WO2015131699A1

WO2015131699A1 - Load balancing method and device

Info

Publication number: WO2015131699A1
Application number: PCT/CN2015/071446
Authority: WO
Inventors: 喻敬海; 张道伟; 吴少勇
Original assignee: 中兴通讯股份有限公司
Priority date: 2014-10-17
Filing date: 2015-01-23
Publication date: 2015-09-11
Also published as: CN105577729B; CN105577729A

Abstract

Disclosed are a load balancing method and device. The method comprises: collecting the data on the feature parameters of multiple exchangers controlled by each software defined network (SDN) controller in a clustered SDN; according to the data, estimating the data of the feature parameters of the multiple exchangers at the next time point; performing, according to the estimated data, load balancing for the SDN controllers in the clustered SDN. The present invention solves the problem in the prior art of load imbalance of controllers in a clustered SDN, and balances the load by using estimated data. The method is implementable and enhances the resources utilization rate for controllers in a clustered SDN.

Description

Load balancing method and device

Technical field

The present invention relates to the field of communications, and in particular to a method and apparatus for load balancing.

Background technique

Software Defined Networks (SDN) technology was originally a new network innovation architecture proposed by the Stanford University clean slate research group. Its core technology Openflow (referred to as OF) protocol realizes flexible control of network traffic by separating the control plane of the network device from the data plane, and provides a good platform for innovation of core network and application.

The SDN network consists of three parts: SDN controller, secure channel and SDN switch. The SDN controller is a network control software designed according to the OpenFlow protocol, and is configured to manage data streams, configure network devices, formulate flow tables, and undertake communication between network services and network devices. There can be multiple controllers in a domain, but generally only one controller is in control and management state. The SDN switch is controlled by the SDN controller, and the flow table defined by the controller is saved, and the flow table entry is queried to determine the processing mode of the packet, including forwarding, caching, submitting the controller, or discarding. The secure channel under the OpenFlow protocol is used to connect the switch to the controller for communication.

With the development of SDN technology, SDN controller cluster technology has also been used more and more. In the SDN network multi-controller architecture, each controller controls a certain number of switches, if the number of packet-in packets reported by a controller controlled by a controller is excessive or the number of flow tables delivered and the amount of OF information exchanged with the switch Too large will cause the controller to generate a large control flow at a certain time. This situation will make the controller become the bottleneck of the entire controller cluster to handle network traffic, affecting the performance of the entire controller cluster.

When the network traffic fluctuates greatly, the switch on a controller reports a large number of packet-in packets (a special packet of the OpenFlow protocol, which is used to match the traffic on the fail or specify the matching traffic. , sent to the controller) or other OF (OpenFlow) messages, the controller may also send a large number of flow tables, while other controllers only receive a small number of packet-in messages and other OF messages, at this time the controller The number of flow tables issued is not large, which may cause some controllers to control overload, and some controllers are far from reaching control load saturation. In order to alleviate this situation, it is necessary to adjust the switches that are controlled by each controller at all times to realize load balancing of the controller at all times and improve the control performance of the entire controller cluster.

The related technology only proposes to increase the shared storage area between multiple controllers to realize inter-controller communication, and does not implement controller load balancing. However, when the network traffic fluctuates greatly, the problem of uneven controller load has not been solved. .

In view of the problem of uneven load of the cluster SDN controller in the related art, an effective solution has not been proposed yet.

Summary of the invention

The embodiment of the invention provides a load balancing method and device to solve at least the problem of uneven load of the cluster SDN controller in the related art.

According to an embodiment of the present invention, a method for load balancing is provided, comprising: collecting data of characterization parameters of a plurality of switches controlled by each of the SDN controllers of the cluster software-defined network SDN controller; Determining, by the data of the characterization parameter of the switch, the data of the characterization parameter of the plurality of switches at a next moment; according to the estimated data of the plurality of switches at the next moment, the cluster SDN Each SDN controller of the controller performs load balancing.

In this embodiment, the characterization parameter includes at least one of the following: a packet-in packet, a flow table number, and other OF messages.

In this embodiment, estimating data of the characterization parameter of the plurality of switches at a next moment includes: estimating data of the characterization parameter of the plurality of switches at a next moment using a least squares method.

In this embodiment, estimating the data of the characterization parameters of the plurality of switches at the next moment by using a least squares method includes: calculating, for each switch, the following each of the switches using a mathematical model of least squares Describe the data of the parameter at the next moment: y = α + β * x, where x is the data of the characterization parameter at the current time, y is the data of the characterization parameter at the next moment; the parameters α and β are according to the following formula Calculation:

Where n is the number of times in the current statistical period, X is the timestamp corresponding to the time, and Y is the value of the characterization parameter of each switch.

In this embodiment, the data of the characterization parameter includes sample historical data of the characterization parameter and data of the characterization parameter at each moment in the current statistical period; and according to the data of the characterization parameters of the plurality of switches, the prediction is performed. The data of the characterization parameters of the switches at the next moment includes: characterization according to the plurality of switches The sample historical data of the parameter and the data of the characterization parameter at each moment in the current statistical period, and the data of the characterization parameter of the plurality of switches at the next moment are estimated.

In this embodiment, estimating data of the characterization parameter of the multiple switches at a next moment includes: estimating the characterization of the multiple switches according to data of the characterization parameters at each moment in a current statistical period The data of the parameter at the next moment; the sample historical data of the characterization parameter is averaged with the estimated data of the characterization parameter at the next moment.

In this embodiment, performing load balancing on each SDN controller of the cluster SDN controller according to the estimated data of the plurality of switches at the next moment includes: estimating the multiple Calculating, by the data of the switch at the next moment, the ratio of the estimated value of each of the controllers to the historical maximum value of the corresponding parameter P _YHi : calculating the total pre-control cluster for the characterization parameter The ratio of the estimate to the total maximum value P _YHv : Compare P _YHi with P _YHv and adjust the controller controlled by the controller according to the result of the comparison.

In this embodiment, P _YHi is calculated by the following formula:

Where n is the number of switches managed by each controller, and Y _inext is an estimated value of the characterization parameters of each switch managed by the controller, and H _iymax is a historical maximum value of each characterization parameter of each switch managed by the controller; And / or, P _YHv is calculated by the following formula:

Wherein m represents the number of controllers, n represents the number of switches controlled by each controller, and Y _inext is an estimated value of the characterization parameters of each switch managed by the controller, and H _iymax is the managed switch controller for the Characterizes the historical maximum of the parameter.

In this embodiment, comparing P _YHi with P _YHv and adjusting the controller controlled by the controller according to the comparison result includes: setting a scaling factor for a plurality of different characterization parameters, wherein the scaling coefficient is set to represent each characterization parameter The proportion of the comparison process.

In this embodiment, comparing P _YHi with P _YHv and adjusting the controller controlled by the controller according to the comparison result includes: for each controller, when the P _{YHi is} greater than P _YHv , calculating the switch that should be called, and according to The calculation result calls the switch; and/or, for each controller, when its P _{YHi is} less than P _YHv , the switch to be transferred is calculated, and is transferred to the switch according to the calculation result.

In this embodiment, the switch that is to be called out is calculated, and the switch is called according to the calculation result, including: calculating for the controller

Set to B _yosize and get B _osize according to the previously set scale factor, where the scale factor is set to represent the proportion of the plurality of characterization parameters in the comparison process; B _{inext is} obtained based on the estimated data for each switch =Y _inext ; according to the value of B _inext , the controllers controlled by the controller are arranged in ascending order to obtain the switch queue; the switch is selected from the top to the bottom of the switch queue, and the total number of B _inext to be removed from the switch is calculated, if not greater than The B _osize continues to select the switch; the selected switch to be mobilized is added to the pool to be mobilized, and the pool to be mobilized is set to store the switch information to be mobilized.

In this embodiment, the switch that is to be input is calculated, and the switch is transferred according to the calculation result, including: calculating for the controller

Set to B _yisize and get B _isize according to the previously set scale factor, where the scale factor is set to represent the proportion of the plurality of characterization parameters in the comparison process; the controller is arranged in descending order according to the value of B _isize , and then Transfer to the switch according to the controller load.

In this embodiment, the switch is sequentially loaded according to the controller load, and the switch is arranged in descending order according to the value of B _inext of each switch to be transferred in the pool to be transferred, and the switch queue is set to be stored. Switch information to be mobilized; for each controller, select the switch from top to bottom from the above queue, compare B _inext with B _isize respectively, and if B _inext > B _isize , select the next switch, if B _inext < B _isize , the transfer controller number of the element corresponding to the switch in the pool to be transferred is set to the local controller number, and B _isize =B _isize -B _inext , and then the next switch is checked until the last switch.

According to another embodiment of the present invention, there is provided a load balancing apparatus, comprising: a collection module configured to collect data of characterization parameters of a plurality of switches controlled by each of the SDN controllers of the cluster software defined network SDN controller And an estimation module configured to estimate data of the characterization parameter of the plurality of switches at a next moment according to data of the characterization parameters of the plurality of switches; and the load balancing module is configured to be configured according to the prediction The characterization parameters of the switches are load balanced at each of the SDN controllers of the cluster SDN controller at the next time.

According to the present invention, the clustering software is used to define data of the characterization parameters of the plurality of switches controlled by each of the SDN controllers of the network SDN controller; and the plurality of switches are estimated according to the data of the characterization parameters of the plurality of switches. And characterizing the data of the parameter at the next moment; performing load balancing on each SDN controller of the cluster SDN controller according to the estimated data of the plurality of switches at the next moment The solution solves the problem of uneven load of the cluster SDN controller in the related technology, and performs load balancing through the estimated data, which is highly implementable and improves the resource utilization efficiency of the cluster SDN controller.

DRAWINGS

The drawings described herein are intended to provide a further understanding of the invention, and are intended to be a part of the invention. In the drawing:

1 is a flow chart of a method of load balancing according to an embodiment of the present invention;

2 is a structural diagram of a device for load balancing according to an embodiment of the present invention;

3 is a block diagram showing the structure of an SDN controller in accordance with a preferred embodiment of the present invention;

4 is a flow diagram of tuning a switch into a controller in accordance with a preferred embodiment of the present invention;

FIG. 5 is a flowchart of cluster controller load balancing scheduling according to a preferred embodiment of the present invention; FIG.

6 is a schematic diagram of allocation of an intra-domain controller and a switch according to a preferred embodiment 1 of the present invention;

Figure 7 is a diagram showing the allocation of intra-domain cascade controllers and switches in accordance with a preferred embodiment 3 of the present invention.

detailed description

The invention will be described in detail below with reference to the drawings in conjunction with the embodiments. It should be noted that the embodiments in the present application and the features in the embodiments may be combined with each other without conflict.

The related technology only proposes to increase the shared storage area between multiple controllers to realize inter-controller communication, and does not implement controller load balancing. However, when the network traffic fluctuates greatly, the problem of uneven controller load has not been solved. In this embodiment, the problem is solved by the method of network traffic prediction and scheduling each controller to control the switch.

In this embodiment, a method for load balancing is provided. FIG. 1 is a flowchart of a method for load balancing according to an embodiment of the present invention. As shown in FIG. 1, the method includes the following steps:

Step S102, collecting data of characterization parameters of multiple switches managed by each SDN controller of the cluster software defined network (SDN) controller;

Step S104: Estimating data of the characterization parameter of the multiple switches at a next moment according to data of the characterization parameters of the multiple switches;

Step S106: Perform load balancing on each SDN controller of the cluster SDN controller according to the estimated data of the plurality of switches at the next moment.

In this embodiment, the data of the characterization parameters of the plurality of switches controlled by each SDN controller are collected through the foregoing steps, and the data at the next moment is estimated according to the collected data, and then according to the estimated characterization parameters. The data of one moment is load-balanced to each SDN controller, thereby realizing the load balancing of the cluster SDN controller, solving the problem of uneven load of the cluster SDN controller in the related technology, and performing load balancing through the estimated data. It is highly achievable and improves the resource utilization efficiency of the cluster SDN controller.

In this embodiment, the data of the characterization parameter includes sample historical data of the characterization parameter and data of the characterization parameter at each moment in the current statistical period; and according to the data of the characterization parameters of the plurality of switches, the prediction is performed. The data of the characterization parameters of the switches at the next moment includes: estimating the plurality of switches according to the sample historical data of the characterization parameters of the plurality of switches and the data of the characterization parameters at each moment in the current statistical period The data characterizing the parameter at the next moment.

In this embodiment, estimating data of the characterization parameter of the multiple switches at a next moment includes: estimating the characterization of the multiple switches according to data of the characterization parameters at each moment in a current statistical period Reference Counting the data at the next moment; summing the sample historical data of the characterization parameters with the estimated data of the characterization parameters at the next moment.

In this embodiment, P _YHi is calculated by the following formula:

Where n is the number of switches managed by each controller, and Y _inext is an estimated value of the characterization parameters of each switch managed by the controller, and H _iymax is a historical maximum value of each characterization parameter of each switch managed by the controller; and / or,

P _YHv is calculated by the following formula:

Wherein m represents the number of controllers, n represents the number of switches controlled by each controller, and Y _inext is an estimated value of the characteristic parameters of each switch managed by the controller, and H _iymax is the switch managed by the controller. Characterizes the historical maximum of the parameter. Among them, Y _{inext is} the result y obtained by the least squares mathematical model.

Corresponding to the load balancing method described above, a device for load balancing is also provided in the embodiment, and the device is configured to implement the foregoing embodiments and preferred embodiments, and details are not described herein. As used below, the term "module" may implement a combination of software and/or hardware of a predetermined function. Although the apparatus described in the following embodiments is preferably implemented in software, hardware, or a combination of software and hardware, is also possible and contemplated.

2 is a structural diagram of a device for load balancing according to an embodiment of the present invention. As shown in FIG. 2, the device includes: a collection module 22, an estimation module 24, and a load balancing module 26. The following describes each module in detail. :

The collecting module 22 is configured to collect data of the characterization parameters of the plurality of switches that are controlled by each of the SDN controllers of the cluster software-defined network SDN controller; the estimating module 24 is connected to the collecting module 22, and is configured to be configured according to the multiple Determining, by the data of the characterization parameter of the switch, the data of the characterization parameter of the plurality of switches at a next moment; the load balancing module 26 is connected to the estimation module 24 and configured to be based on the estimated plurality of switches The characterization parameters are data of the next moment, and load balancing is performed on each SDN controller of the cluster SDN controller.

The following is a detailed description of the preferred embodiment. The following preferred embodiments mainly describe the three types of the packet-in packet number, the outgoing flow table, and other OF information. The preferred embodiment combines the foregoing. Embodiments and preferred embodiments thereof.

In the following preferred embodiments, a method and apparatus for implementing load balancing optimization of an SDN controller are provided, that is, in a distributed SDN network controller cluster scenario, a packet-in packet received by the controller is proposed. The number and the downstream flow table and other OF information are reasonably predicted and optimized, so that each controller achieves load balancing.

The preferred embodiment is mainly applied to an SDN network domain. There are multiple controllers in the domain. Each controller controls a certain number of switches. For each switch, only one controller is in the control state, and the other is in the standby state. According to the packet_in message, The number of flow tables and other OF messages are predicted by the data, and then averaged with the corresponding historical statistics to perform data synthesis, as a basis for adjusting the controller load, and dynamically adjusting the controllers controlled by the controller. The preferred embodiment includes the following two parts, namely, adding a cluster controller scheduling device in the SDN controller and detailed implementation steps:

Cluster controller scheduling device: FIG. 3 is a structural block diagram of an SDN controller according to a preferred embodiment of the present invention. As shown in FIG. 3, a cluster controller scheduling device newly added to the SDN controller mainly includes the following four internal modules:

The information collection module collects packet-in packets, flow table numbers, and sample history data of other OF messages according to the sampling time period, or collects packet-in packets, flow table numbers, and other current switches of each switch. OF message data. The time period may be a time stamp of the same time period of the week or a time stamp of the same time period of the same day in a few weeks, and is determined according to the actual use scenario.

Data processing module: The data is comprehensively processed according to the packet-in, the number of flow tables, and other OF information statistics of each switch collected to obtain the desired data.

Dispatching module: According to the comparison between the load pressure of each controller and the load pressure of the cluster controller obtained by the data processing module, the switches controlled by each controller are adjusted, so that the load pressure of each controller is balanced.

Switching module: According to the scheduling result of each controller to the controlled switch, the slave-master (master-slave mechanism, also can be written as master-slave) mechanism is used to exchange the roles of each controller for the management switch.

Based on the above structure, the implementation steps of the preferred embodiment are as follows:

S1, the cluster controller scheduling starts: the controller sets the historical maximum value of the packet-in number to H _ipmax , the historical maximum value of the flow table number is H _ifmax and the historical maximum value of other OF messages is H _immax , and are respectively set to 0. .

S2: Collecting sample information by using the information collection module in the cluster controller, which may be sample data of the same time period in a week, or sample data of the same time period on the same day in a few weeks, and determining sample data according to the actual application scenario. Collecting sample data information is accomplished by the following substeps:

S2.1: In the same time interval, the super node in the cluster initiates an information collection command of the controller of the controlled cluster; all controllers in the cluster start collecting packet-in packets and statistics from the switch that is controlled by the cluster. Number of flow tables and other OF information;

S2.2, when the controller collects statistics of all the management switches, starts to send statistics of the flow table, packet_in, and other OF messages to the cluster distributed shared storage system, and records them as F _i , P _{i ,} and M _i , respectively. F _i includes: controller number, switch label, switch flow table statistics, time stamp; P _i includes: controller number, switch label, packet-in packet number sent by the switch, time stamp; M _i includes: controller Number, switch label, switch and controller other than the packet_in OF information, timestamp. The timestamp refers to the moment when the response message of all the management switches is collected. All information will be used as a historical reference or timely forecast.

S3, the super node uses the data processing module in the controller to predict the number of packet-in packets reported by the switches under each controller in the next time interval, the number of flow tables sent by the controller, and others according to the mathematical model of the least squares method. The OF message statistics are implemented by step S4, step S5, and step S6, respectively:

S4: According to the timestamp, the super node reads the packet-in packet statistics reported by each controller from the cluster distributed shared storage system, and implements the following substeps:

S4.1, where the number of timestamps N is determined, that is, a complete packet-in packet statistics information set that attempts to acquire N timestamps; the complete packet-in packet statistics information set refers to the same timestamp of the same controller. Packet-in packet statistics reported by all the switches and the timestamp is the most recent;

S4.2: If the number of complete packet-in packet statistics information of the timestamp is less than N, the S2 is cyclically executed until the packet-in packet statistics information set of the N timestamps is obtained.

S4.3, according to the packet-in packet statistics information set, the super node obtains information statistics P ₁ , P ₂ , ..., P _N of the packet-in packets of the N time stamps for each switch. Such data: P _i = {controller number, switch number, timestamp X, packet-in message statistics Y}.

S4.4. If the packet with the highest packet-in packet information of the N timestamps is greater than the historical packet-in packet statistics of the _controller , the _ipmax is

S4.5, using a least squares mathematical model for each switch under the control of the cluster controller:

y=α+β*x, (Equation 1-1)

The regression analysis is used to calculate the packet-in packet statistics P _inext of each switch under the control of the next _timestamp . The calculation formula is as follows:

(Formula 1-2)

(Formula 1-3)

Where: n represents the number of sample data, X is equivalent to the timestamp, and Y is the statistical value of the packet-in packet number of each switch, and the parameters α and β are respectively obtained, and then the packet of the switch at the next moment is estimated. In message information.

S4.6, in order to make the estimated value closer to the actual value, according to the sampling period of the sample data, actually measure the actual value of the number of packet-in packets of the timestamp switch in the first few cycles, and then sum with the estimated value. Take the average to get a more accurate estimate.

S5. According to the timestamp, the super node reads the flow table statistics reported by each controller from the cluster distributed shared storage system, and implements the following substeps:

S5.1, determining a number of timestamps N, that is, a complete flow table statistics information set that attempts to acquire N timestamps; the complete flow table statistics information set refers to flows on all different switches of the same controller with the same timestamp. Table statistics and timestamps are recent;

S5.2, if less than N sets of complete flow table statistics information of the timestamp are obtained, performing S2 cyclically until obtaining a flow table statistical information set of N timestamps;

S5.3. According to the foregoing flow table statistics information set, the super node obtains N time-stamped flow table statistics F ₁ , F ₂ , . . . , F _N for each switch, that is, obtains such data: F _i = {controller number, switch number, timestamp X, flow table statistics Y}.

S5.4, if the one of the N timestamps of the controller flow table statistics is greater than the controller's historical flow table statistics H _ifmax , then

S5.5. For each switch managed by each controller in the cluster controller, use the least squares mathematical model (see Equation 1-1) to calculate the flow table statistics of each switch with the next timestamp by regression analysis. _Inext , the formula is shown in Equation 1-2 and Equation 1-3.

S5.6, in order to make the estimated value closer to the actual value, according to the sampling period of the sample data, take the actual value of the flow table statistical information of the time stamp switch in the first few cycles of the actual measurement, and then sum with the estimated value On average, get a more accurate estimate.

S6. According to the timestamp, the super node reads, from the cluster distributed shared storage system, other OF interaction information (hereinafter referred to as other OF information) reported by each controller except for the packet_in, which is implemented by the following substeps:

S6.1, determining the number of timestamps N, that is, a complete set of other OF information statistics that attempts to acquire N timestamps; the complete other OF information set refers to all the different switches and controllers of the same controller with the same timestamp. All OF interaction statistics except packet_in and the timestamp is the most recent;

S6.2, if less than N complete sets of other OF information statistics are obtained, the S2 is executed cyclically until the other OF information statistics sets of the N timestamps are obtained;

S6.3. According to the foregoing other information collection set of OF information, the super node accumulates other OF information statistics M ₁ , M ₂ , . . . , M _{N of N} time stamps for each switch, that is, obtains such data: M _i = {controller number, switch number, timestamp X, other information statistics Y}.

S6.4. If the other one of the N timestamps has the largest OF information, the larger one is greater than the history of the controller, and the other OF information statistics H _immax , then

S6.5. For each switch managed by each controller in the cluster controller, use the least squares mathematical model (see Equation 1-1) to calculate and calculate other OF statistics of each switch of the next timestamp by regression analysis. _Inext , the formula is shown in Equation 1-2 and Equation 1-3.

S6.6, in order to make the estimated value closer to the actual value, according to the sampling period of the sample data, take the actual value of the other OF statistical information of the timestamp switch in the first few cycles of the actual measurement, and then sum with the estimated value. On average, get a more accurate estimate.

S7, according to the estimation of the number of packet-in packets and the flow table statistics and other OF information of the controller at a certain time, respectively, calculate the pre-preparation of each controller for the packet-in and flow table statistics and other OF information. The ratio of the valuation to its corresponding historical maximum is recorded as P _PHi , P _FHi and P _MHi , respectively:

(Formula 1-4)

(Equation 1-5)

(Equation 1-6)

Where n is the number of switches managed by each controller. Then calculate the ratio of the total number of packet-in packets of the entire controller cluster to the total number of maximum packet-in packets, the ratio of the total estimated statistical flow table to the total maximum flow table statistics, and the total estimate. The ratio of other OF message statistics to the total maximum OF message statistics is recorded as P _PHv , P _FHv and P _MHv , namely:

(Equation 1-7)

(Equation 1-8)

(Equation 1-9)

Where m is the number of controllers and n is the number of switches managed by each controller.

S8, calculating P _PHi , P _FHi and P _MHi according to actual data, respectively comparing with the calculated P _PHv , P _FHv and P _MHv , according to the comparison result size controller using internal scheduling module to adjust each controller control The switch implements load balancing. This step is implemented by the following substeps:

S8.1, to be mobilized, the switch pool is set to be empty, and the switch pool is to be loaded with such elements EF::={transfer controller number, switch number, switch flow table statistics, transfer to controller number}; EP: :={Transfer controller number, switch number, packet-in packet number, transfer to controller number}; EM::={transfer controller number, switch number, other OF message statistics, transfer to controller Numbering}.

S8.2. Set the scaling factor x and y according to the specific application scenario of the cluster controller, and set the metric to adjust the switch controlled by each controller according to the packet-in packet number or the flow table number or other OF message statistics. In order to achieve the purpose of load balancing of each controller. The following formula is obtained:

P _i =(1-y)*(x*P _PHi +(1-x)*P _MHi )+y*P _FHi (Formula 1-10)

P _v =(1-y)*(x*P _PHv +(1-x)*P _MHv )+y*P _FHv (Formula 1-11)

Where: 0<=x<=1, 0<=y<=1;

Referring to P _i , S8.3 is executed for each controller in descending order, and after execution is completed, S8.6 is executed.

S8.3, for each controller, when its P _{i is} greater than P _v , 8.4 is performed, otherwise S8.5 is performed.

S8.4, calculating the list of switches that should be called out, this step is implemented by the following sub-steps:

S8.4.1, separately calculated for the controller

Set to B _posize ,

Set to B _fosize ,

Set to B _mosize , according to the scale factor previously set:

B _osize =(1-y)*(x*B _posize +(1-x)*B _mosize )+y*B _fosize (Equation 1-12)

Where 0<=x<=1, 0<=y<=1;

S8.4.2, based on the estimated data for each switch, the following formula is obtained:

B _inext =(1-y)*(x*P _inext +(1-x)*M _inext )+y*F _inext (Equation 1-13)

According to B _{inext, the} controllers that are controlled by the controller are arranged in ascending order to obtain a switch queue;

S8.4.3, selecting a switch from top to bottom from the above queue;

S8.4.4, cumulatively calculate the total sum of B _inext to be removed from the switch. If it is not greater than B _osize , continue to execute S8.4.4, otherwise execute S8.4.5;

S8.4.5, the number of the switch to be transferred, together with the controller number, B _{inext is} added to the pool to be transferred, where the number of the transferred controller is null.

S8.5, calculating the list of switches to be tuned, this step is implemented by the following sub-steps:

S8.5.1, calculated separately for the controller

Set to B _pisize ,

Set to B _fisize ,

Set to B _misize . According to the previously set scale factor can be obtained:

B _isize =(1-y)*(x*B _pisize +(1-x)*B _misize )+y*B _fisize (Equation 1-14)

Where 0<=x<=1, 0<=y<=1.

S8.5.2, _sorting the controllers in descending order according to the size of _Bize , and then transferring them to the switch according to the load of the controller, which is implemented by the following substeps:

S8.5.2.1: The switches are arranged in descending order according to the size of the _inbound switch pool B _inext to obtain the switch queue.

S8.5.2.2, for each controller, select the switch from top to bottom from the above queue, compare B _inext with B _isize respectively, and if B _inext >B _isize , select the next switch, if B _inext < B _isize , the transfer controller number of the element corresponding to the switch in the pool to be transferred is set to the local controller number, and B _isize =B _isize -B _inext , and then the next switch is checked until the last check is checked. Switch, FIG. 4 is a flow chart of tuning a switch into a controller in accordance with a preferred embodiment of the present invention. The specific steps are shown in FIG.

S8.6: The controller uses an internal switching control module to reconstruct the corresponding switch to implement load balancing. This step is implemented by the following sub-steps:

S8.6.1: The supernode shall transfer each element in the switch pool. If the number of the transferred controller is not empty, transfer the switch command to the switch corresponding to the outgoing controller number and the number of the transferred controller and execute S8.6.2. Otherwise skip this element.

S8.6.2, according to the slave-master mechanism exchange, implements switch scheduling; the slave-master mechanism exchange is set by the OpenFlow protocol, and is set to change the role of different controllers when controlling multiple controllers.

S9, the scheduling ends, and S2 is repeatedly executed at the next time interval.

The preferred embodiment proposes a solution to the cluster controller load imbalance phenomenon. The core idea of the solution is to use the mathematical model of the least squares method to packet-in packets, flow table numbers and other switches of each switch in the future time. The OF message is predicted, and the historical data of the same time period is taken as a reference, and the estimated value is averaged, and then the statistical information of each controller is obtained according to the new average value, and the statistical information of the controller and the historical maximum value are obtained. The ratio is compared with the load pressure value of the entire controller cluster. When the network traffic jitter is large, the number of switches controlled by each controller is adjusted in time, so that all the controllers receive the packet-in packet and the delivered flow table. The other OF messages interacting with the switch reach an equilibrium state, improving the control performance of the entire controller cluster. FIG. 5 is a flowchart of a cluster controller load balancing scheduling according to a preferred embodiment of the present invention. The entire processing flow is shown in FIG. 5.

Example 1:

FIG. 6 is a schematic diagram of the allocation of controllers and switches in the domain according to the preferred embodiment 1 of the present invention. As shown in FIG. 6, the SDN controller nodes A, B, and C are distributed in the domain. The switch controlled by the A controller node includes sw1, sw2, and sw3; the switch controlled by the B controller node includes sw4 and sw5; the switch controlled by the C controller node includes sw6, sw7, and sw8; combined with Table 1, Table 2, and Table 3, Tables 4, 5, 6, and 7 describe the preferred embodiment in detail.

S2: Collecting sample information by using an information collection module in the cluster controller. The sample data collected in this example is statistical information of different timestamps of a time period on a certain day of the week. At the same time, the actual measurement information in the period of the previous week is collected as a reference, so that the estimated information is closer to the actual value. Collecting sample data information is accomplished by the following substeps:

S2.1: In the same time interval, the information collection command of the controller of the controlled cluster is initiated by the super node in the cluster; all controllers A, B, and C of the cluster start collecting packet-in packets from the switches that are controlled by the cluster. Number and statistics of the number of flow tables and other OF messages sent to each switch;

S3, the super node uses the data processing module in the controller to predict the number of packet-in packets reported by the switches in each controller and the number of flow tables sent by the controller to the switch in the next time interval according to the mathematical model of the least squares method. And other OF messages are respectively implemented by step S4, step S5 and step S6:

S4: According to the timestamp, the super node A reads the packet-in packet statistics reported by each controller from the cluster distributed shared storage system, and implements the following substeps:

S4.1, where the number of timestamps is N=5, that is, the complete packet-in packet statistics information set of 5 timestamps is obtained; the complete packet-in packet statistics information set refers to the same controller at the same time. Packet-in packet statistics reported by all the different switches and the timestamp is the most recent;

S4.2: The set of complete packet-in packet statistics information of the timestamp is less than five, and the S2 is cyclically executed until the packet-in packet statistics information set of the five timestamps is obtained, and the obtained packet-in report is obtained. The statistics of the text are shown in Table 1;

Table 1: Initial statistics of packet-in packets

S4.3. According to the packet-in packet statistics information set, the super node obtains information statistics of the packet-in packets of five timestamps for each switch, P ₁ , P ₂ , P ₃ , P ₄ , and P _{5 .} That is, such data is obtained: P _i = {controller number, switch number, time stamp X, packet-in message statistics Y}.

S4.4. If the sum of the packet-in packet statistics in the controller of the five timestamps is greater than the historical packet-in packet statistics of the _controller , H _ipmax ,

S4.5. For each switch under the control of the cluster controller, the least squares mathematical model of Equation 1-1 is used, and the packets of each switch of the next timestamp are calculated according to formulas 1-2 and 1-3. In packet statistics P _inext , the prediction results are shown in Table 2.

The number of packets predicted by packet-in packet T6 in Table 2

S4.6. In order to make the estimated value closer to the actual value, the actual measured value of the received packet-in packet of the timestamp in one week is taken, and the estimated value is summed and averaged, and the result is shown in Table 3.

Table 3 Average number of actual measured values and estimated values of packet-in packets in one week

S5. According to the timestamp, the super node A reads the flow table statistics information reported by each controller from the cluster distributed shared storage system, and implements the following substeps:

S5.1, where the number of timestamps is N=5, that is, a complete flow table statistics information set of 5 timestamps is attempted; the complete flow table statistics information set refers to the same timestamp of the same controller to all the different switches. The flow table statistics delivered and the timestamp is the most recent;

S5.2, if less than 5 sets of complete flow table statistics information of the timestamp are obtained, S2 is executed cyclically until a flow table statistical information set of 5 timestamps is obtained;

S5.3. According to the flow table statistics information set, the super node obtains five time-stamped flow table statistics F ₁ , F ₂ , F ₃ , F ₄ , and F ₅ for each switch, and obtains such data: F _i = {controller number, switch number, timestamp X, flow table statistics Y}.

S5.4, if the sum of the flow table statistics in the above five timestamps is greater than the controller's historical flow table statistics H _ifmax , then

S5.5. For each switch managed by each controller in the cluster controller, use the least squares mathematical model of Equation 1-1, and calculate the switches of the next timestamp according to Equations 1-2 and 1-3. Flow table statistics F _inext ;

S5.6, in order to make the estimated value closer to the actual value, take the flow table of the time stamp within one week to calculate the actual measured value, and sum the average with the estimated value;

S6. According to the timestamp, the super node reads the statistics of the OF interaction information (hereinafter referred to as other OF information), which is reported by each controller, except for the packet_in, from the cluster distributed shared storage system, and is implemented by the following substeps:

S6.3. According to the foregoing other information collection set of OF information, the super node accumulates other OF information statistics M ₁ , M ₂ , . . . , M _{N of N} time stamps for each switch, that is, obtains such data: M _i = {controller number, switch number, timestamp X, other OF information statistics Y}.

S7, P _PHi , P _FHi and P _MHi are calculated according to the actual data using Equations 1-4, 1-5 and 1-6, respectively, and P calculated by Equations 1-7, 1-8 and 1-9, respectively. _PHv , P _FHv and P _MHv are compared, and the switches controlled by each controller are adjusted according to the comparison result to implement load balancing. This step is implemented by the following sub-steps:

S7.1, the embodiment assumes that the coefficient x=1, y=0, that is, according to the packet_in message reported by each switch, the number of controlled switches is adjusted;

S7.2, the buffer pool to be mobilized is set to null; the number of packet-in packets obtained by the five timestamps is calculated: P _PH1 = 1.09, P _PH2 = 0.39, P _PH3 = 1.11; The ratio of the number of received packet-in packets to the total number of maximum packet-in packets is estimated to be P _PHv = 0.95; the load pressure gauges of the controller are shown in Table 4. The above-mentioned to-be-switched switch pool loads such elements EP::={transfer controller number, switch number, packet-in packet number, and transfer to controller number}.

Table 4 Controller load pressure gauge based on packet-in message

控制器编号Controller number	AA	BB	CC	求和Summation
控制器编号Controller number	AA	BB	CC	求和Summation	预测流量Forecast traffic	7575	1313	6262	150150
控制器最大流量Controller maximum flow	6969	3333	5656	158158	预测流量Forecast traffic	7575	1313	6262	150150
控制器最大流量Controller maximum flow	6969	3333	5656	158158	P_PHi P _PHi	1.091.09	0.390.39	1.111.11	0.950.95

S7.3, according to the formulas 1-10 and 1-11, P _i = P _PHi , P _v = P _PHv , P S is performed on each controller in descending order with reference to P _i , and S7.7 is executed after execution is completed.

S7.4, for each controller, when its P _{i is} greater than P _v , S7.5 is performed, otherwise S7.6 is performed;

S7.5: Calculate the list of switches that should be called out. This step is implemented by the following substeps:

S7.5.1. This implementation adjusts the number of managed switches according to the packet_in packet data of each switch received by the controller. According to formula 1-12, B _osize =1*B _posize ;

S7.5.2, and then calculate B _inext according to the formula 1-13, according to the B _inext , the controllers controlled by the controller are arranged in ascending order to obtain the switch queue;

S7.5.3, selecting a switch from top to bottom from the above queue;

S7.5.4, the cumulative calculation should be removed from the switch B _inext sum, if not greater than B _osize then continue to execute S7.5.4, otherwise execute S7.5.5;

S7.5.5, the number of the switch to be transferred, together with the controller number, B _{inext is} added to the pool to be transferred, wherein the number of the transferred controller is empty, and the list to be dropped is obtained as shown in Table 5;

Table 5 Table 1 Before the adjustment, the controller resource sharing area to be mobilized

S7.6, calculating the list of switches to be transferred, this step is implemented by the following sub-steps:

S7.6.1, in this embodiment, according to the packet_in packet data reported by each switch received by the controller, the number of the controlled switches is adjusted, and B _isize =1*B _pisize according to the formula 1-14;

S7.6.2, in descending order of the controller according to the size of B _isize , and then transferred to the switch according to the load of the controller, which is implemented by the following substeps:

S7.6.2.1, the switches are arranged in descending order according to the size of the _inbound switch pool B _inext , and the switch queue is obtained;

S7.6.2.2, for each controller, select the switch from top to bottom from the above queue, compare B _inext with B _isize respectively, and if B _inext >B _isize , select the next switch, if B _inext < B _isize , the transfer controller number of the element corresponding to the switch in the pool to be transferred is set to the local controller number, and B _isize =B _isize -B _inext , and then the next switch is checked until the last check is checked. For the specific steps, see the flowchart shown in Figure 4. The list to be dropped obtained at this time is shown in Table 6.

Table 6 Table 1 to be adjusted after the controller resource sharing area to be mobilized

S7.7: Reconfigure the corresponding switch to implement load balancing. This step is implemented by the following substeps:

In S7.7.1, the supernode shall transfer each element in the switch pool. If the number of the transferred controller is not empty, transfer the switch command to the switch corresponding to the number of the switch-out controller and the number of the switch-in controller and execute S7.7.2. Otherwise skip this element;

S7.7.2, according to the slave-master mechanism exchange, implement switch scheduling; the slave-master mechanism exchange is set by the OpenFlow protocol, and is set to change the role of different controllers when controlling multiple controllers; Shown.

Table 7 in Table 1 according to the packet_in message adjusted after each controller load pressure gauge

S8, the current scheduling ends, and S2 is repeatedly executed at the next time interval.

Example 2:

Taking FIG. 6 as an example, the SDN controller nodes A, B, and C are distributed in the domain. The switch controlled by the A controller node includes sw1, sw2, and sw3; the switch managed by the B controller node includes sw4 and sw5; and the switch managed by the C controller node includes sw6, sw7, and sw8; the preferred embodiment will be described in detail below.

S4.2: If less than five complete packet-in packet statistics information collection time stamps are obtained, S2 is executed cyclically until a packet-in packet statistical information set of five timestamps is obtained.

S4.3. According to the packet-in packet statistics information set, the super node obtains information statistics of the packet-in packets of five timestamps for each switch, P ₁ , P ₂ , P ₃ , P ₄ , and P _{5 .} , that is, obtain such data: P _i = {controller number, switch number, time stamp X, packet-in message statistics Y};

S4.5. For each switch under the control of the cluster controller, use the least squares mathematical model of Equation 1-1, and calculate the packet of each switch of the next timestamp according to Equations 1-2 and 1-3. In message statistics information P _inext ;

In S4.6, in order to make the estimated value closer to the actual value, the actual measured value of the received packet-in packet of the timestamp in one week is taken, and the estimated value is summed and averaged.

S5.3. According to the flow table statistics information set, the super node obtains five time-stamped flow table statistics F ₁ , F ₂ , F ₃ , F ₄ , and F ₅ for each switch, and obtains such data: F _i = {controller number, switch number, timestamp X, flow table statistics Y};

S5.6, in order to make the estimated value closer to the actual value, take the flow table of the time stamp within one week to calculate the actual measured value, and sum the estimated value to average.

S6.5. For each switch managed by each controller in the cluster controller, use the least squares mathematical model (see Equation 1-1) to calculate and calculate other OF statistics of each switch of the next timestamp by regression analysis. _Inext , the formula is shown in Equation 1-2 and Equation 1-3;

S7.1, this embodiment assumes that the coefficient x=0.5, y=0.5, that is, according to the packet_in message reported by each switch, the flow table sent by each switch, and other OF messages, the managed switches are adjusted. number;

S7.2, the pool of the switch to be emptied is set to be empty; the number of packet-in packets obtained by the five timestamps is calculated: P _PH1 , P _{PH2 ,} and P _PH3 ; The table statistics are calculated: P _FH1 , P _FH2 and P _FH3 ; the other OF messages obtained by the 5 time stamps are statistically calculated: P _MH1 , P _MH2 and P _MH3 ; the calculation controller always estimates the received packets The ratio of the number of -in packets to the total number of packet-in packets is P _PHv ; the ratio of the total number of flow tables sent by the controller to the total number of flows is P _FHv ; The ratio of the estimated number of other OF messages to the total number of other largest OF messages is P _MHv ; the above-mentioned to-be-switched switch pool loads such elements EP::={transfer controller number, switch number, packet-in packet number, Transfer to controller number}; EF::={transfer controller number, switch number, flow table number, transfer to controller number}; EM::={transfer controller number, switch number, other OF messages , transfer to controller number};

S7.3, P _i and P _v are obtained according to the formulas 1-10 and 1-11, S7.4 is performed on each controller in descending order with reference to P _i , and S7.7 is executed after the execution is completed.

S7.5.1. In this implementation, the number of controlled switches is adjusted according to the packet_in packet of each switch received by the controller, the number of flow tables sent by the controller, and other OF message data. Available:

B _osize = ( _1-0.5 ) * (0.5 * B _posize + 0.5 * B _fosize ) + 0.5 * B _mosize ;

S7.5.3, selecting a switch from top to bottom from the above queue;

S7.5.4, the cumulative calculation should be removed from the switch B _inext sum, if it is not greater than B _osize , continue to execute S7.5.4, otherwise execute 7.5.5;

S7.5.5, adding the number of the switch to be transferred, together with the controller number, B _inext to the pool of the switch to be transferred, wherein the number of the transferred controller is empty;

S7.6.1, in this embodiment, according to the packet_in message reported by each switch received by the controller, the number of flow tables sent by the controller, and other OF message data, the number of controlled switches is adjusted according to formula 1. -14 available

B _isize = ( _1-0.5 ) * (0.5 * B _pisize + 0.5 * B _fisize ) + 0.5 * B _misize ;

S7.6.2.2, for each controller, select the switch from top to bottom from the above queue, compare B _inext with B _isize respectively, and if B _inext >B _isize , select the next switch, if B _inext < B _isize , the transfer controller number of the element corresponding to the switch in the pool to be transferred is set to the local controller number, and B _isize =B _isize -B _inext , and then the next switch is checked until the last check is checked. Switch, the specific steps are shown in Figure 4.

S7.7.2, according to the slave-master mechanism exchange, implement switch scheduling; the slave-master mechanism exchange is set by the OpenFlow protocol, and is set to change the role of different controllers when controlling multiple controllers.

Example 3:

FIG. 7 is a schematic diagram of the intra-domain cascade controller and switch allocation according to a preferred embodiment 3 of the present invention. As shown in FIG. 7, an SDN controller node A, B, C and a super controller node D are distributed in the domain. The switch controlled by the A controller node includes sw1, sw2, and sw3; the switch managed by the B controller node includes sw4 and sw5; and the switch managed by the C controller node includes sw6, sw7, and sw8; the preferred embodiment will be described in detail below.

S3, the super node D uses the data processing module in the controller to predict the number of packet-in packets reported by the switches in each controller and the flow table sent by the controller to the switch in the next time interval according to the mathematical model of the least squares method. The number and other OF messages are implemented by Step 4, Step 5 and Step 6 respectively:

S4: According to the timestamp, the super node D reads the packet-in packet statistics reported by each controller from the cluster distributed shared storage system, and implements the following substeps:

S5. According to the timestamp, the super node D reads the flow table statistics information reported by each controller from the cluster distributed shared storage system, and implements the following substeps:

S6. According to the timestamp, the super node D reads the statistics of the OF interaction information (hereinafter referred to as other OF information) except for the packet_in reported by each controller from the cluster distributed shared storage system, and implements the following substeps:

S7.2, the pool of the switch to be emptied is set to be empty; the number of packet-in packets obtained by the five timestamps is calculated: P _PH1 , P _{PH2 ,} and P _PH3 ; The table statistics are calculated: P _FH1 , P _FH2 and P _FH3 ; the other OF messages obtained by the 5 time stamps are statistically calculated: P _MH1 , P _MH2 and P _MH3 ; the calculation controller always estimates the received packets The ratio of the number of -in packets to the total number of packet-in packets is P _PHv ; the ratio of the total number of flow tables sent by the controller to the total number of maximum flow tables is P _FHv ; The ratio of the estimated number of other OF messages to the total number of other largest OF messages is P _MHv ; the above-mentioned to-be-switched switch pool loads such elements EP::={transfer controller number, switch number, packet-in packet number, Transfer to controller number}; EF::={transfer controller number, switch number, flow table number, transfer to controller number}; EM::={transfer controller number, switch number, other OF messages , transfer to controller number};

S7.5.3, selecting a switch from top to bottom from the above queue;

S7.6.1, in this embodiment, according to the packet_in message reported by each switch received by the controller, the number of flow tables sent by the controller, and other OF message data, the number of controlled switches is adjusted according to formula 1. -14 available;

In S7.7.1, the supernode shall transfer each element in the switch pool. If the number of the transferred controller is not empty, transfer the switch command to the switch corresponding to the number of the switch-out controller and the number of the switch-in controller and execute S7.7.2. Otherwise skip this element

In another embodiment, a software is provided that is configured to perform the technical solutions described in the above embodiments and preferred embodiments.

In another embodiment, a storage medium is also provided, in which the above software is stored, including but not limited to an optical disk, a floppy disk, a hard disk, an erasable memory, and the like.

It will be apparent to those skilled in the art that the various modules or steps of the present invention described above can be implemented by a general-purpose computing device that can be centralized on a single computing device or distributed across a network of multiple computing devices. Alternatively, they may be implemented by program code executable by the computing device such that they may be stored in the storage device by the computing device and, in some cases, may be different from the order herein. The steps shown or described are performed, or they are separately fabricated into individual integrated circuit modules, or a plurality of modules or steps thereof are fabricated as a single integrated circuit module. Thus, the invention is not limited to any specific combination of hardware and software.

The above description is only the preferred embodiment of the present invention, and is not intended to limit the present invention, and various modifications and changes can be made to the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and scope of the present invention are intended to be included within the scope of the present invention.

Industrial applicability

As described above, the method and apparatus for load balancing provided by the embodiments of the present invention have the following beneficial effects: solving the problem of uneven load of the cluster SDN controller in the related art, and performing load balancing through the estimated data, the manner It is highly achievable and improves the resource utilization efficiency of the cluster SDN controller.

Claims

A method of load balancing, including:

Collecting data of characterization parameters of a plurality of switches controlled by each of the SDN controllers of the cluster software defined network SDN controller;

Estimating data of the characterization parameters of the plurality of switches at a next moment according to data of the characterization parameters of the plurality of switches;

And performing load balancing on each SDN controller of the cluster SDN controller according to the estimated data of the plurality of switches at the next moment.
The method of claim 1, wherein the characterization parameters comprise at least one of: a packet-in message, a flow table number, and other OF messages.
The method of claim 1, wherein estimating data of the characterization parameters of the plurality of switches at a next moment comprises:

The data of the characterization parameters of the plurality of switches at the next moment is estimated using a least squares method.
The method of claim 3, wherein estimating data of the characterization parameters of the plurality of switches at a next moment using a least squares method comprises:

For each switch, the data of the characterization parameter of each switch at the next moment is calculated using the following least squares mathematical model: y = α + β * x, where x is the characterization parameter at the current time Data, y is the data of the characterized parameter at the next moment; the parameters α and β are calculated according to the following formula:

Where n is the number of times in the current statistical period, X is the timestamp corresponding to the time, and Y is the value of the characterization parameter of each switch.
The method of claim 1, wherein the data characterizing the parameter comprises sample history data characterizing the parameter and data of the characterizing parameter at each moment in the current statistical period; data according to the characterizing parameters of the plurality of switches Estimating the data of the characterization parameters of the plurality of switches at the next moment includes:

And estimating, according to the sample historical data of the characterization parameters of the multiple switches and the data of the characterization parameters at each moment in the current statistical period, the data of the characterization parameters of the multiple switches at the next moment.
The method of claim 5, wherein estimating data of the characterization parameters of the plurality of switches at a next moment comprises:

Estimating, according to the data of each time in the current statistical period, the data of the plurality of switches at the next moment;

The sample history data of the characterization parameter is summed with the estimated data of the characterization parameter at the next moment.
The method according to claim 1, wherein load balancing the respective SDN controllers of the cluster SDN controller according to the estimated data of the plurality of switches at the next moment comprises:

Calculating, according to the estimated data of the plurality of switches, the data of the plurality of switches at a next moment, calculating a ratio P YHi of the estimated value of each of the characterization parameters to a corresponding historical maximum value of each controller:

Calculating a ratio P YHv of the total estimated value of the entire controller cluster to the total estimated value and the total maximum value:

P YHi is compared with P YHv , and the controller controlled by the controller is adjusted according to the result of the comparison.
The method of claim 7 wherein

P YHi is calculated by the following formula:

Where n is the number of switches managed by each controller, and Y inext is an estimated value of the characterization parameters of each switch managed by the controller, and H iymax is a historical maximum value of each characterization parameter of each switch managed by the controller; and / or,

P YHv is calculated by the following formula:

Wherein m represents the number of controllers, n represents the number of switches controlled by each controller, and Y inext is an estimated value of the characteristic parameters of each switch managed by the controller, and H iymax is the switch managed by the controller. Characterizes the historical maximum of the parameter.
The method according to claim 7, wherein comparing the P YHi with the P YHv and adjusting the controller controlled by the controller according to the result of the comparison comprises:

A scaling factor is set for a plurality of different characterization parameters, wherein the scaling factor is set to represent a proportion of each characterization parameter in the comparison process.
The method according to claim 7, wherein comparing the P YHi with the P YHv and adjusting the controller controlled by the controller according to the result of the comparison comprises:

For each controller, when its P YHi is greater than P YHv , calculate the switch that should be called out, and call out the switch according to the calculation result; and/or,

For each controller, when its P YHi is less than P YHv , calculate the switch that should be transferred and transfer it to the switch according to the calculation result.
The method of claim 10, wherein calculating the switch that should be invoked and calling the switch according to the calculation result comprises:

For controller calculation
Set to B yosize and obtain B osize according to the previously set scale factor, wherein the scale factor is set to indicate the proportion of the plurality of characterization parameters in the comparison process;

According to the estimated data for each switch, B inext =Y inext ;

According to the value of B inext , the controllers controlled by the controller are arranged in ascending order to obtain a switch queue;

Selecting a switch from top to bottom from the switch queue, and cumulatively calculating a total of B inext to be removed from the switch, and if not greater than B osize , continuing to select the switch;

The switch to be transferred is added to the switch pool to be mobilized, and the pool to be mobilized is set to store switch information to be mobilized.
The method according to claim 10, wherein calculating the switch to be entered and transferring the switch according to the calculation result comprises:

For controller calculation
Set to B yisize and obtain B isize according to the previously set scale factor, wherein the scale factor is set to indicate the proportion of the plurality of characterization parameters in the comparison process;

The controller is sorted in descending order according to the value of B isize , and then transferred to the switch according to the controller load.
The method of claim 12, wherein sequentially switching the switch according to the controller load comprises:

The switches are arranged in descending order according to the value of B inext of each switch to be transferred in the switch pool to be mobilized, and the switch queue is set to store the switch information to be mobilized;

For each controller, select the switch from top to bottom from the above queue, compare B inext and B isize respectively, if B inext >B isize , select the next switch, if B inext <B isize , it will be The transfer controller number of the element corresponding to the switch in the switch pool is set to the local controller number, and B isize =B isize -B inext , and then the next switch is checked until the last switch.
A load balancing device comprising:

a collection module configured to collect data of characterization parameters of a plurality of switches controlled by each of the SDN controllers of the cluster software defined network SDN controller;

The estimating module is configured to estimate data of the characterization parameter of the plurality of switches at a next moment according to data of the characterization parameters of the plurality of switches;

The load balancing module is configured to load balance each SDN controller of the cluster SDN controller according to the estimated data of the plurality of switches at the next moment.