated edge weight between the business flow nodes. A weighted adjacency structure is constructed based on the associated edge weights between all business flow nodes. The business flow node number is used as the row index and column index. The associated edge weights between the corresponding business flow nodes are filled into the corresponding positions to obtain the link disturbance impact matrix. The link disturbance impact matrix is ​​processed by assigning values ​​to the diagonal elements. The sum of the weights of the off-diagonal elements in each row is used as the self-impact value of the corresponding business flow node and filled into the diagonal position. The off-diagonal elements are kept unchanged to obtain the link disturbance impact matrix that represents the link disturbance impact relationship between business flows.

[0010] Optionally, the improved SBM processing module includes: An improved SBM model is constructed, which includes a perturbation effect mapping layer, a local competitive scale generation layer, a diffusion competitive scale generation layer, a multi-scale perturbation fusion layer, a perturbation block probability generation layer, and a flow block response output layer. Input the link disturbance impact matrix into the disturbance impact mapping layer, align the matrix rows and columns according to the service flow node number, extract the off-diagonal elements as the disturbance impact values ​​between service flow nodes, and extract the diagonal elements as the self-impact values ​​of service flow nodes; The local competition scale generation layer reads the disturbance impact value between business flow nodes and uses the disturbance impact value between two business flow nodes as the local competition scale. The diffusion competition scale generation layer reads the disturbance impact values ​​between business flow nodes. For two business flow nodes, it calculates the product of the disturbance impact value from the first business flow node to the intermediate business flow node and the disturbance impact value from the intermediate business flow node to the second business flow node. The cumulative value of the corresponding products of all intermediate business flow nodes is divided by the sum of the disturbance impact values ​​from the first business flow node to all intermediate business flow nodes to generate the diffusion competition scale. The multi-scale disturbance fusion layer performs a weighted summation of the local competition scale and the diffusion competition scale to generate multi-scale disturbance impact values, and arranges all multi-scale disturbance impact values ​​according to the business flow node number to form a multi-scale disturbance impact matrix. The perturbation block probability generation layer converts the multi-scale perturbation impact values ​​in the multi-scale perturbation impact matrix into the same block belonging probability between business flow nodes. The same block belonging probability is the ratio of the exponent value of the multi-scale perturbation impact value between the corresponding business flow nodes to the sum of the exponent values ​​of all multi-scale perturbation impact values ​​in the same business flow node's row. Based on the probability of belonging to the same block, generate the same block connection relationship. Establish the same block connection edge between business flow nodes whose probability of belonging to the same block is greater than the average probability of belonging to the same block in the corresponding row. Divide the connected components formed by the same block connection edge to generate multiple traffic blocks. The traffic block response output layer accumulates the self-impact value of the service flow node within each traffic block and the disturbance impact value between service flow nodes within the traffic block, and divides the accumulated result by the number of service flow nodes within the traffic block to calculate the traffic block-level disturbance response value.

[0011] Optionally, the bandwidth allocation calculation module includes: Extract the remaining data volume, current bandwidth usage, and remaining time until completion for each service flow. Calculate the basic transmission time of the service flow by comparing the remaining data volume with the current bandwidth usage. Calculate the time limit margin of the service flow by comparing the remaining time until completion with the basic transmission time. For each service flow, read the block-level disturbance response value of the corresponding traffic block, multiply the block-level disturbance response value with the basic transmission time of the service flow, and multiply the product result with the reciprocal of the service flow time limit margin to obtain the bandwidth request strength of the service flow. The bandwidth request intensity of all service flows is summed, and the bandwidth request intensity of each service flow is calculated as a ratio to the sum of the bandwidth request intensity of all service flows to obtain the bandwidth allocation ratio of the service flows. Obtain the total allocable bandwidth in the IDC data center, multiply the bandwidth allocation ratio of the service flow by the total allocable bandwidth to calculate the bandwidth allocation for each service flow. Using service flow as the row index and link or time slice as the column index, the bandwidth allocation for each service flow is filled into the corresponding position to generate a bandwidth allocation matrix.

[0012] Optionally, the bandwidth execution and feedback module includes: Read the bandwidth allocation matrix and map the bandwidth allocation of the service flow on the corresponding link and time slice to the port rate limiting parameters or queue scheduling parameters in the network device to form a bandwidth configuration instruction. The bandwidth configuration command is sent to the switching or control equipment, the bandwidth limit value is set on the sending port of the corresponding link, and the scheduling weight is set in the queue of the corresponding port. During the bandwidth allocation process, the actual transmission rate of each link, port queue length, number of lost packets, and service flow completion time are collected to form execution result data. The remaining bandwidth of the link, the port queue length, and the service flow completion time are updated based on the execution result data. The remaining bandwidth of the link is calculated based on the difference between the link bandwidth capacity and the actual transmission rate, and the updated data is written into the corresponding node and link attributes of the IDC data center digital twin link diagram. The updated digital twin link graph will be used as input data for the next round of candidate transmission path generation and link disturbance calculation.

[0013] The beneficial effects of this invention are: This invention constructs a link graph model of an IDC (Internet Data Center) based on digital twins. Combining link disturbance propagation modeling with an improved random block model, it addresses the shortcomings of existing bandwidth allocation methods, such as lack of global coupling modeling capabilities, difficulty in depicting service flow competition, and insufficient dynamic adaptability. It proposes a network state modeling strategy based on candidate path generation, link bandwidth disturbance propagation calculation, and service flow association graph construction, significantly improving the ability to express the coupling effects between links and the interaction relationships of service flows. In the graph modeling stage, a link disturbance impact matrix is ​​introduced. Through the accumulation and structured expression of shared link disturbance propagation, a fine depiction of the competitive relationships between service flows is achieved. At the model level, a multi-scale model driven by link disturbances is constructed. The random block model, through the fusion of local competition scale and diffusion competition scale, characterizes the direct competition and indirect propagation relationships between service flows, and achieves adaptive generation of traffic blocks through a probabilistic block partitioning mechanism. In the bandwidth allocation phase, a bandwidth request intensity calculation method based on service flow status and traffic block-level disturbance response is constructed, and a bandwidth allocation matrix is ​​generated in combination with a proportional allocation mechanism to achieve dynamic optimization of bandwidth resources. In the execution and feedback phase, the link operation status is collected in real time and written back to the digital twin link graph to form a closed-loop update mechanism, which effectively improves the consistency between bandwidth allocation strategy and actual network status, and realizes global perception, structure-driven, and dynamic adaptive allocation of IDC data center bandwidth resources. Attached Figure Description

[0014] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a schematic diagram of the structure of an IDC data center bandwidth resource allocation system based on digital twin proposed in this invention; Figure 2This is a schematic diagram of the improved SBM model structure and traffic block partitioning process in an IDC data center bandwidth resource allocation system based on digital twins proposed in this invention. Detailed Implementation

[0015] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0016] refer to Figures 1-2 A digital twin-based IDC (Internet Data Center) bandwidth resource allocation system includes: The digital twin modeling module is used to collect network topology information, link resource status information, and service flow status information of the IDC data center to construct a digital twin link diagram of the IDC data center. The path generation module is used to generate a set of candidate transmission paths for each service flow based on the digital twin link diagram of the IDC data center. The link disturbance calculation module is used to perform virtual bandwidth disturbance on the links in the candidate transmission path set, update the service flow completion time corresponding to each candidate transmission path, and generate the link disturbance propagation amount. The association graph construction module is used to construct a business flow association graph based on the shared link relationships between business flows and the propagation amount of link disturbances, and to generate a link disturbance impact matrix based on the association edge weights of the business flow association graph. An improved SBM processing module is used to generate local competition scale and diffusion competition scale based on the link disturbance impact matrix, fuse the local competition scale and diffusion competition scale to obtain a multi-scale disturbance impact matrix, and input the multi-scale disturbance impact matrix into the improved SBM model to generate traffic blocks and calculate traffic block-level disturbance response values. The bandwidth allocation calculation module is used to generate the bandwidth request strength of the service flow based on the service flow status information and the traffic block-level disturbance response value, and to generate a bandwidth allocation matrix based on the bandwidth request strength of the service flow. The bandwidth execution and feedback module is used to distribute the bandwidth allocation matrix to network devices to perform bandwidth allocation and collect the execution results to update the digital twin link diagram of the IDC data center.

[0017] In this embodiment, the digital twin modeling module includes: Collect server nodes, switching nodes, link connections, and port connections in the IDC data center, establish connection mapping relationships between nodes and links, and form network topology data; Collect the bandwidth usage and maximum bandwidth capacity of each link, subtract the bandwidth usage from the maximum bandwidth capacity to calculate the remaining bandwidth of the link, and record the amount of data to be transmitted in the port queue as the port queue length to form link resource status data. Collect the source and destination nodes, current bandwidth occupied by the service flow, remaining data volume, and completion time limit of each service flow. Divide the remaining data volume by the current bandwidth to calculate the basic transmission time of the service flow. Subtract the basic transmission time from the completion time limit of the service flow to calculate the time limit margin of the service flow. Based on the basic transmission time and the time limit margin of the service flow, form the service flow status data. Based on network topology data, establish a set of nodes and a set of links, assign link resource status data to the corresponding links, map service flow status data to the corresponding nodes and link paths, and construct a digital twin link graph that includes node connection relationships, remaining link bandwidth, port queue length, and service flow distribution status.

[0018] In this implementation, network topology data is uniformly encoded using a mapping table of node identifiers and link identifiers. Node identifiers include server node identifiers and switch node identifiers, and link identifiers are determined by combining node identifier pairs. Port connection relationships are established through a one-to-one correspondence between port numbers and link identifiers. Link resource status data is updated according to a fixed sampling period. Bandwidth occupancy is calculated based on the actual number of bytes transmitted by the link and the sampling time interval. Port queue length is statistically calculated based on the number of data bytes in the port buffer. In service flow status data, the basic transmission time is calculated as the ratio of the remaining data volume to the currently occupied bandwidth. The time margin is used as the difference between the service flow completion time and the basic transmission time to characterize the time urgency of the service flow under the current bandwidth conditions. The digital twin link graph adopts a graph structure storage method, associating and binding node sets, link sets, and corresponding status data to support data retrieval in subsequent link disturbance calculations and traffic block partitioning processes.

[0019] In this embodiment, the path generation module includes: In the digital twin link diagram of the IDC data center, the source node of each business flow is used as the starting node and the destination node is used as the ending node. An initial set of reachable paths is generated based on the node connection relationship. Each path consists of a node sequence and a corresponding link sequence. For each path in the initial reachable path set, calculate the path length, which is the sum of the lengths of each link in the path. Calculate the average and standard deviation of all path lengths, and retain paths whose path lengths are less than the sum of the average and standard deviations to form candidate path set one. For each path in the candidate path set one, calculate the remaining bandwidth of the links. The path bandwidth is taken as the minimum value of the remaining bandwidth of each link in the path. Calculate the average value of the bandwidth of all paths and retain the paths whose bandwidth is greater than the average value to form the candidate path set two. Queuing delay is calculated for each path in the candidate path set 2. The path queuing delay is the sum of the ratios of the queue length of the corresponding port of each link in the path to the bandwidth of the corresponding link. The minimum and average queuing delays of all paths are calculated. Paths whose queuing delays are less than the sum of the minimum and average values ​​are retained to form a candidate transmission path set.

[0020] In this implementation, the initial reachable path set is generated by limiting the number of path nodes to no more than 8; the path length is calculated by accumulating the physical link lengths, and the upper limit of path filtering is the average of all path lengths plus 1.2 times the standard deviation; the path bandwidth is the minimum remaining bandwidth of the link in the path, and the lower limit of bandwidth filtering is 0.85 times the average bandwidth of all paths; the queuing delay is calculated by accumulating the ratio of port queue length to link bandwidth, and the upper limit of delay filtering is 0.9 times the sum of the minimum and average queuing delays of all paths; the number of candidate transmission paths output by each service flow is limited to 2 to 4.

[0021] In this embodiment, the link disturbance calculation module includes: For each path in the candidate transmission path set, extract the remaining bandwidth, port queue length, and remaining data volume and currently occupied bandwidth of each link in the path, calculate the path bandwidth, which is the minimum of the remaining bandwidth of each link in the path, and calculate the initial completion time of the corresponding service flow based on the path bandwidth. The initial completion time is obtained by summing the queuing delay and transmission delay of each link in the path. The queuing delay is the ratio of the port queue length to the link bandwidth, and the transmission delay is the ratio of the remaining data volume to the path bandwidth. Perform bandwidth perturbation operation on each link in the path, and reduce the remaining bandwidth of the link by a certain proportion of the current remaining bandwidth of the link to obtain the perturbed remaining bandwidth of the link. The path bandwidth is recalculated based on the remaining bandwidth of the link after the disturbance, and the completion time of the corresponding service flow is calculated based on the updated path bandwidth to obtain the disturbance completion time. The difference between the disturbance completion time and the initial completion time is calculated as a ratio to the bandwidth disturbance amount to obtain the disturbance impact value of the path on the corresponding link; The disturbance impact values ​​corresponding to all paths passing through the same link are summed to obtain the link disturbance propagation amount.

[0022] In this implementation, the remaining bandwidth of the link and the length of the port queue are collected periodically through the network device interface, with a sampling period of 1 to 5 seconds. Each link in the path is numbered according to the actual connection order to maintain calculation consistency. The bandwidth disturbance ratio is between 0.05 and 0.15 of the current remaining bandwidth of the link, and disturbances are applied to different links to form a single-link disturbance scenario. When updating the path bandwidth, the minimum remaining bandwidth of the link in the path is reselected as the effective bandwidth. During the completion time calculation, the queuing delay is calculated by accumulating the ratio of queue data volume to link bandwidth link by link, and the transmission delay is calculated by accumulating the ratio of the remaining data volume of the service flow to the path bandwidth. When calculating the disturbance impact value, the time difference is normalized to eliminate the differences in the scale of different service flows. The link disturbance propagation volume is accumulated and stored according to the link dimension, and written into the digital twin link graph with the link identifier as the index, providing unified input data for subsequent traffic block partitioning.

[0023] In this embodiment, the association graph construction module includes: Using each service flow as a graph node, based on the link sequence of the corresponding path of the service flow in the candidate transmission path set, identify the shared link in any two service flow paths, establish the connection relationship between service flow nodes, and form the node set and edge set of the service flow association graph. For any pair of business flow nodes that are connected, extract the link disturbance propagation amount of each link in the corresponding shared link set, and sum up the link disturbance propagation amounts of each link in the shared link set to obtain the associated edge weight between the business flow nodes. A weighted adjacency structure is constructed based on the associated edge weights between all business flow nodes. The business flow node number is used as the row index and column index. The associated edge weights between the corresponding business flow nodes are filled into the corresponding positions to obtain the link disturbance impact matrix. The link disturbance impact matrix is ​​processed by assigning values ​​to the diagonal elements. The sum of the weights of the off-diagonal elements in each row is used as the self-impact value of the corresponding business flow node and filled into the diagonal position. The off-diagonal elements are kept unchanged to obtain the link disturbance impact matrix that represents the link disturbance impact relationship between business flows.

[0024] In this implementation, business flow nodes are numbered using unique identifiers, with the number of nodes denoted as N. The link disturbance impact matrix is ​​an N-row, N-column square matrix. The shared link set is obtained by performing an intersection operation on the link identifiers in the path. The weights of associated edges are processed to have the same dimensions before accumulation, and the link disturbance propagation amount is normalized to the range of 0 to 1 according to the maximum value. For node pairs without shared links, the corresponding matrix elements are assigned a value of 0. Before filling diagonal elements, the summation result of the off-diagonal elements in the row is scaled, and the self-influence value is obtained by dividing the summation result by the total number of nodes N. After the matrix is ​​constructed, it is stored row by row and written into the digital twin link graph with the business flow node number as the index for subsequent multi-scale disturbance calculation and input to the improved SBM model.

[0025] In this embodiment, the improved SBM processing module includes: An improved SBM model is constructed, which includes a perturbation effect mapping layer, a local competitive scale generation layer, a diffusion competitive scale generation layer, a multi-scale perturbation fusion layer, a perturbation block probability generation layer, and a flow block response output layer. Input the link disturbance impact matrix into the disturbance impact mapping layer, align the matrix rows and columns according to the service flow node number, extract the off-diagonal elements as the disturbance impact values ​​between service flow nodes, and extract the diagonal elements as the self-impact values ​​of service flow nodes; The local competition scale generation layer reads the disturbance impact value between business flow nodes and uses the disturbance impact value between two business flow nodes as the local competition scale. The diffusion competition scale generation layer reads the disturbance impact values ​​between business flow nodes. For two business flow nodes, it calculates the product of the disturbance impact value from the first business flow node to the intermediate business flow node and the disturbance impact value from the intermediate business flow node to the second business flow node. The cumulative value of the corresponding products of all intermediate business flow nodes is divided by the sum of the disturbance impact values ​​from the first business flow node to all intermediate business flow nodes to generate the diffusion competition scale. The multi-scale disturbance fusion layer performs a weighted summation of the local competition scale and the diffusion competition scale to generate multi-scale disturbance impact values, and arranges all multi-scale disturbance impact values ​​according to the business flow node number to form a multi-scale disturbance impact matrix. The perturbation block probability generation layer converts the multi-scale perturbation impact values ​​in the multi-scale perturbation impact matrix into the same block belonging probability between business flow nodes. The same block belonging probability is the ratio of the exponent value of the multi-scale perturbation impact value between the corresponding business flow nodes to the sum of the exponent values ​​of all multi-scale perturbation impact values ​​in the same business flow node's row. Based on the probability of belonging to the same block, generate the same block connection relationship. Establish the same block connection edge between business flow nodes whose probability of belonging to the same block is greater than the average probability of belonging to the same block in the corresponding row. Divide the connected components formed by the same block connection edge to generate multiple traffic blocks. The traffic block response output layer accumulates the self-impact value of the service flow node within each traffic block and the disturbance impact value between service flow nodes within the traffic block, and divides the accumulated result by the number of service flow nodes within the traffic block to calculate the traffic block-level disturbance response value.

[0026] In this implementation, the weighting coefficients for the local competition scale and the diffusion competition scale in the multi-scale perturbation fusion layer are 0.6 and 0.4, respectively; during the calculation of the diffusion competition scale, the number of intermediate service flow nodes is no less than 2 and no more than 20; the exponential operation in the perturbation block probability generation layer adopts the natural exponential function, and the input value range is limited to 0 to 10 to avoid numerical overflow; after the calculation of the same block ownership probability is completed, the probability of each row is normalized so that the sum of the probabilities of each row is 1; when establishing the same block connection edge, 0.15 is used as the lower limit of probability judgment, and the average value in the row is used for filtering; the connected component partitioning is implemented by depth-first traversal; the number of service flow nodes in the traffic block is limited to 2 to 50; the traffic block level perturbation response value is linearly normalized after calculation, and the result is mapped to the interval of 0 to 1 for subsequent bandwidth request intensity calculation; Both the improved SBM model and the traditional SBM model group nodes based on graph structures, abstracting business flows into nodes, representing the relationships between nodes as edges, and describing the connection relationships between nodes by constructing an adjacency matrix. At the same time, both models use the strength of the relationship between nodes as the basis for partitioning, grouping nodes with strong relationships into the same structure, thereby forming multiple node sets for subsequent calculations. The improved SBM model introduces a link disturbance driving mechanism and a multi-scale disturbance calculation mechanism in its structure. It replaces the traditional block partitioning method based on connection existence with a continuous weight partitioning method based on the propagation of link disturbances. It also sets local competition scale and diffusion competition scale in the model. By fusing the direct disturbance impact and the indirect disturbance impact propagated through intermediate nodes, a multi-scale disturbance impact matrix is ​​constructed. At the same time, exponential mapping and normalization are introduced in the block generation process to convert the multi-scale disturbance impact value into the probability of belonging to the same block, and traffic block partitioning is performed based on the connection relationship of the same block. The improved SBM model introduces link perturbation propagation, enabling the node partitioning results to reflect the bandwidth competition relationship between service flows. Through multi-scale perturbation calculation, the node partitioning results simultaneously include direct competition and indirect propagation relationships. Through a probabilistic partitioning mechanism, the traffic block structure has continuity and adjustability. These improvements enable the generated traffic blocks to accurately characterize the bandwidth coupling relationship between service flows in the IDC data center, providing a more refined structural basis for subsequent bandwidth allocation.

[0027] In this embodiment, the bandwidth allocation calculation module includes: Extract the remaining data volume, current bandwidth usage, and remaining time until completion for each service flow. Calculate the basic transmission time of the service flow by comparing the remaining data volume with the current bandwidth usage. Calculate the time limit margin of the service flow by comparing the remaining time until completion with the basic transmission time. For each service flow, read the block-level disturbance response value of the corresponding traffic block, multiply the block-level disturbance response value with the basic transmission time of the service flow, and multiply the product result with the reciprocal of the service flow time limit margin to obtain the bandwidth request strength of the service flow. The bandwidth request intensity of all service flows is summed, and the bandwidth request intensity of each service flow is calculated as a ratio to the sum of the bandwidth request intensity of all service flows to obtain the bandwidth allocation ratio of the service flows. Obtain the total allocable bandwidth in the IDC data center, multiply the bandwidth allocation ratio of the service flow by the total allocable bandwidth to calculate the bandwidth allocation for each service flow. Using service flow as the row index and link or time slice as the column index, the bandwidth allocation for each service flow is filled into the corresponding position to generate a bandwidth allocation matrix.

[0028] In this implementation, the remaining data volume of the service flow and the currently occupied bandwidth are obtained using a unified data sampling period, which is set to 1 to 3 seconds. The service flow time limit margin is subjected to non-negativity constraint processing after calculation, and the value is taken as 0.01 when the calculation result is less than 0 to avoid the denominator being zero. The traffic block-level disturbance response value is linearly normalized before participating in the calculation, mapping the value range to between 0.1 and 1. The bandwidth request intensity of the service flow is scaled once after the calculation is completed so that the sum of the bandwidth request intensity of all service flows is equal to 1. The total allocable bandwidth is obtained based on the statistics of the remaining bandwidth of the current link. The bandwidth allocation matrix is ​​updated according to the time slice dimension, with the time slice length set to 1 second. The bandwidth allocation of each service flow remains unchanged in each time slice and is recalculated in the next time slice.

[0029] In this embodiment, the bandwidth execution and feedback module includes: Read the bandwidth allocation matrix and map the bandwidth allocation of the service flow on the corresponding link and time slice to the port rate limiting parameters or queue scheduling parameters in the network device to form a bandwidth configuration instruction. The bandwidth configuration command is sent to the switching or control equipment, the bandwidth limit value is set on the sending port of the corresponding link, and the scheduling weight is set in the queue of the corresponding port. During the bandwidth allocation process, the actual transmission rate of each link, port queue length, number of lost packets, and service flow completion time are collected to form execution result data. The remaining bandwidth of the link, the port queue length, and the service flow completion time are updated based on the execution result data. The remaining bandwidth of the link is calculated based on the difference between the link bandwidth capacity and the actual transmission rate, and the updated data is written into the corresponding node and link attributes of the IDC data center digital twin link diagram. The updated digital twin link graph will be used as input data for the next round of candidate transmission path generation and link disturbance calculation.

[0030] In this implementation, bandwidth configuration instructions are encapsulated in a unified format based on the control protocol, including service flow identifiers, link identifiers, and corresponding bandwidth values. Switching devices perform bandwidth allocation through two methods: port rate limiting and queue weighted scheduling. Port rate limiting is used to constrain the maximum transmission rate, while queue weighted scheduling controls the bandwidth ratio by setting different weights. Execution result data is obtained periodically through the device status interface, with a sampling period of 1 to 2 seconds. The actual link transmission rate is calculated based on the amount of data transmitted per unit time, and the port queue length is read from the number of bytes in the buffer. When updating the digital twin link diagram, the original attribute values ​​are overwritten using the link identifier and node identifier as indexes.

[0031] Example 1: To verify the feasibility of this invention in practice, it was applied to an IDC data center located in East China for a large internet company. This data center contains approximately 2000 server nodes, 48 ​​core switches, and 120 access switches, with a total of over 3500 network links. The service flow types cover various categories, including video streams, database synchronization streams, online transaction streams, and offline data processing streams, with the number of service flows exceeding 12000 during peak periods. Under traditional bandwidth allocation mechanisms, this IDC data center primarily uses priority-based and shortest path scheduling. During peak periods, it suffers from severe link congestion, sudden increases in latency for some service flows, and uneven bandwidth utilization. Especially in scenarios with sudden traffic surges, some links experience continuous congestion while others still have available bandwidth, resulting in low overall resource utilization efficiency.

[0032] In this scenario, the proposed digital twin-based IDC (Internet Data Center) bandwidth resource allocation system is deployed on the data center control plane. By collecting data from existing network device interfaces, real-time mapping of server nodes, switching nodes, link connections, and port connections is achieved, constructing a digital twin link graph. During system operation, the network topology and link status are first modeled to obtain remaining link bandwidth, port queue length, and service flow status information, forming a real-time network mapping environment. Based on this, the system generates a set of candidate transmission paths for each service flow. Through adaptive filtering based on path length, remaining link bandwidth, and queuing delay, multiple feasible candidate paths are obtained.

[0033] Subsequently, in the link disturbance modeling phase, bandwidth disturbance operations are performed on links in the candidate paths. By simulating the impact of link bandwidth changes on the completion time of service flows, the propagation amount of link disturbances is calculated, and a relationship graph between service flows is further constructed. By accumulating the propagation amounts of shared link disturbances, a link disturbance impact matrix is ​​formed, thereby transforming the originally independent service flow scheduling problem into a graph modeling problem with structural relationships. Based on this, an improved multi-scale random block model is introduced. By constructing local competition scales and diffusion competition scales, the direct competition relationship between service flows and the indirect competition relationship propagated through intermediate nodes are characterized, dividing the service flow into multiple traffic blocks so that service flows with similar bandwidth competition characteristics belong to the same set.

[0034] During the bandwidth allocation phase, the system calculates the basic transmission time and time margin of the service flow based on the remaining data volume, current bandwidth usage, and completion time limit. It then generates the bandwidth request intensity of the service flow by combining the traffic block-level disturbance response value. Through normalization calculation, the bandwidth allocation ratio for each service flow is obtained, further generating a bandwidth allocation matrix. This matrix is ​​distributed to the switching equipment and executed through port rate limiting and queue scheduling policies. During execution, the system continuously collects the actual link transmission rate, port queue length, packet loss rate, and service flow completion time, feeding the results back to the digital twin link graph for dynamic closed-loop updates.

[0035] In actual operation testing, the system of this invention was compared with the original bandwidth allocation strategy based on shortest path and priority scheduling. Continuous testing for 72 hours was conducted under three typical scenarios: low load, medium load, and high load. During the testing process, key indicators such as average link utilization, average service flow completion time, maximum link congestion rate, and packet loss rate were monitored. The comparative experiments are shown in Table 1. Table 1 Comparison of IDC Data Center Bandwidth Resource Allocation Effects Experimental Data Table

[0036] As can be seen from the comparison results in Table 1 above, the method of the present invention exhibits significant advantages in all three load scenarios. In the low-load scenario, link utilization is improved by approximately 11%, and service completion time is reduced by approximately 22%; in the medium-load scenario, link congestion rate is reduced by more than 15%, and packet loss rate is reduced by more than 50%; in the high-load scenario, the method of the present invention reduces maximum service latency by approximately 34%, while reducing the idle bandwidth ratio to 4.1%, indicating that bandwidth resources are more fully utilized.

[0037] Further analysis revealed that this invention, by introducing link disturbance propagation modeling, enables the system to identify potentially congested links in advance. Furthermore, it uses a multi-scale random block model to structurally partition service flows, thereby preventing multiple highly competitive service flows from concentrating on the same link. Simultaneously, the bandwidth request intensity calculation mechanism prioritizes bandwidth-demanding and disturbance-sensitive service flows for resource allocation, effectively improving overall scheduling efficiency.

[0038] In summary, this invention can effectively solve the problems of low link utilization, severe congestion, and lack of global scheduling in existing bandwidth allocation methods in actual IDC data center environments, and realize the dynamic optimization allocation of bandwidth resources, verifying the feasibility and significant technical effects of this invention in actual engineering scenarios.

[0039] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A bandwidth resource allocation system for IDC (Internet Data Center) data centers based on digital twins, characterized in that, include: The digital twin modeling module is used to collect network topology information, link resource status information, and service flow status information of the IDC data center to construct a digital twin link diagram of the IDC data center. The path generation module is used to generate a set of candidate transmission paths for each service flow based on the digital twin link diagram of the IDC data center. The link disturbance calculation module is used to perform virtual bandwidth disturbance on the links in the candidate transmission path set, update the service flow completion time corresponding to each candidate transmission path, and generate link disturbance propagation volume. The association graph construction module is used to construct a service flow association graph based on the shared link relationship between service flows and the propagation amount of the link disturbance, and to generate a link disturbance impact matrix based on the association edge weights of the service flow association graph. An improved SBM processing module is used to generate local competition scale and diffusion competition scale based on the link disturbance impact matrix, fuse the local competition scale and diffusion competition scale to obtain a multi-scale disturbance impact matrix, and input the multi-scale disturbance impact matrix into the improved SBM model to generate traffic blocks and calculate traffic block-level disturbance response values. The bandwidth allocation calculation module is used to generate the service flow bandwidth request strength based on the service flow status information and the traffic block-level disturbance response value, and to generate a bandwidth allocation matrix based on the service flow bandwidth request strength. The bandwidth execution and feedback module is used to send the bandwidth allocation matrix to network devices to perform bandwidth allocation, and collect the execution results to update the digital twin link diagram of the IDC data center.

2. The IDC data center bandwidth resource allocation system based on digital twins according to claim 1, characterized in that, The digital twin modeling module includes: Collect server nodes, switching nodes, link connections, and port connections in the IDC data center, establish connection mapping relationships between nodes and links, and form network topology data; Collect the bandwidth usage and maximum bandwidth capacity of each link, subtract the bandwidth usage from the maximum bandwidth capacity to calculate the remaining bandwidth of the link, and record the amount of data to be transmitted in the port queue as the port queue length to form link resource status data. Collect the source and destination nodes, current bandwidth occupied by the service flow, remaining data volume, and completion time limit of each service flow. Divide the remaining data volume by the current bandwidth to calculate the basic transmission time of the service flow. Subtract the basic transmission time from the completion time limit of the service flow to calculate the time limit margin of the service flow. Based on the basic transmission time and the time limit margin of the service flow, form the service flow status data. Based on the network topology data, establish a set of nodes and a set of links, assign link resource status data to the corresponding links, map service flow status data to the corresponding nodes and link paths, and construct a digital twin link graph that includes node connection relationships, remaining link bandwidth, port queue length, and service flow distribution status.

3. The IDC data center bandwidth resource allocation system based on digital twins according to claim 1, characterized in that, The path generation module includes: In the digital twin link diagram of the IDC data center, the source node of each business flow is used as the starting node and the destination node is used as the ending node. An initial set of reachable paths is generated based on the node connection relationship. Each path consists of a node sequence and a corresponding link sequence. For each path in the initial reachable path set, calculate the path length, which is the sum of the lengths of each link in the path. Calculate the average and standard deviation of all path lengths, and retain paths whose path lengths are less than the sum of the average and standard deviations to form candidate path set one. For each path in the candidate path set one, calculate the remaining bandwidth of the links. The path bandwidth is taken as the minimum value of the remaining bandwidth of each link in the path. Calculate the average value of the bandwidth of all paths and retain the paths whose bandwidth is greater than the average value to form the candidate path set two. Queuing delay is calculated for each path in the candidate path set 2. The path queuing delay is the sum of the ratios of the queue length of the corresponding port of each link in the path to the bandwidth of the corresponding link. The minimum and average queuing delays of all paths are calculated. Paths whose queuing delays are less than the sum of the minimum and average values ​​are retained to form a candidate transmission path set.

4. The IDC data center bandwidth resource allocation system based on digital twin as described in claim 1, characterized in that, The link disturbance calculation module includes: For each path in the candidate transmission path set, extract the remaining bandwidth, port queue length, and remaining data volume and currently occupied bandwidth of each link in the path, calculate the path bandwidth, which is the minimum of the remaining bandwidth of each link in the path, and calculate the initial completion time of the corresponding service flow based on the path bandwidth. The initial completion time is obtained by summing the queuing delay and transmission delay of each link in the path. The queuing delay is the ratio of the port queue length to the link bandwidth, and the transmission delay is the ratio of the remaining data volume to the path bandwidth. Perform bandwidth perturbation operation on each link in the path, and reduce the remaining bandwidth of the link by a certain proportion of the current remaining bandwidth of the link to obtain the perturbed remaining bandwidth of the link. The path bandwidth is recalculated based on the remaining bandwidth of the link after the disturbance, and the completion time of the corresponding service flow is calculated based on the updated path bandwidth to obtain the disturbance completion time. The difference between the disturbance completion time and the initial completion time is calculated as a ratio to the bandwidth disturbance amount to obtain the disturbance impact value of the path on the corresponding link; The disturbance impact values ​​corresponding to all paths passing through the same link are summed to obtain the link disturbance propagation amount.

5. The IDC data center bandwidth resource allocation system based on digital twin as described in claim 1, characterized in that, The association graph construction module includes: Using each service flow as a graph node, based on the link sequence of the corresponding path of the service flow in the candidate transmission path set, identify the shared link in any two service flow paths, establish the connection relationship between service flow nodes, and form the node set and edge set of the service flow association graph. For any pair of business flow nodes that are connected, extract the link disturbance propagation amount of each link in the corresponding shared link set, and sum up the link disturbance propagation amounts of each link in the shared link set to obtain the associated edge weight between the business flow nodes. A weighted adjacency structure is constructed based on the associated edge weights between all business flow nodes. The business flow node number is used as the row index and column index. The associated edge weights between the corresponding business flow nodes are filled into the corresponding positions to obtain the link disturbance impact matrix. The link disturbance impact matrix is ​​processed by assigning values ​​to the diagonal elements. The sum of the weights of the off-diagonal elements in each row is used as the self-impact value of the corresponding business flow node and filled into the diagonal position. The off-diagonal elements are kept unchanged to obtain the link disturbance impact matrix that represents the link disturbance impact relationship between business flows.

6. The IDC data center bandwidth resource allocation system based on digital twin as described in claim 1, characterized in that, The improved SBM processing module includes: An improved SBM model is constructed, which includes a perturbation effect mapping layer, a local competitive scale generation layer, a diffusion competitive scale generation layer, a multi-scale perturbation fusion layer, a perturbation block probability generation layer, and a flow block response output layer. Input the link disturbance impact matrix into the disturbance impact mapping layer, align the matrix rows and columns according to the service flow node number, extract the off-diagonal elements as the disturbance impact values ​​between service flow nodes, and extract the diagonal elements as the self-impact values ​​of service flow nodes; The local competition scale generation layer reads the disturbance impact value between business flow nodes and uses the disturbance impact value between two business flow nodes as the local competition scale. The diffusion competition scale generation layer reads the disturbance impact values ​​between business flow nodes. For two business flow nodes, it calculates the product of the disturbance impact value from the first business flow node to the intermediate business flow node and the disturbance impact value from the intermediate business flow node to the second business flow node. The cumulative value of the corresponding products of all intermediate business flow nodes is divided by the sum of the disturbance impact values ​​from the first business flow node to all intermediate business flow nodes to generate the diffusion competition scale. The multi-scale disturbance fusion layer performs a weighted summation of the local competition scale and the diffusion competition scale to generate multi-scale disturbance impact values, and arranges all multi-scale disturbance impact values ​​according to the business flow node number to form a multi-scale disturbance impact matrix. The perturbation block probability generation layer converts the multi-scale perturbation impact values ​​in the multi-scale perturbation impact matrix into the same block belonging probability between business flow nodes. The same block belonging probability is the ratio of the exponent value of the multi-scale perturbation impact value between the corresponding business flow nodes to the sum of the exponent values ​​of all multi-scale perturbation impact values ​​in the same business flow node's row. Based on the probability of belonging to the same block, generate the same block connection relationship. Establish the same block connection edge between business flow nodes whose probability of belonging to the same block is greater than the average probability of belonging to the same block in the corresponding row. Divide the connected components formed by the same block connection edge to generate multiple traffic blocks. The traffic block response output layer accumulates the self-impact value of the service flow node within each traffic block and the disturbance impact value between service flow nodes within the traffic block, and divides the accumulated result by the number of service flow nodes within the traffic block to calculate the traffic block-level disturbance response value.

7. The IDC data center bandwidth resource allocation system based on digital twin as described in claim 1, characterized in that, The bandwidth allocation calculation module includes: Extract the remaining data volume, current bandwidth usage, and remaining time until completion for each service flow. Calculate the basic transmission time of the service flow by comparing the remaining data volume with the current bandwidth usage. Calculate the time limit margin of the service flow by comparing the remaining time until completion with the basic transmission time. For each service flow, read the block-level disturbance response value of the corresponding traffic block, multiply the block-level disturbance response value with the basic transmission time of the service flow, and multiply the product result with the reciprocal of the service flow time limit margin to obtain the bandwidth request strength of the service flow. The bandwidth request intensity of all service flows is summed, and the bandwidth request intensity of each service flow is calculated as a ratio to the sum of the bandwidth request intensity of all service flows to obtain the bandwidth allocation ratio of the service flows. Obtain the total allocable bandwidth in the IDC data center, multiply the bandwidth allocation ratio of the service flow by the total allocable bandwidth to calculate the bandwidth allocation for each service flow. Using service flow as the row index and link or time slice as the column index, the bandwidth allocation for each service flow is filled into the corresponding position to generate a bandwidth allocation matrix.

8. The IDC data center bandwidth resource allocation system based on digital twin as described in claim 1, characterized in that, The bandwidth execution and feedback module includes: Read the bandwidth allocation matrix and map the bandwidth allocation of the service flow on the corresponding link and time slice to the port rate limiting parameters or queue scheduling parameters in the network device to form a bandwidth configuration instruction. The bandwidth configuration command is sent to the switching or control equipment, the bandwidth limit value is set on the sending port of the corresponding link, and the scheduling weight is set in the queue of the corresponding port. During the bandwidth allocation process, the actual transmission rate of each link, port queue length, number of lost packets, and service flow completion time are collected to form execution result data. The remaining bandwidth of the link, the port queue length, and the service flow completion time are updated based on the execution result data. The remaining bandwidth of the link is calculated based on the difference between the link bandwidth capacity and the actual transmission rate, and the updated data is written into the corresponding node and link attributes of the IDC data center digital twin link diagram. The updated digital twin link graph will be used as input data for the next round of candidate transmission path generation and link disturbance calculation.

Images