CN108989122B - Virtual network requests mapping method, device and realization device - Google Patents
Virtual network requests mapping method, device and realization device Download PDFInfo
- Publication number
- CN108989122B CN108989122B CN201810894330.8A CN201810894330A CN108989122B CN 108989122 B CN108989122 B CN 108989122B CN 201810894330 A CN201810894330 A CN 201810894330A CN 108989122 B CN108989122 B CN 108989122B
- Authority
- CN
- China
- Prior art keywords
- matrix
- network
- physical network
- embeded
- physical
- 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
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L41/00—Arrangements for maintenance, administration or management of data switching networks, e.g. of packet switching networks
- H04L41/08—Configuration management of networks or network elements
- H04L41/0893—Assignment of logical groups to network elements
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L41/00—Arrangements for maintenance, administration or management of data switching networks, e.g. of packet switching networks
- H04L41/14—Network analysis or design
- H04L41/142—Network analysis or design using statistical or mathematical methods
Landscapes
- Engineering & Computer Science (AREA)
- Computer Networks & Wireless Communication (AREA)
- Signal Processing (AREA)
- Physics & Mathematics (AREA)
- Algebra (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Analysis (AREA)
- Mathematical Optimization (AREA)
- Mathematical Physics (AREA)
- Probability & Statistics with Applications (AREA)
- Pure & Applied Mathematics (AREA)
- Computer And Data Communications (AREA)
Abstract
The present invention provides a kind of virtual network requests mapping method, device and realization devices;Wherein, this method comprises: receiving the mapping request of virtual network;The mapping request includes one of dummy node, dummy node constraint condition, virtual link and virtual link constraint condition or a variety of;According to mapping request and the physical network distribution model pre-established, physical node and physical link are distributed for virtual network;The physical network distribution model passes through neural network.The present invention reduces the time complexity of virtual network mapping process, improves the allocative efficiency and utilization rate of physical network resource.
Description
Technical field
The present invention relates to virtual networking field, more particularly, to a kind of virtual network requests mapping method, device and
Realization device.
Background technique
Network virtualization technology is considered as constructing the important technology of new generation network architectural framework.Utilize network virtualization skill
Art, Internet Service Provider can create multiple virtual networks on the same bottom physical network, to provide for user more
The customizable of sample services end to end.During virtual network is established, need using virtual network mapping techniques, it will be empty
Quasi- network request is mapped to specific node and link in bottom physical network entity.The virtual network mapping used both at home and abroad at present
Mode be all using didactic algorithm, need to formulate by hand a series of rule and it is assumed that ignore bottom physical network and
The relationship of virtual network requests between the two, causes time complexity higher, efficiency is lower.
Summary of the invention
In view of this, the purpose of the present invention is to provide a kind of virtual network requests mapping method, device and realization device,
To reduce the time complexity of virtual network mapping process, the allocative efficiency and utilization rate of physical network resource are improved.
In a first aspect, the embodiment of the invention provides a kind of virtual network requests mapping methods, comprising: receive virtual network
Mapping request;The mapping request includes dummy node, dummy node constraint condition, virtual link and virtual link constraint condition
One of or it is a variety of;According to mapping request and the physical network distribution model pre-established, physics section is distributed for virtual network
Point and physical link;The physical network distribution model passes through neural network.
With reference to first aspect, the embodiment of the invention provides the first possible embodiments of first aspect, wherein on
It states physical network distribution model to establish in the following manner: establishing the mathematical model of physical network to be allocated;According to preset
Model framework, mathematical model and preset effectiveness indicator, establish the network structure of neural network;Obtain training sample;The training
It include the mapping request of multiple virtual networks in sample;Training sample is input in network structure and is trained, physics is obtained
Network assignment model.
The possible embodiment of with reference to first aspect the first, the embodiment of the invention provides second of first aspect
Possible embodiment, wherein the step of the above-mentioned mathematical model for establishing physical network to be allocated, comprising: by physical network
Nodal information be converted to corresponding attribute matrix, the link information of physical network is converted into corresponding adjacency matrix;It utilizes
Spectral method eliminates the noise of attribute matrix and adjacency matrix, obtains the embeded matrix of attribute matrix and the insertion square of adjacency matrix
Battle array;According to the embeded matrix of attribute matrix and the embeded matrix of adjacency matrix, the embeded matrix of physical network is determined;According to matrix
Perturbation theory solves the embeded matrix of physical network, obtains the mathematical model of physical network;The mathematical model includes quiet
State more new model.
The possible embodiment of with reference to first aspect the first, the embodiment of the invention provides the third of first aspect
Possible embodiment, wherein above-mentioned preset effectiveness indicator include the long-term average yield of operator, income consumption than and it is long
One of phase receptance is a variety of.
Second of possible embodiment with reference to first aspect, the 4th kind the embodiment of the invention provides first aspect can
Can embodiment, wherein it is above-mentioned according to the embeded matrix of attribute matrix and the embeded matrix of adjacency matrix, determine physical network
Embeded matrix the step of, comprising: by following formula obtain when attribute matrix embeded matrix and adjacency matrix insertion square
When the correlation maximum of battle array, the corresponding throwing of embeded matrix of the corresponding projection vector of the embeded matrix of attribute matrix and adjacency matrix
Shadow vector:
Wherein, constraint condition PA (t)′YA (t)′YA (t)PA (t)+PX (t)′YX (t)′YX (t)PX (t)=1.YA (t)、YX (t)Respectively in t
The embeded matrix of moment attribute matrix and the embeded matrix of adjacency matrix, YA (t)′、YX (t)′Respectively YA (t)、 YX (t)Transposition square
Battle array;PA (t)、PX (t)Respectively in the corresponding projection vector of embeded matrix of t moment attribute matrix and the embeded matrix of adjacency matrix
Corresponding projection vector, PA (t)、PX (t)Respectively PA (t)、PX (t)Transposed matrix;It indicates with PA (t)、PX (t)It is asked for variable
Take the maximum value of above-mentioned formula;
Pass through the insertion of the corresponding projection vector of embeded matrix and adjacency matrix that Lagrangian seeks attribute matrix
When the gradient of the corresponding projection vector of matrix is zero, the corresponding projection vector of the embeded matrix of attribute matrix and adjacency matrix it is embedding
Enter the value of the corresponding projection vector of matrix;
Common recognition embeded matrix is obtained by following formula are as follows:
Y(t)=[YA (t),YX (t)]×P(t)
Wherein, P(t)For PA (t)And PX (t)The projection vector of synthesis, P(t)=[PA (t),PX (t)]。
Second aspect, the embodiment of the present invention also provide a kind of virtual network requests mapping device, comprising: request receives mould
Block, for receiving the mapping request of virtual network;The mapping request includes dummy node, dummy node constraint condition, virtual chain
One of road and virtual link constraint condition are a variety of;Physical network distribution module, for building according to mapping request and in advance
Vertical physical network distribution model distributes physical node and physical link for virtual network;The physical network distribution model passes through
Neural network.
In conjunction with second aspect, the embodiment of the invention provides the first possible embodiments of second aspect, wherein on
It states physical network distribution model to establish in the following manner: establishing the mathematical model of physical network to be allocated;According to preset
Model framework, mathematical model and preset effectiveness indicator, establish the network structure of neural network;Obtain training sample;The training
It include the mapping request of multiple virtual networks in sample;Training sample is input in network structure and is trained, physics is obtained
Network assignment model.
In conjunction with the first possible embodiment of second aspect, the embodiment of the invention provides second of second aspect
Possible embodiment, wherein the step of the above-mentioned mathematical model for establishing physical network to be allocated, comprising: by physical network
Nodal information be converted to corresponding attribute matrix, the link information of physical network is converted into corresponding adjacency matrix;It utilizes
Spectral method eliminates the noise of attribute matrix and adjacency matrix, obtains the embeded matrix of attribute matrix and the insertion square of adjacency matrix
Battle array;According to the embeded matrix of attribute matrix and the embeded matrix of adjacency matrix, the embeded matrix of physical network is determined;According to matrix
Perturbation theory solves the embeded matrix of physical network, obtains the mathematical model of physical network;The mathematical model includes quiet
State more new model.
In conjunction with the first possible embodiment of second aspect, the embodiment of the invention provides the third of second aspect
Possible embodiment, wherein above-mentioned preset effectiveness indicator include the long-term average yield of operator, income consumption than and it is long
One of phase receptance is a variety of.
The third aspect, the embodiment of the present invention also provide a kind of virtual network requests Mapping implementation device, including memory and
Processor, wherein memory is held for storing one or more computer instruction, one or more computer instruction by processor
Row, to realize the above method.
The embodiment of the present invention bring it is following the utility model has the advantages that
The embodiment of the invention provides a kind of virtual network requests mapping method, device and realization devices;Receive virtual net
After the mapping request of network, according to the mapping request and the physical network distribution model pre-established, physics is distributed for virtual network
Node and physical link;Which reduces the time complexity of virtual network mapping process, improves point of physical network resource
With efficiency and utilization rate.
Other features and advantages of the present invention will illustrate in the following description, alternatively, Partial Feature and advantage can be with
Deduce from specification or unambiguously determine, or by implementing above-mentioned technology of the invention it can be learnt that.
To enable the above objects, features and advantages of the present invention to be clearer and more comprehensible, better embodiment is cited below particularly, and match
Appended attached drawing is closed, is described in detail below.
Detailed description of the invention
It, below will be to specific in order to illustrate more clearly of the specific embodiment of the invention or technical solution in the prior art
Embodiment or attached drawing needed to be used in the description of the prior art be briefly described, it should be apparent that, it is described below
Attached drawing is some embodiments of the present invention, for those of ordinary skill in the art, before not making the creative labor
It puts, is also possible to obtain other drawings based on these drawings.
Fig. 1 is a kind of flow chart of virtual network requests mapping method provided in an embodiment of the present invention;
Fig. 2 is in a kind of virtual network requests mapping method provided in an embodiment of the present invention, physical network distribution model
The flow chart of method for building up;
Fig. 3 is virtual network requests and physics in a kind of virtual network requests mapping method provided in an embodiment of the present invention
The topological diagram of network;
Fig. 4 is the flow chart of node mapping in a kind of virtual network requests mapping method provided in an embodiment of the present invention;
Fig. 5 is in a kind of virtual network requests mapping method provided in an embodiment of the present invention, physical network distribution model
Process chart;
Fig. 6 is the four kinds provided in an embodiment of the present invention physical network distribution models using algorithms of different to same group of test
Collect the evaluation index comparison diagram tested;
Fig. 7 is a kind of structural schematic diagram of virtual network requests mapping device provided in an embodiment of the present invention;
Fig. 8 is a kind of structural schematic diagram of virtual network requests Mapping implementation device provided in an embodiment of the present invention.
Specific embodiment
In order to make the object, technical scheme and advantages of the embodiment of the invention clearer, below in conjunction with attached drawing to the present invention
Technical solution be clearly and completely described, it is clear that described embodiments are some of the embodiments of the present invention, rather than
Whole embodiments.Based on the embodiments of the present invention, those of ordinary skill in the art are not making creative work premise
Under every other embodiment obtained, shall fall within the protection scope of the present invention.
Currently, virtual networking is fast-developing.During virtual network is established, need to map using virtual network
Technology;The target of virtual network mapping is exactly to realize bottom physical network resource under the premise of meeting resource constraint
Efficiently utilize.The physical node and physical link that virtual network mapping needs to select and virtual network requests match, carrying are rented
Family individual demand.If being able to achieve the virtual network mapping techniques of high efficient and reliable, and the virtual request of dynamic change carries out
Quick response, then the utilization rate of physical resource will be significantly improved, to improve the business revenue ability of Internet Service Provider.
But in practical operation, under the constraint condition of node and link, virtual network mapping is a NP-hard (non-
Deterministic polynomial-hard, nondeterministic polynomial are difficult) problem.
Existing virtual network mapping mode time complexity is higher, to the allocative efficiency and utilization rate of physical network resource
It is low.Based on this, the embodiment of the invention provides a kind of virtual network requests mapping method, device and realization devices, can apply
In the distribution of physical network resource and other field of resource allocation.
To be reflected to a kind of virtual network requests disclosed in the embodiment of the present invention first convenient for understanding the present embodiment
Shooting method describes in detail.
A kind of flow chart of virtual network requests mapping method shown in Figure 1, method includes the following steps:
Step S100 receives the mapping request of virtual network;The mapping request includes dummy node, dummy node constraint item
One of part, virtual link and virtual link constraint condition are a variety of;When establishing virtual network, distribution physical network is proposed
The request of resource;The distribution of the physical network resource need to be carried out according to the node of virtual network, link and corresponding constraint condition,
The constraint condition can be CPU (Central Processing Unit, central processing unit) computing resource and bandwidth resources etc..
Step S102 distributes physics according to mapping request and the physical network distribution model pre-established for virtual network
Node and physical link;The physical network distribution model passes through neural network.
Specifically, above-mentioned physical network distribution model can be established in the following manner:
(1) mathematical model of physical network to be allocated is established;Since physical network resource to be allocated is limited, because
This needs the ability to bear in view of physical network resource to virtual network mapping request, establishes the mathematics of accurate physical network
Model, and be applied in neural network model, it is more advantageous to and efficiently completes virtual network mapping request.
(2) according to preset model framework, mathematical model and preset effectiveness indicator, the network knot of neural network is established
Structure;Specifically, above-mentioned model framework can be able to be the long-term average yield of operator, income consumption for preset effectiveness indicator
Than with one of long-term receptance or a variety of.
(3) training sample is obtained;It include the mapping request of multiple virtual networks in the training sample.
(4) training sample is input in network structure and is trained, obtain physical network distribution model.
The embodiment of the invention provides a kind of virtual network requests mapping methods;After the mapping request for receiving virtual network,
According to the mapping request and the physical network distribution model pre-established, physical node and physical link are distributed for virtual network;
Which reduces the time complexity of virtual network mapping process, improves the allocative efficiency and utilization rate of physical network resource.
The embodiment of the invention also provides in a kind of virtual network requests mapping method, the foundation of physical network distribution model
Method realizes that flow chart is as shown in Fig. 2 on the basis of this method method shown in Fig. 1, comprising the following steps:
The nodal information of physical network is converted to corresponding attribute matrix by step S200, and the link of physical network is believed
Breath is converted to corresponding adjacency matrix;Specifically, the nodal information of physical network can be indicated with an attribute matrix, every a line
A physical node is represented, each column represent a nodal community.Adjacency matrix table can be used in the link information of physical network
Show, A can be respectively adopted(t)And X(t)Indicate the attribute matrix and adjacency matrix of physics bottom-layer network.It is each in following formula
The definition of symbol is as shown in table 1:
Table 1
Step S202 is eliminated the noise of attribute matrix and adjacency matrix using spectral method, obtains the insertion square of attribute matrix
The embeded matrix of battle array and adjacency matrix.
Specifically, the available process of the embeded matrix of adjacency matrix is as follows: regulation DX (t)For X(t)Degree matrix, i.e.,Provide LX (t)For X(t)Laplacian Matrix, i.e. LX (t)=DX (t)-X(t).According to spectral theory,
The n matrix tieed up is mapped to the noise that can be effectively reduced in matrix expression in the embeded matrix of k dimension (k < < n).One universality
Select Yx(t)=[y1,y2,...yn] ' mode be minimize loss functionI.e. in new sky
Between in, distance is closer between node interconnected.Problem has just been degenerated to Generalized-grads Theory at this timeIt is false
IfFor the corresponding eigenvalue 0=λ of the problem1≤ λ2≤ ...≤λnFeature vector, it is easy to verify λ1
=0, and corresponding feature vector is 1.It takesAs adjacency matrix X(t)Embeded matrix YX (t), and
Take their corresponding characteristic value [λ1,λ2..., λk+1] characteristic value as time t.
It is similar with adjacency matrix, attribute matrix A(t)Also available corresponding YA (t), first obtain normalized attribute square
Battle array, then obtains the similarity matrix W of attribute matrix(t).Attribute matrix A is generated in the same way(t)Embedded expression YA (t)。
Step S204 determines the embedding of physical network according to the embeded matrix of attribute matrix and the embeded matrix of adjacency matrix
Enter matrix.
Specifically, which is accomplished by the following way:
(1) correlation of the embeded matrix of the embeded matrix and adjacency matrix of working as attribute matrix is obtained by following formula most
When big, the corresponding projection vector of embeded matrix of the corresponding projection vector of the embeded matrix of attribute matrix and adjacency matrix:
Wherein, constraint condition PA (t)′YA (t)′YA (t)PA (t)+PX (t)′YX (t)′YX (t)PX (t)=1.YA (t)、YX (t)Respectively in t
The embeded matrix of moment attribute matrix and the embeded matrix of adjacency matrix, YA (t)′、YX (t)' it is respectively YA (t)、 YX (t)Transposition square
Battle array;PA (t)、PX (t)Respectively in the corresponding projection vector of embeded matrix of t moment attribute matrix and the embeded matrix of adjacency matrix
Corresponding projection vector, PA (t)、PX (t)Respectively PA (t)、PX (t)Transposed matrix;It indicates with PA (t)、PX (t)It is asked for variable
Take the maximum value of above-mentioned formula.
(2) the corresponding projection vector of the embeded matrix for attribute matrix being sought by Lagrangian and adjacency matrix it is embedding
When the gradient for entering the corresponding projection vector of matrix is zero, the corresponding projection vector of the embeded matrix of attribute matrix and adjacency matrix
The value of the corresponding projection vector of embeded matrix;Specifically, make γ Lagrangian, then passing through Lagrangian pair
PX (t)And PA (t)Seeking gradient is zero, available PX (t)And PA (t)Value.The statement of corresponding Generalized-grads Theory are as follows:
Then we take the preceding L feature vector of above-mentioned Generalized-grads Theory to know together and are embedded in expression, available projection square
Battle array
(3) common recognition embeded matrix is obtained by following formula are as follows:
Y(t)=[YA (t),YX (t)]×P(t) (3)
Wherein, P(t)For PA (t)And PX (t)The projection vector of synthesis, P(t)=[PA (t),PX (t)]。
Step S206 solves the embeded matrix of physical network, obtains physical network according to Matrix Perturbation
Mathematical model;The mathematical model includes static more new model.
In virtual network mapping process, after a virtual request arrives, physical network will change.Therefore
Attribute matrix is the matrix of a high-frequency dynamic change.If passing through feature to each physical network in each time step
The mode of vector sum characteristic value first finds out YX (t)And YA (t), then pass through YX (t)And YA (t)Obtain Y(t), this is unusual elapsed time
, such mode is also unpractical in large scale network.Therefore it proposes in such a way that matrix variables update to YX (t)
And YA (t)Dynamic update is carried out, to reduce time complexity.
A fact of the proposition of the dynamic update algorithm based on virtual network mapping algorithm, i.e., in two continuous times
Step, attribute matrix will not change too much.Because in virtual network mapping, after completing a virtual network requests, Physical Network
Network can be in the resource for subtracting this virtual request needs, and will not usually to account for physical resource very big by one for the resource of this virtual request
Part.The variable quantity of attribute matrix and adjacency matrix in two continuous time step-lengths is represented using Δ A and Δ X, then for
The degree matrix and Laplacian Matrix of the two:
DA (t+1)=DA (t)+ΔDA,LA (t+1)=LA (t)+ΔLA,
DX (t+1)=DX (t)+ΔDX,LX (t+1)=LX (t)+ΔLX. (4)
In static models, the embeded matrix of attribute matrix and adjacency matrix is obtained, is equivalent to solve a generalized character
Problem.Then as embeded matrix, this needs to solve Generalized-grads Theory k feature vector, but is to solve for feature and asks before taking
It is very high to inscribe complexity, so we need a kind of effective mode, obtains the embeded matrix of attribute matrix and adjacency matrix.We
By taking the embeded matrix update method of adjacency matrix as an example, illustrate the specific steps that dynamic updates.From Matrix Perturbation:
For some characteristic value and feature vector to (λi,ai):
How the algorithmic issue that so dynamic updates just is equivalent to by laplace transform matrix Δ L and degree matrix variation square
Battle array Δ D obtains the variation pair of characteristic value and feature vectorThe mode of solution, which is divided into, seeks Δ λiAnd solutionTwo portions
Point.
Δ λ is calculated firsti, (6) formula is unfolded, is obtained:
Such as due to double derivativeDeng being influenced on result small, therefore omitted.And by
InCan be result by (7) abbreviation:
Equation two sides are multiplied by a simultaneouslyi', (8) become:
Because of Laplacian Matrix LX (t)With degree matrix DX (t)It is symmetrical matrix, then:
Therefore equation (9) just becomes:
Therefore the variable quantity of characteristic value are as follows:
It is easy to getTherefore:
Next it calculatesSince in continuous time step-length, the variation of network structure is smooth, it is assumed that feature to
The perturbation of amount is made of preceding k feature vector, i.e.,Wherein αijIt is the weight of j-th of feature vector, it can
Weight is obtained in the following manner:
It willIt is inserted into formula (8), obtains:
Both members are multiplied by simultaneouslyIt is easy to get simultaneously:
Therefore, weight αipIt can find out and:
As p ≠ i, weighted value, which has been found out, to be come, and is described below when seeking p=i, αiiHow to seek:
(17) are unfolded, and cast out double derivative term, are obtained:
It obtains:
Therefore feature vector aiPerturbation are as follows:
Till now, the perturbation pair of characteristic value and eigenmatrix can be found outIn the insertion of physical network
For matrix update process as shown in the pseudocode in algorithm one (Algorithm 1), code is as shown in table 2.
Table 2
The input of algorithm is the characteristic value and spy at initial time (t=1), through traditional solution Generalized-grads Theory
The mode for levying vector, k characteristic value and feature vector pair before obtaining.And the granny rag Lars matrix of each time step and degree square
The perturbation of battle array.Output is in the preceding k characteristic value and feature vector pair of time step T.Wherein the 3rd row and the 4th row call formula
(13) and formula (20) calculates perturbation, then updates characteristic value and feature vector.The above process is shown in a manner of adjacency matrix
How characteristic value and eigenmatrix are updated, then the embeded matrix of attribute matrix also carries out more in the same way
It is new, to obtain the static more new model of physical network.
The above method obtains healthy and strong common recognition matrix using spectral method, this common recognition matrix can effectively indicate bottom
Physical network realizes the static representations of physical network;In addition, also using Matrix Perturbation to capture object in continuous time
The variation of nodes and link is managed, and completes physical network update, the dynamic for realizing physical network updates.
The embodiment of the invention also provides another virtual network requests mapping method, this method methods shown in Fig. 2
On the basis of realize.
Virtual network mapping problems can indicate using virtual network requests and physical network topology as shown in Figure 3,
Wherein (a) and (b) is virtual network requests, is (c) physical network of bottom.Use the model of a non-directed graphIndicate physical network, wherein NSAnd LSThe node set and link set of physical network are respectively indicated,WithRespectively indicate the node of physical network and the attribute set of link.The nodal community of physical network includes the CPU of node
Ability and location information etc., link attribute include the bandwidth and delay of link.Similar, nothing also can be used in virtual network requests
It is indicated to figure.Use a undirected graph modelIndicate virtual network requests, wherein NVAnd LVPoint
Not Biao Shi virtual network requests node set and link set,WithRespectively indicate the node and chain of virtual network requests
The constraint condition on road.In order to which the description for mapping virtual network is succinctly intuitive, the node money of physical resource and virtual network requests
Source mainly considers CPU computing resource, and link circuit resource mainly considers bandwidth resources.
It include seven physical nodes of A to G if (c) in Fig. 3 indicates a physical network, by taking node A and node B as an example,
The computing resource of node A is 30 units, and the computing resource of node B is 40 units, the bandwidth money between node A and node B
Source is 20 units.(b) in (a) and Fig. 3 in Fig. 3 indicates virtual network requests, by taking (a) in Fig. 3 as an example, dummy node
The computing resource that a needs is 5 units, and the computing resource that dummy node b needs is 10 units, dummy node a and virtual section
The virtual link resource needed between point b is 10 units.
Virtual network mapping process is denoted as M:GV(NV,LV)→GS(N', L'), whereinWith
Fig. 3 citing, two nodes of a and b may be mapped on two nodes of A and B, then the link request between ab is mapped to
On physical link between AB;Being also possible to two nodes of a and b may be mapped on two nodes of A and C, then between ab
Link request be mapped between AB on two physical links between BC, being consumed than former mapping mode in this way more
Link circuit resource between BC, with the consumption of more Internet resources completes this virtual network mapping tasks.
When the request of (a) in Fig. 3 is after time t reaches the physical network of (c) in Fig. 3, the money of physical network is occupied
Source duration is denoted as td, entirely occupying duration tdIn time, the physical resource for distributing to the request cannot be asked by other virtual networks
Seek occupancy.Therefore how virtual network mapping algorithm, i.e., make the decision of reasonable distribution virtual network requests, will be to physical resource
Utilization rate have an important influence on.
The main target of virtual network mapping, is exactly mapped to virtual network requests are as much as possible in physical network,
So as to improve the utilization efficiency of physical network resource, increase the income of operator infrastructure.It will be to put down for a long time in the present embodiment
Equal income, income consumption ratio and long-term three indexs of receptance are as the standard for measuring virtual network requests task performance.
According to above-described embodiment, the mathematical modulo of physical network is established according to formula (3), formula (13) and formula (20)
Type, according to the model and combine RDAM (Reinforcement LearningBased Dynamic Attribute Matrix,
Dynamic attribute matrix based on enhancing study indicates) algorithm, physical network distribution model is established, virtual network mapping request is defeated
Enter the network, the probability that virtual network requests node is mapped to each physical network nodes is exported by the model.
RDAM algorithm regards virtual network mapping as node mapping and two stages of link maps.Rank is mapped relative to node
Section, link maps stage are exactly that the node for mapping node mapping phase distributes link according to the method for depth-first traversal.Section
The process of point mapping is as shown in Figure 4, is divided into Problem 1 (problem 1) and Problem 2 (problem 2).What Problem 1 referred to
The problem of being Feature Selection and character representation, Problem 2 refer to the problem of feature dynamic updates.Feature Selection and mark sheet
Show the method using spectrum analysis, obtains the common recognition matrix of a physical network stalwartness;Feature update mode is managed using matrix perturbation
By solving the problems, such as that static update mode time complexity is high.
As shown in figure 4, left side indicates time step t, right side indicates time step t+1.In time step t, physics bottom is obtained first
Layer network (Substrate Network) attribute (attributes) matrix A(t)With adjacency matrix X(t), pass through the side of spectrum analysis
Formula eliminates the noise in two matrixes, obtains the embedded expression Y of attribute matrixA (t)With the embedded expression Y of adjacency matrixX (t).Then
Final common recognition matrix (consensus embedding) Y is obtained by two Embedded matrixes(t).Then in time step t+
1, grey grid indicates change to attributes in figure, by the variation delta A of the attribute matrix and variation delta X of adjacency matrix, exist
Line upgrades (online update), obtains the embedded expression Y of new attribute matrixA (t+1)With the embedded expression Y of adjacency matrixX (t +1), obtain new common recognition matrix Y(t+1);How the problem of above-mentioned node maps as obtains common recognition matrix and by virtual net
New common recognition matrix how is obtained after network mapping;Both of these problems have been solved in the mathematical model establishment process of physical network
Certainly, details are not described herein.
The process chart of the model is as shown in figure 5, the model (also referred to as Policy Network) includes input layer
(input layer), convolutional layer (convolutional layer), softmax (flexible maximum value transfer function) layer and time
Select layer (candidates layer).Wherein, first layer is input layer, and Y is input in model, wherein Yi j(1≤i≤n,1≤
J≤l) it indicates in the hidden layer expression of i-th of node, the value of jth dimension.The second layer is convolutional layer, after convolutional layer, obtains one
The vector of the available physical node of a n dimension indicates.That is:
hi=wYi+b (21)
Wherein, hiIt is the output of i-th of convolutional layer, w is the weight vectors of convolution kernel, and b is biasing.
Third layer is Softmax layers, and the output h of convolutional layer is come into Softmax layers, generates a probability, this probability generation
The table virtual network requests node selects the probability of each physical node.For each physical node, the meter of probability Softmax
Calculation method are as follows:
Softmax is a kind of this several popularization of writing of logic, it can acquire the probability of vector each dimension, each
The range of the probability of a dimension all between (0,1), and the probability of all dimensions and be 1.
4th layer is both candidate nodes layer, and obtaining each physical node by the model has a probability, but actual conditions
In have some physical node and be unsatisfactory for the virtual network requests node, therefore this layer has the filter of a screening node,
Select the physical node of filtered maximum probability.
After establishing model, need to be trained the model using the method for intensified learning.In the supervision of neural network
In study, sample includes the label of sample characteristics and sample, the label that sample characteristics are predicted later by model, with
The distance of the true tag of sample is as loss function, then the process of supervised learning is exactly the continuous mistake for reducing loss function value
Journey.And in intensified learning, sample does not have true label, and intensified learning comparison supervised learning has lacked the label of sample, Duo Liaoyi
A evaluation index function.By sample characteristics by the way that after model, the label predicted chooses the label of prediction, if the mark
Label can bring the value of better evaluation index to whole system, just illustrate that the direction of model prediction is correct, therefore encourage mould
The parameter of type continues to the training of this direction.If the value of bring evaluation index is smaller, even negative value, then just illustrating mould
The direction of type prediction is incorrect, it is therefore desirable to which the parameter training direction of model is adjusted.
Specific to virtual network mapping process, it is assumed that physical network has n node, and the embeded matrix of each node indicates
L dimension obtains each object of virtual network node request selecting by the embeded matrix (n*l) of physical network as incoming model is inputted
Manage the probability of node.The physical network nodes of maximum probability cannot be directly selected at this time, because network model parameter is random first
Beginningization, if the maximum node of select probability, model be have always it is inclined.It therefore will exploration in model and existing mould
A balance is found between the utilization of type.At this point, obtained the probability vector P of each node by the model, then we with
Machine selects i-th of node, constructs the vector of one-hot coding, i.e. the vector has n dimension, and only i-th bit is 1, remaining is complete
It is 0.So loss function can be written as:
Wherein yiAnd piIt is the i-th dimension that randomly selected one-hot vector sum predicts the probability vector P come respectively
Value.Then gradient g just is obtained with the mode of reversed derivation, the training result of the model tends to when this randomly selected node
Can bring the consumption of larger income than when, the direction of model parameter training is to be more likely to make similar decision;And this is random
When the node bring income consumption of selection is relatively low, the direction of model parameter training is not encourage to make similar decision.Cause
This, makes update to gradient:
G:=α rg (24)
Wherein α is learning rate, the speed of Controlling model training.When α is excessive, training process will be unable to restrain, and miss
Globally optimal solution.When α is too small, training process is again too slow, it is therefore desirable to select suitable learning rate.It will reward and gradient phase
Multiply, the decision for obtaining larger reward will generate bigger influence to training agency, therefore can be more likely to generate similar decision;
And the decision for obtaining the either negative reward of lesser reward will generate smaller influence to training agency.
When the model is trained a virtual network requests, due to virtual network requests generally comprise it is multiple virtual
Network request node.After each virtual request node passes through model, gradient is given as security into stacking rather than is applied directly in model, because
Mapping failure is possible to for the virtual network requests.If mapping failure, by the part of virtual network requests node indentation stack
It empties, then handles next virtual network requests.After the number that virtual network requests reach reaches the number of batch processing,
Gradient in stack is applied all into model, then stack is emptied, is counted again.Why decline using batch gradient, first
It is because gradient updating needs to consume a large amount of time, if the decline of batch gradient will save many times.Secondly batch ladder
It spends lower general who has surrendered's gradient to be averaged in batch size, the result trained is more stable.
The code such as algorithm 2 (Algorithm2) institute in table 3 of the training process of primary complete virtual network mapping
Show:
Table 3
7-10 row is the process of node mapping, and 11-13 row is the process of link maps.28th trade mapping is unsuccessful
When, the gradient in stack can be emptied, be trained since next virtual network mapping request.21-23 row, to Batch Size
A gradient is updated, and request counter is reset.
Shown in the pseudocode of test phase such as algorithm 3 (Algorithm 3) in table 4:
Table 4
In test phase, directlys adopt greedy strategy and select physical node to be mapped from maximum probability node.Such as
Fruit node and link all map successfully, then receptance and long-term gain consume ratio to the long-term average yield of final result
There is promotion.
After being trained to physical network distribution model, which can be applied to the processing of virtual network mapping request.
In order to verify the extensive effect of model, we are tested on test set.Check experiment has selected three.First control
Model is using baseline (baseline) algorithm:
Second comparison model is using Node Rank (node sequencing) algorithm.Third comparison model using
Algorithm RLVNE (Reinforcement Learning Algorithm for Virtual equally based on intensified learning
Network Embedding, the enhancing learning algorithm for virtual network insertion).These three algorithms be all by node mapping and
Link maps are divided into the virtual network mapping algorithm in two stages, and link maps are all by the way of depth-first traversal,
Difference is only the algorithm that node mapping phase uses.
In the training process, using long-term gain consumption ratio as evaluation index.In each epoch (temporal one
Point), if long-term average yield is higher than current optimal long-term average yield, optimal long-term average yield is updated for this
The long-term average yield of a epoch, and virtual network mapping is carried out to test set using "current" model.In the training process,
Have many groups of tests to test set.In the case that last obvious group is the result is that model is optimal in training process, to test
The virtual network mapping that collection carries out.In Fig. 6 this algorithm show last group test set as a result, Fig. 6 (a) in show
It is four models to the long-term average yield (long-term average revenue) of same group of test set test result, Fig. 6
(b) display is that four models consume ratio to the income of same group of test set test result in, and display is four in Fig. 6 (c)
Long-term receptance of the model to same group of test set test result;Wherein, RDA-VNE represents the model for using RDAM algorithm, rl
The model for using RLVNE algorithm is represented, NodeRank represents the model for using NodeRank algorithm, and Baseline, which is represented, to be used
The model of baseline algorithm.
In test phase, virtual network request at the beginning of be 22, the end time is 30000, when with 1000 being one
Between unit, Fig. 6 gives within 30 periods, the variation of the evaluation index of virtual network requests.In Fig. 6 (a) and Fig. 6
(b) in, incipient stage long-term average yield and receptance have decline, because with the arrival of virtual network requests, physics money
Source can gradually use up.And incipient stage long-term gain will not significantly decrease with consumption ratio, because of the index and physical resource
Number it is unrelated.Then three indexs all tend to be steady, but can see the convergence of RDAM algorithm in three evaluation indexes
Trend and convergency value will be better than other three kinds of algorithms.By the experiment it may be concluded that in training stage RDN-VNE algorithm
The relationship of physical network nodes has been arrived in middle intensified learning agency study, and can test model generalization in test phase
Superiority is shown on collection.
The present embodiment carries out the physical network distribution model based on neural network by using intensified learning method
Training can effectively find that bottom-layer network indicates the relationship between virtual network requests, to complete effective virtual net
Network mapping.
Corresponding to above-described embodiment, the embodiment of the present invention also provides a kind of virtual network requests mapping device, and structure is shown
Be intended to as shown in fig. 7, comprises: request receiving module 700, for receiving the mapping request of virtual network;The mapping request includes
One of dummy node, dummy node constraint condition, virtual link and virtual link constraint condition are a variety of;Physical network point
With module 702, for distributing physical node for virtual network according to mapping request and the physical network distribution model pre-established
And physical link;The physical network distribution model passes through neural network.
Above-mentioned physical network distribution model can be established in the following manner:
(1) mathematical model of physical network to be allocated is established;Specifically, the nodal information of physical network is converted to pair
The link information of physical network is converted to corresponding adjacency matrix by the attribute matrix answered;Attribute matrix is eliminated using spectral method
And the noise of adjacency matrix, obtain the embeded matrix of attribute matrix and the embeded matrix of adjacency matrix;According to the embedding of attribute matrix
The embeded matrix for entering matrix and adjacency matrix determines the embeded matrix of physical network;According to Matrix Perturbation, to physical network
Embeded matrix solved, obtain the mathematical model of physical network;The mathematical model is the static more new model of physical network.
(2) according to preset model framework, mathematical model and preset effectiveness indicator, the network knot of neural network is established
Structure;The preset effectiveness indicator can be one of the long-term average yield of operator, income consumption ratio and long-term receptance
Or it is a variety of.
(3) training sample is obtained;It include the mapping request of multiple virtual networks in the training sample.
(4) training sample is input in network structure and is trained, obtain physical network distribution model.
Virtual network requests mapping device provided in an embodiment of the present invention, with virtual network requests provided by the above embodiment
Mapping method technical characteristic having the same reaches identical technical effect so also can solve identical technical problem.
Present embodiments provide for a kind of virtual network requests Mapping implementations corresponding with above method embodiment to fill
It sets.Fig. 8 is the structural schematic diagram of the realization device, as shown in fig. 7, the equipment includes processor 1201 and memory 1202;Its
In, memory 1202 is executed by processor for storing one or more computer instruction, one or more computer instruction, with
Realize above-mentioned virtual network requests mapping method.
Realization device shown in Fig. 8 further includes bus 1203 and forwarding chip 1204, processor 1201, forwarding chip 1204
It is connected with memory 1202 by bus 1203.The realization device of the message transmissions can be network edge device.
Wherein, memory 1202 may include high-speed random access memory (RAM, Random Access Memory),
It may also further include non-labile memory (non-volatile memory), for example, at least a magnetic disk storage.Bus
1203 can be isa bus, pci bus or EISA bus etc..The bus can be divided into address bus, data/address bus, control
Bus etc..Only to be indicated with a four-headed arrow in Fig. 8, it is not intended that an only bus or a seed type convenient for indicating
Bus.
Forwarding chip 1204 is used to connect by network interface at least one user terminal and other network units, will seal
The IPv4 message or IPv6 message installed is sent to the user terminal by network interface.
Processor 1201 may be a kind of IC chip, the processing capacity with signal.It is above-mentioned during realization
Each step of method can be completed by the integrated logic circuit of the hardware in processor 1201 or the instruction of software form.On
The processor 1201 stated can be general processor, including central processing unit (Central Processing Unit, abbreviation
CPU), network processing unit (Network Processor, abbreviation NP) etc.;It can also be digital signal processor (Digital
Signal Processing, abbreviation DSP), specific integrated circuit (Application Specific Integrated
Circuit, abbreviation ASIC), ready-made programmable gate array (Field-Programmable Gate Array, abbreviation FPGA) or
Person other programmable logic device, discrete gate or transistor logic, discrete hardware components.It may be implemented or execute sheet
Disclosed each method, step and logic diagram in invention embodiment.General processor can be microprocessor or this at
Reason device is also possible to any conventional processor etc..The step of method in conjunction with disclosed in embodiment of the present invention, can direct body
Now executes completion for hardware decoding processor, or in decoding processor hardware and software module combine and execute completion.It is soft
Part module can be located at random access memory, and flash memory, read-only memory, programmable read only memory or electrically erasable programmable are deposited
In the storage medium of this fields such as reservoir, register maturation.The storage medium is located at memory 1202, and the reading of processor 1201 is deposited
Information in reservoir 1202, in conjunction with its hardware complete aforementioned embodiments method the step of.
Embodiment of the present invention additionally provides a kind of machine readable storage medium, and machine readable storage medium storage is organic
Device executable instruction, for the machine-executable instruction when being called and being executed by processor, machine-executable instruction promotes processor
Realize above-mentioned virtual network requests mapping method, specific implementation can be found in method implementation, and details are not described herein.
Virtual network requests mapping device and realization device provided by embodiment of the present invention, realization principle and generation
Technical effect it is identical with preceding method embodiment, for briefly describe, device embodiments part do not refer to place, can refer to
Corresponding contents in preceding method embodiment.
In several embodiments provided herein, it should be understood that disclosed device and method can also lead to
Other modes are crossed to realize.Device embodiments described above are only schematical, for example, the flow chart in attached drawing and
Block diagram shows the system in the cards of the device of multiple embodiments according to the present invention, method and computer program product
Framework, function and operation.In this regard, each box in flowchart or block diagram can represent a module, program segment or generation
A part of code, a part of the module, section or code include one or more for realizing defined logic function
Executable instruction.It should also be noted that function marked in the box can also be in some implementations as replacement
Occur different from the sequence marked in attached drawing.For example, two continuous boxes can actually be basically executed in parallel, they
Sometimes it can also execute in the opposite order, this depends on the function involved.It is also noted that block diagram and or flow chart
In each box and the box in block diagram and or flow chart combination, can function or movement as defined in executing it is special
Hardware based system is realized, or can be realized using a combination of dedicated hardware and computer instructions.
In addition, each functional module or unit in each embodiment of the present invention can integrate and form one together solely
Vertical part is also possible to modules individualism, can also be integrated to form with two or more modules one it is independent
Part.
It, can be with if the function is realized in the form of SFU software functional unit and when sold or used as an independent product
It is stored in a computer readable storage medium.Based on this understanding, the technical solution of the disclosure is substantially in other words
The part of the part that contributes to existing technology or the technical solution can be embodied in the form of software products, the meter
Calculation machine software product is stored in a storage medium, including some instructions are used so that a computer equipment (can be a
People's computer, server or network equipment etc.) execute all or part of step of each embodiment the method for the disclosure
Suddenly.And storage medium above-mentioned includes: USB flash disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), deposits at random
The various media that can store program code such as access to memory (RAM, Random Access Memory), magnetic or disk.
Finally, it should be noted that embodiment described above, the only specific embodiment of the disclosure, to illustrate this public affairs
The technical solution opened, rather than its limitations, the protection scope of the disclosure are not limited thereto, although referring to aforementioned embodiments pair
The disclosure is described in detail, those skilled in the art should understand that: any technology for being familiar with the art
Personnel can still modify to technical solution documented by aforementioned embodiments in the technical scope that the disclosure discloses
Or variation or equivalent replacement of some of the technical features can be readily occurred in;And these modifications, variation or replacement,
The spirit and scope for disclosure embodiment technical solution that it does not separate the essence of the corresponding technical solution, should all cover in this public affairs
Within the protection scope opened.Therefore, the protection scope of the disclosure shall be subject to the protection scope of the claim.
Claims (6)
1. a kind of virtual network requests mapping method characterized by comprising
Receive the mapping request of virtual network;The mapping request includes dummy node, dummy node constraint condition, virtual link
And one of virtual link constraint condition or a variety of;
According to the mapping request and the physical network distribution model pre-established, for the virtual network distribute physical node and
Physical link;The physical network distribution model passes through neural network;
The physical network distribution model is established in the following manner:
Establish the mathematical model of physical network to be allocated;
According to preset model framework, the mathematical model and preset effectiveness indicator, the network structure of neural network is established;
Obtain training sample;Mapping request, dummy node in the training sample comprising multiple virtual networks constrain item
Part, virtual link and virtual link constraint condition;
The training sample is input in the network structure and is trained, the physical network distribution model is obtained;
The step of mathematical model for establishing physical network to be allocated, comprising:
The nodal information of physical network is converted into corresponding attribute matrix, the link information of physical network is converted to corresponding
Adjacency matrix;
The noise that the attribute matrix and the adjacency matrix are eliminated using spectral method obtains the embeded matrix of the attribute matrix
And the embeded matrix of the adjacency matrix;
According to the embeded matrix of the attribute matrix and the embeded matrix of the adjacency matrix, the insertion of the physical network is determined
Matrix;
According to Matrix Perturbation, the embeded matrix of the physical network is solved, obtains the mathematics of the physical network
Model;The mathematical model includes static more new model.
2. the method according to claim 1, wherein the preset effectiveness indicator includes the long-term flat of operator
Equal income, income consumption ratio and one of receptance or a variety of for a long time.
3. the method according to claim 1, wherein the embeded matrix according to the attribute matrix and described
The embeded matrix of adjacency matrix, the step of determining the embeded matrix of the physical network, comprising:
The correlation when the embeded matrix of the attribute matrix and the embeded matrix of the adjacency matrix is obtained by following formula
When maximum, the corresponding projection of embeded matrix of the corresponding projection vector of the embeded matrix of the attribute matrix and the adjacency matrix
Vector:
Wherein, constraint condition PA (t)′YA (t)′YA (t)PA (t)+PX (t)′YX (t)′YX (t)PX (t)=1;YA (t)、YX (t)Respectively in t moment
The embeded matrix of the embeded matrix of the attribute matrix and the adjacency matrix, YA (t)′、YX (t)′Respectively YA (t)、YX (t)Transposition
Matrix;PA (t)、PX (t)The corresponding projection vector of the embeded matrix of the respectively attribute matrix described in t moment and the adjacency matrix
The corresponding projection vector of embeded matrix, PA (t)′、PX (t)′Respectively PA (t)、PX (t)Transposed matrix;It indicates with PA (t)、PX (t)The maximum value of the formula is sought for variable;
The corresponding projection vector of embeded matrix of seeking the attribute matrix by Lagrangian and the adjacency matrix
When the gradient of the corresponding projection vector of embeded matrix is zero, the corresponding projection vector of the embeded matrix of the attribute matrix and described
The value of the corresponding projection vector of the embeded matrix of adjacency matrix;
Common recognition embeded matrix is obtained by following formula are as follows:
Y(t)=[YA (t),YX (t)]×P(t)
Wherein, P(t)For PA (t)And PX (t)The projection vector of synthesis, P(t)=[PA (t),PX (t)]。
4. a kind of virtual network requests mapping device characterized by comprising
Request receiving module, for receiving the mapping request of virtual network;The mapping request includes dummy node, dummy node
One of constraint condition, virtual link and virtual link constraint condition are a variety of;
Physical network distribution module, for being described according to the mapping request and the physical network distribution model pre-established
Virtual network distributes physical node and physical link;The physical network distribution model passes through neural network;
The physical network distribution model is established in the following manner:
Establish the mathematical model of physical network to be allocated;
According to preset model framework, the mathematical model and preset effectiveness indicator, the network structure of neural network is established;
Obtain training sample;It include the mapping request of multiple virtual networks in the training sample;
The training sample is input in the network structure and is trained, the physical network distribution model is obtained;
The step of mathematical model for establishing physical network to be allocated, comprising:
The nodal information of physical network is converted into corresponding attribute matrix, the link information of physical network is converted to corresponding
Adjacency matrix;
The noise that the attribute matrix and the adjacency matrix are eliminated using spectral method obtains the embeded matrix of the attribute matrix
And the embeded matrix of the adjacency matrix;
According to the embeded matrix of the attribute matrix and the embeded matrix of the adjacency matrix, the insertion of the physical network is determined
Matrix;
According to Matrix Perturbation, the embeded matrix of the physical network is solved, obtains the mathematics of the physical network
Model;The mathematical model includes static more new model.
5. device according to claim 4, which is characterized in that the preset effectiveness indicator includes the long-term flat of operator
Equal income, income consumption ratio and one of receptance or a variety of for a long time.
6. a kind of virtual network requests Mapping implementation device, which is characterized in that including memory and processor, wherein described to deposit
Reservoir is executed for storing one or more computer instruction, one or more computer instruction by the processor, with
Realize the described in any item methods of claims 1 to 3.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810894330.8A CN108989122B (en) | 2018-08-07 | 2018-08-07 | Virtual network requests mapping method, device and realization device |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810894330.8A CN108989122B (en) | 2018-08-07 | 2018-08-07 | Virtual network requests mapping method, device and realization device |
Publications (2)
Publication Number | Publication Date |
---|---|
CN108989122A CN108989122A (en) | 2018-12-11 |
CN108989122B true CN108989122B (en) | 2019-04-16 |
Family
ID=64555183
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810894330.8A Active CN108989122B (en) | 2018-08-07 | 2018-08-07 | Virtual network requests mapping method, device and realization device |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN108989122B (en) |
Families Citing this family (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110365514B (en) * | 2019-05-24 | 2020-10-16 | 北京邮电大学 | SDN multistage virtual network mapping method and device based on reinforcement learning |
CN111106960B (en) * | 2019-12-23 | 2021-03-05 | 北京邮电大学 | Mapping method and mapping device of virtual network and readable storage medium |
CN111200550B (en) * | 2020-01-07 | 2021-11-26 | 中国烟草总公司郑州烟草研究院 | Virtual network mapping method and device |
CN113765683B (en) * | 2020-06-03 | 2022-12-27 | 中国移动通信集团浙江有限公司 | Cutover alarm shielding method and device for network slices |
CN112436992B (en) * | 2020-11-10 | 2022-01-25 | 北京邮电大学 | Virtual network mapping method and device based on graph convolution network |
TR202104311A2 (en) * | 2021-03-05 | 2021-04-21 | Havelsan Hava Elektronik Sanayi Ve Ticaret Anonim Sirketi | METHOD TO SOLVE THE VIRTUAL NETWORK EMBEDDING PROBLEM IN 5G AND BEYOND NETWORKS BY DEEP INFORMATION MAXIMIZATION USING MULTI-PHYSICAL NETWORK |
Family Cites Families (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104702502B (en) * | 2013-12-09 | 2019-11-26 | 中兴通讯股份有限公司 | Network path calculation method and device |
CN105681153B (en) * | 2016-03-16 | 2019-07-05 | 北京邮电大学 | A kind of mapping method of virtual network and device |
CN106209415A (en) * | 2016-06-21 | 2016-12-07 | 北京邮电大学 | A kind of mapping method of virtual network and system |
CN106603293A (en) * | 2016-12-20 | 2017-04-26 | 南京邮电大学 | Network fault diagnosis method based on deep learning in virtual network environment |
CN107566194B (en) * | 2017-10-20 | 2020-07-31 | 重庆邮电大学 | Method for realizing cross-domain virtual network mapping |
CN108111335B (en) * | 2017-12-04 | 2019-07-23 | 华中科技大学 | A kind of method and system of scheduling and link virtual network function |
-
2018
- 2018-08-07 CN CN201810894330.8A patent/CN108989122B/en active Active
Non-Patent Citations (1)
Title |
---|
"SDN-ScaSVNE:可伸缩的SDN生存性虚拟网络映射算法";卢美莲 等;《北京邮电大学学报》;20180630;第41卷(第3期);第76页1.1网络模型 |
Also Published As
Publication number | Publication date |
---|---|
CN108989122A (en) | 2018-12-11 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN108989122B (en) | Virtual network requests mapping method, device and realization device | |
CN107169563B (en) | Processing system and method applied to two-value weight convolutional network | |
Yao et al. | RDAM: A reinforcement learning based dynamic attribute matrix representation for virtual network embedding | |
CN105446988B (en) | The method and apparatus for predicting classification | |
CN109635989A (en) | A kind of social networks link prediction method based on multi-source heterogeneous data fusion | |
CN110473083A (en) | Tree-shaped adventure account recognition methods, device, server and storage medium | |
CN115037628A (en) | Self-adaptive service function chain mapping method based on deep reinforcement learning | |
CN113992259B (en) | Method for constructing time slot resource expansion graph | |
García-Pérez et al. | Precision as a measure of predictability of missing links in real networks | |
CN111626311B (en) | Heterogeneous graph data processing method and device | |
Ulusan et al. | Approximate dynamic programming for network recovery problems with stochastic demand | |
CN115562832B (en) | Multi-resource service function chain scheduling method based on deep reinforcement learning | |
CN111695046A (en) | User portrait inference method and device based on spatio-temporal mobile data representation learning | |
Tang et al. | Recent advance on detecting core-periphery structure: a survey | |
CN111209105A (en) | Capacity expansion processing method, capacity expansion processing device, capacity expansion processing equipment and readable storage medium | |
CN109325655A (en) | Missing characteristic processing method and device in the prediction of crowd's performance feature | |
CN112001786A (en) | Client credit card limit configuration method and device based on knowledge graph | |
Jahani et al. | Green virtual network embedding with supervised self-organizing map | |
CN108173958A (en) | Data-optimized storage method based on ant group algorithm under a kind of cloudy environment | |
CN117155791B (en) | Model deployment method, system, equipment and medium based on cluster topology structure | |
CN109344255A (en) | The fill method and terminal device of label | |
CN110233763B (en) | Virtual network embedding algorithm based on time sequence difference learning | |
WO2023143570A1 (en) | Connection relationship prediction method and related device | |
Yang et al. | Trust-based scheduling strategy for cloud workflow applications | |
US12019712B2 (en) | Enhanced reinforcement learning algorithms using future state prediction scaled reward values |
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 |