CN106992942A - A kind of SDN resource pricing methods based on resource load and user's request - Google Patents

Abstract

The invention discloses a kind of SDN resource pricing methods based on resource load and user's request, it is specially：During SDN resource transactions, cluster subdivision is carried out to user according to user's request and customer consumption model, user type is determined.If user is consumed using plan, calculate user and cancel reservation probability parameter.The formality rate that computing resource provider collects, sets up price model, it is determined that the reservation discounted price of plan transaction.If user is consumed using stock, current resource load situation is calculated, with reference to user's request computing resource initial prices, the final price of by inch of candle model determination.The present invention shifts to an earlier date reserve resource by price incentive user, with consumption habit corresponding favourable price is provided with reference to the degree of belief of each user, user can be effectively avoided to flock together proposition resource request, cause resource load overweight, the problem of network congestion, so as to realize the reasonable distribution of resource and ensure that each meta service all the time or first business can run well.

Description

A kind of SDN resource pricing methods based on resource load and user's request

Technical field

The invention belongs to SDN scheduling of resource field, and in particular to a kind of SDN resources based on resource load and user's request Pricing method

Background technology

As network Development spreads all over the world so that different network applications needs different service providers, and to net Network pricing system proposes more strict requirements.Meanwhile, under the background that network is continued to develop, most optimum distribution of resources is ground Study carefully the connotation for imparting renewal.Therefore, a kind of more reasonably price mechanism and system are taken in research, highlight Internet resources Value, improves resource utilization, reduces the wasting of resources, provides basis for network market running, is that current academia is pursued always The hot issue of research.For this, it is necessary to study a kind of brand-new network service pricing model the problem of a series of, on the one hand, It will rationally and effectively distribute resource：On the other hand, must can support multi-level service mode, meanwhile, also will can from Family there withdraws network service operation cost and provides efficient, high-quality network service.

2010, ONF (Open Networking Foundation, open network foundation) proposed software defined network The concept of network (Software Defined Networks, abbreviation SDN), the characteristics of its is maximum is exactly that datum plane and control are flat The separation in face, supports the control of centralization network, realizes virtualization of the bottom-layer network facility to upper strata, spry and light software programmable Ability, the management and control ability for finally making network obtains huge lifting.SDN is as a kind of new network architecture, and it collects Middle control feature can propose user the resource of demand is transferred to unified transaction platform and is traded.It is mainly reflected at 3 points： First, the control feature of centralization with centralized Control and can integrate all-network resource, it is convenient that resource is effectively managed； Allow layman to add other strategies in scheduling strategy second, flexible software programmability can be realized, realize individual character The customization of change, further improves the performance of scheduling of resource；Third, can to solve current business demand fast for the scalability of height Caused by speed increase the problem of scheduling of resource inefficiency.Can be based on SDN to backbone edges stream from the angle of operator Amount carries out tuning.

Internet resources pricing strategy based on SDN is mainly for following two problems：1) in current network conditions, net is worked as When network resource uses peak period, a large number of users accesses limited Internet resources, will necessarily result in part resource in short supply.Meanwhile, it is excellent Matter resource inherently turn into user access preferred object, secondary resources must visit capacity it is sparse.So easily causing resource makes With imbalance, high-quality resource high capacity causes Qos to decline；Secondary resources utilization rate is too low, causes resources idle to waste.2) such as What formulates a rational Internet resources pricing strategy so that resource can be distributed reasonably, excludes the dry of malicious user Disturb so that service provider, in the case where meeting user's request, result in maximum value.

The content of the invention

It is an object of the invention to concentrate to propose that resource load is overweight and net caused by resource request for user in network The problem of network congestion, there is provided a kind of SDN resource pricing methods based on resource load and user's request.

The purpose of the present invention is achieved through the following technical solutions：It is a kind of based on resource load and user's request SDN resource pricing methods, this method comprises the following steps：

Step 1：During SDN resource transactions, user is carried out according to user's request and customer consumption model PTMQDC Cluster subdivision, the type that subdivision result determines user is clustered according to gained；

Step 2：If user is consumed using plan, the type of the user according to obtained by step 1, with reference to the consumption of subscriber Custom, calculates user and cancels reservation probability parameter.

Step 3：Probability parameter is preengage according to the cancellation that step 2 is obtained, with reference to the subscription time and users to trust degree of user, The formality rate that computing resource provider collects, sets up price model, and different according to the different time sections of reserve resource correspondence The characteristics of price, it is determined that the reservation discounted price of plan transaction.

Step 4：If user is consumed using stock, current resource load situation is calculated, with reference to user's request, at the beginning of computing resource Beginning price, the final price of by inch of candle model determination.

Further, the step 1 is specially：

The user's request is divided into three classes：Time delay sensitive type, bandwidth sensitive type, cost sensitive.

The PTMQDC models include six attributes：P:Pattern, represents the resource transaction pattern that user takes, stock Consumption or plan consumption；T：Time, represents that user proposes the time of resource request；M：Money, represents the consumption of user every time The amount of money；Q：Quantity, represents each resource transaction quantity of user；D：Discount, represents whether customer consumption enjoys folding Button； C：Credit, represents users to trust degree；The significance level and customer type of client is described by six property values.

K-Means clustering algorithm instrument WEKA are utilized with reference to PTMQDC models, user are divided into four classes, according to user's Transaction record and users to trust degree are ranked up to four class users, are followed successively by outstanding client, common customer, limitation client, malice Client.

Further, the step 2 is specially：Gained user type determines users to trust degree in step 1.User is taken The behavior of reservation of disappearing is analyzed, it is considered to two aspects：The behavior prediction of individual character and the probability analysis based on statistics.It is individual Bulk properties behavior prediction mainly investigates the consumption habit of subscriber, the factor of its cancellation reservation of analyzing influence.Based on statistics Probability analysis is divided, to the rate of failing to keep an appointment of a certain class user based on counting using the model for considering heterogeneous to user It is predicted.By determining that user does not have to cancel the probability preengage in time span t, determine to use with reference to users to trust degree Cancel reservation probability parameter in family.

Further, the step 3 is specially：The discount that the formality rate and user for cancelling reservation are enjoyed when preengaging in advance Relevant, the discount that user is enjoyed is bigger, and the service charge ratio that resource provider is collected to it when user breaks one's promise is higher.According to The price and service charge of presell resource, determine that resource provider releases the price of each period.

Further, the step 4 specifically includes following sub-step：

(4.1) SDN resource is divided into forwarding, key-course and application layer, its service function block difference corresponding element energy Power, meta service and first business.

First ability, is the fine granularity functional unit of forwarding capability in Resource Abstract layer.It is holding substantially in network Component is carried, all first abilities provide diversified basic bearing capacity in network-wide basis for meta service.

The meta service is characteristic and the requirement according to Network and clusters the basic network that multiple first abilities are formed Service function component.

Characteristic and the requirement of upper layer application are contained in first business, it according to the characteristic of application and can require abstract The most basic network service function gone out needed for business.

(4.2) each index weights of resource are determined, each data target is normalized.

(4.3) calculate each yuan of ability and account for all first ability ratio, coefficient of variation and weight numbers for including correspondence index；Pass through Above-mentioned parameter assesses resource utilization, and decision-making system is overload, the normal or free time.

(4.4) probability selected using Logit model analysis user meta service, and determine demand of the user to resource.

(4.5) earnings pattern is set up, the initial prices of resource are determined, by determining final resource based on combinational auction model Price.

The beneficial effects of the invention are as follows：By price incentive, user shifts to an earlier date reserve resource, with reference to the degree of belief of each user Corresponding favourable price is provided with consumption habit, can effectively avoid user from flocking together proposition resource request, cause resource load It is overweight, the problem of network congestion, so as to realize the reasonable distribution of resource and ensure each meta service all the time or first business all It can run well.

Brief description of the drawings

Fig. 1 resource pricing general frames；

The selection of the different trade modes between layers of Fig. 2；

Fig. 3 user clusterings result is exported, and it is 8 clusters that (a), which is that user gathers, and it is 4 clusters that (b), which is that user gathers,；

Fig. 4 subscription times and the probabilistic relation for cancelling reservation；

Different time sections corresponding preferential price when Fig. 5 different users shift to an earlier date reserve resource, (a) is the subscription time of user 1 section Corresponding price, (b) is the corresponding price of the subscription time of user 2 section, and (c) is the corresponding price of the subscription time of user 3 section.

Embodiment

The present invention is described in further detail with specific embodiment below in conjunction with the accompanying drawings.

As shown in figure 1, a kind of SDN resource pricing methods based on resource load and user's request that the present invention is provided, should Method comprises the following steps：

Step 1：User's request is assessed, cluster subdivision is carried out to user using PTMQDC models, is clustered and segmented according to gained As a result the type of user is determined.

The user's request is divided into three classes：Time delay sensitive type, bandwidth sensitive type, cost sensitive.The use of time delay sensitive type Family, it usually needs be the business such as voice, call, it is more sensitive to time delay, pay close attention to network link real-time；And bandwidth is quick Sense type user, pursuit is the network bandwidth, is such as used for foradownloaded video business；Cost sensitive user, only focuses on valency substantially Lattice.

Using PTMQDC models, its implication of PTMQDC is：

P：Pattern, represents the resource transaction pattern that user takes, and stock is consumed or plan consumption；T：Time, represents to use Family proposes the time of resource request；M：Money, represents each spending amount of user；Q：Quantity, represents that user is each Resource transaction quantity；D：Discount, represents whether customer consumption enjoys discount；C：Credit, represents users to trust degree. The significance level and customer type of client is described by six property values.

The consumption demand of user directly affects resource price, for ease of more intuitively being analyzed, and we are by PTMQDC six Individual parameter is handled as follows：For parameter P, SPOT represents that stock is consumed, and PLANNED represents plan consumption；For parameter T, Real resource exchange hour is determined according to user's request, is also continuous, and four discrete times are divided time into here Section：Morning M (00:00——06:59)；Morning A (07:00——11:59)；Afternoon B (12:00——17:59)；Evening E (18:00——23:59)；For parameter M and C, each user can have a plurality of resources consumption to record, the resources consumption amount of money Quantity required and degree of belief are all different, and the present embodiment takes the average value of all records of the user as reference value, and discount D is then used Discount whether is enjoyed to represent.To the resource transaction parameter such as table 1 of each user.

The resource transaction parameter signal table of the user of table 1

Division classification is carried out to client with this, and is that it predicts different resource transaction modes for different users.

K-Means clustering algorithm instrument WEKA are utilized with reference to PTMQDC models, user are divided into four classes, according to user's Transaction record and users to trust degree are ranked up to four class users, are followed successively by outstanding client, common customer, limitation client, malice Client.

After subscriber segmentation, the classification to user is realized using algorithm.Clustering to client mainly utilizes data K-Means algorithms in mining algorithm.K-means is the classical clustering algorithm of comparison, compared with other clustering algorithms, is had It can be readily appreciated that the features such as fast convergence rate, and it applies comparative maturity.The algorithm weighs two objects using distance Between similarity, when two object distances are nearer, it is believed that their similarity is bigger, using Euclidean distance formula.

Step 2：If user is consumed using plan, according to step 1 gained subscriber segmentation result and the type of user, with reference to pre- About the consumption habit of user, calculates user and cancels reservation probability parameter.

The behavior prediction of individual character and the probability based on statistics are essentially consisted in the behavioural analysis that user cancels reservation Analysis.Individual character behavior prediction, the consumption habit of primary concern subscriber, factor of its cancellation reservation of analyzing influence etc.. Probabilistic model is then based on statistics, to be divided into the probabilistic model for not considering heterogeneity and consider heterogeneous probabilistic model.Its In do not consider heterogeneity probabilistic model be the rate of failing to keep an appointment that is averaged obtained based on all historical datas in resource provider, will All users uniformly treat, according to identical fail to keep an appointment rate processing；And consider that heterogeneous model is based on the base divided to user On plinth, the prediction for rate of being failed to keep an appointment to a certain class user, this model more meets reality.

In general, it is assumed that the cancellation reservation behavior of user is random, and the cancellation process of the user is also independent 's.User is in moment h, and the event for not cancelling reservation is chance event, after time interval s, does not still cancel reservation Probability it is unrelated with moment h, it is to be understood that the chance event is necessarily satisfying for exponential distribution.If P₀(t) it is to be used in time span t The probability of reservation is not cancelled at family, and over time, the probability that user cancels reservation will be more and more lower, that is, have：

P₀(t)=a^t(formula 1)

Wherein a meets a >=0, if a=0, P₀(t) be constantly equal to 0, illustrate no matter the time how much, someone always cancel reservation； If a=1, P₀(t) be constantly equal to 1, illustrate no matter time length, all nobody cancel reservation.Obviously both of which is and reality Border situation is not inconsistent.Therefore there may be a λ ＞ 0, make P₀(t) meet：

P₀(t)=e^-λt(formula 2)

The probability that user cancels reservation is not only relevant with the time of reserve resource in advance, also with user's history consumer record institute The users to trust degree of generation is closely related, for the high user of degree of belief, its cancel reservation probability compare it is relatively low, And the relatively low user of degree of belief, its probability for cancelling reservation is compared with other users, relatively high.That is, user cancels The probability of reservation, it is not only related to meeting the natural law of exponential distribution, it is also closely bound up in itself with user.By degree of belief It is attached in the probability that user cancels reservation, it may be determined that user user in time span t does not cancel the probability P of reservation₀ (t) it is：

P₀(t)=Te^-λt(formula 3)

Wherein T represents the degree of belief of user, then in time span t, and the probability that user cancels resource reservation is q (t)：

Q (t)=1-P₀(t)=1-Te^-λt(formula 4)

By above-mentioned analysis, it is known that λ is the parameter for cancelling reservation probability description to user, the determination of the parameter is main Drawn by the way that a large amount of subscribers are cancelled with reservation progress probability analysis.

Step 3：According to cancelling reservation probability parameter obtained by step 2, with reference to user subscription time and with users to trust degree, The formality rate that computing resource provider collects, sets up price model, and when more the different prices of reserve resource are corresponding different Between section the characteristics of, it is determined that the reservation discounted price that obtains of plan exchange.

The formality rate for cancelling reservation combines reality, it should which the discount enjoyed when being preengage in advance with user is relevant, Yong Husuo The discount of enjoyment is bigger, that is, the price for obtaining resource is lower, then the service charge that resource provider is collected to it when user breaks one's promise Ratio then can be higher.So service charge ratio should be a function r=r (P) related to resource price.

The inversely proportional relations of price P when then having formality rate r and the resource to preengage in advance, are shown below：

Wherein f is a constant more than zero, is determined by resource provider oneself.This relation also illustrate that simultaneously, no By when reserve resource, once cancel reservation, the service charge to be paid of user is identical, but enjoyment discount More, the ratio shared by service charge is higher.

It is as follows that final resource preengages the price model changed over time during sale in advance：

Wherein C is the operation cost of resource provider；P₀Represent the original prices of resource.

If the lowest price that resource presell is gone out is P, then must is fulfilled for P (t) >=P, then has：

I.e. according to the acceptable lowest price of resource provider, and according to preengaging in advance, the resource that user enjoys The characteristics of price is lower, the time that user preengages in advance can not be more than

It is then determined the corresponding different time sections of reserve resource difference price：

Upper one walks the dynamic pricing models for obtaining resource, and obtains the maximum time that resource is preengage in advance, and in reality The price of border resource is discrete, not consecutive variations.N-th of period is represented with n, then P_n(t) discrete price is represented Functional relation, the price of n-th of period resource, q_n(t) user cancels the probability of reservation when representing n-th of period, that ：

Because the consumption time of user distance reservation is longer, the probability for cancelling reservation is bigger, that is to say, that time interval is got over Long, probability is higher, so q_n(t) it is an increasing function, then in period (t_n-1,t_n), q_n-1(t) ＜ q_n(t) ＜ q_n+1(t), And then have above formula rewriting：

So according to the price and service charge of presell resource, it may be determined that resource provider releases the time of each price Section.The purpose for allowing resource provider to sell resource in advance with proper price in the abundant suitable period is reached, simultaneously Sales network resource, it is possible to prevente effectively from user concentrates the situation for proposing resource request, causes network to gather around for the moment in this way The problem of plug, overload.

Step 4：If user is consumed using stock, current resource load situation is calculated, with reference to user's request, at the beginning of computing resource Beginning price, the final price of by inch of candle model determination.

(4.1) each index weights of resource are determined：

1) assume in SDN service abstraction layer, there are multiple meta services in system, some meta service is by m first energy Power is constituted, and all first abilities for constituting all these meta services have n kinds, all first abilities that current time is obtained Achievement data is n as sample data, i.e. sample size, and the index of the first capability resource of measurement has p, then utilizes entropy assessment Parameter weight, system input index be：

Wherein X_j=(x_1j x_2j … x_nj), j=1,2 ..., p, X_jRepresent the sample value of j-th of index, x_ijRepresent i-th J-th of finger target value of individual first ability.

2) x after data normalization, data processing_ijValue be：

A desired value big index as far as possible) is calculated,

B desired value small index as far as possible) is calculated,

3) i-th comprising j-th of index yuan ability is asked to account for the proportion of all first abilities for including j-th of index：

Calculate the comentropy of jth index：

BecauseIt is a constant relevant with sample size n sample range, and 0≤P of proportion_ij≤ 1, then 0≤E_ij≤1。

4) coefficient of variation of jth index is calculated：

Because 0≤E_ij≤ 1, according to the entropy size of a certain item index in system with its degree of variation on the contrary, then jth refers to Target coefficient of variation is G_j：G_j=1-E_j。

For this index, difference value is bigger, illustrates that the index is more important, and entropy is with regard to smaller, i.e. G_jIt is bigger, E_ijIt is smaller, Index is more important.

The weight number of parameter:

Then for p indexs, every weight sets is combined into W={ W₁,W₂,W₃,…,W_P}。

(4.2) utilization rate of resource is assessed：

The jth index of i-th yuan of ability is designated as in time t service conditionSo this yuan all of ability use Situation data acquisition system is

The data of meta service are the data acquisition systems of bottom member ability, and a meta service is made up of m kind member abilities, then all members Service data collection is combined into：D={ D₁,D₂,…,D_m}。

Requirement of the different meta services to resource is different, then the first ability for constituting different meta services is also different, and same unitary First ability configuration performance under service differs, and weight when different first abilities constitute the meta service is now represented with preset parameter, is weighed Set expression is again：R={ R₁,R₂,…,R_m}。

By previous step, the weight W={ W of a certain index of a first ability₁,W₂,W₃,…,W_P}.Define member ability i when The actual loading situation for carving t is L_i, theoretical duty ability Cs of first ability i on index j_ijRepresent, C_ijFor one in theory Value, such as first ability can normally run in memory usage below 80%, then C_ij80% is taken, then index j theoretical duty ability It is expressed as RLoad_j。

First capacity unit in time t, index j load factor for currently practical load and this yuan of ability load capacity it Than representing first ability in actual use situation of the t to index i.So constituting first ability of the meta service in the unit interval T, index j load factor is：

Finally give meta service unit is to index i instream factor in tIf business is transported Have N number of meta service when turning to participate in, then resource utilization ratio can be designated as:For multiple meta service utilizations of resources The average value of rate.

By comparing Rate size, the system of can determine that is overload, the normal or free time.

(4.3) user's request is analyzed：

It is time delay sensitive type, bandwidth respectively because the difference of demand can be classified as three classes here in resources consumption Responsive type, cost sensitive.The user of time delay sensitive type, it usually needs be the business such as voice, call, it is quicker to time delay Sense, pays close attention to the real-time of network link；And bandwidth sensitive type user, pursuit is the network bandwidth, such as foradownloaded video etc. Business；Cost sensitive user, only focuses on price substantially, for two phase same-action resources, any less expensive to select It is any, almost do not mind other factors.

Therefore, selection of the user to resource, is weighed with Logit preference patterns here.In Science of Economics, Logit moulds Type is earliest Discrete Choice Model, and it is that user is when being selected, it will usually be inclined to based on the effectiveness described in economics In the higher option of the effectiveness for them.

The Logit models, when user has two kinds of selection schemes, U₁With U₂Two kinds of different choice sides of user are represented respectively Efficiency in effectiveness during case, wherein Logit models is expressed as：U=V+ σ, wherein V represent it is observed that influence factor structure Into effectiveness, σ represent not it is observed that factors composition random entry.P₁With P₂The general of user's selection two schemes is represented respectively Rate, during user's selection consumption plan, inherently selects the higher scheme of effectiveness, i.e. P₁Mean that U₁＞ U₂Probability, P₂Represent U₂＞ U₁Probability.

By Logit model inferences, it can obtain：

Wherein θ is a positive coefficient, commonly uses Maximum-likelihood estimation.

By the selection of the loading condition of resource and user, double factor binding analysis, with regard to that can draw to resource transaction both sides Equal reasonable prices, can also reach the purpose of resource reasonable distribution.

Comparison between different meta services, we represent the price of meta service with P, and the meta service currently negative is represented with L Load situation, the extra returns obtained during customer consumption are represented with R, represent that user obtains the effectiveness obtained during meta service with U.Effect Paid cost is subtracted with the extra returns obtained when being customer consumption.

So earnings pattern can be defined as：U=ω₁R-(ω₂P+ω₃L)；

Wherein, ω={ ω₁,ω₂,ω₃Null constant is greater than, the factor of influence parameter of each single item is represented respectively, Different users, the requirement to each single item index is different, therefore its parameters weighting is different.According to Logit models, user The probability for selecting meta service is q, then

Ass represents all meta services.

In addition meta service is made up of multiple first abilities, however, it is determined that is selected the probability of each first ability, can also be obtained The selected probability of meta service, is designated as

For resource provider, it is therefore desirable to which it is their final purpose to obtain maximum value, and cost is constant, then Meta service i is to the maximum return that resource provider is brought：R_i(P)：

R_i(P)=q_i×P_i× Q-C (formula 22)

Wherein Q represents the quantity of this meta service of user's selection, and C represents the intrinsic cost of the meta service, then to ask Go out the most suitable price of the meta service.Single order and second order derivation are carried out to the function on P.Then have：

If P is the optimal solution for meeting above formula, P necessarily meets first derivation equal to zero, that is, meets following formula：

At the same time, to expect maximum revenue, also need to meet second order derivation less than zero, that is, meet following formula：

Because the key for solving optimal solution is relevant with the probability that user selects the meta service, and the key for influenceing user to select It is the effectiveness that user finally obtains again, the parameter of the effectiveness after all with each factor of influence is closely bound up, that is, solves Optimal solution, the most important parameter size for being just to determine each factor of influence.

Embodiment

For PTMQDC models, using all-network resource user historical transaction record, data set as shown in table 2 is taken Sample, is emulated.

The sample data collection of table 2

To simplify simulation process, using the 1000 groups of user data randomly generated, user is analyzed, number is first directed to According to obtaining the data message shown in table 2.

Primary data mainly contains the data attribute included in tables of data, data type, be numeric type data or The data bulk of classifying type data and each attribute etc. information.

K-Means algorithms are selected using WEKA Clustering tool, when the numCluster of selection is 8, user is gathered is 8 classes, it can be seen that such as (b)

Fig. 3 output results.As can be seen that the first cluster includes 312 people, the 31% of total user is accounted for, the people of the second cluster 158 accounts for 16%, the 3rd cluster includes 169 people, accounts for 17%, the 4th cluster includes 361 people, accounts for 36%.

Resource load is analyzed：

Before being fixed a price to specific resource, the situation of resource load should be determined first, and 10 members are given below The resource behaviour in service of ability, wherein weigh this yuan of ability performance index include internal memory (RAM) utilization rate, CPU usage, Disk (Disk IO) utilization rate, network usage (Network IO) etc. 4, and each index dimensionless.

The use ability of normal work is kept on 3 yuan of each Index Theories of ability of table

First ability numbering	Memory usage	CPU usage	Disk utilization rate	Network usage
					1	0.80	0.80	0.95	0.95
2	0.80	0.80	0.95	0.95
					3	0.70	0.75	0.95	0.85
4	0.85	0.85	0.90	0.90
					5	0.85	0.85	0.90	0.90
6	0.70	0.75	0.95	0.85
					7	0.70	0.75	0.95	0.85
8	0.80	0.80	0.95	0.95
					9	0.80	0.80	0.95	0.95
10	0.85	0.85	0.90	0.90

4 yuan of abilities of table each achievement data in certain moment actual motion

First ability performance indications Weights Example analysis

The first capacity index data of the items listed in table 4 are normalized first, and for first ability, respectively The data of individual index should be the smaller the better, then have table 5.

Data set after the normalization of table 5

First ability numbering	Memory usage	CPU usage	Disk utilization rate	Network usage
					1	0.40	0.67	0.67	0.43
2	0.00	0.33	0.33	0.14
					3	1.00	1.00	1.00	1.00
4	0.60	0.67	0.50	0.43
					5	0.20	0.17	0.00	0.00
6	0.60	0.83	0.67	0.57
					7	0.80	1.00	0.83	0.71
8	0.40	0.17	0.33	0.29
					9	0.20	0.50	0.67	0.57
10	0.00	0.00	0.00	0.00

Include the proportion P for accounting for all first abilities comprising this index of i-th yuan of ability each index_ijAs shown in table 6.

The ratio shared by each index in each first ability of table 6

First ability numbering	Memory usage	CPU usage	Disk utilization rate	Network usage
					1	0.10	0.13	0.13	0.10
2	0.00	0.06	0.07	0.03
					3	0.24	0.19	0.20	0.24
4	0.14	0.13	0.10	0.10
					5	0.05	0.03	0.00	0.00
6	0.14	0.16	0.13	0.14
					7	0.19	0.19	0.17	0.17
8	0.10	0.03	0.07	0.07
					9	0.05	0.09	0.13	0.14
10	0.00	0.00	0.00	0.00

Then show that the comentropy and differentiation coefficient of each index are as shown in table 7.

The comentropy of each index of table 7 and differentiation coefficient

Finally give each index weights as shown in table 8.

Weight shared by each index of table 8

It can be seen that, although shared weight size is close for four indices, but still has nuance, this use with first ability State is closely related, therefore during actual resource use, according to first ability service condition data of more magnanimity, can be with Draw more accurately index weightses.The loading condition of each first ability is can be determined that according to this weighted value.

Meta service resource utilization instance analysis：

If a meta service A is made up of this 3 first abilities in the 1st, the 3rd and the 5th, the weight of each yuan of ability is 0.3,0.2, 0.5, the weighted value is determined by the importance of each first ability for constituting meta service A, here process to simplify the analysis, directly finger It is fixed.And assume that four indexs of these three first abilities in theory keep the maximum load situation C of normal work_ijAs shown in table 9.

The use ability of normal work is kept in the meta service A of table 9 on each Index Theory of first ability

First ability numbering	Memory usage	CPU usage	Disk utilization rate	Network usage
					1	0.80	0.80	0.95	0.95
3	0.70	0.75	0.95	0.85
					5	0.85	0.85	0.90	0.90

And at a time the service condition of each index of resource is as shown in table 10 in real work.

The service condition of each index real work of first ability in the meta service A of table 10

First ability numbering	Memory usage	CPU usage	Disk utilization rate	Network usage
					1	0.40	0.30	0.30	0.50
3	0.10	0.10	0.10	0.10
					5	0.50	0.60	0.70	0.80

Comprehensive two tables, each Index Theory load capacity of meta service A and actual loading condition can be drawn according to formula It is as shown in table 11 respectively.

The meta service A of table 11 each index load capacity

So for meta service A, the indices data of all first abilities by constituting meta service A pass through reality The contrast of loading condition and gross data, obtains the loading condition of each single item index in the meta service, and every with reference to what is calculated The weight of index is respectively：0.27,0.21,0.23,0.29.

The resource utilization that meta service A may finally be calculated is：

Rate_A=0.48 × 0.27+0.50*0.21+0.50*0.23+0.63*0.29=0.53

Similarly, if being made up of in the presence of another meta service B the 2nd, the 5th and the 10th 4 first ability, and three first energy are assumed The weight of power is respectively 0.3,0.3,0.4, then, process according to the above analysis, it can be deduced that, indices are real in meta service B Border load and the accounting of theoretical duty ability are：0.68,0.73,0.70,0.84, then have meta service B resource utilization For：

Rate_B=0.68 × 0.27+0.73*0.21+0.70*0.23+0.84*0.29=0.74

Obviously it is single from the analysis of meta service A, B, it is possible to find out that the resource utilization of two meta services is different, meta service B's Load is relatively heavy.Therefore judge that meta service is overload, idle or normal condition by comparing Rate size.

Price-setting process instance analysis based on user's request：

According to resource request user type demand：Cost sensitive, bandwidth sensitive type, time delay sensitive type, ordinary consumption type, Wherein bandwidth sensitive can be attributed to meta service loading condition under the sensitivity to resource performance, i.e. current state with delay sensitive Sensitivity, therefore user is divided into three classes, and the parameter being directed to is (ω₁,ω₂,ω₃, θ), wherein (ω₁,ω₂,ω₃) be The factor of influence of user utility is influenceed, the data consumed by user's history are drawn using entropy assessment；θ is the discrete choosings of Logit The parameter of model is selected, Maximum Likelihood Estimation is commonly used and calculates, it is 0.25 that definite value is taken here.The corresponding ginseng of so every kind of user Number, as shown in table 12.

The factor of influence weight that 12 3 kinds of users of table are related to

User type	ω1(extra returns)	ω2(meta service price)	ω3(meta service loading condition)
				Cost sensitive	0.2	0.5	0.3
Load-sensitive type	0.1	0.3	0.6
				Plain edition	0.3	0.4	0.3

For meta service A and meta service B, it is assumed that the extra returns that user obtains during selection meta service A and meta service B are distinguished For 60,100, and the two current load is respectively 0.53,0.74, is converted into the amount of money to represent, that is, loads and pay more greatly Cost it is bigger, respectively 53,74.The wherein equal dimensionless of parameters.So three kinds users select the effectiveness of two kinds of meta services It is as shown in table 13 respectively.

13 3 kinds of user selection meta service A and meta service B of table effectiveness

User type	Meta service A effectiveness	Meta service B effectiveness
			Cost sensitive	-3.9-0.5P	-2.2-0.5P
Load-sensitive type	-25.8-0.3P	-34.4-0.3P
			Plain edition	2.1-0.4P	7.8-0.4P

The effectiveness of meta service is substituted into user's select probability formula, and is calculated in MATLAB environment and works as resource provider During Income Maximum, different users select the price of two kinds of meta services as shown in the table.

Best price during 14 3 kinds of user selection meta service A and meta service B of table

User type	Meta service A price	Meta service B effectiveness
			Cost sensitive	14.9	17.2
Load-sensitive type	39.7	20.8
			Plain edition	16.3	25.9

As can be seen from Table 14, when different user selects different meta service resources, desired best price is different.For For cost sensitive user, relative to the performance load of resource, it is preferential therefore first that they are more desirable to resource price The price for servicing A and meta service B is all relatively low, and wherein meta service B price is higher than meta service A, because meta service B is current Loading condition is higher than meta service A, and this, when resource load is heavier, by properly increasing resource price, is pierced with proposed by the present invention Sharp user's selection price is more preferential, light load meta service resource theory is mutually unified.Similarly, for plain edition user this Rule is still met.And for load-sensitive type user, they more take notice of any meta service resource current performance more It is good, thus two kinds of meta service prices are of a relatively high, but in order to select the resource meta service A that performance is more preferable, load is lighter, The user of this type is accomplished by paying higher price.

Based on booking-mechanism resource Dynamic Pricing algorithm simulating test, including user cancel reservation probability Case Simulation and Resource pricing process instance emulation for preengaging risk, it is specific as follows：

User cancels reservation probability Case Simulation：

The time that user preengages in advance is longer, and the price that user obtains resource is lower, but because more shift to an earlier date, from specifically making More remote with the date of resource, the probability that user cancels reservation is also bigger.According to historical data analysis, the value that λ is determined here is 0.1.And hypothesis there are three class users of different degree of beliefs, in the case of not reserve resource, resource original prices such as table 15 is obtained It is shown.

The different degree of belief user resources initial transaction prices of table 15

User type	Degree of belief	Cancel the probability q (t) of reservation	Meta service A price
				User 1	1	1-e-0.1t	14.9
User 2	0.8	1-0.8e-0.1t	39.7
				User 3	0.7	1-0.7e-0.1t	16.3

When so these users are using the purchase resource of appointment mode in advance, the probability for cancelling reservation is as shown in Figure 4.

Understand, for the user of different degree of beliefs, when their same 10 days reserve resources in advance, and pushing away over time Enter, the user that the probability of the high user's cancellation reservation of degree of belief is always lower than degree of belief is low, when the time is 0, that is, arrives The same day of consumption resources, the low user people of degree of belief are it is possible to cancel the resource of reservation.Illustrate that user cancels resource reservation Probability and users to trust degree are closely bound up, while being also able to demonstrate that the resource transaction process of degree of belief introducing booking-mechanism Reasonability.

Resource pricing process instance emulation for preengaging risk：

Complete to cancel user and preengage after probability analysis emulation, will further confirm that below under booking-mechanism, based on use The specific price-setting process emulation of resource of family reservation risk.According to price-setting process, be directed to cancel the formality rate of reservation with The lowest price of the acceptable meta service resource of provider.Meta service A situations about preengaging in advance are divided with three class users Analysis.Assuming that being not required to consuming cost, i.e. C=0 in meta service A lowest price 10, and user's cancellation reservation process, it need to only pay Certain service charge, and service charge ratio to reserve resource when price it is related, service charge be resource provider oneself definition One constant, can customize as 2.5.

Price model and earliest time that user can preengage in advance.

The user's dynamic price model of table 16 and earliest subscription time

As shown in Table 16, each user, when the consumption attribute according to user itself preengages meta service resource with degree of belief, User is different, and its time span that can be preengage in advance is different.Such as user 1, as the cost sensitive user that degree of belief is 1, If buying meta service resource A by way of stock immediately is consumed, the best price that can be struck a bargain is 14.9, if however, it is selected Select the plan by way of reservation and consume same resource, it can in advance be preengage with more preferential price, and price is not minimum low In the minimum price 10 of the meta service, time interval is not above 10；For user 2, user 3 similarly.

So according to the price model, the meta service resource dynamic price of each user and the relation of time such as Fig. 5 institutes Show.

The time length that different users can preengage in advance it can be seen from Fig. 5 (a), (b), (c) is differed.Carry The time gap use time of preceding reservation is longer, and resource price is cheaper, and close to the acceptable lowest price of provider, The time preengage in advance nearer it is to the time using meta service resource, and resource price is more expensive.(a) show that the user can carry Preceding 10 days reserve resources, the user of (b) can about 23 days reserve resources in advance, the user of (c) can about 9 days reserve resources in advance. By price incentive, user shifts to an earlier date reserve resource, and corresponding preferential price is provided with reference to degree of belief and the consumption habit of each user Lattice, can effectively avoid user from flocking together proposition resource request, cause resource load overweight, the problem of network congestion, so that real Each meta service or first business can run well all the time for the reasonable distribution of existing resource and guarantee.

Claims (5)

1. a kind of SDN resource pricing methods based on resource load and user's request, it is characterised in that this method includes following step Suddenly：

Step 1：During SDN resource transactions, user is clustered according to user's request and customer consumption model PTMQDC Subdivision, the type that subdivision result determines user is clustered according to gained.

Step 2：If user is consumed using plan, the type of the user according to obtained by step 1 is practised with reference to the consumption of subscriber It is used, calculate user and cancel reservation probability parameter.

Step 3：Probability parameter is preengage according to the cancellation that step 2 is obtained, with reference to the subscription time and users to trust degree of user, calculated The formality rate that resource provider is collected, sets up price model, and according to the different prices of the different time sections of reserve resource correspondence The characteristics of, it is determined that the reservation discounted price of plan transaction.

Step 4：If user is consumed using stock, current resource load situation, with reference to user's request, the initial valency of computing resource are calculated Lattice, the final price of by inch of candle model determination.

2. a kind of SDN resource pricing methods based on resource load and user's request according to claim 1, its feature exists In the step 1 is specially：

The user's request is divided into three classes：Time delay sensitive type, bandwidth sensitive type, cost sensitive.

The PTMQDC models include six attributes：P:Pattern, represents the resource transaction pattern that user takes, and stock is consumed Or plan consumption；T：Time, represents that user proposes the time of resource request；M：Money, represents each spending amount of user； Q：Quantity, represents each resource transaction quantity of user；D：Discount, represents whether customer consumption enjoys discount；C： Credit, represents users to trust degree；The significance level and customer type of client is described by six property values.

K-Means clustering algorithm instrument WEKA are utilized with reference to PTMQDC models, user is divided into four classes, are remembered according to the transaction of user Record and users to trust degree are ranked up to four class users, are followed successively by outstanding client, common customer, limitation client, Malicious clients.

3. a kind of SDN resource pricing methods based on resource load and user's request according to claim 1, its feature exists In the step 2 is specially：Gained user type determines users to trust degree in step 1.The behavior that user cancels reservation is entered Row analysis, it is considered to two aspects：The behavior prediction of individual character and the probability analysis based on statistics.Individual character behavior prediction master Investigate the consumption habit of subscriber, the factor of its cancellation reservation of analyzing influence.Probability analysis based on statistics is to count Basis, is divided, the rate of failing to keep an appointment to a certain class user is predicted using the model for considering heterogeneous to user.By determining User does not have to cancel the probability preengage in time span t, determines that user cancels reservation probability parameter with reference to users to trust degree.

4. a kind of SDN resource pricing methods based on resource load and user's request according to claim 1, its feature exists In the step 3 is specially：The discount that the formality rate for cancelling reservation is enjoyed when being preengage in advance with user is relevant, and user is enjoyed The discount received is bigger, and the service charge ratio that resource provider is collected to it when user breaks one's promise is higher.According to the price of presell resource With service charge, determine that resource provider releases the price of each period.

5. a kind of SDN resource pricing methods based on resource load and user's request according to claim 1, its feature exists In the step 4 specifically includes following sub-step：

(4.1) SDN resource is divided into forwarding, key-course and application layer, its service function block difference corresponding element ability, member Service and first business.First ability, is the fine granularity functional unit of forwarding capability in Resource Abstract layer, it is the base in network This bearing assembly, all first abilities provide diversified basic bearing capacity in network-wide basis for meta service.The meta service Be according to Network characteristic and require and cluster the basic network service function component that multiple first abilities are formed.The member Characteristic and the requirement of upper layer application are contained in business, can be most basic according to needed for the characteristic of application and requirement take out business Network service function.

(4.2) each index weights of resource are determined, each data target is normalized.

(4.3) calculate each yuan of ability and account for all first ability ratio, coefficient of variation and weight numbers for including correspondence index；By above-mentioned Parameter evaluation resource utilization, decision-making system is overload, the normal or free time.

(4.4) probability selected using Logit model analysis user meta service, and determine demand of the user to resource.

(4.5) earnings pattern is set up, the initial prices of resource are determined, by determining final resource valency based on combinational auction model Lattice.