CN109831386B - Optimal path selection algorithm based on machine learning under SDN - Google Patents
Optimal path selection algorithm based on machine learning under SDN Download PDFInfo
- Publication number
- CN109831386B CN109831386B CN201910174856.3A CN201910174856A CN109831386B CN 109831386 B CN109831386 B CN 109831386B CN 201910174856 A CN201910174856 A CN 201910174856A CN 109831386 B CN109831386 B CN 109831386B
- Authority
- CN
- China
- Prior art keywords
- data
- network
- link
- data set
- packet loss
- 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.)
- Active
Links
Images
Landscapes
- Data Exchanges In Wide-Area Networks (AREA)
Abstract
An optimal path selection algorithm based on machine learning under an SDN is used for building an SDN platform, simulating a real network environment, collecting discrete real-time network state data, classifying different QoS index requirements according to different services in network transmission, sorting experimental data to obtain a sample data set, screening optimal paths of the sample data set according to consideration standards of the different services for each index by using a heuristic algorithm, marking the optimal paths corresponding to each group of data, and finally training the data set by using the machine learning algorithm to obtain a classifier, so that the purpose of rapid dynamic routing is achieved. The result of the method is basically the same as the optimization result of the heuristic algorithm, and the calculation time of the model is far shorter than that of the heuristic algorithm, so that the necessary condition of quick decision in the actual network operation is met. Compared with the particle swarm algorithm, the CPU operation time required by the calculation of the extreme learning machine algorithm is greatly shortened, and the real network deployment requirement can be completely met.
Description
Technical Field
The invention relates to the problem that a general routing method under a software defined network architecture can generate higher time cost, provides a multi-constraint QoS routing planning method with low time consumption, and particularly relates to an optimal path selection algorithm based on machine learning under an SDN.
Background
According to the 42 th statistical report of the development conditions of the Chinese Internet published by the information center of the Chinese Internet, the method comprises the following steps: by 6 months in 2018, the scale of the netizens in China reaches 8.02 hundred million, and the mobile terminal users reach 7.88 hundred million. With the rapid development of information technology and the continuous emergence of emerging technologies such as cloud computing and big data, network data is explosively increased in scale and variety. Ubiquitous network access and large bandwidth make dynamic management of the network more important. Meanwhile, the current internet service characteristics and service requirements have changed greatly, that is, the original single point-to-point transmission mode gradually evolves into a communication mode supporting multiple service types and high quality requirements, such as multi-service network requirements of Web browsing, e-commerce, video conference and the like. To achieve this, an internet provider or a network maintenance manager has to provide different services according to various different traffic types, for example, to achieve high real-time in a data center network to satisfy voice services, video phones, etc.; the congestion is reduced as much as possible to keep the communication of services such as E-mail, SMS, multimedia short message and the like normal. Software-defined networking (SDN) has become a very attractive solution in recent years and is sought after. Software-defined networks have two important characteristics: one is the separation between the control plane and the data plane, and the other is the network with editability. Therefore, the SDN can improve more efficient configuration, better performance and higher flexibility, and is more adaptable to the development requirements of future networks. The routing problem under the multi-constraint condition of the SDN has been proved to be an NP problem at present, and in the solution of such problem, the precise mathematical algorithm either cannot obtain a credible result, or can only work normally in a small network environment, or when a relatively large network size is considered, the algorithm appears to be ineffective. Therefore, much research work has been focused on research of approximation algorithms to obtain a solution that is as optimal as possible and to replace the optimal solution, and classical heuristic algorithms have been proposed for effectively managing network traffic and balancing network environments, where the research of routing algorithms in multiple SDN environments is included, and although the current heuristic algorithms have significant effects and obtain an approximate optimal solution, the heuristic algorithms have fatal defects, and the heuristic algorithms need to consume much time during operation, and cannot complete selection of dynamic routing within the time required by a real network.
Disclosure of Invention
The invention aims to remarkably reduce the time consumption of an algorithm aiming at meeting different requirements of a plurality of QoS indexes in different services, solve the problem of introducing higher time cost in the existing algorithm, provide an optimal path selection algorithm based on machine learning in an SDN and solve the problem of rapid dynamic routing of different service flows.
In order to achieve the purpose, the invention adopts the following technical scheme:
an optimal path selection algorithm based on machine learning under an SDN comprises the following steps:
firstly, a software defined network platform is built, a real network environment is simulated, a network topology is built, real-time network state data are collected, and a network state data set is formed;
secondly, dividing different services in the network into four categories, namely session services, streaming services, interactive services and background services according to different requirements of the services on time delay, jitter and packet loss rate;
thirdly, preprocessing the data in the network state data set, defining the time delay rate and packet loss rate indexes by adopting an analytic hierarchy process to eliminate the dimensional influence among the original data, and calculating by utilizing an Euclidean distance formula to obtain a sample data set;
fourthly, screening the optimal path of the sample data set X by using a particle swarm algorithm according to the consideration standards of different services on each index, and labeling the optimal path corresponding to each group of data to form a path label data set;
and fifthly, training the original data set and the path label data set by adopting an extreme learning machine algorithm, and classifying the data to be classified by using a trained model after the extreme learning machine algorithm is converged so as to achieve the purpose of rapid routing dynamics.
The invention is further improved in that in the first step, a controller for constructing the software-defined network platform selects floodlight, openanyght, ryu controller or onos controller to construct fat-tree type network topology and barrier-free full-phase network topology to simulate real network environment.
In a further development of the invention, in a first step, discrete data sets are acquired for the real-time network status data according to the same time interval, of which discrete data setsEach individual sample is noted as: skK ∈ 1,2,3.. m, each link in each sample is denoted xijI ∈ (1,2,3.. n), j ∈ (1,2,3.. n), wherein n is the number of terminals, xijRepresenting the end-to-end time delay, jitter and packet loss rate data from the source switch to the destination switch, which are sequentially marked as xi1,xi2,xi3.., the sample data can be represented in matrix form.
The invention is further improved in that in the third step, the data in the network status data set is preprocessed, and each link x in each sample is analyzed by an analytic hierarchy processijThe important index processing obtains corresponding weight according to different business requirements, and then the sample data set X is obtained by calculating according to the weight and utilizing a distance formulaijWherein i ∈ (1,2,3.. n), j ∈ (1,2,3.. m), i.e., Xi＝(xi1,xi2,...xim) N represents the number of samples in the experiment, m represents the number of switches in the network environment, and the sample data set X is expressed as:
a further improvement of the present invention is that the sample data set X is obtained by specifically performing the following process:
step 1: preprocessing raw data:
and (3) standardization: scaling the original data to fall into a specific region of [0,1 ]; removing unit limitation of original data, and converting the unit limitation into a dimensionless pure numerical value;
step 2: obtaining experimental data
And (3) obtaining a sample data set X from the dimensionless data obtained in the step (1) through an Euclidean distance calculation formula.
The invention is further improved in that the specific process of the step 1 is as follows:
(1) bandwidth utilization ηijIs defined as the current link node v in the networkiTo node vjThe ratio of the used link bandwidth to the maximum bandwidth of the link in the system, bandwidth utilization ηijCalculated by the following formula:
here, loadijIndicating the current link node viTo node vjLink bandwidth, load, already in useklIndicating the current link node vkTo node vlThe bandwidth of the link that is already in use,
which represents the maximum bandwidth capacity, i.e. the maximum transmission rate,
represents the minimum bandwidth capacity, i.e. the minimum transmission rate, on the link under the current network state;
(2) delay rate trijIs defined as the current link node v in the networkiTo link node vjThe ratio of the transmission delay to the maximum delay in the link under the current network state, the delay rate trijCalculated by the following formula:
here, the first and second liquid crystal display panels are,
representing the maximum time delay on a link in the network under the current network state;
represents the minimum time delay, td, on the link in the network under the current network stateijRepresenting a link node viTo link node vjTime delay of tdklRepresenting a link node vkTo link node vlTime delay of (2);
(3) packet loss ratio lrijIs defined as the current link node v in the networkiTo node vjWhen transmitting, the ratio of the difference between the packet loss rate on the link and the minimum packet loss rate in the current network environment to the difference between the maximum and minimum packet loss rates; packet loss ratio lrijCalculated by the following formula:
here, the first and second liquid crystal display panels are,
representing the maximum time delay on a link in the network under the current network state;
indicating the minimum delay, loss, on the link in the network at the current network stateijDenotes viTo vjThe packet loss rate of; lossklDenotes vkTo vlThe packet loss rate of;
using the bandwidth utilization η in the above equationijDelay rate trijAnd packet loss ratio lrijAnd replacing the bandwidth, time delay and packet loss rate in the original data, eliminating unit limitation among the original data, and converting the original data into dimensionless data.
The invention has the further improvement that the concrete process of the fourth step is as follows: screening an optimal path by using a heuristic algorithm on the obtained sample data set X, and repeatedly operating the heuristic algorithm to obtain the optimal path meeting the QoS index under each service requirement; due to the diversity of network states, a plurality of optimal paths exist in the same network state, and since all the optimal paths are the optimal paths and only the sequences of the nodes of the passed links are different, the link of the minimum node in all the optimal paths is taken as the optimal path from the same node.
The further improvement of the invention is that the concrete process of the fifth step is as follows: each group of sample data in the path label data set corresponds to an optimal path, then the optimal paths corresponding to all the sample data are labeled to form a single mapping relation between the optimal paths and the optimal paths, a group of labeled data sets are obtained, each data in the data sets has a definite label, and the data sets are divided into a training set and a testing set; then, a supervised learning model is used, an extreme learning machine algorithm is used for training the training set, and each parameter of the model is adjusted to reach a preset accuracy rate by continuously comparing a prediction result with an actual result; and further correcting the parameters of the supervised learning model by using the data of the test set until the model converges, and classifying the data to be classified by using the trained model to achieve the aim of rapid routing dynamics.
Compared with the prior art, the invention has the following beneficial effects:
firstly, an SDN platform is built, a real network environment is simulated, and discrete real-time network state data are collected. Under the network environment, the different requirements of the QoS index are classified according to different services in network transmission. And (5) collating the experimental data according to the classification requirement, and calculating the weight to obtain a sample data set. Secondly, screening the optimal path of the sample data set by using a heuristic algorithm according to the consideration standard of different services on each index, and labeling the optimal path corresponding to each group of data. And finally, training a data set by using a machine learning algorithm to obtain a classifier so as to achieve the purpose of rapid and dynamic routing.
Because a software defined network platform is built, a real network environment is simulated, a network topology is built, and real-time network state data is collected, the data is real and reliable, and the credibility is achieved.
According to the data set obtained by the service type, unit limitation of data is removed in the standardization process, and the data is converted into a dimensionless pure numerical value, so that unit influence between each index is avoided.
And screening the sorted data set by using a heuristic algorithm according to the consideration standards of different services on each index, marking labels on the optimal paths corresponding to each group of data, and training the data set by using a machine learning algorithm to obtain a classifier so as to achieve the purpose of rapid and dynamic routing. At present, the heuristic algorithm has obvious effect of routing planning, obtains approximate optimal solution, has more confidence in the path optimization process, and then trains a data set by using a machine learning algorithm to obtain a classifier so as to achieve the aim of rapid and dynamic routing. The result given by the method of the invention is basically the same as the optimization result of the heuristic algorithm, but the calculation time of the model is far shorter than that of the heuristic algorithm, thereby meeting the necessary condition of quick decision-making in the actual network operation. Compared with a particle swarm algorithm, the operation time of the CPU required by calculation is greatly shortened, and the real network deployment requirement can be completely met.
Furthermore, the service is more targeted because the QoS indexes are classified according to different requirements of different services.
Drawings
FIG. 1 is a detailed flow chart of the present invention for solving an optimal path using machine learning.
Detailed Description
The invention is described in detail below with reference to the figures and examples, but the scope of protection of the invention is not limited to the examples.
Referring to fig. 1, the present invention provides an optimal path selection algorithm based on machine learning under SDN, which includes the following steps:
firstly, a software defined network platform is built, a real network environment is simulated, a network topology is built, real-time network state data are collected, and a network state data set is formed;
the controller is selected from the group consisting of: the system comprises floodlight, openanyight, ryu, onos and the like, wherein a floodlight controller is used for building simulation real network environments such as fat-tree type network topology, barrier-free full-phase network topology and the like and acquiring real-time network state data.
For real-time network state data, discrete data sets are acquired according to the same time interval, and each single sample in the discrete data sets is recorded as: sk(k ∈ 1,2,3.. m), each link in each sample is denoted xijI ∈ (1,2,3.. n), j ∈ (1,2,3.. n), wherein n is the number of terminals, xijExpressing the end-to-end time delay, jitter and packet loss rate from the source switch to the destination switch, and recording the data in sequenceIs xi1,xi2,xi3.., the sample data may each be represented in the form of a matrix, e.g., the first sample s1Expressed as:
secondly, for different services in the network, according to different requirements on time delay, jitter and packet loss rate, the services are divided into four categories, which are respectively: the specific classification method of the conversational service, the streaming service, the interactive service and the background service is shown in the following table 1:
table 1 specific classification method of service
Since different services have different requirements on the transmission capability of the network, that is, the QoS indexes of service sensitivity are different, the services are roughly classified into four categories according to the real-time performance of the services, the congestion condition in the transmission process, the response time during data interaction, and the like.
And thirdly, preprocessing the data in the network state data set, considering different influence degrees of each index, defining indexes such as time delay rate, packet loss ratio and the like by using an analytic hierarchy process to eliminate dimensional influence among the original data, and calculating by using an Euclidean distance formula to obtain a sample data set.
Specifically, because each index has different influence degrees on the source and destination nodes in the data transmission process, the originally acquired data is preprocessed, and each link x in each sample is analyzed by using an analytic hierarchy processijThe important index processing obtains corresponding weight according to different business requirements, and then the sample data set X is obtained by calculating according to the weight and utilizing a distance formulaijWherein i ∈ (1,2,3.. n), j ∈ (1,2,3.. m), i.e., Xi＝(xi1,xi2,...xim) N represents the number of samples in the experiment, m represents the number of switches in the network environment, and the sample data set X is expressed as:
the sample data set X is obtained through the following process:
step 1: preprocessing raw data:
and (3) standardization: the original data is scaled to fall within a specific region of [0,1 ]. Unit limitation of original data is removed, and the original data is converted into a dimensionless pure numerical value, so that different unit or magnitude indexes can be compared and weighted conveniently. The specific process is as follows:
(1) bandwidth utilization ηijIs defined as the current link node v in the networkiTo node vjThe ratio of the used link bandwidth to the maximum bandwidth of the link in the system, bandwidth utilization ηijCalculated by the following formula:
here, loadijIndicating the current link node viTo node vjLink bandwidth, load, already in useklIndicating the current link node vkTo node vlThe bandwidth of the link that is already in use,
which represents the maximum bandwidth capacity, i.e. the maximum transmission rate,
indicating the minimum bandwidth capacity, i.e., minimum transmission rate, on the link under the current network conditions ηijThe expression of (2) is simple, but can intuitively reflect the load of each link.
(2) Delay rate trijIs defined as the current link node v in the networkiTo link node vjThe ratio of the transmission delay to the maximum delay in the link under the current network state, the delay rate trijIs calculated by the following formulaCalculating:
here, the first and second liquid crystal display panels are,
representing the maximum delay on the link in the network at the current network state.
Represents the minimum time delay, td, on the link in the network under the current network stateijRepresenting a link node viTo link node vjTime delay of tdklRepresenting a link node vkTo link node vlTime delay of (2).
(3) Packet loss ratio lrijIs defined as the current link node v in the networkiTo node vjAnd during transmission, the ratio of the difference between the packet loss rate on the link and the minimum packet loss rate in the current network environment to the difference between the maximum packet loss rate and the minimum packet loss rate. Packet loss ratio lrijCalculated by the following formula:
here, the first and second liquid crystal display panels are,
representing the maximum delay on the link in the network at the current network state.
Indicating the minimum delay, loss, on the link in the network at the current network stateijDenotes viTo vjThe packet loss rate of (1). lossklDenotes vkTo vlThe packet loss rate of (1).
Bandwidth utilization η using the formula defined aboveijDelay rate trijAnd packet loss ratio lrijReplaces the bandwidth, time delay, packet loss rate and the like in the original data,the unit limitation between the original data is eliminated, and the original data is converted into dimensionless data, so that the dimensionless data can be compared and weighted conveniently.
Step 2: obtaining experimental data
And (3) obtaining a sample data set X by using the dimensionless data obtained in the step (1) through an Euclidean distance calculation formula.
And fourthly, screening the optimal path of the sample data set X by using a particle swarm algorithm according to the consideration standards of different services on each index, and labeling the optimal path corresponding to each group of data.
And screening the optimal path of the obtained sample data set X by using a heuristic algorithm, and repeatedly operating the heuristic algorithm to obtain the optimal path meeting the QoS index under each service requirement. Due to the diversity of network states, multiple optimal paths exist in the same network state, and the processing method comprises the following steps: since all the optimal paths are the optimal paths, only the nodes of the passed links have different sequences, the link of the minimum node in all the optimal paths is defined as the optimal path after the same node.
And fifthly, training a path label data set by adopting an extreme learning machine algorithm, and classifying the data to be classified by using a trained model after the extreme learning machine algorithm is converged to achieve the purpose of rapid routing dynamics.
Each group of sample data corresponds to an optimal path, and then the optimal paths corresponding to all the sample data are labeled, so that a single mapping relation is formed between the optimal paths. A set of labeled data sets is obtained, each data set having an unambiguous label therein, and the data sets are partitioned into a training set and a test set. Then, a supervised learning model is used for training the data set, and each parameter of the model is adjusted to reach a preset accuracy rate by continuously comparing a prediction result with an actual result; and further correcting the parameters of the model by using the data of the test set until the model converges, and classifying the label-free data (namely the data to be classified) by using the trained model. When the controller receives a new transmission request, the controller can acquire real-time network data, independently calculate an optimized path similar to a heuristic algorithm, and then the routing process is very quick, so that the requirement of real network deployment is met.
After the extreme learning machine algorithm is converged, the heuristic particle swarm optimization can be effectively replaced, and the time-consuming problem in the process of solving the optimal path is weakened.
The method is implemented based on an SDN network architecture environment, an SDN network platform is built, a real network environment is simulated, a network topology is built, discrete network state data are collected, data are preprocessed, influences among different dimensions are removed, a weight is determined by using an analytic hierarchy process thought, influences of various QoS indexes on an optimal path are comprehensively considered, and an experiment sample data set is built. And performing route planning on the whole sample data set by using a heuristic algorithm to obtain a sample route database, wherein the sample data set and the sample route database are in a single mapping relation so as to ensure the uniqueness of the label. And sorting the sample data set and the route data set, equally dividing into k parts to serve as a machine learning data set, then operating a machine learning algorithm on the new data set by using k-fold cross validation, and performing routing decision so as to achieve the capability of calculating and optimizing a route by using a particle swarm algorithm according to the current network state. The result given by the method of the invention is basically the same as the optimization result of the heuristic algorithm, but the calculation time of the model is far shorter than that of the heuristic algorithm, thereby meeting the necessary condition of quick decision-making in the actual network operation. Simple verification is performed on the optimal path selection model based on machine learning under the SDN, and the effect is shown in the following table 2.
TABLE 2 Algorithm runtime comparison
Table 2 shows that, in the same software and hardware environment, the particle swarm algorithm and the extreme learning machine algorithm calculate the average cpu running time spent on the optimal path for the same current network state, and it can be seen from the table that, compared with the particle swarm algorithm, the extreme learning machine algorithm greatly shortens the cpu running time required for calculation, and can completely meet the requirements of real network deployment.
Claims (3)
1. An optimal path selection algorithm based on machine learning under an SDN is characterized by comprising the following steps:
firstly, a software defined network platform is built, a real network environment is simulated, a network topology is built, real-time network state data are collected, and a network state data set is formed;
secondly, dividing different services in the network into four categories, namely session services, streaming services, interactive services and background services according to different requirements of the services on time delay, jitter and packet loss rate;
thirdly, preprocessing the data in the network state data set, defining the time delay rate and packet loss rate indexes by adopting an analytic hierarchy process to eliminate the dimensional influence among the original data, and calculating by utilizing an Euclidean distance formula to obtain a sample data set;
fourthly, screening the optimal path of the sample data set X by using a particle swarm algorithm according to the consideration standards of different services on each index, and labeling the optimal path corresponding to each group of data to form a path label data set; preprocessing data in the network state data set, and adopting an analytic hierarchy process to carry out link analysis on each link x in each sampleijThe important index processing obtains corresponding weight according to different business requirements, and then the sample data set X is obtained by calculating according to the weight and utilizing a distance formulaijWherein i ∈ (1,2,3.. n), j ∈ (1,2,3.. m), i.e., Xi＝(xi1,xi2,...xim) N represents the number of samples in the experiment, m represents the number of switches in the network environment, and the sample data set X is expressed as:
the sample data set X is obtained through the following process:
step 1: preprocessing raw data:
and (3) standardization: scaling the original data to fall into a specific region of [0,1 ]; removing unit limitation of original data, and converting the unit limitation into a dimensionless pure numerical value; the specific process is as follows:
(1) bandwidth utilization ηijIs defined as the current link node v in the networkiTo node vjThe ratio of the used link bandwidth to the maximum bandwidth of the link in the system, bandwidth utilization ηijCalculated by the following formula:
here, loadijIndicating the current link node viTo node vjLink bandwidth, load, already in useklIndicating the current link node vkTo node vlThe bandwidth of the link that is already in use,
which represents the maximum bandwidth capacity, i.e. the maximum transmission rate,
represents the minimum bandwidth capacity, i.e. the minimum transmission rate, on the link under the current network state;
(2) delay rate trijIs defined as the current link node v in the networkiTo link node vjThe ratio of the transmission delay to the maximum delay in the link under the current network state, the delay rate trijCalculated by the following formula:
here, the first and second liquid crystal display panels are,
representing the maximum time delay on a link in the network under the current network state;
represents the minimum time delay, td, on the link in the network under the current network stateijRepresenting a link node viTo link node vjTime delay of tdklRepresenting a link node vkTo link node vlTime delay of (2);
(3) packet loss ratio lrijIs defined as the current link node v in the networkiTo node vjWhen transmitting, the ratio of the difference between the packet loss rate on the link and the minimum packet loss rate in the current network environment to the difference between the maximum and minimum packet loss rates; packet loss ratio lrijCalculated by the following formula:
here, the first and second liquid crystal display panels are,
representing the maximum packet loss rate on a link in the network under the current network state;
represents the minimum packet loss rate, loss, on the link in the network under the current network stateijDenotes viTo vjThe packet loss rate of; lossklDenotes vkTo vlThe packet loss rate of;
using the bandwidth utilization η in the above equationijDelay rate trijAnd packet loss ratio lrijReplacing bandwidth, time delay and packet loss rate in the original data, eliminating unit limitation among the original data, and converting the original data into dimensionless data;
step 2: obtaining experimental data
Obtaining a sample data set X from the dimensionless data obtained in the step 1 through an Euclidean distance calculation formula;
screening an optimal path by using a heuristic algorithm on the obtained sample data set X, and repeatedly operating the heuristic algorithm to obtain the optimal path meeting the QoS index under each service requirement; due to the diversity of network states, a plurality of optimal paths exist in the same network state, and since the optimal paths are all the optimal paths and only the sequences of the nodes of the passed links are different, the link of the minimum node in all the optimal paths is used as the optimal path after the same node;
fifthly, training an original data set and a path label data set by adopting an extreme learning machine algorithm, and classifying data to be classified by using a trained model after the extreme learning machine algorithm is converged so as to achieve the purpose of rapid routing dynamics; the specific process is as follows: each group of sample data in the path label data set corresponds to an optimal path, then the optimal paths corresponding to all the sample data are labeled to form a single mapping relation between the optimal paths and the optimal paths, a group of labeled data sets are obtained, each data in the data sets has a definite label, and the data sets are divided into a training set and a testing set; then, a supervised learning model is used, an extreme learning machine algorithm is used for training the training set, and each parameter of the model is adjusted to reach a preset accuracy rate by continuously comparing a prediction result with an actual result; and further correcting the parameters of the supervised learning model by using the data of the test set until the model converges, and classifying the data to be classified by using the trained model to achieve the aim of rapid routing dynamics.
2. The optimal path selection algorithm based on machine learning in the SDN according to claim 1, wherein in the first step, a controller for building the software-defined network platform selects floodlight, opendataright, ryu controller or onos controller, and builds a fat-tree network topology and an unobstructed full-phase network topology to simulate a real network environment.
3. The optimal path selection algorithm based on machine learning in SDN according to claim 2, wherein in the first step, for real-time network state data, discrete data sets are collected according to the same time interval, and each single sample in the discrete data sets is recorded as: skK ∈ 1,2,3.. m, each link in each sample is denoted xijI ∈ (1,2,3.. n), j ∈ (1,2,3.. n), wherein n is the number of terminals, xijRepresenting the end-to-end time delay, jitter and packet loss rate data from the source switch to the destination switch, which are sequentially marked as xi1,xi2,xi3.., the sample data can be represented in matrix form.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910174856.3A CN109831386B (en) | 2019-03-08 | 2019-03-08 | Optimal path selection algorithm based on machine learning under SDN |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910174856.3A CN109831386B (en) | 2019-03-08 | 2019-03-08 | Optimal path selection algorithm based on machine learning under SDN |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109831386A CN109831386A (en) | 2019-05-31 |
| CN109831386B true CN109831386B (en) | 2020-07-28 |
Family
ID=66865655
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910174856.3A Active CN109831386B (en) | 2019-03-08 | 2019-03-08 | Optimal path selection algorithm based on machine learning under SDN |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109831386B (en) |
Families Citing this family (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110287124B (en) * | 2019-07-03 | 2023-04-25 | 大连海事大学 | Method for automatically marking software error report and carrying out severity identification |
| CN110493068B (en) * | 2019-09-05 | 2023-01-10 | 烽火通信科技股份有限公司 | Network route generation method and system |
| CN110888744B (en) * | 2019-11-29 | 2022-06-14 | 杭州电子科技大学 | Load balancing method based on automatic adjustment and optimization of workload |
| CN112311670B (en) * | 2019-12-04 | 2024-01-26 | 重庆邮电大学 | Software defined network machine learning route optimization method |
| CN111083051B (en) * | 2019-12-20 | 2020-12-25 | 北京邮电大学 | Path planning method and device based on multiple intelligent agents and electronic equipment |
| CN111447114A (en) * | 2020-03-25 | 2020-07-24 | 深信服科技股份有限公司 | Data packet transmission method and device, electronic equipment and storage medium |
| CN112671648A (en) * | 2020-12-22 | 2021-04-16 | 北京浪潮数据技术有限公司 | SDN data transmission method, SDN, device and medium |
| US11368401B1 (en) | 2021-01-20 | 2022-06-21 | Cisco Technology, Inc. | Globally avoiding simultaneous reroutes in a network |
| CN114884879B (en) * | 2022-05-06 | 2023-07-14 | 鹏城实验室 | Software defined network hybrid switching method based on label technology |
| CN114938530B (en) * | 2022-06-10 | 2023-03-21 | 电子科技大学 | Wireless ad hoc network intelligent networking method based on deep reinforcement learning |
| CN116578924A (en) * | 2023-07-12 | 2023-08-11 | 太极计算机股份有限公司 | Network task optimization method and system for machine learning classification |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106953802A (en) * | 2017-03-01 | 2017-07-14 | 浙江工商大学 | A kind of network optimal route selection method based on deep learning |
| CN108259367A (en) * | 2018-01-11 | 2018-07-06 | 重庆邮电大学 | A kind of Flow Policy method for customizing of the service-aware based on software defined network |
| CN108667734A (en) * | 2018-05-18 | 2018-10-16 | 南京邮电大学 | It is a kind of that the through street with LSTM neural networks is learnt by decision making algorithm based on Q |
| CN108923975A (en) * | 2018-07-05 | 2018-11-30 | 中山大学 | A kind of traffic behavior analysis method of Based on Distributed network |
Family Cites Families (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US9577782B2 (en) * | 2014-06-13 | 2017-02-21 | Ciena Corporation | Systems and methods for statistical multiplexing with OTN and DWDM |
| CN106529667B (en) * | 2016-09-23 | 2018-10-30 | 中国石油大学(华东) | Well-log facies recognition and analysis method based on fuzzy deep learning under a kind of big data environment |
| CN108418756B (en) * | 2018-01-23 | 2021-03-02 | 南京邮电大学 | Software defined backhaul network access selection method based on similarity measurement |
-
2019
- 2019-03-08 CN CN201910174856.3A patent/CN109831386B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106953802A (en) * | 2017-03-01 | 2017-07-14 | 浙江工商大学 | A kind of network optimal route selection method based on deep learning |
| CN108259367A (en) * | 2018-01-11 | 2018-07-06 | 重庆邮电大学 | A kind of Flow Policy method for customizing of the service-aware based on software defined network |
| CN108667734A (en) * | 2018-05-18 | 2018-10-16 | 南京邮电大学 | It is a kind of that the through street with LSTM neural networks is learnt by decision making algorithm based on Q |
| CN108923975A (en) * | 2018-07-05 | 2018-11-30 | 中山大学 | A kind of traffic behavior analysis method of Based on Distributed network |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109831386A (en) | 2019-05-31 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109831386B (en) | Optimal path selection algorithm based on machine learning under SDN | |
| Pei et al. | Optimal VNF placement via deep reinforcement learning in SDN/NFV-enabled networks | |
| Wang et al. | Machine learning for networking: Workflow, advances and opportunities | |
| Sevgican et al. | Intelligent network data analytics function in 5G cellular networks using machine learning | |
| Le et al. | Applying big data, machine learning, and SDN/NFV to 5G traffic clustering, forecasting, and management | |
| CN103379002B (en) | The adaptive monitoring of telecommunication network | |
| US11669751B2 (en) | Prediction of network events via rule set representations of machine learning models | |
| CN111371644B (en) | Multi-domain SDN network traffic situation prediction method and system based on GRU | |
| CN107872457B (en) | Method and system for network operation based on network flow prediction | |
| CN110290077B (en) | Industrial SDN resource allocation method based on real-time service configuration | |
| CN111431819A (en) | Network traffic classification method and device based on serialized protocol flow characteristics | |
| CN107704868A (en) | Tenant group clustering method based on Mobile solution usage behavior | |
| CN114827021B (en) | Multimedia service flow acceleration system based on SDN and machine learning | |
| Chen et al. | ALBLP: Adaptive Load‐Balancing Architecture Based on Link‐State Prediction in Software‐Defined Networking | |
| Yu et al. | Accurate fault location using deep belief network for optical fronthaul networks in 5G and beyond | |
| Qiao et al. | Trace-driven optimization on bitrate adaptation for mobile video streaming | |
| CN118055024A (en) | Dynamic self-adaptive network flow management system and method | |
| CN112260953A (en) | Multi-channel data forwarding decision method based on reinforcement learning | |
| Hongvanthong | Novel four-layered software defined 5g architecture for ai-based load balancing and qos provisioning | |
| Tong et al. | Machine learning based root cause analysis for SDN network | |
| CN111478800A (en) | DSE optimization method and device based on K-NN algorithm | |
| Aziz et al. | Content-aware network traffic prediction framework for quality of service-aware dynamic network resource management | |
| Tajiri et al. | Optimizing edge-cloud cooperation for machine learning accuracy considering transmission latency and bandwidth congestion | |
| CN115695280A (en) | Routing method and device based on edge node, electronic equipment and storage medium | |
| Etengu et al. | AI-assisted traffic matrix prediction using GA-enabled deep ensemble learning for hybrid SDN |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |