CN105282242A - Multiattribute information-based inter-data center data transmission scheduling method - Google Patents

Abstract

The invention relates to a multiattribute information-based inter-data center data transmission scheduling method. The method uses a hierarchical method for analyzing the subordination relation among different attributes and uses an information entropy theory for determining weights of different attributes, thereby providing comprehensive evaluation of a multiattribute information-based transfer data center. By establishing a time expansion graph model, the multiattribute information-based inter-data center data transmission scheduling problem is converted into the minimum cost flow problem, and an algorithm for solving the problem is provided. The beneficial effects are that the problem that the multiattribute information-based inter-data center data transmission scheduling is low in efficiency is solved; and by means of obtained weights, the multiattribute information-based method has the best optimization performance in the dimension of the highest weight.

Description

Data transmission scheduling method between a kind of data center based on multiattribute information

Technical field

The present invention relates to data center's technical field, particularly relate to the data transmission scheduling method between a kind of data center based on multiattribute information.

Background technology

In recent years, the developing rapidly at line service along with cloud computing and the Internet, the Internet giant establishes multiple large-scale data center in all parts of the world, by the data Replica of user to multiple geographical position to improve QoS of customer.Need between these distributed data centers to transmit a large amount of data.Service traffics between data center have occupied the data center's outlet bandwidth close to half.But because data center is in the service condition of busy inequality, cause the link between data center not to be utilized effectively.Therefore, how carrying out transmitting and scheduling to the data between data center is current problem demanding prompt solution.Because the access bandwidth service condition of data center changed along with the time, traditional routing technology can not solve existing problem.

Therefore, by using the data center of access bandwidth free time to carry out transfer, again data are forwarded when interim data center possesses the bandwidth affluence path to destination data center.Utilize such store-and-forward mechanism, to improve data center's access bandwidth utilance, reduce traffic transport cost between data center, to reduce the Path selection Optimization Mechanism that between territory, delivery flow rate is target.But in the data transfer path selection course of reality, often can run into manager and not only be concerned about data center's access bandwidth service condition, and also there is requirement for the expense of link, service quality; Or manager is higher and be not very high situation to other attribute specifications to a certain item attribute specification wherein.So only consider that the Path selection of single attribute or single optimization aim is incomplete.

Summary of the invention

The present invention overcomes above-mentioned weak point, object is to provide the data transmission scheduling method between a kind of data center based on multiattribute information, this method utilizes the subordinate relation between the methods analyst different attribute of stratification, and use the weight of information entropy theory determination different attribute, thus the overall merit at interim data center under providing multiattribute information; By expander graphs model settling time, data transmission scheduling problem form between the data center based on multiattribute information is turned to minimum price flux problem.The inefficient problem of data transmission scheduling between the data center that present method solves multiattribute information.

The present invention achieves the above object by the following technical programs: the data transmission scheduling method between a kind of data center based on multiattribute information, comprises the steps:

(1) utilize analytic hierarchy process (AHP), the subordinate relation between multiple attribute is analyzed;

(2) according to analyzing the property value difference distribution obtained, utilizing the relative weighting that comentropy sets a property, obtaining overall merit formula, and obtain evaluation of estimate by formula;

(3) temporal extension figure is obtained to evaluation of estimate modeling, by temporal extension figure, the data transmission scheduling problem between the data center based on multiattribute information is converted into minimum price flux problem;

(4) calculate solution minimum price flux problem by negative cost loop algorithm, obtain the path of overall merit optimum, i.e. minimum cost stream.

As preferably, multiple attribute information is divided into different levels according to dependence by the analytic hierarchy process (AHP) of described step (1), using the structure structure Multiple Attribute Decision Problems of stratification, quantitatively obtaining property value by monitoring.

As preferably, described Multiple Attribute Decision Problems is the order of preference evaluating different alternate data center based on the decision matrix of multiattribute information, and decision matrix is as follows:

Wherein, the individual core attribute of number n " n by the 1st layer " of decision attribute determines, the row representative of matrix has m decision-making alternative { A _i, row represent n different decision attribute { X _j, each element x of matrix _ijthe jth item property value of representative data center i.

As preferably, it is as follows that described step (2) obtains overall merit formula process:

1) all types of attribute is converted into cost type attribute;

2) based on decision matrix, will about attribute X _jproperty value be normalized to evaluation of estimate, normalization formula is $X_{ij}=\frac{x_{ij}}{\underset{i=1}{\overset{m}{\Σ}}{x}_{ij}};$

3) according to shannon formula, computation attribute X _jcomentropy power wherein, k is constant, for making 0≤Ej≤1, and order

4) degree of deviation of information is established to be defined as d _j=1-E _j;

5) relative weighting is made and use arithmetic weight averaging operator to be weighted evaluation of estimate, solving overall merit formula is $U_i^{final}(X_{i1},X_{i2},\mathrm{...},X_{in})={\Σ}_{j=1}^{n^{*}}{w}_j{X}_{ij},\∀i.$

As preferably, described all types of attributes are total up to two kinds, cost type attribute and profit evaluation model attribute.

As preferably, the negative cost loop algorithm of described step (4) is data scheduling algorithm between the data center based on multiattribute information; This algorithm is calculated, for transfer of data selects the path of an overall merit optimum when data dispatch according to the attribute value information of data center's monitoring tool collection by the manager of data center.

As preferably, described algorithm flow is as follows:

A () calculates a feasible flow by Edmonds-karp algorithm;

B () generates corresponding residual network based on feasible flow, whether there is negative cost loop with in Ballman-ford algorithm cycle criterion figure;

If c () exists negative cost loop, eliminate negative ring and upgrade residual network, until there is not negative cost loop, obtaining final stream, be minimum cost stream; If do not exist, this feasible flow is minimum cost stream.

Beneficial effect of the present invention is: (1) solves the inefficient problem of data transmission scheduling between the data center of multiattribute information; (2) according to the weight obtained, the method based on multiattribute information has the highest Optimal performance in the dimension that weight is the highest.

Accompanying drawing explanation

Fig. 1 is the structural representation based on multiattribute information data scheduling problem attribute hierarchies of the embodiment of the present invention.

Embodiment

Below in conjunction with specific embodiment, the present invention is described further, but protection scope of the present invention is not limited in this:

Embodiment: the data transmission scheduling method between a kind of data center based on multiattribute information, comprises the steps:

The first step, utilizes analytic hierarchy process (AHP), analyzes the subordinate relation between multiple attribute;

Multiple attribute information is divided into different levels according to dependence by described analytic hierarchy process (AHP), using the structure structure Multiple Attribute Decision Problems of stratification, quantitatively obtaining property value by monitoring.As shown in Figure 1, wherein the 0th layer is the final overall assessment to multiple attribute information to the attribute hierarchies structure based on data dispatch problem between multiattribute information data center of the present embodiment.The 1st layer of core attribute being manager and needing to be concerned about, as the access bandwidth cost at interim data center, the service quality, data center's energy use efficiency (PUE, powerusageeffectiveness) etc. of access link.2nd layer is the decomposition to the 1st layer of core attribute, as service quality can be broken down into bandwidth sum time delay 2 sub-attributes.The decomposition of attribute can obtain according to the daily management experience of the classification of attribute and administrative staff, such as, the service quality of link is as relevant in the bandwidth of link and time delay etc. with several basic key element, and the popularity of content is relevant with the geographical position at user place and the population size of position etc.If the 2nd layer of attribute also has sub-attribute, the 3rd layer can be constructed further, the 4th Rotating fields is until kth-1 layer.Kth layer is each alternative, represents the source of above-mentioned attribute value, i.e. each alternative interim data center.

Below based on the hierarchical structure of attribute, the decision matrix constructed based on multiattribute information evaluates the order of preference at different alternate data center.First original decision matrix is provided such as formula shown in (1).

The row representative of matrix has m decision-making alternative { A _i, be m alternative interim data center in the problem studied herein.Row represent n different decision attribute { X _j, comprise bandwidth cost, bandwidth, round-trip delay, data center's energy use efficiency etc.Each element x of matrix _ijthe jth item property value of representative data center i.According to the subordinate relation between Hierarchical structure model proposed above and attribute, herein by original decision matrix D _originresolve into some sub-decision matrixs.Each sub-decision matrix can top-downly successively construct.Such as, in illustrated structural model, the 1st layer can be become by the sub-attribute construction of the 2nd layer such as formula shown in (2) and formula (3) with the decision matrix of content popularit about service quality QoS.

Service quality decision matrix D _qoSwith content popularit decision matrix D _poprow be still made up of m alternate data center, but decision attribute is respectively by 2 sub-attributes of quality of service attribute: n' sub-attribute of bandwidth, time delay and content popularit: population, geographical position etc. are formed.Each decision matrix is normalized, and calculates the weight of the sub-attribute of each decision-making, the evaluation result to current attribute can be generated, be labeled as the value of current attribute, as according to D _qoSand D _popthe property value of different alternative decision making scheme for quality of service attribute and content popularit attribute can be generated.Can construct the decision matrix of the 0th layer after obtaining this value further, namely final decision matrix is

Wherein, number n " only by the n of the 1st layer " the individual core attribute of decision attribute determines.This matrix is normalized again and weight setting, final Decision-Making Evaluation result can be obtained.In like manner, having under the attribute structure of more multi-layered time, also by sub-decision matrix, the upwards final comprehensive evaluation result of the structure of recurrence.

Second step, according to analyzing the property value difference distribution obtained, utilizing the relative weighting that information entropy theory sets a property, and providing the computing formula of evaluation of estimate.

Its basic thought is: when different alternative decision making scheme differs greatly for the value of same attribute, this attribute is larger for the impact of whole Decision-Making Evaluation, and when difference is less, this attribute even can be ignored for Decision-Making Evaluation role is just very little.According to the extremum property of comentropy, the size of entropy just can reflect different alternative on same property value close to situation, property value is more close, and entropy is larger.Therefore, this method adopts information entropy to arrange different attribute weight.When the arranging of relative weighting, decision matrix is normalized.First, all types of attribute is converted into cost type attribute.In the model of this method, only have 2 generic attributes, a class be profit evaluation model as bandwidth, popularity etc., its property value is higher better to its evaluation; The another kind of cost type that belongs to is as bandwidth price, time delay etc., and the lower evaluation result of its property value is better.This method is by using 1/x _ijreplace x _ijthe attribute of profit evaluation model is converted into cost type.Then, by alternative A _iabout attribute X _jproperty value be normalized to evaluation of estimate.

$X_{ij}=\frac{x_{ij}}{\underset{i=1}{\overset{m}{\Σ}}{x}_{ij}}---\left(5\right)$

Use X _ijreplace the x in a upper decision matrix one by one _ijthus construct Evaluations matrix (P)=(X _ij) mn; According to shannon formula, computation attribute X _jcomentropy power $E_j=-k\underset{i=1}{\overset{m}{\Σ}}{X}_{ij}{\mathrm{lnX}}_{ij},\∀j.$ Wherein, k is constant, for making 0≤Ej≤1, and order $k=\frac{1}{\ln m}.$ The degree of deviation of information is defined as d _j=1-E _j.This method tentation data central administrator does not have attribute bias.A kind of method simply arranging different attribute weight is order here use the evaluation of estimate of arithmetic weight averaging operator to alternate data center to calculate, its final evaluation of estimate can be with analytic hierarchy process (AHP) recursive resolve

$U_i^{final}(X_{i1},X_{i2},\mathrm{...},X_{in})={\Σ}_{j=1}^{n^{*}}{w}_j{X}_{ij},\∀i$

(3) temporal extension figure is obtained to evaluation of estimate modeling, by temporal extension figure, the data transmission scheduling problem between the data center based on multiattribute information is converted into minimum price flux problem;

When preparing to transmit data in t, need to consider: select access bandwidth cheapest data center interim data to save bandwidth cost simultaneously; The data center's interim data selecting access link service quality best is to improve data transmission performance; Select energy use efficiency the maximum data center interim data to save energy resource consumption etc.This just means that all properties of data center changed along with the time and makes this problem become more complicated.Here use hierarchical parsing approach provides the evaluation to each interim data center, and the data center internet along with time variations is modeled as a temporal extension figure.Due to the final evaluation function U set up _it () is the function of a cost type, this just means that functional value is lower, obtain to alternative A _ievaluation result better.When a selection transfer link, the factors such as the access bandwidth of link receiving terminal data center, access bandwidth price, energy use efficiency be considered.Thus expression formula (11) can be rewritten as again

$\begin{array}{cc}\min imize& z=\underset{t=0}{\overset{T-1}{\Σ}}\underset{eij\∈E}{\Σ}{f}_{ij}\left(t\right)*U_j^{final}\left(t\right)\end{array}$

$\begin{array}{cc}st& \underset{t=0}{\overset{T-1}{\Σ}}\underset{eij\∈E}{\Σ}{f}_{1j}\left(t\right)=F\end{array}$

$\underset{eij\∈E}{\Σ}{f}_{ij}(t-1)-\underset{eij\∈E}{\Σ}{f}_{ji}\left(t\right)=0(i\≠1,j\≠n)$

$\underset{t=0}{\overset{T-1}{\Σ}}\underset{eij\∈E}{\Σ}{f}_{jn}\left(t\right)=F$

$0\≤f_{ij}\left(t\right)\≤u_{ij}\left(t\right),\∀e_{ij}\∈E$

(4) calculate solution minimum price flux problem by negative cost loop algorithm, obtain the path of overall merit optimum, i.e. minimum cost stream.

This algorithm is needed to calculate during scheduling in data block according to the attribute value information of data center's monitoring tool collection by the manager of data center, and the path of an overall merit optimum is selected in the transmission for data block.Its basic thought first calculates a feasible flow (being also max-flow usually) by Edmonds-karp algorithm, then corresponding residual network is generated, whether negative cost loop is there is with in Ballman-ford algorithm cycle criterion figure, if any then eliminating negative ring and upgrading residual network, until there is not cost is negative ring, the stream finally obtained, is minimum cost stream.

Specific as follows step:

Input: G (N, E) is the data center with N, the interconnected figure of data center of E bar link.Wherein every bar limit e _ij∈ E

Each node i ∈ N has cost function c _it () // link cost function, gets different function respectively in distinct methods

T: maximum expectation transmission time; S: source data center; D: destination data center

Export: by the data transmission scheduling path based on multiattribute information of s to d

1) corresponding temporal extension figure G ' (N ', E ') is generated according to original graph G (N, E)

2) in G ' (N ', E '), a feasible flow x is set up

3)E ₁＝e _i′j′∈E′:F _i′j′＜u _i′j′

4)E ₂＝e _i′j′∈E′:F _j′i′＞0

5)G′(x)＝{N′,E ₁∪E ₂}

6) whileG ' (x) comprises a negative cost loop do

7) one of them negative cost loop W is found

8) $\mathrm{\δ}:=min\{r_{i^{\′}{j}^{\′}}:\∀e_{i^{\′}{j}^{\′}}\∈W\}$

9) along the flow of the direction increase δ unit of ring W, and G ' (x) is upgraded:

10) $f_{ij}=f_{i^{\′}{j}^{\′}}+\mathrm{\δ},\∀e_{i^{\′}{j}^{\′}}\∈E_1\∩W$

11) $f_{ij}=f_{i^{\′}{j}^{\′}}-\mathrm{\δ},\∀e_{j^{\′}{i}^{\′}}\∈E_2\∩W$

12)endwhile

13) path of returns to d.

The know-why being specific embodiments of the invention and using described in above, if the change done according to conception of the present invention, its function produced do not exceed that specification and accompanying drawing contain yet spiritual time, must protection scope of the present invention be belonged to.

Claims (7)

1. based on multiattribute information data center between a data transmission scheduling method, it is characterized in that comprising the steps:

(1) utilize analytic hierarchy process (AHP), the subordinate relation between multiple attribute is analyzed;

(2) according to analyzing the property value difference distribution obtained, utilizing the relative weighting that comentropy sets a property, obtaining overall merit formula, and obtain evaluation of estimate by formula;

(3) temporal extension figure is obtained to evaluation of estimate modeling, by temporal extension figure, the data transmission scheduling problem between the data center based on multiattribute information is converted into minimum price flux problem;

(4) calculate solution minimum price flux problem by negative cost loop algorithm, obtain the path of overall merit optimum, i.e. minimum cost stream.

2. the data transmission scheduling method between a kind of data center based on multiattribute information according to claim 1, it is characterized in that: multiple attribute information is divided into different levels according to dependence by the analytic hierarchy process (AHP) of described step (1), using the structure structure Multiple Attribute Decision Problems of stratification, quantitatively obtaining property value by monitoring.

3. the data transmission scheduling method between a kind of data center based on multiattribute information according to claim 2, it is characterized in that: described Multiple Attribute Decision Problems is the order of preference evaluating different alternate data center based on the decision matrix of multiattribute information, and decision matrix is as follows:

Wherein, the number n of decision attribute " by the n of the 1st layer " individual core attribute decision, the row representative of matrix has m decision-making alternative { A _i, row represent n different decision attribute { X _j, each element x of matrix _ijthe jth item property value of representative data center i.

4. the data transmission scheduling method between a kind of data center based on multiattribute information according to claim 1, is characterized in that: it is as follows that described step (2) obtains overall merit formula process:

1) all types of attribute is converted into cost type attribute;

2) based on decision matrix, will about attribute X _jproperty value be normalized to evaluation of estimate, normalization formula is $X_{ij}=\frac{x_{ij}}{\underset{i=1}{\overset{m}{\Σ}}{x}_{ij}};$

3) according to shannon formula, computation attribute X _jcomentropy power

Wherein, k is constant, for making 0≤Ej≤1, and order

4) degree of deviation of information is established to be defined as d _j=1-E _j;

5) relative weighting is made and use arithmetic weight averaging operator to be weighted evaluation of estimate, solving overall merit formula is $U_i^{final}(X_{i1},X_{i2},...,X_{in})={\Σ}_{j=1}^{n^{*}}{w}_j{X}_{ij},\∀i.$

5. the data transmission scheduling method between a kind of data center based on multiattribute information according to claim 4, is characterized in that: described all types of attributes are total up to two kinds, cost type attribute and profit evaluation model attribute.

6. the data transmission scheduling method between a kind of data center based on multiattribute information according to claim 1, is characterized in that: the negative cost loop algorithm of described step (4) is data scheduling algorithm between the data center based on multiattribute information; This algorithm is calculated, for transfer of data selects the path of an overall merit optimum when data dispatch according to the attribute value information of data center's monitoring tool collection by the manager of data center.

7. the data transmission scheduling method between a kind of data center based on multiattribute information according to claim 1 or 6, is characterized in that: described algorithm flow is as follows:

A () calculates a feasible flow by Edmonds-karp algorithm;

B () generates corresponding residual network based on feasible flow, whether there is negative cost loop with in Ballman-ford algorithm cycle criterion figure;

If c () exists negative cost loop, eliminate negative ring and upgrade residual network, until there is not negative cost loop, obtaining final stream, be minimum cost stream; If do not exist, this feasible flow is minimum cost stream.