CN109921991B

CN109921991B - SDN controller deployment method based on Dinkelbach

Info

Publication number: CN109921991B
Application number: CN201910031586.0A
Authority: CN
Inventors: 邓清勇; 黄少波; 马玲; 王振振; 曾艳; 田淑娟; 朱容
Original assignee: Xiangtan University
Current assignee: Xiangtan University
Priority date: 2019-01-14
Filing date: 2019-01-14
Publication date: 2021-06-04
Anticipated expiration: 2039-01-14
Also published as: CN109921991A

Abstract

The invention provides a method for deploying an SDN controller based on Dinkelbach. Firstly, abstracting an SDN network topological structure into an undirected graph by obtaining switch node information, forming a parent domain by all nodes, then performing domain division by judging controller capacity conditions, iteratively searching and storing the optimal node positions of controller deployment in the domain by using a Dinkelbach method, and finally determining the deployment positions and the quantity of all controllers under the topological structure. According to the invention, the load condition of the controller is dynamically adjusted according to the size of the domain, the deployment result of inter-domain load balance can be realized, the network flexibility is increased, the utilization rate of the SDN controller is greatly improved, in addition, the Dinkelbach method improves the convergence rate, and the method can be suitable for network topological structures with different scales.

Description

SDN controller deployment method based on Dinkelbach

Technical Field

The invention belongs to the field of software defined networks, and particularly relates to a deployment method of a software defined network controller.

Background

Software Defined Networking (SDN) supports centralized Network state control by decoupling a control plane and a data plane, realizes transparency of underlying Network infrastructure to upper layer applications, provides a flexible open interface and Software programming capability for the Network, and can support scheduling and allocation of Network resources as required, thereby effectively solving the problems of blocked resource expansion, poor networking flexibility, difficulty in meeting emerging service requirements and the like faced by the current Network system.

With the expansion of a network topology structure and the increase of network flow, a single SDN controller is limited by aspects such as expansibility and reliability when managing a network, and thus, the network quality under a large-scale network topology structure is difficult to guarantee. Managing deployment locations of multiple SDN controllers for a given network topology is therefore an efficient way to achieve fast network response. In order to solve the SDN controller deployment problem, the deployment problem can be optimized by researching factors such as delay overhead, load, capacity limitation, node similarity and the like in a network structure.

Disclosure of Invention

The invention provides a method for deploying an SDN controller based on Dinkelbach. Dinkelbach is an improved binary search algorithm with a higher convergence rate compared to the binary method.

The deployment of the SDN controller at least comprises the following steps:

step 1: and acquiring the information of the switch nodes and abstracting the SDN network topology structure into an undirected graph. And acquiring the average message sending quantity and the average message sending rate of the nodes of the switch, wherein all the nodes form a parent domain. The parent domain represents a domain containing all switch nodes under the current SDN network topology structure, and other domains obtained by dividing the parent domain are sub-domains. Since the SDN controller is generally selected to be deployed on a certain switch node when the SDN controller is deployed, a node with the optimal controller deployment generated in any sub-domain in an iterative process is hereinafter referred to as an optimal deployment node, and a node that ultimately determines to deploy the controller is referred to as a controller deployment node. The initialization controller deploys node information, including location information and quantity information.

Step 2: it is determined whether the domain satisfies the controller capacity condition. If the controller capacity condition is met, go to step 4, otherwise go to step 3.

And step 3: and for each domain, searching for the optimal node deployed in the domain based on the deployment benefits, dividing the domain according to the association condition of the optimal node deployed, and recording the information of the optimal node deployed. The method specifically comprises the following substeps:

step 3.1: deleting original optimal node deployment information of the domain, and adding 1 to the total number of the nodes of the controller;

step 3.2: optionally, 2 switch nodes are used as a possible controller deployment situation to generate all possible deployment situations;

step 3.3: calculating the time delay overhead of each deployment situation;

step 3.4: calculating the switch node concentration of each deployment situation;

step 3.5: calculating deployment benefits, and searching for deployment conditions for deploying optimal nodes through Dinkelbach;

step 3.6: dividing the domain into 2 sub-domains according to the association condition of the 2 optimal deployment nodes, and recording the position information and the association information of the 2 optimal deployment nodes;

step 3.7: the 2 sub-fields are returned to step 2.

And 4, step 4: judging whether the domain is a mother domain, and if the domain is the mother domain, executing the step 4.1; if the field is a subdomain, step 4.2 is performed. The method specifically comprises the following substeps:

step 4.1: searching 1 optimal node for deployment in the parent domain, storing the position information and the associated information of the optimal node for deployment, and storing the number of the nodes of the controller as 1;

step 4.2: and 3.6, acquiring and storing the position information and the associated information of the current sub-domain deployment optimal node, finishing the division of the current domain, and entering the step 5 after the division of all the domains is finished.

And 5, outputting the stored controller deployment node position information and quantity information, and finishing deployment.

The invention has the advantages that:

1. when the domain is divided into two domains, the load condition of the controller is dynamically adjusted, the purpose of load balancing can be achieved, the network processing efficiency is improved, the flexibility of the network is improved to a certain extent, and the utilization rate of the SDN controller is improved.

2. The Dinkelbach algorithm is used when the optimal deployment condition is searched, convergence is accelerated, and the advantage of the method is more obvious along with the enlargement of the network scale.

Drawings

FIG. 1 is a flow chart of the invention.

Figure 2 is a parent domain of an SDN network topology.

Fig. 3 is an intra-domain optimal deployment scenario.

Fig. 4 is the result of the first time domain division.

FIG. 5 is the result of the second time domain division.

Fig. 6 is a search result of the optimal controller directly on the parent domain.

Detailed Description

As shown in fig. 1, the method comprises the following specific steps:

step 1: obtaining switch node information and abstracting SDN network topology structure into an undirected graph G (V, L), wherein V is a set of switches in the network, and V is { V ═ V }₁,v₂,…,v_nN is the number of switch nodes in the SDN network structure, L is a link distance matrix between switches, and_ijrepresenting a switch node v_iTo the switch node v_jThe link distance of (c). Obtaining average message sending rate t of each switch node_iAnd average message transmission amount u_iAll switch nodes constitute a parent domain. Initializing the controller deployment node information, and setting the controller deployment node position information to be null, and setting the number to be 0. It is assumed that the switches used are all OpenFlow switches. The network topologies are shown by fig. 2 to 6, wherein fig. 2 to 5 belong to the same network topology and fig. 6 belongs to another smaller scale network topology.

Step 2: it is determined whether the domain satisfies the controller capacity condition. Assuming that the capacity of each controller to the message volume is D and the domain is SV_now. The capacity condition of the controller is

I.e. the controller capacity is not less than the domain SV_nowTotal amount of messages in. If the controller capacity condition is met, go to step 4, otherwise go to step 3.

step 3.2: optionally 2 exchanger sectionsThe point acts as a controller node, creating one possible deployment scenario for the controller, and similarly all deployment scenarios within the domain may be created. Assuming the domain is SV_nowSV can be recorded_nowThe set of all possible deployment scenarios within is { X₁,X₂,…,X_mAnd m is the total number of possible deployment cases.

Step 3.3: the delay overhead for each deployment case is calculated.

Defining deployment scenario X_dThe delay overhead of (d ═ 1,2, … m) is: d_d＝Dt_u+Dt_p。Dt_u、Dt_pRespectively represent deployment scenarios X_d(d ═ 1,2, … m) next 2 controller nodes v_u、v_pThe overhead of (a). Controller node v_u、v_pThe overhead representation method is the same with controller node v_uFor example, assuming that the processing rate of each controller to the message is H, and given that k is the number of switch nodes in the domain, the controller node v is represented by the following formula_uThe delay overhead of (2):

wherein Dp is_iuRepresenting a switch node v_iAnd controller node v_uInter propagation delay, wherein L_iuFor switch node v_iTo the controller node v_uControl link length of, L_iuCan be obtained by a link distance matrix L in an undirected graph G, and a c tableThe propagation rate of electromagnetic waves is shown, and since the controller is generally selected to be deployed on a certain switch node when the controller is deployed, the controller node v is not calculated_uPropagation delay to the present controller, Dp_uu＝0；Dtr_iRepresenting a switch node v_iTransmission delay of (u)_iAnd t_iRespectively a switch node v_iAverage message sending amount and average message sending rate; dh_iRepresenting a switch node v_iProcessing latency of the transmitted message at the controller. y is_iuRepresenting a switch node v_iAnd controller node v_uIs associated with y_iu1 denotes a switch node v_iAnd controller node v_uIs associated with y_iuDenotes a switch node v by 0_iAnd controller node v_uAre not associated.

The method of determining the association of a switch node with a controller node is as follows: the total cost from 2 controller nodes in the domain to the switch nodes associated with the controller nodes is required to be minimum, and the following model is established:

D_iuand D_ipRespectively representing switch nodes v_iTo the controller node v_uAnd v_pInter-node overhead of (1), wherein D_iuCan be expressed as: d_iu＝2·Dp_iu+Dtr_i+Dh_i，D_ipAnd D_iuThe expression mode is consistent.

Indicating controller load balancing conditions, i.e. controller node v_uAmount of messages processed and controller node v_pThe absolute value of the difference of the processed message quantity is less than delta (0 < delta < 1) in the occupation ratio of the total message quantity in the domain, and the delta tableThe allowable maximum load difference value accounts for the proportion of the total message volume in the domain, and the lower the delta is, the smaller the load difference value between the domains is, and the better the load balance of the controller is.

y_iu∈{0,1}And y_ip＝1-y_iuMeaning that each switch node can only be associated with 1 of 2 controller nodes. y is_uu＝1,y_pp1 denotes associating a controller node with itself. The association condition of the switch node and the controller node can be obtained through the model.

Step 3.4: the switch node concentration for each deployment case is calculated. Defining deployment scenario X_dThe switch node concentration of (d ═ 1,2, … m) is:

where the denominator represents the sum of the link lengths of the 2 controller nodes to the associated switch node, L_iu,L_ipRespectively representing switch nodes v_iTo the controller node v_u,v_pThe link length of (c). Since the number k of switch nodes in the domain does not change, the larger the denominator is, the larger the controller node v is_uAnd v_pThe larger the sum of the link lengths to the switch nodes with which they are associated, results in a switch node concentration of medium Cr_dAnd (4) descending.

Step 3.5: and calculating deployment benefits, and searching the deployment condition of the optimal deployment node through Dinkelbach to obtain 2 optimal deployment nodes. The deployment benefit is expressed as

D_dFor deployment case X_dLower delay overhead, Cr_dIs deployment case X_dLower switch node concentration. Thus for D_dLarger and larger Cr_dDeployment benefits for smaller deployment scenarios

The lower the objective function is set to

Searching using Dinkelbach

The 2 controller nodes corresponding to the obtained optimal deployment situation are the optimal deployment nodes, and the result is shown in fig. 3, which represents that the 2 optimal deployment nodes in the domain have been determined.

Step 3.6: and dividing the domain into 2 subdomains according to the association conditions of the 2 optimal nodes for deployment. If the domain is SV_nowThe obtained 2 optimal nodes for deployment are v_uAnd v_pThen SV is set_nowDivision into 2 sub-fields SV_uAnd SV_p. Then will be reacted with v_uAnd v_pAssociated nodes are respectively divided into SV_uAnd SV_pIn, e.g. node v_iIs y_iu＝1,y_ipWhen v is equal to 0, v is_iDivision into sub-fields SV_uAnd (4) the following steps. The first iteration result is shown in fig. 4, the parent domain is divided into 2 sub-domains, and the second iteration result is shown in fig. 5, the 2 sub-domains are divided into 4 sub-domains. And recording the position information and the association condition of the optimal node for deployment.

Step 3.7: the 2 sub-fields are returned to step 2.

step 4.1: the time domain is a mother domain, the network topology structure can be known to be small in scale, and the capacity requirement of the controllers in the domain can be met by deploying 1 controller. 1 optimal controller deployment node is directly searched in the parent domain, and the searching method is Dinkelbach. The search objective function is replaced by

Wherein

The deployment benefit when only 1 switch node is selected as the controller deployment node is shown. Assume controller deploys node v_u，

Representing the switch node concentration when only 1 controller is deployed. D_s＝Dt_uThe delay overhead, Dt, when only 1 controller is deployed_uThe expression is as follows:

at this time y_iu≡ 1, meaning that the controller deployment node is associated with all switch nodes of the parent domain. An optimal controller deployment node for the parent domain is searched, and an example result is shown in fig. 6. And (5) storing the position information of the optimal controller deployment node, wherein the number of the controller deployment nodes is stored as 1, and entering the step 5.

Step 4.2: if the time domain is a sub-domain, it indicates that a final sub-domain has been obtained through one or more iterations, and the position information of the deployment optimal node of the final sub-domain can be obtained through step 3.6 in the previous iteration, and is used as a controller deployment node, the position information and the association condition are stored, and the division of the current sub-domain is finished. And 3, after all the subdomain partitions are finished, acquiring and storing the quantity information of the controller deployment nodes obtained in the step 3.1, and entering the step 5.

And 5, outputting the stored controller deployment node position and quantity information, and finishing deployment.

Claims

1. A method of deploying an SDN controller, the method comprising at least the steps of:

step 1: abstracting an SDN network topological structure into an undirected graph, acquiring information of each switch, including average message sending quantity and average message sending rate, dividing all nodes into a parent domain, and initializing node information of a controller;

step 2: judging whether the domain meets the capacity condition of the controller, if so, entering a step 4, otherwise, entering a step 3;

and step 3: for each domain, searching for the optimal node deployed in the domain based on the deployment benefits, dividing the domain according to the association condition of the optimal node deployed, and recording the information of the optimal node deployed, specifically comprising the following substeps:

step 3.3: calculating the time delay overhead of each deployment situation;

step 3.7: returning 2 sub-fields to step 2;

and 4, step 4: judging whether the domain is a mother domain, if so, executing a step 4.1, otherwise, executing a step 4.2, wherein the specific substeps are as follows:

step 4.2: acquiring and storing the position information and the associated information of the current sub-domain deployment optimal node through the step 3.6, finishing the division of the current domain, and entering the step 5 after the division of all the domains is finished;

and 5: and outputting the position, the quantity information and the associated information of the controller nodes, and finishing the deployment.

2. The method of claim 1, wherein the controller capacity condition in step 2 is an iterative determination condition, and the controller is configured to deploy the SDN controllerThe capacity condition is

D represents the capacity of the controller to the message volume, and the domain is SV_nowI.e. the controller capacity is not less than the domain SV_nowAverage total message size, u_iIs a switch node v_iAverage message transmission amount.

3. The SDN controller deployment method of claim 1, wherein the objective function of Dinkelbach search in step 3.5 is

Wherein

And expressing the deployment benefit, specifically expressing as follows:

Cr_dis deployment case X_dThe lower switch node concentration is specifically represented as:

k is the number of switch nodes in the domain, L_iu,L_ipRespectively representing switch nodes v_iTo the controller node v_u,v_pLink length of y_iu,y_ipRespectively representing switch nodes v_iAnd controller node v_u,v_pThe association of (2);

D_dfor deployment case X_dOverhead of D_d＝Dt_u+Dt_p，Dt_u、Dt_pRespectively represent deployment scenarios X_dThe next 2 controller nodes v_u、v_pOverhead of where Dt_uThe concrete expression is as follows:

Dtr_irepresenting a switch node v_iThe transmission delay is specifically expressed as:

wherein u is_iAnd t_iRespectively a switch node v_iAverage message sending amount and average message sending rate;

Dh_irepresenting a switch node v_iThe processing delay of the sent message at the controller is specifically expressed as:

wherein, H is the processing rate of the controller to the message;

Dp_iurepresenting a switch node v_iAnd controller node v_uThe propagation delay of (a) is specifically expressed as:

wherein c represents an electromagnetic wave propagation rate;

similarly, according to Dt_uThe calculation formula of (a) obtains the node v of the controller_pCost Dt of_p。