CN105791151A - Dynamic flow control method and device - Google Patents

Abstract

The invention discloses a dynamic flow control method and device. The method comprises the following implementation steps that control layer equipment obtains a clustering criterion and sets classification labels for activity flows according to the clustering criterion and the service-level agreement SLA levels of the activity flows; the control layer equipment adopts a feature search algorithm to select features from a feature set of the activity flows as a feature subset, wherein the correlation between the classification labels and the features meets a scheduled rule; the control layer equipment adopts a correlation-based feature selection CFS algorithm to compute a score of the feature subset, ends the feature search algorithm when the score of the feature subset meets a predetermined condition and takes the feature subset as an optimal feature subset; and the control layer equipment utilizes the optimal feature subset and the current network status information to compute feature vector values in the activity flows and executes a clustering operation to obtain a flow group to which each activity flow belongs according to the feature vector values of the activity flows. According to the dynamic flow control method and device, the overall flow feature under the actual network condition can be accurately described, and the method and the device are suitable for controlling the overall flow of the network.

Description

A kind of dynamic flow control method, and device

Technical field

The present invention relates to communication technical field, particularly to a kind of dynamic flow control method, and device.

Background technology

Software defined network (SoftwareDefinedNetwork, SDN) is the technology of the design proposed in a kind of about about 2008, structure and management network.The core concept of SDN is to forward face to separate with physics network control forwarding logic.Wherein, the entity being responsible for controlling forwarding logic is called controller, and it can provide abstract global network view to top level control program, and network manager can forward the network traffics in face based on this view how Treated Base that quickly makes a policy.Being stripped the switch controlling logic and be only responsible for the instruction forwarding packet according to controller, agreement the most frequently used between controller and switch is called open flows (OpenFlow, OF).

SDN is the research field that current industry is the most popular, has occurred in that the privately owned controller of increase income controller and some producers.The whole network view and centralized management ability based on controller, investment and the operation cost of Virtual network operator can be reduced, improve the programmability of network, operator can for meeting its business objective rapid deployment new application, service and corresponding infrastructure, the upper layer application of SDN such as resource is distributed, and traffic engineering etc. relies on the controller real-time management and control to network flow, is better meeting QoS (QualityofService, QoS), while, network resource utilization is improved.Further, SDN technology makes network application can utilize the whole network management and control ability of controller, perceive the flow of the whole network, and and then the result of real-time network state and traffic characteristic analysis is applied to the multiple application such as traffic engineering, Internet resources distribution, fault recovery.

About the scheme that dynamic flow controls, there is the method detecting net abuse behavior that load is unrelated.The method is: first gather normal discharge log information and the traffic log by net abuse behavior forms traffic log information training set；The characteristic vector composition characteristic vector set of net abuse behavior is extracted from traffic log information training set；Utilize machine learning algorithm that set of eigenvectors is learnt, obtain abuse detection grader；Finally arrange net abuse behavioral value grader, traffic log is carried out on-line checking, detect net abuse behavior.

Wherein, the characteristic vector extracted from traffic log information training set includes two parts: reference characteristic and auxiliary candidate feature.Wherein, reference characteristic includes: Internet protocol (InternetProtocol, IP) the address sum communicated with destination host in certain time interval；The total quantity that in certain time interval, transmission control protocol (TransmissionControlProtocol, the TCP)/user datagram protocol (UserDatagramProtocol, UDP) relevant to destination host connects；The total flow that certain time interval internal object main frame sends and receives.Auxiliary candidate feature includes: the total quantity of the short connection of TCP/UDP relevant to destination host in certain time interval；The flow that certain time interval internal object main frame sends；20 supplemental characteristics such as the flow that certain time interval internal object main frame receives.Its reference characteristic extracted is to detect necessary feature, according to real network situation and needs, it is also possible to select some or all of auxiliary candidate feature to test Detection accuracy.

Above scheme characteristic vector derives from traffic log and lays particular emphasis on net abuse, and uses fixing characteristic vector parameter, it is impossible to the feature of overall flow in accurate description real network situation；Therefore it is not particularly suited for network entirety flow is controlled.

Summary of the invention

Embodiments provide a kind of dynamic flow control method and device, for provide can bulk flow measure feature in accurate description real network situation, and be applicable to network entirety flow is controlled.

The embodiment of the present invention provides a kind of dynamic flow control method on the one hand, including:

Key-course equipment obtains clustering criteria, is that described active flow arranges key words sorting according to the service-level agreement SLA grade of described clustering criteria and active flow；Described key words sorting is the active flow degree of association to described clustering criteria described in labelling；

Described key-course equipment adopts feature searching algorithm to select from the feature set of described active flow and described key words sorting degree of association meets the feature of pre-defined rule, as character subset；Use the Attributions selection CFS algorithm based on dependency to calculate the score of described character subset, when the score of described character subset meets predetermined condition, terminate feature searching algorithm, and using described character subset as optimal feature subset；

Described key-course equipment uses described optimal feature subset and current network state information, calculates the characteristic vector value being in active flow, and performs cluster operation according to the characteristic vector value of active flow, obtains the stream group belonging to each active flow.

In conjunction with implementation on the one hand, in the implementation that the first is possible, before arranging key words sorting for described active flow, described method also includes:

Key-course monitoring of equipment current network state information, described network state information includes the statistical information of network topology, the status information of network topology, the packet information of described active flow and described active flow；If monitoring current network state information to meet predetermined trigger condition, then described key-course equipment obtains clustering criteria, is that described active flow arranges key words sorting according to the service-level agreement SLA grade of described clustering criteria and active flow.

In conjunction with the first possible implementation on the one hand, in the implementation that the second is possible, described in monitor current network state information and meet predetermined trigger condition and include:

Current network state meets the regular expression based on threshold value required with the parameter of described network state at least one network performance index that benchmark sets.

In conjunction with the first possible implementation on the one hand, in the implementation that the third is possible, described network topology, network topology status information include: at least one in the division of node adjacency relation, network, link capacity, port type, port statistics info, link and port status；

The packet information of described active flow and the statistical information of described active flow include: data package size, packet interval time, Internet protocol IP five-tuple, packet loss and sent at least one in data volume.

In conjunction with implementation on the one hand, in the 4th kind of possible implementation, described key-course equipment adopts feature searching algorithm to select from the feature set of described active flow and described key words sorting degree of association meets the feature of pre-defined rule, as character subset, including:

Adopt feature searching algorithm to select a feature the highest with described key words sorting degree of association from the feature set of described active flow each time, add in character subset.

In conjunction with the 4th kind of possible implementation on the one hand, in the 5th kind of possible implementation, when the described score at described character subset meets predetermined condition, terminate feature searching algorithm, including:

When the score of described character subset no longer increases because increasing new feature, terminate feature searching algorithm.

In conjunction with on the one hand, on the one hand the first, the second, the third, the 4th kind or the 5th kind of possible implementation, in the 6th kind of possible implementation, after obtaining the stream group belonging to each active flow, described method also includes:

Have when newly flowing into network, described new stream is subdivided into the stream group that the characteristic vector value of described new stream is corresponding.

In conjunction with on the one hand, on the one hand the first, the second, the third, the 4th kind or the 5th kind of possible implementation, in the 7th kind of possible implementation, described feature searching algorithm includes: the one in search, heuristic search or random search completely.

In conjunction with implementation on the one hand, in the 8th kind of possible implementation, described current network state information is obtained by south orientation application programming interfaces API monitoring by described key-course；Described key-course equipment obtains clustering criteria and includes:

Key-course equipment obtains current clustering criteria by north orientation API from controlling application.

In conjunction with implementation on the one hand, in the 9th kind of possible implementation, using described optimal feature subset and described network state information, also including before calculating the characteristic vector value of the active flow being currently at active state:

The weights of each characteristic component are set according to the degree of association of the feature in described optimal feature subset with the stream key words sorting of described activity.

The embodiment of the present invention two aspect provides a kind of dynamic flow control device, including:

Criterion acquiring unit, is used for obtaining clustering criteria；

Flag setting unit, being used for according to the service-level agreement SLA grade of described clustering criteria and active flow is that described active flow arranges key words sorting；Described key words sorting is the active flow degree of association to described clustering criteria described in labelling；

Feature searching unit, for adopting feature searching algorithm to select from the feature set of described active flow and described key words sorting degree of association meets the feature of pre-defined rule, as character subset；Use the Attributions selection CFS algorithm based on dependency to calculate the score of described character subset, when the score of described character subset meets predetermined condition, terminate feature searching algorithm, and using described character subset as optimal feature subset；

Vector calculation unit, is used for using described optimal feature subset and current network state information, calculates the characteristic vector value being in active flow；

Group determines unit, performs cluster operation for the characteristic vector value according to active flow, obtains the stream group belonging to each active flow.

In conjunction with the implementation of two aspects, in the implementation that the first is possible, described device also includes:

Information monitoring unit, for before arranging key words sorting for described active flow, monitoring current network state information, described network state information includes the statistical information of network topology, the status information of network topology, the packet information of described active flow and described active flow；

Described flag setting unit, if monitor current network state information meet predetermined trigger condition for described information monitoring unit, is then that described active flow arranges key words sorting according to the service-level agreement SLA grade of described clustering criteria and active flow.

In conjunction with the first possible implementation of two aspects, in the implementation that the second is possible, described in monitor current network state information and meet predetermined trigger condition and include:

Current network state meets the regular expression based on threshold value required with the parameter of described network state at least one network performance index that benchmark sets.

The first possible implementation in conjunction with two aspects, in the implementation that the third is possible, described network topology, network topology status information include: at least one in the division of node adjacency relation, network, link capacity, port type, port statistics info, link and port status；

The packet information of described active flow and the statistical information of described active flow include: data package size, packet interval time, Internet protocol IP five-tuple, packet loss and sent at least one in data volume.

Implementation in conjunction with two aspects, in the 4th kind of possible implementation, described feature searching unit, specifically for adopting feature searching algorithm to select a feature the highest with described key words sorting degree of association from the feature set of described active flow each time, adds in character subset.

In conjunction with the 4th kind of possible implementation of two aspects, in the 5th kind of possible implementation, described feature searching unit, specifically for when the score of described character subset no longer increases because increasing new feature, terminating feature searching algorithm.

In conjunction with two aspects, two aspects the first, the second, the third, the 4th kind or the 5th kind of possible implementation, in the 6th kind of possible implementation, described group determines unit, it is additionally operable to after obtaining the stream group belonging to each active flow, have when newly flowing into network, described new stream is subdivided into the stream group that the characteristic vector value of described new stream is corresponding.

In conjunction with two aspects, two aspects the first, the second, the third, the 4th kind or the 5th kind of possible implementation, in the 7th kind of possible implementation, described feature searching algorithm includes: the one in search, heuristic search or random search completely.

In conjunction with the implementation of two aspects, in the 8th kind of possible implementation, described current network state information is obtained by south orientation application programming interfaces API monitoring by described information monitoring unit；

Criterion acquiring unit, specifically for obtaining current clustering criteria by north orientation API from controlling application.

In conjunction with the implementation of two aspects, in the 9th kind of possible implementation, described device also includes:

Weights arrange unit, for using described optimal feature subset and described network state information in described vector calculation unit, before calculating the characteristic vector value of active flow being currently at active state, the weights of each characteristic component are set according to the degree of association of the feature in described optimal feature subset with the stream key words sorting of described activity.

As can be seen from the above technical solutions, the embodiment of the present invention has the advantage that by the real-time management and control ability of network is obtained in that clustering criteria and the equipment of real-time network status information, dynamically obtain character subset and characteristic vector, and carry out cluster operation according to the characteristic vector of the active flow dynamically obtained, it is thus possible to take into full account the overall distribution of current network state and flow, it is to avoid the one-sidedness that single current processes.Therefore, the character subset that embodiment of the present invention scheme dynamically obtains can bulk flow measure feature in accurate description real network situation, and can be applicable to network entirety flow is controlled.

Accompanying drawing explanation

In order to be illustrated more clearly that the technical scheme in the embodiment of the present invention, below the accompanying drawing used required during embodiment is described is briefly introduced, apparently, accompanying drawing in the following describes is only some embodiments of the present invention, for those of ordinary skill in the art, under the premise not paying creative work, it is also possible to obtain other accompanying drawing according to these accompanying drawings.

Fig. 1 is embodiment of the present invention method flow schematic diagram；

Fig. 2 is the main module structure schematic diagram of embodiment of the present invention system；

Fig. 3 is embodiment of the present invention system architecture schematic diagram；

Fig. 4 is embodiment of the present invention method flow schematic diagram；

Fig. 5 is embodiment of the present invention system architecture schematic diagram；

Fig. 6 is embodiment of the present invention apparatus structure schematic diagram；

Fig. 7 is embodiment of the present invention apparatus structure schematic diagram；

Fig. 8 is embodiment of the present invention apparatus structure schematic diagram；

Fig. 9 is embodiment of the present invention apparatus structure schematic diagram.

Detailed description of the invention

In order to make the object, technical solutions and advantages of the present invention clearly, below in conjunction with accompanying drawing, the present invention is described in further detail, it is clear that described embodiment is only a part of embodiment of the present invention, rather than whole embodiments.Based on the embodiment in the present invention, all other embodiments that those of ordinary skill in the art obtain under not making creative work premise, broadly fall into the scope of protection of the invention.

Embodiments provide a kind of dynamic flow control method, it is possible to be applied to SDN and can also be applied to non-SDN field, as it is shown in figure 1, include:

101: key-course equipment obtains clustering criteria, is that above-mentioned active flow arranges key words sorting according to the service-level agreement SLA grade of above-mentioned clustering criteria and active flow；Above-mentioned key words sorting is used for the labelling above-mentioned active flow degree of association to above-mentioned clustering criteria；

Further, the embodiment of the present invention additionally provides the condition citing starting cluster operation, it should be noted that, entry condition can have other schemes, such as periodically cluster does not then need the entry condition in the present embodiment, the entry condition of the present embodiment can as a preferred implementation scheme, but should not be construed as the uniqueness to the embodiment of the present invention to limit, it is specific as follows: before for above-mentioned active flow, key words sorting is set, said method includes: key-course monitoring of equipment current network state information, above-mentioned network state information includes network topology, the status information of network topology, the packet information of above-mentioned active flow and the statistical information of above-mentioned active flow；If monitoring current network state information to meet predetermined trigger condition, then above-mentioned key-course equipment obtains clustering criteria, it is that above-mentioned active flow arranges key words sorting according to service-level agreement (Service-LevelAgreement, the SLA) grade of above-mentioned clustering criteria and active flow.

Alternatively, aforementioned entry condition can use predetermined trigger condition, predetermined trigger condition can be determined according to different application scenarios or technical need, embodiments provides a kind of optional scheme as follows: the above-mentioned current network state information that monitors meets predetermined trigger condition and includes:

Current network state meets the regular expression based on threshold value required with the parameter of above-mentioned network state at least one network performance index that benchmark sets.

Alternatively, in embodiments of the present invention, additionally provide the concrete optional parameters of network state information, specific as follows: above-mentioned network topology, network topology status information include: at least one in the division of node adjacency relation, network, link capacity, port type, port statistics info, link and port status；

The packet information of above-mentioned active flow and the statistical information of above-mentioned active flow include: data package size, packet interval time, Internet protocol IP five-tuple, packet loss and sent at least one in data volume.

102: above-mentioned key-course equipment adopts feature searching algorithm to select from the feature set of above-mentioned active flow and above-mentioned key words sorting degree of association meets the feature of pre-defined rule, as character subset；Use the Attributions selection (Correlation-basedFeatureSelection based on dependency, CFS) algorithm calculates the score of features described above subset, when the score of features described above subset meets predetermined condition, terminate feature searching algorithm, and using features described above subset as optimal feature subset；

Alternatively, in embodiments of the present invention, it is selective that active flow has a lot of feature, the set of these alternative features is exactly feature set, how to select from feature set feature calculate characteristic vector can be disposable selection several can also be once select one can be determined according to being actually needed, embodiments provide following preferred version: above-mentioned key-course equipment adopts feature searching algorithm to select from the feature set of above-mentioned active flow and above-mentioned key words sorting degree of association meets the feature of pre-defined rule, as character subset, including:

Adopt feature searching algorithm to select a feature the highest with above-mentioned key words sorting degree of association from the feature set of above-mentioned active flow each time, add in character subset.

Alternatively, after calculating the score of character subset in embodiments of the present invention, how to determine that terminating feature searching algorithm can determine as required, such as the threshold value of score more than one mark terminates, or other schemes, the embodiment of the present invention give one can implementation preferably as follows: when the above-mentioned score in features described above subset meets predetermined condition, terminate feature searching algorithm, including:

When the score of features described above subset no longer increases because increasing new feature, terminate feature searching algorithm.

Alternatively, in embodiments of the present invention, feature searching algorithm can carry out arbitrarily selected as desired, embodiments provides optional several scheme as follows: features described above searching algorithm includes: the one in search, heuristic search or random search completely.

103: above-mentioned key-course equipment uses above-mentioned optimal feature subset and current network state information, calculate the characteristic vector value being in active flow, and perform cluster operation according to the characteristic vector value of active flow, obtain the stream group belonging to each active flow.

Adopt embodiment of the present invention scheme, by the real-time management and control ability of network is obtained in that clustering criteria and the equipment of real-time network status information, dynamically obtain character subset and characteristic vector, and carry out cluster operation according to the characteristic vector of the active flow dynamically obtained, it is thus possible to take into full account the overall distribution of current network state and flow, it is to avoid the one-sidedness that single current processes.Therefore, the character subset that embodiment of the present invention scheme dynamically obtains can bulk flow measure feature in accurate description real network situation, and can be applicable to network entirety flow is controlled.

Further, after obtaining the stream group belonging to each active flow, said method also includes: have when newly flowing into network, above-mentioned new stream is subdivided into the stream group that the characteristic vector value of above-mentioned new stream is corresponding.

Alternatively, above-mentioned current network state information is obtained by south orientation application programming interfaces API monitoring by above-mentioned key-course；Above-mentioned key-course equipment obtains clustering criteria and includes:

Key-course equipment obtains current clustering criteria by north orientation API from controlling application.

Further, using above-mentioned optimal feature subset and above-mentioned network state information, also including before calculating the characteristic vector value of the active flow being currently at active state:

The weights of each characteristic component are set according to the degree of association of the feature in above-mentioned optimal feature subset with the stream key words sorting of above-mentioned activity.

Following example will with regard in SDN, and controller realizes the scheme of cluster operation and is specifically described.

In SDN, the controller with overall management and control ability has the real-time perception ability of bearer traffic on network.The embodiment of the present invention is focused mainly on flow information and the network state of the overall situation how having used controller, and in conjunction with the different demands of user, it is possible to it is dynamically generated and extracts the characteristic vector of stream, and the characteristic vector that non-usage is fixed describes stream.And then according to the characteristic vector dynamically generated, mark off, by performing cluster operation, the stream group that current network exists, in order to unify to dredge or process to flow in group.

The scheme that the embodiment of the present invention proposes is a kind of scheme of dynamic flow cluster and classification in SDN.This method key step has several as follows:

1, controller judges whether to trigger cluster operation according to real-time network state；

2, by network topology and state feature, network flow data bag feature, statistical flow characteristic composition candidate characteristic set；

3, utilize the predefined clustering criteria of control program and feature selecting algorithm to concentrate in candidate feature and choose optimal feature subset, generate characteristic vector；

4, calculate the characteristic vector value of every active flow, and all active flows are carried out cluster operation.

5, after newly flowing into network, controller, according to the flow group information divided, puts new stream under particular demographic.

The main modular of embodiment of the present invention system is as shown in Figure 2.Main modular describes as follows:

Management module: management network state and flow status, network information storehouse (NetworkInformationBase is left in distribution in, and flow statistic storehouse (FlowStatisticBase NIB), FSB), wherein, NIB primary responsibility storage network topology and state characteristic information, the appreciable global information of controller such as including node adjacency relation, network division, link capacity, port type, port statistics info, link and port status；FSB then saves network flow data bag feature and statistical flow characteristic, including data package size, packet interval time, VLAN (VirtualLocalAreaNetwork, VLAN), IP five-tuple, packet loss, send the various features information of the packets such as data volume and stream.Management module controls program principally for the resource distribution on controller upper strata, flow scheduling etc. to be provided data access and safeguards service.

Cluster module: controller generates candidate characteristic set according to the flow status of real-time network state and active flow, therefrom choose optimal flux character subset, generation traffic characteristic vector, calculate the characteristic component value of all movable flow of current network, and adopt clustering algorithm to perform cluster operation, generate flow group.

It addition, have when newly flowing into, it is possible to newly entering single current is carried out sort operation by the flow group according to generating, it is determined that the stream group belonging to new stream.

Concrete, the embodiment of the present invention specifically comprises 3 key steps, and each step describes as follows:

Step 1, threshold triggers cluster operation.The embodiment of the present invention indicates controller to judge whether to cluster operation according to real-time network state by arranging a threshold value in the upper layer application of SDN, is embodied as details as follows:

Controller utilizes the OF message such as port_feature and port_statistics_request to obtain network state information in real time from switch, such as link capacity, port transceiving data amount statistics etc.；

Computing network performance indications, include but not limited to: the whole network maximum link utilization L_max, the maximum packet loss D of the whole network_max, network jitter time delay J_max.It is respectively provided with a threshold value for above-mentioned performance indications, by logical "and" "or" relation composition threshold triggers regular expression between multiple performance indications, as:

(L_max>L_r)&&(D_max>D_r)&&(J_max>J_r)&&(…)

When threshold triggers condition is set up, it was shown that current network state is not ideal enough, it is necessary to by operation adjustment resource allocation policy or the flow scheduling of again clustering.

Step 2, the generation of behavioral characteristics vector.After controller triggers cluster operation, by collecting the real-time network information and user's request, dynamically carry out the generation of characteristic vector.Its main process of characteristic vector describes as follows:

Controller passes through north orientation application programming interfaces (NorthboundApplicationProgrammingInterface, NorthboundAPI) to controlling application (such as flow scheduling) request current cluster SLA criterion, above-mentioned cluster SLA criterion includes but not limited to end-to-end time delay, occupied bandwidth, stream persistent period, average request response time etc..Controller also needs to obtain real-time network state information；

Controller determines candidate characteristic set according to real-time network state information.Real-time network state information includes: the network topology preserved in the network state storehouse of Dynamic Maintenance and flow statistic storehouse and state feature, network flow data bag feature, statistical flow characteristic etc..Wherein network topology and state characteristic information include node adjacency relation, network division, link capacity, port type, port statistics info, link and port status etc. may be used for generate characteristic vector status information；The feature of packet and stream includes: data package size, packet interval time, VLAN, IP five-tuple, packet loss, send the characteristic informations such as data volume.

Controller concentrates selection according to the SLA information of current clustering criteria and active flow from candidate feature, obtains character subset, and determines optimal feature subset.Method particularly includes: according to the SLA information of clustering criteria and active flow, by arranging threshold value, active flow being divided into some grades, different brackets reflects this stream degree of association to current criterion；Utilizing feature searching algorithm to obtain candidate feature subset, feature searching algorithm can adopt the one in search completely, heuristic search or random search.

Utilize Attributions selection (Correlation-basedFeatureSelection, the CFS) algorithm based on dependency that produced character subset is estimated, select the character subset composition characteristic vector for cluster operation of highest scoring；Furthermore it is also possible to calculate the dependency of the inter-stages such as each characteristic component and SLA further, and according to this dependency, each characteristic component is arranged weights.

Step 3, performs cluster operation and the classification of new stream.Controller, after determining the characteristic vector being made up of optimal feature subset as each component, calculates the characteristic vector value of every active flow according to flow statistic storehouse, and performs clustering algorithm and obtain the stream group belonging to each active flow.Afterwards, when there being newly stream arrival, it is classified among nearest stream group according to the distance of the characteristic vector value of new stream and Ge Liu group characteristic of correspondence vector.

The flow process of the embodiment of the present invention, by conjunction with the system architecture diagram shown in Fig. 3, is described in detail by following example.The network architecture as shown in Figure 3 is divided into three layers, is followed successively by application layer (controlling the layer that application is residing), key-course and physical layer；Wherein can comprise router-level topology, traffic engineering, virtual network etc. in application layer；Key-course is primarily related to controller, safeguards NIB and FSB in the controller, possesses clustering/classification function.Physical layer is hardware layer, it is possible to comprise the equipment such as switch.

The dynamic flow that the embodiment of the present invention proposes clusters and sorting technique, mainly realizes at key-course.By expanding the controller process to global flow information, utilize network information storehouse (NetworkInformationBase, and flow statistic storehouse (FlowStatisticBase NIB), FSB) network state information and flow statistic are stored respectively, coordinate the traffic engineering application of the application layer on upper strata, the operation that active flow network currently carried when meeting pre-provisioning request clusters, and utilize the stream group marked off to provide the foundation of resource distribution and traffic grooming for upper strata flow engineer applied.

As shown in Figure 4, for the embodiment of the present invention, the present embodiment can be specifically divided into following step:

401: traffic engineering application configuration triggers the threshold value of network flow reunion bunch, generate threshold triggers expression formula；Initialize cluster SLA criterion, such as the network bandwidth；And threshold value is updated controller.

402: new stream arrives access switch, and the first packet newly flowed is reported to controller by access switch, and controller issues stream list item for it.

403: controller passes through south orientation API (such as: Openflow agreement) real time monitoring network state, update NIB and FSB.

South orientation API refers to the communication protocol of chain of command and data surface, Openflow agreement the most well-known in SDN, open virtual switch database protocol (OpenvSwitchDatabase for configuration switch, OVSDB), border network management protocol (BorderGatewayProtocol in legacy network, BGP), Simple Network Management Protocol (SimpleNetworkManagementProtocol, SNMP) etc. broadly fall into south orientation API.

404: controller is true time determining that predetermined threshold value triggers expression formula, it is determined that needs to re-start active flow group and divides.Controller obtains current clustering criteria by north orientation API from controlling application, and the SLA information according to clustering criteria and stream is that stream arranges key words sorting.Key words sorting can be as: SLA bandwidth request arranged from high to low, levelling is divided into the bigger and less two-stage of bandwidth demand.

North orientation API refers to the communication interface between management application and the controller on controller upper strata, the presentation layer state transfer interface (RepresentationalStateTransferApplicationProgramInterface of SDN open source projects OpenDaylight, RestAPI) OSS (OpertationalSupportSystem, OSS) etc. and in legacy network broadly falls into north orientation API.

405: select a feature the highest with stream SLA rank correlation degree according to feature searching algorithm every time, join character subset；Feature searching algorithm can be the sequence forward direction selection algorithm in heuristic search.Every time to after character subset adds a new feature, by the Attributions selection (Correlation-basedFeatureSelection based on dependency, CFS) score of evaluation of algorithm character subset, searching algorithm is terminated when adding new feature and character subset score does not improve, and export current signature subset and obtain characteristic vector as optimal subset, it can in addition contain the weight of each component is set according to each component of characteristic vector and the degree of association of the key words sorting of above-mentioned active flow.The characteristic vector that this step obtains is optimal characteristics vector.

Will be given below relatedness computation and the method for character subset score calculating, specific as follows:

Assume there is discrete random variable Y, then the comentropy H (Y) of Y is defined as:

$H\left(Y\right)=-\underset{i=1}{\overset{m}{\mathrm{\Σ}}}{P}_i{\log }_2{P}_i$

Wherein P_iPossible value Y for Y_iThe probability occurred, m is the possible value number of Y.

After given discrete random variable X, the conditional information entropy of Y is expressed as:

$H\left(Y\right|X)=\underset{x=U}{\mathrm{\Σ}}p\left(x\right)H\left(Y\right|X=u)--\underset{x=U}{\mathrm{\Σ}}\underset{y=V}{\mathrm{\Σ}}p(x,y)\log p\left(y\right|x)$

The wherein codomain of U, V respectively stochastic variable X, Y.

Calculate two variable X, Y degree of association method as follows:

$r_{\mathrm{xy}}=2\frac{H\left(X\right)-H\left(X\right|Y)}{H\left(X\right)+H\left(Y\right)}$

Calculate the method (for assessing the quality of character subset) of character subset score:

${\mathrm{Merit}}_s=\frac{k\stackrel{\‾}{r_{\mathrm{cf}}}}{\sqrt{k+k(k-1)\stackrel{\‾}{r_{\mathrm{ff}}}}}$

WhereinIt is all features and the meansigma methods of key words sorting degree of association,Being the meansigma methods of degree of association between all features, K refers to the number of features in subset.

406: controller utilizes the statistical information in FSB to calculate each characteristic component value of active flow characteristic vector, then each characteristic component value adopts K average (K-Means) algorithm perform cluster operation and obtains the stream group belonging to active flow the division result of storage flow group.

407: after having and newly flowing into access switch, switch performs new stream and reports, and controller performs new flow point class, and according to group's routing policy route.Wherein new flow point class particularly as follows: put closest stream group under according to the characteristic vector value of new stream by new stream.

The present embodiment method relates to three main bodys, and namely (such as traffic engineering) is applied in bottom switch, the controller of SDN and the control on upper strata.Switch and being mainly reflected in alternately of controller are newly flowed when arrival, ask route flow list item to controller；Controller controls application configuration meet again after bunch threshold value the update notification to controller with controlling being mainly reflected in alternately of application, and controls, at the controller bunch forward direction that carries out meeting again, the clustering criteria that application request is current；Further relate to the renewal of NIB and FSB inside controller, bunch front optimal characteristics vector of meeting again is chosen, the sort operation of the cluster of active flow and new stream.

In above example, what adopt is that the controller side in SDN realizes, in the present embodiment, controller could alternatively be have traffic management and control ability, can the NMS of all flows of sensing network or the network management software, thus using the present invention to carry out cluster and the sort operation of flow, it is achieved the Group Decision of flow.

Functional structure according to NMS or software, the traffic engineering application being arranged in SDN upper strata in previous embodiment also can be grouped into NMS or software goes.And then the method for the dynamic flow cluster of the present embodiment proposition and classification can both complete in NMS or software.As shown in Figure 5, when controller is replaced by NMS, NMS is safeguarded not only according to network state and updates NIB and FSB, also can according to the Traffic Engineering capabilities in system, the division operation of stream group, the unified flow dredging different group is carried out after threshold triggers.Implementing flow process and be referred to previous embodiment, the present embodiment no longer repeats one by one.

The present embodiment can apply to non-SDN field, and now the role of SDN controller can be replaced by the equipment of an energy management and control the whole network or software, for instance Network Management Equipment, and the control on upper strata application can also be replaced by load balancing or flow-control software or device.Therefore the citing of above example should not be construed as the uniqueness to the embodiment of the present invention and limits.

Adopt embodiment of the present invention scheme, by the controller real-time management and control ability to network, obtain network state information and traffic statistics in real time, dynamically carry out cluster and the sort operation of flow according to network state and service strategy.Utilize the traffic characteristic dynamically generated the active flow of current network to be clustered as some groups, take into full account the overall distribution of current network state and flow, it is to avoid the one-sidedness that single current processes.By arranging the threshold value that heavily divides of stream group, it is to avoid stream group division operation frequently.

The embodiment of the present invention additionally provides a kind of dynamic flow control device, as shown in Figure 6, and including:

Criterion acquiring unit 601, is used for obtaining clustering criteria；

Flag setting unit 602, being used for according to the service-level agreement SLA grade of described clustering criteria and active flow is that described active flow arranges key words sorting；Described key words sorting is the active flow degree of association to described clustering criteria described in labelling；

Feature searching unit 603, for adopting feature searching algorithm to select from the feature set of described active flow and described key words sorting degree of association meets the feature of pre-defined rule, as character subset；Use the Attributions selection CFS algorithm based on dependency to calculate the score of described character subset, when the score of described character subset meets predetermined condition, terminate feature searching algorithm, and using described character subset as optimal feature subset；

Vector calculation unit 604, is used for using described optimal feature subset and current network state information, calculates the characteristic vector value being in active flow；

Group determines unit 605, performs cluster operation for the characteristic vector value according to active flow, obtains the stream group belonging to each active flow.

Further, the embodiment of the present invention additionally provides the condition citing starting cluster operation, it should be noted that, entry condition can have other schemes, such as periodically cluster does not then need the entry condition in the present embodiment, and the entry condition of the present embodiment as a preferred implementation scheme, but can should not be construed as the uniqueness to the embodiment of the present invention and limit, specific as follows: as it is shown in fig. 7, described device also includes:

Information monitoring unit 701, for before arranging key words sorting for described active flow, monitoring current network state information, described network state information includes the statistical information of network topology, the status information of network topology, the packet information of described active flow and described active flow；

Described flag setting unit 602, if monitor current network state information meet predetermined trigger condition for described information monitoring unit 701, is then that described active flow arranges key words sorting according to the service-level agreement SLA grade of described clustering criteria and active flow.

Alternatively, aforementioned entry condition can use predetermined trigger condition, predetermined trigger condition can be determined according to different application scenarios or technical need, embodiments provides a kind of optional scheme as follows: described in monitor current network state information and meet predetermined trigger condition and include:

Current network state meets the regular expression based on threshold value required with the parameter of described network state at least one network performance index that benchmark sets.

Alternatively, in embodiments of the present invention, additionally provide the concrete optional parameters of network state information, specific as follows: described network topology, network topology status information include: at least one in the division of node adjacency relation, network, link capacity, port type, port statistics info, link and port status；

The packet information of described active flow and the statistical information of described active flow include: data package size, packet interval time, Internet protocol IP five-tuple, packet loss and sent at least one in data volume.

Alternatively, in embodiments of the present invention, it is selective that active flow has a lot of feature, the set of these alternative features is exactly feature set, how to select from feature set feature calculate characteristic vector can be disposable selection several can also be once select one can be determined according to being actually needed, embodiments provide following preferred version: described feature searching unit 603, specifically for adopting feature searching algorithm to select a feature the highest with described key words sorting degree of association from the feature set of described active flow each time, add in character subset.

Alternatively, after calculating the score of character subset in embodiments of the present invention, how to determine that terminating feature searching algorithm can determine as required, such as the threshold value of score more than one mark terminates, or other schemes, the embodiment of the present invention give one can implementation preferably as follows: described feature searching unit 603, specifically for when the score of described character subset no longer increases because increasing new feature, terminating feature searching algorithm.

Further, described group determines unit 605, is additionally operable to after obtaining the stream group belonging to each active flow, has when newly flowing into network, described new stream is subdivided into the stream group that the characteristic vector value of described new stream is corresponding.

Alternatively, in embodiments of the present invention, feature searching algorithm can carry out arbitrarily selected as desired, embodiments provides optional several scheme as follows: described feature searching algorithm includes: the one in search, heuristic search or random search completely.

Alternatively, described current network state information is obtained by south orientation application programming interfaces API monitoring by described information monitoring unit 701；

Criterion acquiring unit 601, specifically for obtaining current clustering criteria by north orientation API from controlling application.

Further, as shown in Figure 8, described device also includes:

Weights arrange unit 801, for using described optimal feature subset and described network state information in described vector calculation unit 604, before calculating the characteristic vector value of active flow being currently at active state, the weights of each characteristic component are set according to the degree of association of the feature in described optimal feature subset with the stream key words sorting of described activity.

The embodiment of the present invention additionally provides another kind of dynamic flow control device, as it is shown in figure 9, include: receptor 901, emitter 902, processor 903 and memorizer 904；Described memorizer 904 may be used for providing in the data handling procedure of processor 903 spatial cache；

Wherein, described processor 903, it is used for obtaining clustering criteria, is that above-mentioned active flow arranges key words sorting according to the service-level agreement SLA grade of above-mentioned clustering criteria and active flow；Above-mentioned key words sorting is used for the labelling above-mentioned active flow degree of association to above-mentioned clustering criteria；Adopt feature searching algorithm to select from the feature set of above-mentioned active flow and above-mentioned key words sorting degree of association meets the feature of pre-defined rule, as character subset；Use the score calculating features described above subset based on CFS algorithm, when the score of features described above subset meets predetermined condition, terminate feature searching algorithm, and using features described above subset as optimal feature subset；Use above-mentioned optimal feature subset and current network state information, calculate the characteristic vector value being in active flow, and perform cluster operation according to the characteristic vector value of active flow, obtain the stream group belonging to each active flow.

Further, the embodiment of the present invention additionally provides the condition citing starting cluster operation, it should be noted that, entry condition can have other schemes, such as periodically cluster does not then need the entry condition in the present embodiment, the entry condition of the present embodiment can as a preferred implementation scheme, but should not be construed as the uniqueness to the embodiment of the present invention to limit, specific as follows: described processor 903, it is additionally operable to before for above-mentioned active flow, key words sorting is set, monitoring current network state information, above-mentioned network state information includes network topology, the status information of network topology, the packet information of above-mentioned active flow and the statistical information of above-mentioned active flow；If monitoring current network state information to meet predetermined trigger condition, then above-mentioned key-course equipment obtains clustering criteria, it is that above-mentioned active flow arranges key words sorting according to service-level agreement (Service-LevelAgreement, the SLA) grade of above-mentioned clustering criteria and active flow.

Alternatively, aforementioned entry condition can use predetermined trigger condition, predetermined trigger condition can be determined according to different application scenarios or technical need, embodiments provides a kind of optional scheme as follows: the above-mentioned current network state information that monitors meets predetermined trigger condition and includes:

Current network state meets the regular expression based on threshold value required with the parameter of above-mentioned network state at least one network performance index that benchmark sets.

Alternatively, in embodiments of the present invention, additionally provide the concrete optional parameters of network state information, specific as follows: above-mentioned network topology, network topology status information include: at least one in the division of node adjacency relation, network, link capacity, port type, port statistics info, link and port status；

The packet information of above-mentioned active flow and the statistical information of above-mentioned active flow include: data package size, packet interval time, Internet protocol IP five-tuple, packet loss and sent at least one in data volume.

Alternatively, in embodiments of the present invention, it is selective that active flow has a lot of feature, the set of these alternative features is exactly feature set, how to select from feature set feature calculate characteristic vector can be disposable selection several can also be once select one can be determined according to being actually needed, embodiments provide following preferred version: described processor 903, the feature of pre-defined rule is met with above-mentioned key words sorting degree of association for adopting feature searching algorithm to select from the feature set of above-mentioned active flow, as character subset, including:

Adopt feature searching algorithm to select a feature the highest with above-mentioned key words sorting degree of association from the feature set of above-mentioned active flow each time, add in character subset.

Alternatively, after calculating the score of character subset in embodiments of the present invention, how to determine that terminating feature searching algorithm can determine as required, such as the threshold value of score more than one mark terminates, or other schemes, the embodiment of the present invention give one can implementation preferably as follows: described processor 903, for when the score of features described above subset meets predetermined condition, terminate feature searching algorithm, including:

When the score of features described above subset no longer increases because increasing new feature, terminate feature searching algorithm.

Alternatively, in embodiments of the present invention, feature searching algorithm can carry out arbitrarily selected as desired, embodiments provides optional several scheme as follows: features described above searching algorithm includes: the one in search, heuristic search or random search completely.

Further, described processor 903, it is additionally operable to after obtaining the stream group belonging to each active flow, has when newly flowing into network, above-mentioned new stream is subdivided into the stream group that the characteristic vector value of above-mentioned new stream is corresponding.

Alternatively, described current network state information is obtained by south orientation application programming interfaces API monitoring by above-mentioned key-course；Described processor 903, is used for obtaining clustering criteria and includes: obtain current clustering criteria by north orientation API from controlling application.

Further, described processor 903, it is additionally operable to using above-mentioned optimal feature subset and above-mentioned network state information, before calculating the characteristic vector value of active flow being currently at active state, the weights of each characteristic component are set according to the degree of association of the feature in above-mentioned optimal feature subset with the stream key words sorting of above-mentioned activity.

Those skilled in the art is it can be understood that arrive, for convenience and simplicity of description, the system of foregoing description, the specific works process of device and unit, it is possible to reference to the corresponding process in preceding method embodiment, do not repeat them here.

In several embodiments provided herein, it should be understood that disclosed system, apparatus and method, it is possible to realize by another way.Such as, device embodiment described above is merely schematic, such as, the division of described unit, being only a kind of logic function to divide, actual can have other dividing mode when realizing, for instance multiple unit or assembly can in conjunction with or be desirably integrated into another system, or some features can ignore, or do not perform.Another point, shown or discussed coupling each other or direct-coupling or communication connection can be through INDIRECT COUPLING or the communication connection of some interfaces, device or unit, it is possible to be electrical, machinery or other form.

The described unit illustrated as separating component can be or may not be physically separate, and the parts shown as unit can be or may not be physical location, namely may be located at a place, or can also be distributed on multiple NE.Some or all of unit therein can be selected according to the actual needs to realize the purpose of the present embodiment scheme.

It addition, each functional unit in each embodiment of the present invention can be integrated in a processing unit, it is also possible to be that unit is individually physically present, it is also possible to two or more unit are integrated in a unit.Above-mentioned integrated unit both can adopt the form of hardware to realize, it would however also be possible to employ the form of SFU software functional unit realizes.

If described integrated unit is using the form realization of SFU software functional unit and as independent production marketing or use, it is possible to be stored in a computer read/write memory medium.Based on such understanding, part or all or part of of this technical scheme that prior art is contributed by technical scheme substantially in other words can embody with the form of software product, this computer software product is stored in a storage medium, including some instructions with so that a computer equipment (can be personal computer, server, or the network equipment etc.) perform all or part of step of method described in each embodiment of the present invention.And aforesaid storage medium includes: USB flash disk, portable hard drive, read only memory (ROM, Read-OnlyMemory), the various media that can store program code such as random access memory (RAM, RandomAccessMemory), magnetic disc or CD.

The above, above example only in order to technical scheme to be described, is not intended to limit；Although the present invention being described in detail with reference to previous embodiment, it will be understood by those within the art that: the technical scheme described in foregoing embodiments still can be modified by it, or wherein portion of techniques feature is carried out equivalent replacement；And these amendments or replacement, do not make the essence of appropriate technical solution depart from the spirit and scope of various embodiments of the present invention technical scheme.

Claims (20)

1. a dynamic flow control method, it is characterised in that including:

Key-course equipment obtains clustering criteria, is that described active flow arranges key words sorting according to the service-level agreement SLA grade of described clustering criteria and active flow；Described key words sorting is the active flow degree of association to described clustering criteria described in labelling；

Described key-course equipment adopts feature searching algorithm to select from the feature set of described active flow and described key words sorting degree of association meets the feature of pre-defined rule, as character subset；Use the Attributions selection CFS algorithm based on dependency to calculate the score of described character subset, when the score of described character subset meets predetermined condition, terminate feature searching algorithm, and using described character subset as optimal feature subset；

Described key-course equipment uses described optimal feature subset and current network state information, calculates the characteristic vector value being in active flow, and performs cluster operation according to the characteristic vector value of active flow, obtains the stream group belonging to each active flow.

2. method according to claim 1, it is characterised in that before arranging key words sorting for described active flow, described method also includes:

Key-course monitoring of equipment current network state information, described network state information includes the statistical information of network topology, the status information of network topology, the packet information of described active flow and described active flow；If monitoring current network state information to meet predetermined trigger condition, then described key-course equipment obtains clustering criteria, is that described active flow arranges key words sorting according to the service-level agreement SLA grade of described clustering criteria and active flow.

3. method according to claim 2, it is characterised in that described in monitor current network state information and meet predetermined trigger condition and include:

Current network state meets the regular expression based on threshold value required with the parameter of described network state at least one network performance index that benchmark sets.

4. method according to claim 2, it is characterised in that

Described network topology, network topology status information include: at least one in the division of node adjacency relation, network, link capacity, port type, port statistics info, link and port status；

The packet information of described active flow and the statistical information of described active flow include: data package size, packet interval time, Internet protocol IP five-tuple, packet loss and sent at least one in data volume.

5. method according to claim 1, it is characterised in that described key-course equipment adopts feature searching algorithm to select from the feature set of described active flow and described key words sorting degree of association meets the feature of pre-defined rule, as character subset, including:

Adopt feature searching algorithm to select a feature the highest with described key words sorting degree of association from the feature set of described active flow each time, add in character subset.

6. method according to claim 5, it is characterised in that when the described score at described character subset meets predetermined condition, terminates feature searching algorithm, including:

When the score of described character subset no longer increases because increasing new feature, terminate feature searching algorithm.

7. method according to claim 1 to 6 any one, it is characterised in that after obtaining the stream group belonging to each active flow, described method also includes:

Have when newly flowing into network, described new stream is subdivided into the stream group that the characteristic vector value of described new stream is corresponding.

8. method according to claim 1 to 6 any one, it is characterised in that described feature searching algorithm includes: the one in search, heuristic search or random search completely.

9. method according to claim 1, it is characterised in that described current network state information is obtained by south orientation application programming interfaces API monitoring by described key-course；Described key-course equipment obtains clustering criteria and includes:

Key-course equipment obtains current clustering criteria by north orientation API from controlling application.

10. method according to claim 1, it is characterised in that using described optimal feature subset and described network state information, also including before calculating the characteristic vector value of the active flow being currently at active state:

The weights of each characteristic component are set according to the degree of association of the feature in described optimal feature subset with the stream key words sorting of described activity.

11. a dynamic flow control device, it is characterised in that including:

Criterion acquiring unit, is used for obtaining clustering criteria；

Flag setting unit, being used for according to the service-level agreement SLA grade of described clustering criteria and active flow is that described active flow arranges key words sorting；Described key words sorting is the active flow degree of association to described clustering criteria described in labelling；

Feature searching unit, for adopting feature searching algorithm to select from the feature set of described active flow and described key words sorting degree of association meets the feature of pre-defined rule, as character subset；Use the Attributions selection CFS algorithm based on dependency to calculate the score of described character subset, when the score of described character subset meets predetermined condition, terminate feature searching algorithm, and using described character subset as optimal feature subset；

Vector calculation unit, is used for using described optimal feature subset and current network state information, calculates the characteristic vector value being in active flow；

Group determines unit, performs cluster operation for the characteristic vector value according to active flow, obtains the stream group belonging to each active flow.

12. device according to claim 11, it is characterised in that described device also includes:

Information monitoring unit, for before arranging key words sorting for described active flow, monitoring current network state information, described network state information includes the statistical information of network topology, the status information of network topology, the packet information of described active flow and described active flow；

Described flag setting unit, if monitor current network state information meet predetermined trigger condition for described information monitoring unit, is then that described active flow arranges key words sorting according to the service-level agreement SLA grade of described clustering criteria and active flow.

13. device according to claim 12, it is characterised in that described in monitor current network state information and meet predetermined trigger condition and include:

Current network state meets the regular expression based on threshold value required with the parameter of described network state at least one network performance index that benchmark sets.

14. device according to claim 12, it is characterised in that

Described network topology, network topology status information include: at least one in the division of node adjacency relation, network, link capacity, port type, port statistics info, link and port status；

The packet information of described active flow and the statistical information of described active flow include: data package size, packet interval time, Internet protocol IP five-tuple, packet loss and sent at least one in data volume.

15. device according to claim 11, it is characterised in that

Described feature searching unit, specifically for adopting feature searching algorithm to select a feature the highest with described key words sorting degree of association from the feature set of described active flow each time, adds in character subset.

16. device according to claim 15, it is characterised in that

Described feature searching unit, specifically for when the score of described character subset no longer increases because increasing new feature, terminating feature searching algorithm.

17. device according to claim 11 to 16 any one, it is characterised in that

Described group determines unit, is additionally operable to after obtaining the stream group belonging to each active flow, has when newly flowing into network, described new stream is subdivided into the stream group that the characteristic vector value of described new stream is corresponding.

18. device according to claim 11 to 16 any one, it is characterised in that described feature searching algorithm includes: the one in search, heuristic search or random search completely.

19. device according to claim 11, it is characterised in that described current network state information is obtained by south orientation application programming interfaces API monitoring by described information monitoring unit；

Criterion acquiring unit, specifically for obtaining current clustering criteria by north orientation API from controlling application.

20. device according to claim 11, it is characterised in that described device also includes:

Weights arrange unit, for using described optimal feature subset and described network state information in described vector calculation unit, before calculating the characteristic vector value of active flow being currently at active state, the weights of each characteristic component are set according to the degree of association of the feature in described optimal feature subset with the stream key words sorting of described activity.