CN117749631A - Isolation method and device for dynamic network topology resources - Google Patents
Isolation method and device for dynamic network topology resources Download PDFInfo
- Publication number
- CN117749631A CN117749631A CN202311812377.2A CN202311812377A CN117749631A CN 117749631 A CN117749631 A CN 117749631A CN 202311812377 A CN202311812377 A CN 202311812377A CN 117749631 A CN117749631 A CN 117749631A
- Authority
- CN
- China
- Prior art keywords
- data
- network
- resource isolation
- resource
- resources
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000002955 isolation Methods 0.000 title claims abstract description 173
- 230000008859 change Effects 0.000 claims abstract description 78
- 238000000034 method Methods 0.000 claims abstract description 38
- 238000013468 resource allocation Methods 0.000 claims abstract description 28
- 238000011156 evaluation Methods 0.000 claims abstract description 18
- 238000013528 artificial neural network Methods 0.000 claims description 26
- 230000009471 action Effects 0.000 claims description 25
- 125000004122 cyclic group Chemical group 0.000 claims description 21
- 238000012549 training Methods 0.000 claims description 21
- 238000005265 energy consumption Methods 0.000 claims description 12
- 230000003416 augmentation Effects 0.000 claims description 11
- 238000012544 monitoring process Methods 0.000 claims description 10
- 238000005516 engineering process Methods 0.000 claims description 9
- 238000004458 analytical method Methods 0.000 claims description 7
- 230000003139 buffering effect Effects 0.000 claims description 7
- 238000011478 gradient descent method Methods 0.000 claims description 6
- 238000010606 normalization Methods 0.000 claims description 6
- 239000012634 fragment Substances 0.000 claims description 4
- 238000013467 fragmentation Methods 0.000 claims description 2
- 238000006062 fragmentation reaction Methods 0.000 claims description 2
- 230000000875 corresponding effect Effects 0.000 description 28
- 230000006870 function Effects 0.000 description 16
- 238000005457 optimization Methods 0.000 description 15
- 238000007726 management method Methods 0.000 description 11
- 230000003190 augmentative effect Effects 0.000 description 9
- 230000000306 recurrent effect Effects 0.000 description 9
- 238000003062 neural network model Methods 0.000 description 8
- 230000003321 amplification Effects 0.000 description 7
- 238000003199 nucleic acid amplification method Methods 0.000 description 7
- 230000008569 process Effects 0.000 description 6
- 238000004364 calculation method Methods 0.000 description 3
- 238000004891 communication Methods 0.000 description 3
- 230000008878 coupling Effects 0.000 description 3
- 238000010168 coupling process Methods 0.000 description 3
- 238000005859 coupling reaction Methods 0.000 description 3
- 230000006872 improvement Effects 0.000 description 3
- 230000002776 aggregation Effects 0.000 description 2
- 238000004220 aggregation Methods 0.000 description 2
- 230000008901 benefit Effects 0.000 description 2
- 230000015556 catabolic process Effects 0.000 description 2
- 238000006731 degradation reaction Methods 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 239000011159 matrix material Substances 0.000 description 2
- 230000007246 mechanism Effects 0.000 description 2
- 230000008520 organization Effects 0.000 description 2
- 230000002787 reinforcement Effects 0.000 description 2
- 238000003860 storage Methods 0.000 description 2
- 230000002123 temporal effect Effects 0.000 description 2
- 230000002159 abnormal effect Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 230000000295 complement effect Effects 0.000 description 1
- 230000001276 controlling effect Effects 0.000 description 1
- 230000002079 cooperative effect Effects 0.000 description 1
- 238000007405 data analysis Methods 0.000 description 1
- 230000006866 deterioration Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000009826 distribution Methods 0.000 description 1
- 230000003203 everyday effect Effects 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000013439 planning Methods 0.000 description 1
- 230000002035 prolonged effect Effects 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000035945 sensitivity Effects 0.000 description 1
- 238000004904 shortening Methods 0.000 description 1
- 230000003068 static effect Effects 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
Landscapes
- Data Exchanges In Wide-Area Networks (AREA)
Abstract
The invention discloses a method and a device for isolating dynamic network topology resources, wherein the method comprises the steps of obtaining topology data of a dynamic network according to a preset period; after positioning flow hot spots according to topology data and performing resource allocation, determining network change prediction data by performing change prediction on a dynamic network; performing demand evaluation on the network change prediction data to generate an initial resource isolation strategy; performing resource isolation by adopting an initial resource isolation strategy to generate a resource isolation result; and adjusting the initial resource isolation strategy according to feedback information corresponding to the resource isolation result to generate a final resource isolation strategy, thereby effectively improving the utilization efficiency of resources.
Description
Technical Field
The present invention relates to the field of resource isolation technologies, and in particular, to a method and an apparatus for isolating dynamic network topology resources.
Background
With the rapid development of network intelligence and internet of things, dynamic network topology gradually becomes one of the network topologies which are more commonly used at present. Compared with the traditional static network topology, the dynamic network topology can be more effectively adapted to various network change conditions, such as network node faults, link interruption or addition of new nodes.
In the everyday use of dynamic network topologies, resource allocation complexity and traffic load imbalance are often important issues faced by network resource management and optimization. In the process of resource allocation, the mode and the utilization efficiency of various resources need to be considered, but the research on multi-mode resource isolation in the prior art is limited, but the reasonable allocation and isolation of the multi-mode resources are very important for improving the network efficiency and ensuring the communication quality.
The traditional resource isolation method generally divides network resources into fixed parts, each fixed part allocates certain resources, meanwhile, the change of network topology cannot be timely perceived and adapted, and the resource re-planning and allocation cannot be carried out according to new topology information, so that the utilization efficiency of the resources is reduced.
Disclosure of Invention
The invention provides a method and a device for isolating dynamic network topology resources, which solve the technical problems that the traditional method for isolating the resources generally divides the network resources into fixed parts, each fixed part is allocated with certain resources, meanwhile, the change of the network topology cannot be timely sensed and adapted, and the resource can not be re-planned and allocated according to new topology information, so that the utilization efficiency of the resources is reduced.
The invention provides a method for isolating dynamic network topology resources, which comprises the following steps:
obtaining topology data of a dynamic network according to a preset period;
after positioning flow hot spots and carrying out resource allocation according to the topology data, determining network change prediction data by carrying out change prediction on the dynamic network;
performing demand evaluation on the network change prediction data to generate an initial resource isolation strategy;
performing resource isolation by adopting the initial resource isolation strategy to generate a resource isolation result;
and adjusting the initial resource isolation strategy according to the feedback information corresponding to the resource isolation result to generate a final resource isolation strategy.
Optionally, after locating the traffic hot spot according to the topology data and performing resource allocation, determining network change prediction data by performing change prediction on the dynamic network includes:
performing dynamic network topology analysis according to the topology data, and determining flow hot spots in the dynamic network;
performing resource allocation according to the service level corresponding to the traffic hot spot, and updating the topology data;
and predicting the change of the dynamic network according to the updated topology data through a preset target cyclic neural network, and determining network change prediction data.
Optionally, the topology data includes node information and network structure information; the step of performing dynamic network topology analysis according to the topology data to determine traffic hot spots in the dynamic network includes:
selecting a source node and a sink node according to the node information;
retrieving the augmentation paths between the source node and the sink node one by one according to the network structure information, and updating the flow information of each augmentation path in real time until the augmentation paths cannot be retrieved;
and selecting the node connection with the traffic information in the augmentation path as a traffic hot spot in the dynamic network.
Optionally, the method further comprises:
acquiring historical topology data of the dynamic network; the historical topological data is provided with historical change prediction data;
sequentially selecting outlier data one by one from the historical topological data, and respectively calculating an average value among a plurality of historical topological data adjacent to each outlier data;
zero-mean normalization is respectively carried out on each average value, and a historical data set is constructed;
and training a preset cyclic neural network by adopting the historical data set and the historical change prediction data and combining a standard back propagation algorithm and a gradient descent method to obtain a target cyclic neural network.
Optionally, the network change prediction data comprises a plurality of data streams of different modalities; the step of carrying out demand evaluation on the network change prediction data to generate an initial resource isolation strategy comprises the following steps:
respectively setting priorities of various data streams from high to low according to the quantity of required resources respectively corresponding to the various data streams;
setting the resource quota ratio of various data flows from high to low according to the priority;
and distributing the available resources of the dynamic network at the current moment to various data flows according to the resource quota proportion to obtain an initial resource isolation strategy.
Optionally, the method further comprises:
periodically monitoring various of said data streams and said dynamic network;
when new available resources appear and the allocated available resources do not meet the quantity of the required resources corresponding to the data stream with the highest priority, the new available resources are allocated to the data stream with the highest priority;
and if any one of the data streams is received to update the required resource quantity or the dynamic network has bandwidth change, executing the jump according to the required resource quantity respectively corresponding to the various data streams, and respectively setting the priorities of the various data streams from high to low.
Optionally, the performing resource isolation by using the initial resource isolation policy, generating a resource isolation result includes:
analyzing the initial resource isolation strategy, determining resource information of the multi-mode resources, and setting a time window according to the resource information;
dividing the time window by adopting dynamic fragmentation to obtain a plurality of time fragments, and respectively storing the multi-mode resources in each time fragment by adopting a resource buffering technology;
according to the overlapping degree of execution time between the multi-mode resources in each time segment, adjusting the segment length of the time segment and recording;
and when the time window is finished, generating a resource isolation result by adopting the segment length, the starting time and the ending time of each time segment.
Optionally, before the generating the resource isolation result by adopting the segment length, the start time and the end time of each time segment when the time window ends, the method further includes:
and adjusting the priority of the multi-mode resources according to the emergency degree of each multi-mode resource.
Optionally, the feedback information includes network delay, packet loss rate, and energy consumption index attribute; the step of adjusting the initial resource isolation strategy according to the feedback information corresponding to the resource isolation result to generate a final resource isolation strategy comprises the following steps:
Respectively acquiring a reward function according to network delay, a data packet loss rate and energy consumption index attributes corresponding to the resource isolation result;
and taking the initial resource isolation strategy as an action, and adjusting according to the reward function and the Q-Learning algorithm to generate a final resource isolation strategy.
The invention also provides an isolation device of the dynamic network topology resource, which comprises:
the topology data acquisition module is used for acquiring topology data of the dynamic network according to a preset period;
the change prediction module is used for determining network change prediction data by performing change prediction on the dynamic network after positioning flow hot spots and performing resource allocation according to the topology data;
the demand evaluation module is used for carrying out demand evaluation on the network change prediction data and generating an initial resource isolation strategy;
the resource isolation module is used for carrying out resource isolation by adopting the initial resource isolation strategy to generate a resource isolation result;
and the feedback adjustment module is used for adjusting the initial resource isolation strategy according to the feedback information corresponding to the resource isolation result to generate a final resource isolation strategy.
From the above technical scheme, the invention has the following advantages:
The topology data of the dynamic network are obtained according to a preset period; after positioning flow hot spots according to topology data and performing resource allocation, determining network change prediction data by performing change prediction on a dynamic network; performing demand evaluation on the network change prediction data to generate an initial resource isolation strategy; performing resource isolation by adopting an initial resource isolation strategy to generate a resource isolation result; and adjusting the initial resource isolation strategy according to feedback information corresponding to the resource isolation result to generate a final resource isolation strategy, thereby effectively improving the utilization efficiency of resources.
Drawings
In order to more clearly illustrate the embodiments of the invention or the technical solutions of the prior art, the drawings which are used in the description of the embodiments or the prior art will be briefly described, it being obvious that the drawings in the description below are only some embodiments of the invention, and that other drawings can be obtained from these drawings without inventive faculty for a person skilled in the art.
Fig. 1 is a flowchart of steps of a method for isolating dynamic network topology resources according to an embodiment of the present invention;
FIG. 2 is a flowchart illustrating a method for isolating dynamic network topology resources according to another embodiment of the present invention;
fig. 3 is a block diagram of a dynamic network topology resource isolation device according to an embodiment of the present invention.
Detailed Description
The embodiment of the invention provides a method and a device for isolating dynamic network topology resources, which are used for solving the technical problems that the traditional method for isolating the resources generally divides the network resources into fixed parts, each fixed part is allocated with certain resources, meanwhile, the change of the network topology cannot be timely perceived and adapted, and the resource can not be re-planned and allocated according to new topology information, so that the utilization efficiency of the resources is reduced.
In the embodiment of the invention, the scheme can be applied to the scenes of network performance optimization, security monitoring or automatic network management and the like. The network performance optimization is closely related to the cloud network optimization, and the network performance optimization and the cloud network optimization complement each other to improve network efficiency, stability and user experience. Cloud network optimization includes the ability to dynamically allocate resources to accommodate demands, which is closely related to network performance optimization. Through cloud services, the network can more flexibly allocate and adjust resources to meet constantly changing traffic and loads. Cloud network optimization requires a balance between security and performance to be considered. The optimized network not only needs to guarantee high performance, but also needs to implement effective security measures to protect data and user privacy. The security monitoring is closely related to cloud network optimization, and the cooperative action between the security monitoring and the cloud network optimization is beneficial to establishing a safer and more efficient cloud network environment. The security monitoring includes analysis of network traffic to detect abnormal activity. In combination with cloud network optimization, network traffic management can be optimized, overall performance is improved, and potential network congestion or malicious traffic is reduced. The automatic network management is closely related to cloud network optimization, and the association between the automatic network management and the cloud network optimization can improve network efficiency, reduce management cost and better adapt to dynamic service requirements. Automated network management helps to ensure consistency of network devices and configurations, while cloud network optimization requires that the consistency of the overall network architecture be ensured. Configuration errors can be reduced and network reliability can be improved through version control and automatic configuration management.
In order to make the objects, features and advantages of the present invention more comprehensible, the technical solutions in the embodiments of the present invention are described in detail below with reference to the accompanying drawings, and it is apparent that the embodiments described below are only some embodiments of the present invention, but not all embodiments of the present invention. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
Referring to fig. 1, fig. 1 is a flowchart illustrating a step of a method for isolating dynamic network topology resources according to an embodiment of the present invention.
The invention provides a method for isolating dynamic network topology resources, which comprises the following steps:
step 101, obtaining topology data of a dynamic network according to a preset period;
topology data refers to topology data representing positions of various nodes of a dynamic network and connection information, and can be represented in the form of an adjacency matrix or adjacency table of the dynamic network, and mainly comprises node information and connection information, wherein the node information is used for indicating the positions of the nodes and node identifications, and the connection information is used for indicating connection relations among the nodes and data capacity.
In the embodiment of the application, in order to realize multi-mode resource isolation in the dynamic network, topology data of the dynamic network can be acquired according to a preset period or in real time, so as to obtain a data base for predicting dynamic network variation.
102, after positioning flow hot spots and performing resource allocation according to topology data, determining network change prediction data by performing change prediction on a dynamic network;
traffic hotspots refer to traffic aggregation connections after the topology of the dynamic network cannot find an augmented path. For example, the dynamic network topology data is as follows, the node information includes node 1, node 2, node 3 and node 4, the connection information includes connection 1 from node 1 to node 2, the capacity is 10, connection 2 from node 1 to node 3, the capacity is 5; connection 3 from node 2 to node 3 has a capacity of 10, connection 4 from node 2 to node 4 has a capacity of 15, and connection 5 from node 3 to node 4 has a capacity of 10. And taking the node 1 as a source node and the node 4 as a sink node to calculate the Ford-Fulkerson algorithm. Initially, the traffic in the network is 0, starting from the source node 1, finding an augmented path. One possible augmented path is node 1 to node 2 to node 4, on which the minimum capacity is the capacity 10 of connection 1, thus increasing the traffic on connection 1 by 10. After updating, the traffic of connection 1 is 10, and the search for an augmented path is continued, another possible augmented path being node 1 to node 2 to node 3 to node 4, on which path the minimum capacity is the capacity 10 of connection 3, thus increasing the traffic on connection 3 by 10. After updating, the traffic of connection 3 is 10, and the path is continuously searched for, and no other path is available at present. As no more amplification paths can be found, traffic hotspots in the network are obtained, traffic is concentrated on connection 1 and connection 3. Finally, the flow rate of connection 1 is 10, the flow rate of connection 3 is 10, and the flow rates of the other connections are 0.
After the topology data is acquired, traffic hot spots in the dynamic network can be positioned based on the topology data to determine node connection needing to keep resource allocation preferentially. After the resource allocation of the flow hot spot is completed, the dynamic network is subjected to change prediction through a preset target cyclic neural network, so that the change condition of the dynamic network is determined, and network change prediction data are obtained.
Step 103, carrying out demand evaluation on the network change prediction data to generate an initial resource isolation strategy;
after generating the network variation prediction data, the network variation prediction data can be subjected to demand assessment to assess the demand degree of the data stream of each mode on the resources, so that after the priority of the data stream of each mode is set, an initial resource isolation strategy is generated to generate the resource allocation strategy for the data stream of each mode.
104, carrying out resource isolation by adopting an initial resource isolation strategy to generate a resource isolation result;
after the initial resource isolation strategy is obtained, the initial resource isolation strategy can be adopted to conduct resource isolation on resources required by various modal data amounts, the resource isolation is adjusted and executed by judging time conflict or interference among the resources, and finally, when a time window is over, a resource isolation result is generated according to the execution result of the initial resource isolation strategy.
And 105, adjusting the initial resource isolation strategy according to feedback information corresponding to the resource isolation result to generate a final resource isolation strategy.
And after each time of resource isolation is completed, collecting feedback information of a resource isolation result in real time, and evaluating and adjusting an initial resource isolation strategy by using Q-Learning so as to generate a final resource isolation strategy.
In the embodiment of the application, topology data of a dynamic network are acquired according to a preset period; after positioning flow hot spots according to topology data and performing resource allocation, determining network change prediction data by performing change prediction on a dynamic network; performing demand evaluation on the network change prediction data to generate an initial resource isolation strategy; performing resource isolation by adopting an initial resource isolation strategy to generate a resource isolation result; and adjusting the initial resource isolation strategy according to feedback information corresponding to the resource isolation result to generate a final resource isolation strategy, thereby effectively improving the utilization efficiency of resources.
Referring to fig. 2, fig. 2 is a flowchart illustrating a step of a method for isolating dynamic network topology resources according to an embodiment of the present invention.
The invention provides a method for isolating dynamic network topology resources, which comprises the following steps:
Step 201, obtaining topology data of a dynamic network according to a preset period;
in the embodiment of the present application, the implementation process of step 201 is similar to that of step 101, and will not be described herein.
Step 202, performing dynamic network topology analysis according to topology data, and determining flow hot spots in a dynamic network;
in one example of the invention, the topology data includes node information and network structure information; step 202 comprises the sub-steps of:
selecting a source node and a sink node according to the node information;
retrieving the amplification paths between the source node and the sink node one by one according to the network structure information, and updating the flow information of each amplification path in real time until the amplification paths cannot be retrieved;
and selecting the node connection with the flow information in the augmentation path as the flow hot spot in the dynamic network.
In the embodiment of the application, an enhanced path, namely a path from a source node to a sink node, is searched from the source node by using a Ford-Fulkerson algorithm, and the minimum traffic on the path is recorded. Adding the minimum traffic on the path to the existing traffic and updating the traffic of each connection in the network. And continuously searching the augmented path and updating the traffic until the augmented path cannot be found, thereby obtaining traffic hot spots in the network. There is a dynamic network topology data as follows, the node information includes node 1, node 2, node 3 and node 4, the connection information includes connection 1 from node 1 to node 2, capacity 10, connection 2 from node 1 to node 3, capacity 5, connection 3 from node 2 to node 3, capacity 10, connection 4 from node 2 to node 4, capacity 15, connection 5 from node 3 to node 4, capacity 10. And taking the node 1 as a source node and the node 4 as a sink node to calculate the Ford-Fulkerson algorithm. Initially, the traffic in the network is 0, starting from the source node 1, finding an augmented path. The core of finding an augmented path by using the FordFulkerson algorithm can be expressed as:
FordFulkerson(1,2,3,4)
for each path (1, 2) ∈1do
f(1,2)←0
An augmentation path 1do exists in while Gf
cf (p) =min { cf (1, 2) |12 is edge on 1 }
for each edge(1,2)in 1do
f(1,2)←f(1,2)+cf(p)
The Ford-Fulkerson algorithm (FFA) refers to a method for solving the maximum flow problem, an iterative algorithm, for finding the maximum flow from the source node to the sink in a directed graph, while considering the capacity of each edge.
Step 203, performing resource allocation according to the service level corresponding to the traffic hot spot, and updating topology data;
in the embodiment of the application, the service level protocol requirements related to the traffic hotspot are acquired, for example, the service level protocol requires that the bandwidth of the traffic hotspot is at least 1Gbps, and the delay is less than 50ms. According to this requirement, it can be ensured that the resource allocation meets the service level agreement requirements. Depending on the bandwidth usage of traffic hotspots in the network, for example, the bandwidth usage of a certain traffic hotspot is found to reach 90%. From this, it can be determined that the traffic hot spot demand for bandwidth is 10%. Depending on the delay sensitivity requirements of the traffic hotspot, the traffic hotspot requires a delay of less than 100ms, for example. If the current bandwidth fails to meet this delay requirement, it needs to be allocated more resources.
In one example of the present invention, prior to performing step 204, training of the target recurrent neural network may be performed by:
acquiring historical topology data of a dynamic network; the historical topological data is provided with historical change prediction data;
sequentially selecting outlier data one by one from historical topological data, and respectively calculating average values among a plurality of historical topological data adjacent to each outlier data;
zero mean normalization is respectively carried out on each average value, and a historical data set is constructed;
and training a preset cyclic neural network by adopting a historical data set and historical change prediction data and combining a standard back propagation algorithm and a gradient descent method to obtain the target cyclic neural network.
In this embodiment, according to historical dynamic network topology data, node attributes, edge attributes and network structure attributes are acquired, wherein the node attributes include type, importance and centrality attributes of the nodes, the edge attributes include weight, connectivity and direction attributes of the edges, and the network structure attributes include connection relationships among the nodes, network density and clustering coefficient attributes. And acquiring the time attribute through the time stamp and the time interval information in the historical dynamic network topology data. And constructing a model for predicting network variation by using a cyclic neural network algorithm according to the node attribute, the edge attribute, the network structure attribute and the time attribute as inputs. And training the cyclic neural network model by using the historical dynamic network topology data as a training set through a standard back propagation algorithm and a gradient descent method so as to obtain the target cyclic neural network.
For example, there is a historical dynamic network topology dataset, such as a social network, that contains information about nodes and edges of a social network. It is desirable to construct a model from this data to predict future changes in the social network. First, node attributes may be obtained from historical data. The type of node may be a user, an organization, an article, or the like, which may be encoded into a numeric attribute, 0 representing the user, and 1 representing the organization. The importance of a node can be measured by calculating the centrality of the node, which represents the number of nodes connected to other nodes. Node A has 5 nodes connected and node B has 3 nodes connected, so that the centrality of node A is 5 and the centrality of node B is 3. Centrality may be considered one of the attributes of a node. Next, edge attributes may be obtained. The weight of an edge may represent the degree of strength of a relationship between nodes. If there are many friends or people of interest in common between two users, the weight of the edges between them may be higher. Connectivity attributes may indicate whether an edge is directed or undirected. An edge is directional if it represents a relationship of interest between users, because the interest is unidirectional. The direction attributes may represent the start and end nodes of an edge. If an edge indicates that user A is interested in user B, then the direction of the edge may be directed from node A to node B. The recurrent neural network may process the sequence data and may consider the temporal properties. The time attribute may be represented by a time stamp and time interval information. Each time stamp may be encoded as a fixed length vector with the time interval as part of the input. By taking as input node attributes, edge attributes, network structure attributes, and time attributes, a recurrent neural network model can be trained to predict future network changes. And training a cyclic neural network model according to the historical dynamic network topology data.
And acquiring a training data set according to the given historical dynamic network topology data. From the training dataset, input and output layers of the recurrent neural network model are determined. And initializing the weight and deviation of the cyclic neural network model. And obtaining a prediction result of the model on the input data through forward propagation of the cyclic neural network model. And calculating the error between the model prediction result and the actual label, and determining the value of the loss function. From the values of the loss function, a back-propagation algorithm is used to calculate the gradient of the model parameters relative to the loss function. The weight and bias of the model are updated using a gradient descent method. Judging whether the training reaches a stopping condition, if not, repeating the prediction, and updating the weight and the deviation of the model; otherwise, finishing training to obtain the training-completed cyclic neural network model. The weight matrix and the bias vector of the hidden layer and the output layer are updated according to the gradient and the learning rate. Then, it is determined whether the training has reached a stop condition. A maximum number of training rounds or a convergence threshold may be set and training may be stopped if the number of training rounds reaches a maximum or the change in the loss function is less than the convergence threshold. If the training does not reach the stopping condition, the prediction is repeated, and the weight and the deviation of the model are updated. For example, new input data may be used for forward propagation, loss functions calculated and model parameters updated. Finally, when the training reaches the stopping condition, a trained cyclic neural network model is obtained and can be used for predicting the future network topology.
Acquiring a data set X according to historical dynamic network topology data i (x 1 、x 2 、…、x n ). Obtaining an average of data in the dataset from nearest neighbor data of k near the dataset, which can be expressed as
Wherein x is i-k K pieces of data representing the front of the outlier; x is x i+k Represented as k data after outliers; the 0 mean normalization method can convert the original dataset into a dataset with a mean of 0 and a variance of 1. The conversion formula is:
expressed as normalized weights, ++>Expressed as the average of the data; z is Z i Represented as representative variance; after the data is trained by the neural network to obtain the predicted value, the output data of the neural network is still [0,1 ]]Within a range of (2). In this case, the output data needs to be denormalized to be dimensional data again. The inverse normalized calculation formula is:
wherein x represents output sequence data of the neural network; x is x max And x min Representing the maximum and minimum values of the original sequence data; x' represents the denormalized data value; and obtaining a network change prediction result through model aggregation.
Step 204, performing change prediction on the dynamic network according to the updated topology data through a preset target cyclic neural network, and determining network change prediction data;
Among other things, the target recurrent neural network (Recurrent Neural Network, RNN) generally refers to a temporal recurrent neural network rather than a structural recurrent neural network (Recursive Neural Network), which is primarily used to model sequence data. RNN is called a recurrent neural network in the sense that a sequence's current output is related to the previous output. The specific expression is that the network will memorize the previous information and apply it to the calculation of the current output, i.e. the nodes between the hidden layers are no longer connectionless but connected, and the input of the hidden layer includes not only the output of the input layer but also the output of the hidden layer at the previous moment. In theory, RNNs are able to process sequence data of any length. In practice, however, it is often assumed for complexity reduction that the current state is only related to the previous states.
Step 205, performing demand assessment on the network change prediction data to generate an initial resource isolation strategy;
in one example of the invention, the network change prediction data comprises multiple data streams of different modalities; step 205 may comprise the sub-steps of:
respectively setting priorities of various data streams from high to low according to the quantity of required resources respectively corresponding to the various data streams;
Setting the resource quota ratio of various data flows from high to low according to the priority;
and distributing the available resources of the dynamic network at the current moment to various data flows according to the resource quota proportion to obtain an initial resource isolation strategy.
In the embodiment of the application, the demand level of each mode on resources, such as a data stream and an energy source stream, is evaluated according to the network change prediction result. Assuming that the data stream modality needs to occupy 50Mbps of bandwidth, the energy source stream modality needs to occupy 100W of power. And designing a corresponding resource allocation strategy according to the resource demand evaluation result. In order to meet the requirements of different modes, the priority of the data stream mode can be set to be high, and the priority of the energy source stream mode can be set to be medium. In terms of quota, the bandwidth quota of the data stream modality may be set to 50Mbps, and the power quota of the energy source stream modality is 100W. The allocation mode can be adjusted according to actual conditions, for example, allocation is performed according to priority order. And implementing multi-mode resource isolation setting according to the resource allocation strategy. A bandwidth management policy may be implemented on the network device to prioritize the data flow through QoS mechanisms to ensure that it has sufficient bandwidth. And effectively distributing the energy source flow by using a dynamic priority scheduling algorithm, and adjusting the resource distribution according to the real-time requirement so as to ensure the effective utilization of the energy source. The resource use condition and the network environment change are monitored regularly, statistics of the resource use condition is carried out once per hour, and the network environment is evaluated once per day. And adjusting and optimizing a resource allocation strategy according to the actual situation, adjusting quota according to the actual bandwidth requirement of the data stream, and adjusting priority according to the actual requirement of the energy source stream.
For example, there are three modalities, video, audio and text, based on the network change prediction results.
Assessing the extent of demand for resources by each modality can be quantitatively assessed by assuming that one video stream needs to occupy 10Mbps of bandwidth and 100MB of memory, one audio stream needs to occupy 2Mbps of bandwidth and 10MB of memory, and one text stream needs to occupy 5Mbps of bandwidth and 1MB of memory. According to the resource demand evaluation result, designing a corresponding resource allocation strategy can be performed by determining the priority of the data stream, wherein the priority can be determined according to the importance of the mode, the video mode is set to be the highest priority, the audio is inferior, and the text is the lowest priority. And according to the actual situation and the resource limitation, corresponding bandwidth and storage space quota are allocated for each mode. 50% of available bandwidth and storage space are allocated for video modalities, 30% for audio modalities and 20% for text modalities.
Further, step 205 may further comprise the sub-steps of:
periodically monitoring various data flows and dynamic networks;
when new available resources appear and the allocated available resources do not meet the quantity of the required resources corresponding to the data stream with the highest priority, the new available resources are allocated to the data stream with the highest priority;
If the number of the updated required resources of any data stream is received or the bandwidth of the dynamic network is changed, the jump execution sets the priorities of the various data streams from high to low according to the number of the required resources respectively corresponding to the various data streams.
In this embodiment, by periodically monitoring the usage of various data flows and dynamic networks, a dynamic allocation manner may be adopted, and dynamic adjustment may be performed according to actual requirements and network conditions. When there is a surplus of network bandwidth, the surplus bandwidth may be allocated to the higher priority modality. To implement multi-modal resource isolation settings, bandwidth management policies may be implemented on the network device, prioritizing data flows through QoS mechanisms. Different bandwidth limits and priority flags are set for data streams of different priorities to ensure that higher priority data streams can get better transmission quality. And (3) periodically monitoring the use condition of resources and the change of network environment, and adjusting and optimizing the resource allocation strategy according to the actual condition. If an increase in demand or change in network bandwidth is found for a modality, the resource demand can be re-assessed and the quota and priority settings adjusted accordingly.
Step 206, adopting an initial resource isolation strategy to conduct resource isolation, and generating a resource isolation result;
in one example of the invention, step 206 may comprise the sub-steps of:
analyzing an initial resource isolation strategy, determining resource information of the multi-mode resources, and setting a time window according to the resource information;
dividing a time window by adopting dynamic slicing to obtain a plurality of time slices, and respectively storing multi-mode resources in each time slice by adopting a resource buffering technology;
according to the overlapping degree of execution time between the multi-mode resources in each time segment, adjusting the segment length of the time segment and recording;
and when the time window is finished, generating a resource isolation result by adopting the segment length, the starting time and the ending time of each time segment.
In the embodiment of the application, according to the multi-mode resource isolation setting, the types and the number of the multi-mode resources are determined, and the length of a preset time window is set. And dividing the preset time window into a plurality of time slices by utilizing dynamic slicing. For the multi-mode resources in each time segment, the resources are stored and managed through a resource buffering technology. By determining time conflicts or interference between resources, the resource isolation policy is adjusted, including adjusting the length of the time slices or adjusting the order of the priority of the resources. And after the time window is finished, obtaining a real-time resource isolation result according to the execution result of the resource isolation strategy.
For example, according to the multi-modal resource isolation setting, the kind and number of multi-modal resources can be determined, and the length of the preset time window can be set. According to the multi-mode resource isolation setting, the types and the quantity of the multi-mode resources are determined, and the length of a preset time window is set. And dividing the preset time window into a plurality of time slices by utilizing dynamic slicing. For the multi-mode resources in each time segment, the resources are stored and managed through a resource buffering technology. Adjusting the resource isolation strategy by judging time conflict or interference among the resources, wherein the adjustment of the resource isolation strategy comprises the step of prolonging the time slicing length according to the ratio of the overlapping degree of the execution time to the preset overlapping degree value when the overlapping degree of the execution time among the resources is larger than the preset overlapping degree value; when the overlapping degree of the execution time among the resources is smaller than a preset overlapping degree value, shortening the length of time slicing according to the ratio of the overlapping degree of the execution time to the preset overlapping degree value; and adjusting the sequence of the priority of the resources according to the emergency degree of the resources. And after the time window is finished, obtaining a time segment adjustment record according to the execution result of the resource isolation strategy, wherein the time segment adjustment record comprises the adjustment conditions of the starting time, the ending time and the length of each time segment.
For example, there are three multi-modal resources, audio, video and text, according to the multi-modal resource isolation setting. The length of the preset time window is set to 10 seconds. The preset time window is divided into 5 time slices by dynamic slicing, and the length of each time slice is 2 seconds. For the multi-mode resources in each time segment, the resources are stored and managed through a resource buffering technology.
In the first time segment, there may be 2 seconds of audio, 1 second of video, and 3 seconds of text resources. And adjusting the resource isolation strategy by judging the time conflict or interference between the resources. The preset overlap degree of the execution time is greater than the overlap degree value by 50%. If the overlapping degree of the execution time of the audio and video resources exceeds 50% in a certain time slice, the length of the time slice is prolonged according to the ratio of the overlapping degree of the execution time to a preset value larger than the overlapping degree. If the execution time of the audio and video resources overlaps 60%, the length of the time slicing can be extended to 2.4 seconds. If the overlapping degree of the execution time of the audio and video resources is smaller than the preset value smaller than the overlapping degree by 30% in a certain time segment, the length of the time slicing is shortened according to the ratio of the overlapping degree of the execution time to the preset value smaller than the overlapping degree. If the execution time of the audio and video resources overlaps by 20%, the length of the time slicing can be shortened to 1.34 seconds. In addition, the order of resource priorities may be adjusted according to the urgency of the resources. Audio resources may be prioritized if there is an urgent audio resource and a non-urgent video resource. And after the time window is finished, according to the execution result of the resource isolation strategy, obtaining a time slice adjustment record.
Further, before generating the resource isolation result using the segment length, the start time and the end time of each time segment at the end of the time window, step 206 may further comprise the sub-steps of:
and adjusting the priority of the multi-mode resources according to the emergency degree of each multi-mode resource.
Step 207, according to the feedback information corresponding to the resource isolation result, the initial resource isolation strategy is adjusted, and the final resource isolation strategy is generated.
In one example of the invention, the feedback information includes network delay, packet loss rate, and energy consumption index attributes; step 207 may comprise the sub-steps of:
respectively acquiring a reward function according to network delay, a data packet loss rate and energy consumption index attributes corresponding to the resource isolation result;
and taking the initial resource isolation strategy as an action, and adjusting according to a reward function and a Q-Learning algorithm to generate a final resource isolation strategy.
The Q-Learning algorithm is a value-based reinforcement Learning algorithm, Q is an abbreviation for quality, and Q function Q (state, action) represents quality of performing action at state, i.e., what Q value is available.
Q-tables, also known as Q-tables, are a data structure used in reinforcement learning to store and manage action values. The purpose of the Q-table is to assign a value to each state-action to represent the expected return for taking a particular action in a particular state.
QoS is a quality of service that is used to provide different priorities for different traffic to control delay and jitter and reduce packet loss. QoS is a combination of techniques for controlling bandwidth, delay, jitter, and packet loss in a network.
In this embodiment, the resource isolation policy is defined as an action space. A bonus function is defined for each state and action combination based on the respective attributes in the feedback information.
A Q-table is created for storing the Q value for each state and action combination. And after each resource isolation period is finished, updating the Q value in the Q-table according to the feedback information. The resource isolation policy is adjusted by selecting the action with the highest Q value. And repeatedly executing the resource isolation period and the Q value updating process until the Q value converges or reaches the preset training times, and obtaining the estimated and adjusted resource isolation strategy.
For example, a resource isolation policy is being designed to optimize the performance of the server. A resource isolation policy may be defined as an action space that includes allocation of server resources to different tasks or applications. Based on the various attributes in the feedback information, a bonus function may be defined for each state and action combination. And after each resource isolation period is finished, updating the Q value in the Q-table according to the feedback information.
The following reward function may be defined to evaluate each state and action combination, for states of low latency, actions of increased bandwidth, giving forward rewards according to the extent of network latency improvement.
If the network delay is reduced by 20%, the prize is +2.
Action is to reduce bandwidth, giving negative rewards according to the network delay degradation level.
If the network delay increases by 30%, then the prize is-3.
For a state-in-delay, the action is to increase bandwidth, and a forward prize is given according to the network delay improvement degree.
If the network delay is reduced by 10%, the prize is +1.
Action is to reduce bandwidth, giving negative rewards according to the network delay degradation level.
If the network delay increases by 20%, then the prize is-2.
For a state of high delay, the action is to increase the bandwidth, and a forward direction reward is given according to the improvement degree of the energy consumption index.
If the energy consumption is reduced by 15%, the prize is +15.
The action is to reduce the bandwidth and give a negative prize according to the deterioration degree of the energy consumption index.
If the energy consumption increases by 25%, the prize is-25.
When action a is selected and feedback information in state S is observed, the Q value in the Q-table may be updated using the following equation:
Q n+1 (S,A)＝Q n (S,A)+α*(R+γ*max(Q n (S′,A′)))-Q n (S,A)
Where α is the learning rate, γ is the discount factor, R is the reward, S 'is the next state, and A' is the best action to select in the next state. By selecting the action with the highest Q value, the resource isolation policy can be adjusted.
If Q (S, A1) > Q (S, A2), action A1 will be selected to be performed.
And repeatedly executing the resource isolation period and the Q value updating process until the Q value converges or reaches the preset training times. Eventually, a final resource isolation policy will be obtained.
In the embodiment of the application, topology data of a dynamic network are acquired according to a preset period; after positioning flow hot spots according to topology data and performing resource allocation, determining network change prediction data by performing change prediction on a dynamic network; performing demand evaluation on the network change prediction data to generate an initial resource isolation strategy; performing resource isolation by adopting an initial resource isolation strategy to generate a resource isolation result; and adjusting the initial resource isolation strategy according to feedback information corresponding to the resource isolation result to generate a final resource isolation strategy, thereby effectively improving the utilization efficiency of resources. And resources can be effectively isolated in dynamic network topology, firstly, the traffic hot spot in the network is identified in real time by utilizing advanced data analysis, and the resource allocation is dynamically adjusted to avoid network bottleneck. By combining with a machine learning algorithm, the technology can predict the upcoming network change and conduct resource isolation in advance, thereby realizing prospective network management.
Referring to fig. 3, fig. 3 is a block diagram illustrating a dynamic network topology resource isolation device according to an embodiment of the present application.
The embodiment of the invention provides an isolation device for dynamic network topology resources, which comprises the following components:
a topology data acquisition module 301, configured to acquire topology data of a dynamic network according to a preset period;
the change prediction module 302 is configured to determine network change prediction data by performing change prediction on the dynamic network after locating the traffic hot spot according to the topology data and performing resource allocation;
the requirement evaluation module 303 is configured to perform requirement evaluation on the network variation prediction data, and generate an initial resource isolation policy;
the resource isolation module 304 is configured to perform resource isolation by adopting an initial resource isolation policy, and generate a resource isolation result;
and the feedback adjustment module 305 is configured to adjust the initial resource isolation policy according to feedback information corresponding to the resource isolation result, and generate a final resource isolation policy.
Optionally, the change prediction module 302 includes:
the traffic hot spot positioning sub-module is used for carrying out dynamic network topology analysis according to the topology data and determining traffic hot spots in the dynamic network;
the topology updating sub-module is used for carrying out resource allocation according to the service level corresponding to the traffic hot spot and updating topology data;
And the network change prediction sub-module is used for predicting the change of the dynamic network according to the updated topology data through a preset target cyclic neural network and determining network change prediction data.
Optionally, the topology data includes node information and network structure information; the traffic hotspot positioning sub-module is specifically configured to:
selecting a source node and a sink node according to the node information;
retrieving the amplification paths between the source node and the sink node one by one according to the network structure information, and updating the flow information of each amplification path in real time until the amplification paths cannot be retrieved;
and selecting the node connection with the flow information in the augmentation path as the flow hot spot in the dynamic network.
Optionally, the apparatus further comprises:
the historical topological data acquisition module is used for acquiring historical topological data of the dynamic network; the historical topological data is provided with historical change prediction data;
the average value calculation module is used for sequentially selecting the outlier data one by one from the historical topological data and respectively calculating the average value among a plurality of historical topological data adjacent to each outlier data;
the normalization module is used for performing zero-mean normalization on each average value respectively and constructing a historical data set;
The training module is used for training a preset cyclic neural network by adopting a historical data set and historical change prediction data and combining a standard back propagation algorithm and a gradient descent method to obtain a target cyclic neural network.
Optionally, the network change prediction data comprises a plurality of data streams of different modalities; the demand assessment module 303 is specifically configured to:
respectively setting priorities of various data streams from high to low according to the quantity of required resources respectively corresponding to the various data streams;
setting the resource quota ratio of various data flows from high to low according to the priority;
and distributing the available resources of the dynamic network at the current moment to various data flows according to the resource quota proportion to obtain an initial resource isolation strategy.
Optionally, the requirement evaluation module 303 is specifically further configured to:
periodically monitoring various data flows and dynamic networks;
when new available resources appear and the allocated available resources do not meet the quantity of the required resources corresponding to the data stream with the highest priority, the new available resources are allocated to the data stream with the highest priority;
if the number of the updated required resources of any data stream is received or the bandwidth of the dynamic network is changed, the jump execution sets the priorities of the various data streams from high to low according to the number of the required resources respectively corresponding to the various data streams.
Optionally, the resource isolation module 304 is specifically configured to:
analyzing an initial resource isolation strategy, determining resource information of the multi-mode resources, and setting a time window according to the resource information;
dividing a time window by adopting dynamic slicing to obtain a plurality of time slices, and respectively storing multi-mode resources in each time slice by adopting a resource buffering technology;
according to the overlapping degree of execution time between the multi-mode resources in each time segment, adjusting the segment length of the time segment and recording;
and when the time window is finished, generating a resource isolation result by adopting the segment length, the starting time and the ending time of each time segment.
Optionally, the resource isolation module 304 is further specifically configured to, at the end of the time window, use the segment length, the start time and the end time of each time segment, and before generating the resource isolation result:
and adjusting the priority of the multi-mode resources according to the emergency degree of each multi-mode resource.
Optionally, the feedback information includes network delay, packet loss rate, and energy consumption index attributes; the feedback adjustment module 305 is specifically configured to:
respectively acquiring a reward function according to network delay, a data packet loss rate and energy consumption index attributes corresponding to the resource isolation result;
And taking the initial resource isolation strategy as an action, and adjusting according to a reward function and a Q-Learning algorithm to generate a final resource isolation strategy.
It will be clearly understood by those skilled in the art that, for convenience and brevity of description, specific working procedures of the apparatus, modules and sub-modules described above may refer to corresponding procedures in the foregoing method embodiments, which are not described herein again.
In the several embodiments provided in this application, it should be understood that the disclosed apparatus and method may be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative, and for example, the division of the modules is merely a logical function division, and there may be additional divisions when actually implemented, for example, multiple modules or components may be combined or integrated into another system, or some features may be omitted or not performed. Alternatively, the coupling or direct coupling or communication connection shown or discussed with each other may be an indirect coupling or communication connection via some interfaces, devices or modules, which may be in electrical, mechanical, or other forms.
The above embodiments are only for illustrating the technical solution of the present invention, and not for limiting the same; although the invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical scheme described in the foregoing embodiments can be modified or some technical features thereof can be replaced by equivalents; such modifications and substitutions do not depart from the spirit and scope of the technical solutions of the embodiments of the present invention.
Claims (10)
1. A method for isolating dynamic network topology resources, comprising:
obtaining topology data of a dynamic network according to a preset period;
after positioning flow hot spots and carrying out resource allocation according to the topology data, determining network change prediction data by carrying out change prediction on the dynamic network;
performing demand evaluation on the network change prediction data to generate an initial resource isolation strategy;
performing resource isolation by adopting the initial resource isolation strategy to generate a resource isolation result;
and adjusting the initial resource isolation strategy according to the feedback information corresponding to the resource isolation result to generate a final resource isolation strategy.
2. The method according to claim 1, wherein determining network change prediction data by performing change prediction on the dynamic network after locating traffic hot spots and performing resource allocation according to the topology data comprises:
performing dynamic network topology analysis according to the topology data, and determining flow hot spots in the dynamic network;
performing resource allocation according to the service level corresponding to the traffic hot spot, and updating the topology data;
and predicting the change of the dynamic network according to the updated topology data through a preset target cyclic neural network, and determining network change prediction data.
3. The method of claim 2, wherein the topology data includes node information and network structure information; the step of performing dynamic network topology analysis according to the topology data to determine traffic hot spots in the dynamic network includes:
selecting a source node and a sink node according to the node information;
retrieving the augmentation paths between the source node and the sink node one by one according to the network structure information, and updating the flow information of each augmentation path in real time until the augmentation paths cannot be retrieved;
and selecting the node connection with the traffic information in the augmentation path as a traffic hot spot in the dynamic network.
4. The method as recited in claim 2, further comprising:
acquiring historical topology data of the dynamic network; the historical topological data is provided with historical change prediction data;
sequentially selecting outlier data one by one from the historical topological data, and respectively calculating an average value among a plurality of historical topological data adjacent to each outlier data;
zero-mean normalization is respectively carried out on each average value, and a historical data set is constructed;
and training a preset cyclic neural network by adopting the historical data set and the historical change prediction data and combining a standard back propagation algorithm and a gradient descent method to obtain a target cyclic neural network.
5. The method of claim 1, wherein the network change prediction data comprises a plurality of data streams of different modalities; the step of carrying out demand evaluation on the network change prediction data to generate an initial resource isolation strategy comprises the following steps:
respectively setting priorities of various data streams from high to low according to the quantity of required resources respectively corresponding to the various data streams;
setting the resource quota ratio of various data flows from high to low according to the priority;
and distributing the available resources of the dynamic network at the current moment to various data flows according to the resource quota proportion to obtain an initial resource isolation strategy.
6. The method as recited in claim 5, further comprising:
periodically monitoring various of said data streams and said dynamic network;
when new available resources appear and the allocated available resources do not meet the quantity of the required resources corresponding to the data stream with the highest priority, the new available resources are allocated to the data stream with the highest priority;
and if any one of the data streams is received to update the required resource quantity or the dynamic network has bandwidth change, executing the jump according to the required resource quantity respectively corresponding to the various data streams, and respectively setting the priorities of the various data streams from high to low.
7. The method of claim 1, wherein performing resource isolation using the initial resource isolation policy generates a resource isolation result, comprising:
analyzing the initial resource isolation strategy, determining resource information of the multi-mode resources, and setting a time window according to the resource information;
dividing the time window by adopting dynamic fragmentation to obtain a plurality of time fragments, and respectively storing the multi-mode resources in each time fragment by adopting a resource buffering technology;
according to the overlapping degree of execution time between the multi-mode resources in each time segment, adjusting the segment length of the time segment and recording;
and when the time window is finished, generating a resource isolation result by adopting the segment length, the starting time and the ending time of each time segment.
8. The method of claim 7, wherein prior to generating a resource isolation result using the segment length, the start time and the end time of each of the time segments when the time window ends, further comprising:
and adjusting the priority of the multi-mode resources according to the emergency degree of each multi-mode resource.
9. The method of claim 1, wherein the feedback information includes network delay, packet loss rate, and energy consumption index attributes; the step of adjusting the initial resource isolation strategy according to the feedback information corresponding to the resource isolation result to generate a final resource isolation strategy comprises the following steps:
respectively acquiring a reward function according to network delay, a data packet loss rate and energy consumption index attributes corresponding to the resource isolation result;
and taking the initial resource isolation strategy as an action, and adjusting according to the reward function and the Q-Learning algorithm to generate a final resource isolation strategy.
10. An isolation device for dynamic network topology resources, comprising:
the topology data acquisition module is used for acquiring topology data of the dynamic network according to a preset period;
the change prediction module is used for determining network change prediction data by performing change prediction on the dynamic network after positioning flow hot spots and performing resource allocation according to the topology data;
the demand evaluation module is used for carrying out demand evaluation on the network change prediction data and generating an initial resource isolation strategy;
The resource isolation module is used for carrying out resource isolation by adopting the initial resource isolation strategy to generate a resource isolation result;
and the feedback adjustment module is used for adjusting the initial resource isolation strategy according to the feedback information corresponding to the resource isolation result to generate a final resource isolation strategy.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311812377.2A CN117749631A (en) | 2023-12-26 | 2023-12-26 | Isolation method and device for dynamic network topology resources |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311812377.2A CN117749631A (en) | 2023-12-26 | 2023-12-26 | Isolation method and device for dynamic network topology resources |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN117749631A true CN117749631A (en) | 2024-03-22 |
Family
ID=90279457
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311812377.2A Pending CN117749631A (en) | 2023-12-26 | 2023-12-26 | Isolation method and device for dynamic network topology resources |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117749631A (en) |
-
2023
- 2023-12-26 CN CN202311812377.2A patent/CN117749631A/en active Pending
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112970228B (en) | Method and system for performance assurance with conflict management when providing network slicing service | |
| US11153229B2 (en) | Autonomic resource partitions for adaptive networks | |
| CN112153700B (en) | Network slice resource management method and equipment | |
| Bunyakitanon et al. | End-to-end performance-based autonomous VNF placement with adopted reinforcement learning | |
| Singh et al. | Machine-learning-based prediction for resource (re) allocation in optical data center networks | |
| CN110381541B (en) | Smart grid slice distribution method and device based on reinforcement learning | |
| Kim et al. | Multi-agent reinforcement learning-based resource management for end-to-end network slicing | |
| CN112822050B (en) | Method and apparatus for deploying network slices | |
| US10833934B2 (en) | Energy management in a network | |
| GB2539993A (en) | Energy management in a network | |
| Rezazadeh et al. | On the specialization of fdrl agents for scalable and distributed 6g ran slicing orchestration | |
| Yang et al. | Automatic guarantee scheme for intent-driven network slicing and reconfiguration | |
| WO2015106795A1 (en) | Methods and systems for selecting resources for data routing | |
| Othman et al. | Efficient admission control and resource allocation mechanisms for public safety communications over 5G network slice | |
| Mason et al. | Using distributed reinforcement learning for resource orchestration in a network slicing scenario | |
| Luo et al. | Online deadline-aware bulk transfer over inter-datacenter WANs | |
| Guan et al. | Slice reconfiguration based on demand prediction with dueling deep reinforcement learning | |
| Cai et al. | SARM: service function chain active reconfiguration mechanism based on load and demand prediction | |
| Eghbali et al. | Bandwidth allocation and pricing for SDN-enabled home networks | |
| Li et al. | DQN-enabled content caching and quantum ant colony-based computation offloading in MEC | |
| US20220232584A1 (en) | Method and apparatus for bandwidth allocation | |
| Abdellatif et al. | Intelligent-slicing: An AI-assisted network slicing framework for 5G-and-beyond networks | |
| Tekiyehband et al. | An efficient dynamic service provisioning mechanism in fog computing environment: A learning automata approach | |
| Bedda et al. | Efficient wireless network slicing in 5G networks: An asynchronous federated learning approach | |
| Kamboj et al. | A policy based framework for quality of service management in software defined networks |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| TA01 | Transfer of patent application right | ||
| TA01 | Transfer of patent application right |
Effective date of registration: 20240326 Address after: Unit 1, Building 1, China Telecom Zhejiang Innovation Park, No. 8 Xiqin Street, Wuchang Street, Yuhang District, Hangzhou City, Zhejiang Province, 311100 Applicant after: Tianyi Shilian Technology Co.,Ltd. Country or region after: China Address before: Room 1423, No. 1256 and 1258, Wanrong Road, Jing'an District, Shanghai 200072 Applicant before: Tianyi Digital Life Technology Co.,Ltd. Country or region before: China |