CN108599994A - A kind of SDN slice building methods based on flow cluster - Google Patents

Abstract

The purpose of the present invention is to construct network slices suitable for each type of traffic transmission by counting and clustering the service quality requirements of all network flows, so as to better utilize network resources and ensure network service quality; in each slice update period T, Its main steps include: collecting the user's SLA (service level agreement) and the physical link resource capacity C and available resources R of the entire network; counting the QoS characteristics of each traffic demand to obtain the OD traffic characteristics; clustering the OD traffic characteristics to obtain typical The type and scale of QoS requirements; assemble the traffic of each type of QoS requirements into physical links to obtain network slices suitable for carrying corresponding QoS traffic, and the SDN controller sends the slice information to the slice maintenance table of the forwarding device; when SDN When the forwarding device receives the customer's data stream, it searches for a suitable slice for forwarding according to the QoS request, and monitors slice performance such as resource utilization and SLA violations within the slice.

Description

A SDN slice construction method based on traffic clustering

technical field

The present invention relates to the technical field of communication, in particular to an SDN slice construction method based on traffic clustering.

Background technique

With the continuous enrichment of Internet business types, the types of service reliability required by network applications are increasing day by day. Guaranteed quality of service (QoS, Quality of Service) requires reliable contract transmission for different network flows, especially end-to-end reliable contract transmission, which is one of the hotspots that continue to be concerned by academia and industry. This requires analyzing the characteristics of each business flow and defining a good QoS mechanism in the network. However, traditional networks provide best-effort services and cannot guarantee reliable transmission of traffic. QoS models represented by IntServ and DiffServ often only consider bandwidth, and resource reservation is blind. Network flows that reserve resources in the early stage are likely to block resource requests for subsequent network flows. In addition, the natural distributed characteristics of traditional networks determine that network resources cannot be allocated globally, and it is difficult to accurately meet the QoS requirements of different services or customer service level contracts (SLA, Service-Level Agreement). In addition, in the traditional network, equipment vendors' agreement support is not uniform, and the policies and interests of operators or countries on network services are inconsistent, which also hinders the promotion of network QoS guarantee.

The emerging network architecture of software-defined network (SDN) provides new ideas and favorable infrastructure conditions for QoS guarantee. It decouples the control plane and the forwarding plane, and can centrally program and control the topology and forwarding resources of the entire network to dynamically optimize traffic and manage resources. In addition, in the 5G planning proposed by the Next Generation Mobile Network Alliance (NGMN), the concept of network slicing (Network Slicing) is introduced, which divides the network into different subnets according to the characteristics of physical bearable resources, and each subnet has an independent logical topology view , so that different services are transmitted using different slices to ensure end-to-end transmission performance. However, the current papers and forum materials on network slicing are still limited to the concept definition discussion of slicing and the underlying technical implementation details, lacking an overall slicing construction plan and scientific analysis. Therefore, it is particularly important to solve this type of problem.

Contents of the invention

Aiming at the deficiencies of the prior art, the present invention provides a method for constructing SDN slices based on traffic clustering, which collects real-time statistics of network flow requirements of users for different types of service quality, dynamically adjusts the settings of network slices, and achieves resource utilization optimization. Sophisticated management of network slicing.

In order to solve the above problems, the present invention provides a method for constructing SDN slices based on traffic clustering, which is characterized in that, within each slice update period T, the following steps are included: Step 1: Register the information of each access network terminal user Network service quality level agreement (SLA), which includes network data transmission characteristics such as access bandwidth, maximum delay, packet loss rate, and jitter rate;

Step 2: Statistics of the physical link resource information of the entire network in the SDN control domain to obtain the physical resource capacity matrix C and the physical remaining available resource matrix R;

Step 3: Collect the real-time traffic demand of the whole network in the SDN control domain, obtain the QoS characteristics of the OD traffic demand of the whole network, and summarize it into an OD traffic set Set;

Step 4: Calculate the hotspot traffic in the OD traffic, cluster the traffic in the set Set, and obtain the typical QoS requirement category and scale under the current network;

Step 5: Using the self-designed resource assembly algorithm, for each type of QoS requirements obtained by clustering, divide the network slice that is most suitable for carrying them in the physical link, and the SDN controller sends the flow table to the slice of the forwarding device Maintain the table, and the slices carrying each type of QoS requirements can be calculated in parallel and independently, without waiting for all calculations to be issued;

Step 6: After the SDN forwarding device receives the customer's data flow, it searches for a network slice suitable for transmission according to its QoS request characteristics and forwards it. At the same time, it monitors the resource utilization rate and SLA violation in the slice. , to decide whether to repeat steps 1 to 6 to adjust the slice division.

A further improvement is: in step 1, in the network within the SDN control domain, the network composed of all forwarding device nodes and user access nodes is represented as a graph G=(V, E), wherein the number of nodes is |V|=N, and the number of links is |E|=M.

Further improvement lies in: in step 2, the QoS performance that can be carried on each link and the resource usage of each link are reported to the SDN controller by the forwarding device of the SDN network, and the physical available resources of the entire network are recorded as the performance matrix C (h):

Among them, C(h) is the QoS performance boundary constraint that the h-th dimension (1≤h≤l) physical link can carry, specifically, it can be bandwidth (upper bound), delay (lower bound), frame length, packet loss Rate and other QoS indicators, and at least bandwidth characteristics are required, and generally there are delay characteristics. c _uv (h) can also be written as c(h)<u,v>, which is the h-th dimension QoS characteristic on the link e<u,v>; at the same time, the physical remaining available resources of the entire network are recorded as R(h), Each element has the same meaning as C(h), and R(h)=C(h) initially.

The further improvement is: the step of collecting the OD traffic demand set Set of the whole network in step 3 is: firstly, the user terminal connected to the network sends a QoS guarantee request message to the adjacent SDN forwarding device, and the forwarding device communicates with the secure SDN controller The southbound link between the forwarding devices is submitted to the SDN controller to register the QoS performance index of the data flow to be initiated; the SDN controller checks whether the identity of the client is legal and whether the QoS request exceeds the SLA contract. An OD network flow from network node u to node v is denoted as

q<u,v>＝[q ₁ ,q ₂ ,…q _l ],u,v∈V (2)

Among them, nodes u and v respectively represent the source node and sink node of this traffic, and q _h is the hth QoS index of this traffic.

In the current slice management period T, the m-th network flow detected by the network controller can be recorded as q _m <u,v>, 1≤m≤FS, and added to the set Set, where FS is the statistics in the current slice management period Number of incoming network streams:

Set＝{q _m (u,v)|1≤m≤FS,u,v∈V} (3)

A further improvement lies in: in step 4, an unlimited and mainstream clustering method can be selected to cluster the OD traffic set Set, and the cluster number K _opt is determined. Optionally, the present invention may use k-means to cluster the vectors in the Set, and at the same time use the GapStatistic method to automatically determine K _opt . In the clustering results, the kth (1≤k≤K _opt ) traffic subset is denoted as Set ^(k) , the number of set elements is FS _k , and its cluster centroid reflects the typical QoS requirements of this type of traffic and attribute upper and lower bounds In addition, each type of traffic is sorted in descending order according to its frequency of occurrence. The frequency can also reflect the scale of traffic, and the calculation method is:

Therefore, the performance of the kth traffic component in the OD traffic demand is:

Write down matrix OD ^(k) (h) according to each QoS dimension h:

in e _h is a column vector in which only the hth element is 1 and the rest are 0.

The further improvement is: in step 5, use the self-designed resource assembly algorithm to find the most suitable network slice for carrying each type of QoS requirements obtained by clustering. The specific method is as follows:

(1) Process each type of traffic sequentially according to the descending order of formula (4). For the clustered k-th traffic OD ^(k) , it is necessary to find a suitable resource allocation scheme, map the traffic to the physical link in graph G, and obtain a pre-allocation scheme suitable for carrying the k-th traffic slice. 2) Readjust. In the present invention, the goal of resource allocation is to ensure that the QoS performance of the slice meets the traffic carrying requirements within the slice, and at the same time meet the goal of the lowest slice deployment cost, that is, the following optimization mathematical model:

(a) Objective function:

Where f(u,v) is the cost per unit bandwidth on the link e<u,v>; It is a continuous decision variable: the bandwidth usage of the i-th flow on the link e<u,v>.

(b) Constraints:

From the perspective of each link that any destination traffic depends on in the slice, the constraints are:

From the perspective of network slicing operation, in order to ensure end-to-end performance, there are also constraints:

in Belonging to Set ^(k) , it is the i-th flow in the k-th type of flow in the OD flow, originating from node u and ending at node v; bw, d, ω and data represent the bandwidth, delay and loss rate of QoS performance respectively and the subscript of the byte length of the data stream; α ^(k) is the resource allocation adjustment factor for the kth type of traffic, which can be set by the network administrator in combination with the load of the slice to which the various types of traffic belong, to perform manual intervention in resource allocation. The default is 1, 0<α ^(k) ≤ 1; V ^(k) is a subset of V, including the nodes involved in the k-th type of OD flow; D ^(k) <u, v> is the average frame length of the kth traffic when it is transmitted through the link e<u, v>. P _D (k), P _BW (k) and P _ω (k) respectively represent the lower bound of the probability that the delay, bandwidth and loss rate in the network slice QoS performance meet the user performance requirements, ideally 1, It is set by the network slice manager and will be stored in the slice maintenance table. When adding more types of QoS constraints to the model, formulas (10)~(12) can be compared to write contract probability constraints for other QoS indicators.

(2) According to the optimization model solved in the above step (1), a preliminary scheme suitable for carrying the kth traffic slice is obtained: by It can be known whether each link carries the kth type of traffic, and thus the topology diagram of these links can be known

However need to check the graph The connectivity of , that is, using the eigenvalues of the Laplacian matrix to calculate the number of connected components:

picture The Laplacian matrix L is:

The eigenvalues of the matrix are L is positive semi-definite, λ _i ≥ 0, λ ₀ = 0, the number of 0 in the eigenvalue of the Laplacian matrix is the number of connected components of the graph

If A(k)=1, then the graph is the slice topology suitable for carrying traffic of the kth class; otherwise, Each connected component of forms a slice, and there are slices

Among them, for the slice The weight on the edge e<u,v> is Represents the bandwidth allocated by the slice on this link.

(3) The SDN controller slices the above 1≤k≤K _opt . The information is delivered to the SDN forwarding device. Specifically, the slice maintenance table TS(v) of forwarding device node v is a set:

The further improvement lies in: in step 6, after the traffic request initiated by each user, the forwarding layer device first determines the clustering type of the traffic, performs routing in the corresponding network slice, and monitors the resource utilization rate and SLA violation in the slice at the same time, which can be flexible Adjust the slice construction for the next slice construction cycle.

The beneficial effects of the present invention are:

1. The QoS performance requirements of all users and the QoS carrying capacity of links are considered, and when constructing slices, the main categories and requirements of QoS requirements of the entire network are first obtained through clustering, and then each type of QoS is separately analyzed in the algorithm Plan appropriate slices for required traffic instead of considering all QoS requirements at one time, so that for each expected slice, the traffic performance it carries is similar, the time complexity of the slice planning algorithm is lower, and the realization possibility is higher;

2. The network slice constructed according to the algorithm can effectively guarantee the end-to-end QoS characteristics in the slice;

3. The scale of the slice can be directly related to the scale of various types of traffic. By monitoring the performance of each slice, the various demands of the network can be known, and the solution speed is faster;

4. Slices carrying each type of QoS requirements can be calculated in parallel. After each slice is planned, the SDN controller can deliver the flow table to the forwarding device. There is no need to wait for all slices to be calculated. The construction wait time is shortened and the practicality of the technology is better.

Description of drawings

FIG. 1 is a schematic diagram of an SDN network architecture applicable to an embodiment of the present invention;

FIG. 2 is an operation flowchart of a network controller in an embodiment of a method for constructing a network slice according to the present invention;

FIG. 3 is an operation flowchart of a network forwarding device in an embodiment of a network slicing construction method according to the present invention;

Fig. 4 is a flow chart of an algorithm for finding a suitable slice for each type of traffic in Step 5 of the present invention;

FIG. 5 is a schematic diagram of functional modules included in a user interface, an SDN controller, and an SDN forwarding device according to an embodiment of the present invention;

Fig. 6 is an example of communication interaction applicable to the embodiment of the present invention.

Detailed ways

In order to deepen the understanding of the present invention, the present invention will be further described below in conjunction with the examples, which are only used to explain the present invention, and do not constitute a limitation to the protection scope of the present invention.

As shown in Figure 1, the network includes an SDN controller and several SDN forwarding devices. The controller has a centralized perspective and can uniformly collect and control the network and device status information of the SDN forwarding device. The forwarding device is responsible for receiving The sent flow table performs data routing and forwarding. Different network end users are connected to SDN forwarding devices, and each forwarding device can maintain communication with the SDN controller through the secure southbound link of SDN, and there is a network link connection between each forwarding device to transmit data flow.

Figure 5 shows the functional modules that the user-side interface, the SDN controller, and the SDN forwarding device should respectively have according to the embodiment of the present invention. The user-side interface should have functions such as registering SLA and registering QoS requirements; the SDN controller should have functions such as topology resource management, routing calculation module, traffic characteristic analysis, slice performance monitoring, slice resource management, and user QoS registration; the SDN forwarding device should have functions such as Traffic characteristic analysis, slice maintenance table and other functions.

When the SDN slice construction method described in the present invention works, it is carried out according to the following steps:

Step 1: Register the network service level agreement (SLA) of each end user accessing the network. This is realized by the network user sending an SLA registration request to the adjacent SDN forwarding device. The forwarding device sends the SLA registration request to the SDN controller through a secure southbound link. The QoS registration function module of the controller checks the user identity and the legality of the request. Register this SLA. The contract includes network data transmission characteristics such as access bandwidth, maximum delay, packet loss rate, and jitter rate, and is only registered once when an update is required;

Step 2: Statize the resource information of physical links in the entire network in the SDN control domain, and obtain the physical resource capacity matrix C and the physical remaining available resource matrix R. The QoS performance boundary and remaining QoS performance that can be carried on each link are reported to the SDN controller by the forwarding device of the SDN network;

Step 3: Collect the real-time traffic demand of the whole network in the SDN control domain, obtain the QoS characteristics of the OD traffic demand of the whole network, and summarize it into an OD traffic set Set. Firstly, the user terminal connected to the network sends a QoS guarantee request message to the adjacent SDN forwarding device, and the forwarding device submits the SDN controller through the secure southbound link, and registers the QoS performance index of the data flow to be initiated; the SDN controller Check whether the client identity is legal and whether the QoS request exceeds the SLA contract;

Step 4: Calculate the hotspot traffic in the OD traffic, cluster the traffic in the set Set, and obtain the typical QoS requirement category and scale under the current network;

Step 5: Using the self-designed resource assembly algorithm, for each type of QoS requirements obtained by clustering, divide the network slice that is most suitable for carrying them in the physical link, and the SDN controller sends the flow table to the slice of the forwarding device maintenance table. Moreover, the slices carrying each type of QoS requirements can be calculated independently in parallel, without waiting for all calculations to be completed before delivery;

Step 6: After receiving the customer's data flow, the SDN forwarding device searches for a suitable network slice for forwarding according to its QoS request characteristics, and monitors resource utilization and SLA violations in the slice at the same time. It can be decided whether to repeat S1-S6 to adjust slice division according to slice resource utilization rate and SLA breach rate.

Wherein the network in the SDN control domain in the step 1, the network composed of all forwarding device nodes and user access nodes is denoted as graph G=(V, E), wherein the number of nodes is |V|=N, and the number of links is | E|=M.

The QoS performance that can be carried on each link in the step 2 and the resource usage of each link are reported to the SDN controller by the forwarding device of the SDN network, and the physical available resources of the whole network are recorded as the performance matrix C( h):

Among them, C ^(h) is the QoS performance boundary constraint that the h-th dimension (1≤h≤l) physical link can carry, specifically, it can be bandwidth (upper bound), delay (lower bound), frame length, packet loss Rate and other QoS indicators, and at least bandwidth characteristics are required, and generally there are delay characteristics. c _uv (h) is the h-th dimension QoS characteristic on the link e<u,v>;

At the same time, the physical remaining available resources of the entire network are recorded as R(h), and the meanings of each element are the same as C(h), and R(h)=C(h) initially.

The step of collecting the OD traffic demand set Set of the whole network in the step 3 is as follows: firstly, the user terminal connected to the network sends a QoS guarantee request message to the adjacent SDN forwarding device, and the forwarding device passes through the secure SDN controller and the forwarding device The southbound link between them is submitted to the SDN controller to register the QoS performance index of the data flow to be initiated; the SDN controller checks whether the client identity is legal and whether the QoS request exceeds the SLA contract. An OD network flow from network node u to node v is denoted as

q<u,v>＝[q ₁ ,q ₂ ,…q _l ],u,v∈V (2)

Among them, nodes u and v respectively represent the source node and sink node of this traffic, and q _h is the hth QoS index of this traffic.

In the current slice management period T, the m-th network flow detected by the network controller can be recorded as q _m <u,v>, 1≤m≤FS, and added to the set Set, where FS is the statistics in the current slice management period Number of incoming network streams:

Set＝{q _m (u,v)|1≤m≤FS,u,v∈V} (3)

In step four, an unrestricted, mainstream clustering method can be selected to cluster the OD traffic. Here, k-means is selected to cluster the vectors in the set, and the number of clusters K _opt is automatically determined by using gap statistics. In the clustering results, the kth (1≤k≤K _opt ) traffic subset is denoted as Set ^(k) , the number of set elements is FS _k , and its cluster centroid reflects the typical QoS requirements of this type of traffic and attribute upper and lower bounds In addition, each type of traffic is sorted in descending order according to its frequency of occurrence. The frequency can also reflect the scale of traffic, and the calculation method is

Then, the performance of the kth traffic component in the OD traffic demand is

Write down matrix OD ^(k) (h) according to each QoS dimension h:

in e _h is a column vector in which only the hth element is 1 and the rest are 0.

The fifth step uses the self-designed resource assembly algorithm to find the most suitable network slice for each type of QoS requirement obtained by clustering. The specific method is as follows:

(1) Process each type of traffic sequentially according to the descending order of formula (4). For the clustered k-th traffic OD ^(k) , it is necessary to find a suitable resource allocation scheme, map the traffic to the physical link in graph G, and obtain a pre-allocation scheme suitable for carrying the k-th traffic slice. 2) Readjust. In the present invention, the goal of resource allocation is to ensure that the QoS performance of the slice meets the traffic carrying requirements within the slice, and at the same time meet the goal of the lowest slice deployment cost, that is, the following optimization mathematical model:

(a) Objective function:

Where f(u,v) is the cost per unit bandwidth on the link e<u,v>, It is a continuous decision variable: the bandwidth usage of the i-th flow on the link e<u,v>.

(b) Constraints:

From the perspective of each link that any destination traffic depends on in the slice, the constraints are:

From the perspective of network slicing operation, in order to ensure end-to-end performance, there are also constraints:

in Belonging to Set ^(k) , it is the i-th flow in the k-th type of flow in the OD flow, originating from node u and ending at node v; bw, d, ω and data represent the bandwidth, delay and loss rate of QoS performance respectively and the subscript of the byte length of the data stream; α ^(k) is the resource allocation adjustment factor for the kth type of traffic, which can be set by the network administrator in combination with the load of the slice to which the various types of traffic belong, to perform manual intervention in resource allocation. The default is 1, 0<α ^(k) ≤ 1; V ^(k) is a subset of V, including the nodes involved in the k-th type of OD flow; D ^(k) <u, v> is the average frame length of the kth traffic when it is transmitted through the link e<u, v>. P _D (k), P _BW (k) and P _ω (k) respectively represent the lower bound of the probability that the delay, bandwidth and loss rate in the network slice QoS performance meet the user performance requirements, ideally 1, It is set by the network slice manager and will be stored in the slice maintenance table. When adding more types of QoS constraints to the model, formulas (10)~(12) can be compared to write contract probability constraints for other QoS indicators.

(2) According to the optimization model solved in the above step (1), a preliminary scheme suitable for carrying the kth traffic slice is obtained: by It can be known whether each link carries the kth type of traffic, and thus the topology diagram of these links can be known

However need to check the graph The connectivity of , that is, using the eigenvalues of the Laplacian matrix to calculate the number of connected components:

picture The Laplacian matrix L is:

The eigenvalues of the matrix are L is positive semi-definite, λ _i ≥ 0, λ ₀ = 0, the number of 0 in the eigenvalue of the Laplacian matrix is the number of connected components of the graph

If A(k)=1, then the graph is the slice topology suitable for carrying traffic of the kth class; otherwise, Each connected component of forms a slice, and there are slices

Among them, for the slice The weight on the edge e<u,v> is Represents the bandwidth allocated by the slice on this link.

(3) The SDN controller slices the above 1≤k≤K _opt . The information is delivered to the SDN forwarding device. Specifically, the slice maintenance table TS(v) of forwarding device node v is a set:

The steps (1) to (3) of the above method are organized into a flow chart, as shown in Figure 2.

In step six, after each user initiates a traffic request, the forwarding layer device first determines the clustering category of the traffic, determines the slice attribution, identifies the slice in the corresponding field in the network flow data packet, and performs routing in the corresponding network slice, Simultaneously monitor resource utilization and SLA violations within slices.

In Step 6, after each user initiates a traffic request, the forwarding layer device first determines the clustering category of the traffic, routes it in the corresponding network slice, and monitors the resource utilization rate and SLA violation in the slice, which can be flexibly adjusted A slice construction of a slice construction cycle.

Wherein, for a specific g-th slice, the resource utilization rate η(g) is the ratio of the used bandwidth of each link in the slice to the total bandwidth;

The SLA default rate φ(g) is the proportion of the number of network flows sampled in the slice that do not meet their requested QoS performance;

In specific applications, slices can be flexibly adjusted for η(g) and φ(g). This embodiment mentions a feasible scheme, but not as the limitation of the present invention:

When φ(g) is greater than a certain threshold, it means that the load of this slice is heavy and the default is high. To find out the traffic category k supported by this slice, α ^(k) in formula (9) should be increased in the next slice construction cycle ;

When φ(g) is less than a certain threshold and η(g) is greater than a certain threshold, it indicates that this slice is working well, find out the traffic category k supported by this slice, and α ^(k) can be maintained;

When φ(g) is less than a certain threshold and η(g) is also less than a certain threshold, it means that this slice is under light load. If other slices are heavily loaded, α ^(k) can be reduced.

Wherein, when the foregoing steps 1 to 6 are performed, the communication process among the network user, the SDN controller, and the SDN forwarding device is shown in FIG. 6 .

Claims (6)

1. a kind of SDN based on flow cluster is sliced building method, which is characterized in that in each slice update cycle T, including Following steps：

Step 1：The network classes of service contract (SLA) of each access network terminal user of registration, the contract contain access band The network data transmissions feature such as width, highest time delay, packet loss, jitter rate；

Step 2：The whole network physical link resource information in SDN control domains is counted, it is surplus with physics to obtain physical resource capacity matrix C Remaining available resources matrix R；

Step 3：The whole network real-time traffic demand in SDN control domains is acquired, the QoS characteristics of the whole network OD flow demands are obtained, Summarize for OD flow set Set；

Step 4：The hot spot flow in OD flows is calculated, the flow in set Set is clustered, is obtained under current network Typical QoS demand classification and scale；

Step 5：Every class QoS demand that cluster obtains is drawn in physical link using the resource assembly algorithm of designed, designed It separates and is most suitable for carrying their network slice, and flow table is issued to the slice Maintenance Table of forwarding unit by SDN controllers, and hold Carrying can independently calculate per the slice of class QoS demand parallel；

Step 6：After SDN forwarding units receive the data flow of client, the network for being suitble to transmission is found according to its QoS request characteristic Slice is forwarded, while monitoring resource utilization and SLA violations of agreement in slice, can be disobeyed according to slice resource utilization, SLA About rate decides whether that repeat step 1 adjusts slice division to step 6.

2. a kind of SDN based on flow cluster as described in claim 1 is sliced building method, it is characterised in that：In step 2 The resource service condition of the QoS performances and each of the links that can be carried in middle each of the links is converged by the forwarding unit of SDN network It reports to SDN controllers.

3. a kind of SDN based on flow cluster as described in claim 1 is sliced building method, it is characterised in that：In step 3 The step of middle acquisition the whole network OD flow demand set Set is first to be sent from the user terminal of access network to SDN forwarding units are closed on QoS ensures request message, and the south orientation link between the SDN controllers and forwarding unit for passing through safety by forwarding unit submits SDN Controller registers the QoS performance indicators for the data flow that will be initiated, and SDN controllers check whether client identity is legal, and Whether QoS request is more than SLA contracts.

4. a kind of SDN based on flow cluster as described in claim 1 is sliced building method, it is characterised in that：In step 4 Clustering method that middle selection does not limit, mainstream clusters OD flow set Set, and determines cluster numbers K_opt, cluster knot In fruit, kth (1≤k≤K_opt) class flow subset is denoted as Set^(k), set element number is FS_k, its cluster barycenter reflects this The typical QoS demand of class flowWith attribute boundFlow of all categories is according to its frequency of occurrence descending Arrangement.

5. a kind of SDN based on flow cluster as described in claim 1 is sliced building method, it is characterised in that：In step 5 The middle resource assembly algorithm using designed, designed, the every class QoS demand obtained to cluster in SDN controllers, calculates most suitable It closes and carries its network slice, and slice information is issued to SDN forwarding units, be stored in slice Maintenance Table.

6. a kind of SDN based on flow cluster as described in claim 1 is sliced building method, it is characterised in that：In step 6 In after each user initiates traffic requests, forwarding equipment first judges the cluster classification of flow, is carried out in corresponding network slice Routing, while resource utilization and SLA violations of agreement in slice are monitored, the slice structure in next slice construction period is adjusted flexibly It makes.