A method for balanced allocation of computing network slice resources based on elastic optical network

Abstract

The present invention discloses a method for balanced allocation of computing power network slice resources based on an elastic optical network, which belongs to the field of network slicing and resource allocation, and includes the following steps: obtaining computing power network node computing resources and link spectrum resource information and network slice request information through an SDN controller; queuing according to service transmission priority, and judging whether the network slice request is allowed to access the computing power network; using a heuristic algorithm to allocate resources for various types of network slice requests that are allowed to access, and obtaining a corresponding computing power network slice resource allocation strategy; using an integrated neural network model to predict the traffic changes of each slice in the next unit time to assist in slice reconfiguration. The present invention proposes a method for balanced allocation of computing power network slice resources based on an elastic optical network, which can make full use of computing power network resources, while achieving balanced allocation of computing power network resources and improving the success rate of network slice requests.

Description

Technical Field

[0001] The present invention relates to the technical field of network slicing and resource allocation, and in particular to a method for balanced allocation of computing power network slicing resources based on an elastic optical network. Background Art

[0002] With the rapid development of next-generation information technology, network services are experiencing a new trend of differentiated demand. Business demands for computing and communication resources are growing exponentially, leaving existing cloud computing architectures facing challenges of limited transmission, computation, and storage. To address this issue, the industry has proposed using computing networks to integrate existing computing, spectrum, and other resources, flexibly allocating and scheduling computing and spectrum resources within the network based on business needs. Traditional wavelength-division multiplexing optical networks, due to their one-size-fits-all approach to spectrum spacing and modulation levels, suffer from low spectrum utilization and flexibility, making them inadequate for computing network requirements. However, elastic optical networks, with their flexible spectrum resource allocation and transmission configuration adjustments, can meet the differentiated business needs of computing networks. Therefore, computing networks can leverage the flexibility of elastic optical networks to provide users with customized network slices, support dynamic business transmission based on differentiated needs, and achieve efficient utilization of computing and spectrum resources.

[0003] Computing networks can effectively promote the integrated development of computing power and networks, fully utilizing computing and spectrum resources within the network. Given the diverse nature of services within computing networks, including latency-sensitive and non-latency-sensitive services, simply providing computing and spectrum resources is insufficient to meet service needs. Appropriate routing for these services is also necessary. Existing network slicing routing research has mostly adopted a single shortest path algorithm. While this can meet the latency requirements of diverse service types, this single routing strategy does not meet the long-term goals of resource allocation. Secondly, the key to supporting dynamic, latency-sensitive services in network slicing lies in how to handle the lack of prior knowledge of slice arrival and departure times. Without this knowledge, slice resource allocation can become uneven over time and lead to spectrum fragmentation. For these reasons, resource allocation to slices must be re-optimized to minimize fragmentation and ensure optimal resource utilization. Existing slice reconfiguration research often performs slice reconfiguration frequently, resulting in unnecessary operational complexity and failing to consider the timing of slice reconfiguration. Most current slice reconfiguration research utilizes traffic prediction to assist in resource allocation. Traffic prediction is a method that uses current network traffic information to reasonably infer future network traffic changes. Existing traffic prediction research is primarily categorized into two approaches: model-driven and data-driven. Model-driven approaches struggle to handle the high dynamics and nonlinearity of network traffic data. Data-driven approaches, on the other hand, are better able to handle the high dynamics of network traffic data and extract nonlinear data features. However, a single prediction model struggles to cope with complex and changing network environments and suffers from unstable training results. Summary of the Invention

[0004] The purpose of the present invention is to provide a method for balanced allocation of computing power network slice resources based on elastic optical networks, which combines access control, path planning, traffic prediction, calculation and spectrum resource balanced allocation algorithms. It can achieve load balancing of computing power network nodes and links while making full use of computing power network resources, and improve the success rate of network slice requests, solve the problem of load imbalance between the underlying nodes, links and network slices of the computing power network, avoid wasting resources, and even affect network slice mapping.

[0005] To achieve the above objectives, the present invention provides a method for balancing computing network slice resources based on an elastic optical network, comprising the following steps:

[0006] Step 1: Obtain computing resources and link spectrum resource information of computing network nodes and network slice request information through the SDN controller;

[0007] Step 2: Queue according to service transmission priority and determine whether the network slice request is allowed to access the computing network;

[0008] Step 3: Use the heuristic algorithm to allocate resources for each type of network slice request allowed to access, and obtain the corresponding computing power network slice resource allocation strategy;

[0009] Step 4: Use the integrated neural network model to predict the traffic changes of each slice in the next unit time to assist in slice reconfiguration.

[0010] Preferably, the specific steps of step 2 are as follows:

[0011] Step 21: Calculate the service transmission priority Q r , the formula is as follows

[0012]

[0013]

[0014]

[0015] Among them, N r The network service level requested for the network slice, τ r Request the maximum tolerable delay for this network slice, The waiting time of the network slice request in the queue, t r Indicates the urgency of business transmission and avoids t through the min-max standardization formula r The sharp fluctuation of the value affects the business transmission priority Q r performance;

[0016] Step 22: Arrange the network slice request set in descending order according to the service priority to obtain a sorted network slice request set;

[0017] Step 23. When the node computing resources and link spectrum resources requested by the network slice are less than or equal to the node computing resources and link spectrum resources available in the computing power network, the network slice request is allowed to access the computing power network. Otherwise, the network slice request needs to be queued. When the number of queued network slice requests is greater than the maximum capacity of the queue, the network slice request access is denied.

[0018] Preferably, the specific steps of step 3 are as follows:

[0019] Step 31, initialize the number of iterations to k+1;

[0020] Step 32: Calculate the virtual node weights and arrange them in descending order. Starting with the node with the largest virtual node weight, select the node with the largest weight from its adjacent nodes. Then, use this node to find the adjacent node with the largest weight. If there is no adjacent node, go back to the previous node and find the node with the largest weight among its adjacent nodes. Repeat the above process until all virtual nodes are traversed. The virtual node traversal order is the mapping order. The following formula is used to calculate the virtual node weight:

[0021]

[0022] in, Calculate resource requirements for node i, is the maximum spectrum resource requirement of the adjacent links of node i, is the node degree of node i;

[0023] Step 33: Based on whether the available computing resources of the physical node are greater than the computing resources requested by the virtual node, determine whether the node degree of the physical node is greater than or equal to the node degree of the virtual node. Finally, determine whether the maximum spectrum gap of the adjacent links of the physical node can meet the spectrum resource requirements of the adjacent links of the virtual node. Nodes that meet the above conditions are selected as candidate physical nodes for the virtual node.

[0024] Step 34: Select the physical node to which the virtual node is mapped from the candidate physical node set according to the node selection probability, and map the virtual node to the physical node. The node selection probability is calculated using the following formula:

[0025]

[0026]

[0027]

[0028]

[0029] Among them, P i r The probability of node selection, is the node availability, is a binary variable, which is 1 when the virtual node is mapped to the physical node i, and 0 otherwise. is the concentration of available frequency slots on both sides of the link at node i, is the sth frequency slot of the link between physical nodes i and j, is the node load, is the computing resources used on the node, C i is the total computing resources of the node;

[0030] Step 35: Use the k shortest path algorithm to deploy virtual links between virtual nodes to ensure link availability. As the link weight, the link is selected with the goal of minimizing the weight;

[0031]

[0032]

[0033] in, is the link load, is the bandwidth occupied on the link, B i,j is the total bandwidth of the link. S is the number of hops between physical nodes. i,j Indicates the total number of frequency slots in the physical link. Indicates the maximum number of idle frequency slots on the physical link;

[0034] Step 36: Record the computing power network node mapped by the network slice request, add 1 to the number of iterations, and determine whether the number of node selection iterations is equal to k+1. If it is equal to k+1, the shortest link availability is used. As the link weight, select the link with the goal of minimizing the weight. If it is less than k+1, go to step 32.

[0035]

[0036] in, is the distance of the link between physical nodes i and j, Yes, the number of hops between physical nodes;

[0037] Step 37: Arrange the k+1 paths in descending order of the objective function and pre-allocate service resources from bottom to top. If the SLA is met, map the network slice request to the current candidate physical node and physical link and accept the current network slice request. Otherwise, try the next candidate path. If all candidate paths cannot be mapped, reject the current network slice request.

[0038] Preferably, the specific steps of step 4 are as follows:

[0039] Step 41: When the slice request success rate per unit time of the computing network is less than the threshold, the traffic change of each slice in the next unit time is predicted using an integrated neural network model composed of a long short-term memory neural network, a gated recurrent neural network, and a convolutional neural network based on the historical traffic information of each slice of the computing network.

[0040] Step 42: Slice reconfiguration is performed based on the traffic prediction value of each slice in the next unit time, so as to reallocate available resources to each slice in accordance with its resource requirements and reduce resource waste caused by spectrum fragmentation in the underlying computing network.

[0041] Step 43: To prevent the integrated neural network model from predicting excessive fluctuations, which may cause the resources allocated to the slice to affect subsequent service transmission, the change in the resources allocated to each slice during the slice reconfiguration phase shall not exceed M% of the original allocated resources.

[0042] Preferably, in step 37, the objective function value of successfully mapping the virtual network to the computing power network is calculated, and the specific formula is as follows:

[0043]

[0044] Here, α and β are constants whose sum is 1. is the load of node i, Indicates the average node utilization. is the load of the physical link between node i and node j, Indicates the average link utilization.

[0045] Preferably, in step 41, the integrated neural network model time series predicts the flow change of each slice in the next unit time in the following specific steps:

[0046] Step 411: Using the computing network unit time as the sampling period, collect the traffic information of each slice, construct a training set and a test set, and pre-process the training set data;

[0047] Step 412: Using the long short-term memory neural network, the doorframe recurrent neural network, and the convolutional neural network as basic models, three separate models are trained on the training set.

[0048] Step 413: Use the boosting algorithm to combine the three basic models into a powerful integrated model, and continuously improve the performance of the weak classifiers through iteration to make them strong classifiers. Use the three trained models to predict the test set and obtain three prediction result sequences;

[0049] Step 414: Calculate a weighted average value based on the given weight coefficients w1, w2, and w3, and use a cross-validation method to select the optimal weighting parameters w1_opt, w2_opt, and w3_opt. Finally, use the optimal weighting parameters to determine the final prediction result.

[0050] Step 415: reallocate node computing resources and link spectrum resources based on the final prediction result ratio of each slice, so that the resources allocated to each slice are adapted to the demand.

[0051] Therefore, the present invention adopts the above-mentioned method for balancing computing network slice resources based on elastic optical networks, which has the following beneficial effects:

[0052] (1) Based on the two-layer resource allocation scheme between slices and within slices, computing network resources are allocated by the main SDN controller to the SDN sub-controllers that control different slices, and then allocated to the services within the slice by the SDN sub-controllers. The present invention is divided into two stages: computing network slice mapping and computing network slice reconfiguration.

[0053] (2) In the computing network slice mapping phase, in order to increase the correlation between the mapped nodes and links, the physical nodes are selected by comprehensively considering the node load and the degree of idle frequency slot aggregation on the node's adjacent links. Different routing strategies are used to select links for different types of services, so as to achieve the effect of balanced resource allocation while meeting business needs.

[0054] (3) In the computing network slice reconfiguration phase, in order to avoid frequent slice reconfiguration and unnecessary operational complexity, only when the slice request success rate per unit time drops below the threshold, the integrated neural network model prediction information is used to assist in changing the resource boundaries of each slice to avoid uneven resource allocation of each slice and minimize spectrum fragmentation, thereby achieving balanced allocation of computing network resources and improving the success rate of network slice requests.

[0055] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1 A schematic diagram of a method for balancing computing network slice resources based on an elastic optical network provided by the present invention;

[0057] Figure 2 A schematic diagram of a flow chart of a method for balancing computing network slice resources based on an elastic optical network provided by the present invention;

[0058] Figure 3 A schematic diagram of the computing power network slice mapping provided by the present invention;

[0059] Figure 4 This is a schematic diagram of the integrated neural network model training process provided by the present invention. DETAILED DESCRIPTION

[0060] Example

[0061] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0062] In response to the problem of load imbalance between the underlying nodes, links and network slices of the computing power network under resource constraints mentioned in the above background technology. An embodiment of the present invention provides a method for balanced allocation of computing power network slice resources based on an elastic optical network. In this invention example, the computing power network node computing resources and link spectrum resource information and network slice request information are obtained through the SDN controller; the computing power network node computing resources and link spectrum resource information and network slice request information are obtained; the priority is queued and it is determined whether the network slice request is allowed to access the computing power network; the heuristic algorithm is used to allocate resources for various types of network slice requests that are allowed to access, and the corresponding computing power network slice resource allocation strategy is obtained; the integrated neural network model is used to predict the traffic changes of each slice in the next unit time to assist in slice reconfiguration. While meeting business needs, the load balancing of computing power network resources is achieved, the computing power network resources are fully utilized, and the success rate of network slice requests is improved. This solves the problem of resource waste caused by unreasonable allocation of computing power network slice resources and spectrum fragmentation.

[0063] like Figure 1-2 A method for balancing computing network slice resources based on an elastic optical network includes the following steps:

[0064] Step 1: Obtain computing network node computing resources and link spectrum resource information as well as network slicing request information through the SDN controller.

[0065] Step 2: Queue according to the service transmission priority and determine whether the network slicing request is allowed to access the computing network.

[0066] The first step is to calculate the service transmission priority P by comprehensively considering the service network service level, queue waiting time and maximum tolerable delay to avoid long-term waiting of low-priority services. r , the specific formula is as follows:

[0067]

[0068]

[0069]

[0070] Among them, N r The network service level requested for the network slice, τ r Request the maximum tolerable delay for this network slice, The waiting time of the network slice request in the queue, t r Indicates the urgency of business transmission. Use the min-max normalization formula to avoid t r The sharp fluctuation of the value affects the service transmission priority P r performance.

[0071] In the second step, the network slice request set is sorted in descending order according to the service transmission priority to obtain the sorted network slice request set, and the network slice requests in the set are processed from bottom to top.

[0072] When the node computing resources and link spectrum resources requested by the network slice are less than or equal to the node computing resources and link spectrum resources available in the computing power network, the network slice request is allowed to access the computing power network. Otherwise, the network slice request needs to be queued. When the number of queued network slice requests is greater than the maximum capacity of the queue, the network slice request access is denied.

[0073] Step 3: Use the heuristic algorithm to allocate resources for various network slice requests that are allowed to access the computing network, and obtain the corresponding computing network slice resource allocation strategy.

[0074] Step 1: Initialize the number of iterations to k+1;

[0075] Step 2: Calculate the virtual node weights and sort them in descending order. Take the node with the largest virtual node weight as the starting node, select the node with the largest weight from the adjacent nodes of the node, and then use this node to find the adjacent node with the largest weight. If there is no adjacent node, go back to the previous node and find the node with the largest weight among its adjacent nodes. Repeat the above process until all virtual nodes are traversed. The virtual node traversal order is its mapping order, such as Figure 3 The following formula is used to calculate the virtual node weight:

[0076]

[0077] in, Calculate resource requirements for node i, is the maximum spectrum resource requirement of the adjacent links of node i, is the node degree of node i.

[0078] Step 3: Determine whether the available computing resources of the physical node are greater than the computing resources requested by the virtual node, then determine whether the node degree of the physical node is greater than or equal to the node degree of the virtual node, and finally determine whether the maximum spectrum gap of the adjacent links of the physical node can meet the spectrum resource requirements of the adjacent links of the virtual node. Nodes that meet the above conditions are considered as candidate physical nodes for the virtual node.

[0079] Step 4: In the candidate physical node set, select the physical node to which the virtual node is mapped based on the node selection probability, and map the virtual node to the physical node. The node selection probability is specifically calculated using the following formula:

[0080]

[0081]

[0082]

[0083]

[0084] Among them, Pir is the node selection probability, is the node availability, is a binary variable, which is 1 when the virtual node is mapped to the physical node i, and 0 otherwise. is the concentration of available frequency slots on both sides of the link at node i, is the sth frequency slot of the link between physical nodes i and j, is the node load, is the computing resources used on the node, C i The total computing resources of the node.

[0085] Step 5: Use the k shortest path algorithm to deploy virtual links between virtual nodes to ensure link availability. Links are selected with the goal of minimizing the link weight.

[0086]

[0087]

[0088] in, is the link load, is the bandwidth occupied on the link, B i,j is the total bandwidth of the link. S is the number of hops between physical nodes. i,j Indicates the total number of frequency slots in the physical link. Indicates the maximum number of idle frequency slots on a physical link.

[0089] Step 6: Record the computing power network node mapped by the network slice request, add 1 to the number of iterations, and determine whether the number of node selection iterations is equal to k+1. If it is equal to k+1, the shortest link availability is used. As the link weight, a link is selected with the goal of minimizing the weight. If it is less than k+1, a physical node is reselected from the candidate physical node set based on the node selection probability.

[0090]

[0091] in, is the distance of the link between physical nodes i and j, is the number of hops between physical nodes.

[0092] Step 7: Arrange the k+1 paths in descending order of objective function value and pre-allocate service resources from bottom to top. If the SLA is met, the network slice request is mapped to the current candidate physical node and physical link and the current network slice request is accepted. Otherwise, the next candidate path is tried. If all candidate paths cannot be mapped, the current network slice request is rejected.

[0093] The network slice traffic information of each slice is calculated, and the success rate of network slice requests per unit time is calculated to determine whether it is less than the threshold value. The network slice request success rate per unit time is specifically calculated using the following formula:

[0094]

[0095] Among them, R succ is the number of successful network slice requests per unit time, R total It is the total number of network slice requests per unit time of the computing network.

[0096] Step 4: Use the integrated neural network model to predict the flow change of each slice in the next unit time to assist in slice reconfiguration.

[0097] In the first step, to avoid the impact of frequent network slice reconfiguration on service transmission, only when the slice request success rate per unit time of the computing power network is less than the threshold value, the traffic changes of each slice in the next unit time are predicted based on the historical traffic information of each slice of the computing power network, using an integrated neural network model with long short-term memory neural network, gated recurrent neural network and convolutional neural network as components.

[0098] like Figure 4 , where the steps of integrating the neural network model time series to predict the flow change of each slice in the next unit time are as follows:

[0099] a. Using the computing network unit time as the sampling period, collect traffic information for each slice, construct training and test sets, and preprocess the training set data.

[0100] b. Use the long short-term memory neural network, the gate frame recurrent neural network, and the convolutional neural network as the basic models, and train three separate models on the training set.

[0101] c. Use the boosting algorithm to combine the three basic models into a powerful ensemble model. This iterative approach continuously improves the performance of the weak classifiers, turning them into strong classifiers. Use the three trained models to predict the test set, generating three sequences of prediction results.

[0102] d. Calculate the weighted average based on the given weight coefficients w1, w2, and w3, and use the cross-validation method to select the optimal weighting parameters w1_opt, w2_opt, and w3_opt. Finally, use the optimal weighting parameters to determine the final prediction result.

[0103] e. Based on the final prediction result ratio of each slice, reallocate node computing resources and link spectrum resources to make the resources allocated to each slice adapt to demand.

[0104] The second step is to assist in slice reconfiguration based on the traffic prediction value of each slice in the next unit time, reallocate available resources to each slice in accordance with its resource requirements, and reduce resource waste caused by the fragmentation of the underlying spectrum of the computing power network.

[0105] In the third step, during the slice reconfiguration process, in order to avoid the integrated neural network model's prediction fluctuations being too large, which may cause the resources allocated to the slice to affect subsequent service transmission, the change in the resources allocated to each slice during the slice reconfiguration phase shall not exceed M% of the original allocated resources.

[0106] Calculate the objective function value of successfully mapping the virtual network to the computing power network using the following formula:

[0107]

[0108] Here, α and β are constants whose sum is 1. is the load of node i, Indicates the average node utilization. is the load of the physical link between node i and node j, Indicates the average link utilization.

[0109] Therefore, the present invention adopts the above-mentioned method for balanced allocation of computing power network slice resources based on elastic optical network. In this embodiment of the invention, the computing power network node computing resources and link spectrum resource information and network slice request information are obtained through the SDN controller; the computing power network node computing resources and link spectrum resource information and network slice request information are obtained; the priority is queued, and it is determined whether the network slice request is allowed to access the computing power network; the heuristic algorithm is used to allocate resources for various types of network slice requests that are allowed to access, and the corresponding computing power network slice resource allocation strategy is obtained; the integrated neural network model is used to predict the traffic changes of each slice in the next unit time to assist in slice reconfiguration. While meeting business needs, the load balancing of computing power network resources is achieved, the computing power network resources are fully utilized, and the success rate of network slice requests is improved. This solves the problem of resource waste caused by unreasonable allocation of computing power network slice resources and spectrum fragmentation.

[0110] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit the same. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that they can still modify or replace the technical solutions of the present invention with equivalents, and these modifications or equivalent replacements cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for balanced allocation of computing network slice resources based on elastic optical networks, characterized by: The following steps are involved: Step 1: Obtain computing resources and link spectrum resource information of computing network nodes and network slice request information through the SDN controller; Step 2: Queue according to service transmission priority and determine whether the network slice request is allowed to access the computing network; Step 3: Use the heuristic algorithm to allocate resources for each type of network slice request allowed to access, and obtain the corresponding computing power network slice resource allocation strategy; The specific steps are as follows: Step 31, initialize the number of iterations to k+1; Step 32: Calculate the virtual node weights and arrange them in descending order. Take the node with the largest virtual node weight as the starting node, select the node with the largest weight from the adjacent nodes of the node, and then use this node to find the adjacent node with the largest weight. If there is no adjacent node, go back to the previous node and find the node with the largest weight among its adjacent nodes. Repeat the above process in step S32 until all virtual nodes are traversed. The virtual node traversal order is its mapping order. The virtual node weight is calculated using the following formula: in, Calculate resource requirements for node i, is the maximum spectrum resource requirement of the adjacent links of node i, is the node degree of node i; Step 33: Based on whether the available computing resources of the physical node are greater than the computing resources requested by the virtual node, determine whether the node degree of the physical node is greater than or equal to the node degree of the virtual node. Finally, determine whether the maximum spectrum gap of the adjacent links of the physical node meets the spectrum resource requirements of the adjacent links of the virtual node. Nodes that meet the conditions in step 33 are selected as candidate physical nodes for the virtual node. Step 34: Select the physical node to which the virtual node is mapped from the candidate physical node set according to the node selection probability, and map the virtual node to the physical node. The specific formula for calculating the node selection probability is as follows: in, The probability of node selection, is the node availability, is a binary variable, is the concentration of available frequency slots on both sides of the link at node i, is the sth frequency slot of the link between physical nodes i and j, is the node load, is the computing resources used on the node, is the total computing resources of the node; Step 35: Use the k shortest path algorithm to deploy virtual links between virtual nodes to ensure link availability. As the link weight, the link is selected with the goal of minimizing the weight. Link availability The formula is as follows in, is the link load, is the bandwidth occupied on the link, is the total bandwidth of the link; is the number of hops between physical nodes, Indicates the total number of frequency slots in the physical link. Indicates the maximum number of idle frequency slots on the physical link; Step 36: Record the computing power network node mapped by the network slice request, add 1 to the number of iterations, and determine whether the number of node selection iterations is equal to k+1. If it is equal to k+1, the shortest link availability is used. As the link weight, select the link with the goal of minimizing the weight. If it is less than k+1, go to step 32. in, is the distance of the link between physical nodes i and j, Yes, the number of hops between physical nodes; Step 37: Arrange the k+1 paths in descending order according to the size of the objective function, and pre-allocate service resources from bottom to top. If the SLA is met, map the network slicing request to the current candidate physical node and physical link, and accept the current network slicing request. Otherwise, try the next candidate path. If all candidate paths cannot be mapped, reject the current network slicing request. Step 4: Use the integrated neural network model to predict the traffic changes of each slice in the next unit time to assist in slice reconfiguration.

2. The method for balanced allocation of computing network slice resources based on elastic optical network according to claim 1 is characterized in that , the specific steps of step 2 are as follows: Step 21: Calculate service transmission priority , the formula is as follows in, The network service level requested for the network slice, Request the maximum tolerable delay for this network slice, The time that the network slice request waits in the queue. Indicates the urgency of business transmission and avoids it through the min-max normalization formula Sharp fluctuations in values ​​affect service transmission priorities performance; Step 22: Arrange the network slice request set in descending order according to the service priority to obtain a sorted network slice request set; Step 23. When the node computing resources and link spectrum resources requested by the network slice are less than or equal to the node computing resources and link spectrum resources available in the computing power network, the network slice request is allowed to access the computing power network. Otherwise, the network slice request needs to be queued. When the number of queued network slice requests is greater than the maximum capacity of the queue, the network slice request access is denied.

3. The method for balanced allocation of computing network slice resources based on elastic optical network according to claim 1, characterized in that: The specific steps of step 4 are as follows: Step 41: When the slice request success rate per unit time of the computing network is less than the threshold, the traffic change of each slice in the next unit time is predicted using an integrated neural network model composed of a long short-term memory neural network, a gated recurrent neural network, and a convolutional neural network based on the historical traffic information of each slice of the computing network. Step 42: Slice reconfiguration is performed based on the traffic prediction value of each slice in the next unit time, so as to reallocate available resources to each slice in accordance with its resource requirements and reduce resource waste caused by spectrum fragmentation in the underlying computing network. Step 43: To avoid excessive fluctuations in the prediction of the integrated neural network model, which may cause the resources allocated to the slice to affect subsequent service transmission, the change in the resources allocated to each slice during the slice reconfiguration phase shall not exceed M% of the original allocated resources.

4. The method for balanced allocation of computing network slice resources based on elastic optical network according to claim 1, characterized in that: In step 37, the objective function value of successfully mapping the virtual network to the computing power network is calculated. The specific formula is as follows: in, and are constants that sum to 1, is the load of node i, represents the average node utilization, is the load of the physical link between node i and node j, Indicates the average link utilization.

5. The method for balanced allocation of computing network slice resources based on elastic optical network according to claim 3, characterized in that: In step 41, the integrated neural network model time series predicts the flow change of each slice in the next unit time in the following specific steps: Step 411: Using the computing network unit time as the sampling period, collect the traffic information of each slice, construct a training set and a test set, and pre-process the training set data; Step 412: Using the long short-term memory neural network, the doorframe recurrent neural network, and the convolutional neural network as basic models, three separate models are trained on the training set. Step 413: Use the boosting algorithm to combine the three basic models into a powerful integrated model. Iteratively improve the performance of the weak classifiers to make them strong classifiers. Use the three trained models to predict the test set and obtain three prediction result sequences. Step 414: Calculate the weighted average value based on the given weight coefficients w1, w2 and w3, and use the cross-validation method to select the best weighting parameter. w1_opt, w2_opt, and w3_opt, Finally, the optimal weighted parameters are used to determine the final prediction results; Step 415: reallocate node computing resources and link spectrum resources based on the final prediction result ratio of each slice, so that the resources allocated to each slice are adapted to the demand.

Images