CN103530801A - Method for optimizing costs of multiple data centers based on dynamic pricing strategy - Google Patents

Abstract

The invention discloses a method for optimizing the costs of multiple data centers based on a dynamic pricing strategy. The method comprises the following steps: 1) modeling for the user benefit situation and the service quality of the multiple data centers based on a queuing theory; 2) determining the optimal pricing strategy of each data center according to the models established in the step 1); 3) determining the load routing strategy among the multiple data centers according to the determined optimal pricing strategy. According to the method provided by the invention, load distribution, service pricing and energy consumption cost among the multiple centers are optimized and unified based on the benefits of service providers; on the basis of meeting the benefits of the service providers, the energy consumption costs of the data centers are optimized through a load balancing algorithm.

Description

A kind of many data centers Cost Optimization Approach based on Dynamic Pricing strategy

Technical field

The present invention relates to cloud computing and data center's management domain, relate in particular to a kind of many data centers Cost Optimization Approach based on Dynamic Pricing strategy.

Background technology

，Yun provider generally can adopt a kind of pricing model flexibly (Flexible pricing scheme) to regulate services request amount at present, makes the income of self reach maximum.As Amazon sells cloud resource at a bargain when demand is lower, to improve the utilization factor of resource, realize the optimum of its business earnings.

Obviously, different price levels has direct impact to market demand.As shown in Fig. 1 (a), when price rises, demand reduces thereupon, and both are inverse relation, and this relation presents certain stochastic volatility.Fig. 1 (b) shows that under three kinds of different price levels, demand is situation over time.As can be seen here, price level is determining demand (but not being unique determinative), and demand has significant impact to the managing power consumption of data center.Current, many clouds provider, as how Amazon, Google etc. all sets up ，Jiang data center of a plurality of data center in the whole world, be deployed in diverse geographic location and be not only convenient to expand cloud service market, and be conducive to cut down energy consumption cost (utilizing the temperature difference, electricity price difference etc.).Consider many data center environment as shown in Figure 2: the supposition P of cloud provider has set up a plurality of data centers in zones of different, due to areal variation, power price and towards user market all different.For example, moment t ₀, the service price of data center is p ₀, to t ₁the P of ，Yun provider, in order to boost consumption, turns down service price to p1 constantly, and now the market demand of various places may have different reactions to price adjustment.Meanwhile, also may there is change (as adopted step price or tou power price) in the power price of various places.Now, if can transfer load to all lower data centers of demand and electricity price, can effectively improve the utilization factor of resource and reduce the energy consumption cost of data center.

Because different pieces of information center has the features such as isomerism is large, trans-regional, above-mentioned optimisation strategy faces the challenge of following several aspects:

1, load is shifted and need to be spent the regular hour, may cause the decline of user QoS performance;

2, different pieces of information center service ability is not identical, and scale is limited, and load distributes the isomerism that must consider data center;

3, the object of data center's price adjustment is that energy optimization strategy should be consistent with price adjustment policy co-ordination in order to obtain the maximization of number one, and both do not exist conflict;

4, the load flow in actual environment is subject to various factors, has stochastic volatility, and energy optimization strategy should be able to effectively adapt to the dynamic change of load flow.

Summary of the invention

Goal of the invention: in order to overcome the deficiencies in the prior art, the invention provides a kind of many data centers Cost Optimization Approach based on Dynamic Pricing strategy, for there being the feature of the differences such as electricity price cost, flow of services between many data centers, energy consumption cost Optimization Mechanism based on Dynamic Pricing strategy under a kind of many data center environment is proposed, by the unified consideration of the mechanism such as service pricing, load distribution and server state control, make a harmonious system optimization scheme.

Technical scheme: for achieving the above object, the technical solution used in the present invention is:

A kind of many data centers Cost Optimization Approach based on Dynamic Pricing strategy, by the unified consideration of the corresponding strategies such as service pricing, load distribution and server state adjusting, make a harmonious system optimization scheme, thereby reach effective object of cutting down data center's energy consumption cost.Model plays the variation model of the market demand, utilize the incidence relation of the performance index such as queuing theory Analysis Service price, task response-time and service ability, design accordingly rational load routing mechanism and there is the energy consumption cost optimisation strategy that QoS guarantees.Specifically comprise the steps:

Step 1) based on waiting line theory to majority according to central user be benefited situation and service quality modeling;

Step 2) model of setting up according to step 1), establishes the optimal pricing strategy of each data center;

Step 3), according to determined optimal pricing strategy, is determined the load routing policy between many data centers.

Concrete, in described step 1), the foundation of model need to be considered the following condition of data center: energy consumption, service ability, number of servers, power price (cost price of data center), the scale upper limit, deferred constraint that per unit service ability consumes.

Concrete, described step 2) comprise the following steps:

Step 201) determining data center demand function;

Step 202) determine the mapping relations one by one between the same service price of demand function (selling price), number of servers;

Step 203) according to 202) in determined mapping relations, solve and obtain optimal pricing strategy.

Concrete, described step 3) comprises the following steps:

Step 301) solve the load sharing policy of data center, i.e. load routing policy;

Step 302), according to known routing vector, solve the integrity service ability that obtains data center;

Step 303) postponing and serving under the condition of price constraints, solve the optimal service device quantity of each data center, finally obtain optimum energy consumption solution.

Described data center is provided with standby server.A lot of because have due to server energy consumption, such as server inefficacy, burst flow etc., arrange standby server (server in Idle state) and can guarantee the service quality to user.

Beneficial effect: the many data centers Cost Optimization Approach based on Dynamic Pricing strategy provided by the invention, the prior art of comparing, has following advantage:

Between 1 ,Duo data center, according to demand for services amount and electricity price difference in the heart in different pieces of information, by by dynamic Service price, load sharing policy is unified considers, has made a harmonious system optimization scheme; Not only reduce the whole energy consumption cost of data center, and made the income of data center reach optimum;

2, by Analysis Service, fix a price and data center's energy consumption cost between relation, set up service pricing and the load route matrix of data center; From the load between service provider's interests ，Jiang Duo data center, distribute, service pricing and energy consumption cost optimization are united; On the basis that meets service provider's interests, by load-balancing algorithm, realize the optimization to data center's energy consumption cost;

3 ,Jiang data center problems of management and pricing problem organically combine, and for effectively solving the problem of management of cloud computing data center, provide new approaches;

4, data center's pricing method and load routing policy are simple effectively, accuracy is high, go for large-scale data center environment.

Accompanying drawing explanation

Fig. 1 is service price and market demand variation relation schematic diagram; Wherein (a) is the relation between quantity and price, is (b) quantity and the relation between the time;

Fig. 2 is most according to center configuration diagram.

Embodiment

Below in conjunction with accompanying drawing, the present invention is further described.

Many data centers Cost Optimization Approach of Dynamic Pricing strategy, comprises the steps:

Step 1) based on waiting line theory to majority according to central user be benefited situation and service quality modeling;

Step 2) model of setting up according to step 1), establishes the optimal pricing strategy of each data center;

Step 3), according to determined optimal pricing strategy, is determined the load routing policy between many data centers.

Data center sets a service price to obtain certain income to the user of each access service, and user accesses data center, not only needs to consider service price, also will consider service quality simultaneously.For simplified model, this case is described the satisfaction of user to service quality with delay cost.Task response-time is longer, and delay cost is just higher, and user's service quality satisfaction is just lower.Delay cost can be expressed as the function of task response-time, and conventional cost function has linear function, exponential function etc.Only have when service price and delay cost sum are less than user's reservation price (customer expected price), user just can select visit data center.For the many data center environment shown in Fig. 2, adopt based on the approximate large scale queuing theory of heavy duty below, be modeled as queuing model.Based on this model, research incidence relation between different system parameter (service price, delay cost, number of servers etc.) under stable system state.In order clearly to explain this case content, the notation of variable in this case and description are listed among table 1.The specific definition of variable will elaborate in style of writing.

The symbol of form 1 variable

Suppose that the service price that ISP sets is p > 0, task postpones for ED.This case adopts linear function q*ED to represent user's delay cost, and wherein q is a constant, is the coefficient of user's delay cost.User's reservation price is a stochastic variable v, obeys probability cumulative distribution function F.When given p, user can, according to the reservation price v of self, determine whether visit data center by probability P (v >=p+qED).Only have when user's reservation price is greater than delay cost and service price sum, user just can select visit data center.The system action of Main Analysis data center when steady state (SS) herein, user's request rate should meet following demand relation:

λ(p)=ψP(v≥p+qED)=ψ(1-F(p+qED))

Wherein ψ is the potential demand amount in market.Be that the potential demand amount that user's actual access amount equals market is multiplied by access probability.Above-mentioned Requirement model is common in the middle of various economic models.Below, the flexible concept of definition demand function, to weigh the sensitivity of demand for services amount to service price change in market.Demand function only has the elastic condition of meeting, and the related conclusions of this case could be set up.If the market demand does not meet elastic condition,, when price reduces or raise, demand does not change, and the related conclusions of this paper is no longer applicable.

Definition 1: make λ (p) for client's demand function, p is service price, and the elasticity of λ (p) may be defined as:

$\mathrm{\ϵ}\left(p\right)=-\frac{\∂\mathrm{\λ}\left(p\right)}{\∂p}\·\frac{p}{\mathrm{\λ}\left(p\right)}---\left(1\right)$

If λ (p) meets ε (p) on interval [a, b] > 1, claim that λ (p) is flexible.Visual interpretation to elasticity of demand is: due to the effect of economic law, between price and demand, exist inverse relation.When price raises, demand declines, and vice versa.Flexibility concept is intended to illustrate the relation between price change ratio and demand changing ratio.This simple and important relation between price and demand function, is applied in the middle of various forms of economic models widely.

Although data center build different regions in, power price and service ability are different, and service price should be consistent, and price has region independence.Though such as user wherever, all need equal integration to download the data of certain website, we suppose that the demand function in different markets all meets elastic condition defined above.User, by being distributed in the Agent visit data center of various places, makes user's reservation price distribution function in zones of different market be respectively F _i, demand function is

i=1,2 ..., the task that M(generates arrives intensity), p is service price.According to above analyzing,

Order for various places user's aggregate demand, and defined function

$F={\mathrm{\Σ}}_{i=1}^M{\mathrm{\ψ}}_i{F}_i/{\mathrm{\Σ}}_{i=1}^M{\mathrm{\ψ}}_i$

So can there be following derivation:

Order

for the routing probability of task, to be sent to the probability of the j of data center be p to the task of Agent i _ij.Therefore, [p _ij] formed a route matrix.Order

so the arrival intensity of the j of data center is

when demand function meets resilient relationship, there is following theorem:

Theorem 1: the potential greatest requirements total amount of supposition Agent

with the total service ability of data center ${\mathrm{\Σ}}_{j=1}^N{n}_j{\mathrm{\μ}}_j$ Retention wire sexual intercourse, ${\mathrm{\Σ}}_{j=1}^N{n}_j{\mathrm{\μ}}_j=\mathrm{\κ\ψ},$ So, for any, meet ${\mathrm{\Σ}}_{i=1}^M{p}_{\mathrm{ij}}=n_j{\mathrm{\μ}}_j/\mathrm{\ψ\κ}$ Probability route matrix P[p _ij], and service price p>0, following equation is set up:

P ₁(congestion)=...=P _j(congestion) → ν ∈ (0,1), n _j→ ∞, wherein P _j(congestion) be the congestion probability of the j of data center, and

User asks total arrival rate to be:

Service price p has following structure:

$p=p^{*}+{\mathrm{\Σ}}_{j=1}^N{p}_j\frac{\mathrm{\π}}{\sqrt{n_j}}+o(1/\sqrt{n_j})---\left(3\right)$

Wherein,

be independent of the number of servers of data center,

π is the function of η.

The average retardation of the j of data center is:

${\mathrm{ED}}_j=\frac{d}{\sqrt{n_j}}+o(1/\sqrt{n_j})---\left(4\right)$

Wherein, (π, η, d) is by the unique decision of ν.

Below, we are divided into three steps based on the approximate thinking of heavy duty and consider how to solve this problem.First, solve the load sharing policy of data center, load routing policy, is essentially solving of routing vector.(reason is that, when service price p determines, the aggregate demand of system is also determined.Therefore, a user market can be merged in different user markets, so route matrix P[p _ij] deteriorate to a routing vector).For the delay of M/M/n queuing system, accurate expression formula should be

ν is the congestion probability of system.

Under the condition of delay and price constraints, solve the optimal service device quantity of each data center, finally obtain optimum energy consumption solution.Detailed step is as follows:

The first step: work as n _jduring → ∞, according to above analysis, following equation is set up:

$D_k(L_k{u}_k-{\mathrm{\λ}}_k)\≈D_j(L_j{u}_j-{\mathrm{\λ}}_j)\≈1,\∀1\≤j,k\≤N$

Can obtain approximate routing vector.

Second step: making C is the total service ability of data center, by p ^*(λ _j)=p ^*(λ _k)=p ^*(C):

$\underset{j=1}{\overset{N}{\mathrm{\Σ}}}(p^{*}\left({\mathrm{\λ}}_j\right)-{\mathrm{\ω}}_j{e}_j){\mathrm{\λ}}_j=p^{*}\left(C\right)C-C\underset{j=1}{\overset{N}{\mathrm{\Σ}}}{\mathrm{\ω}}_j{e}_j{p}_j$

P wherein _jfor flow of services is assigned to the probability of the j of data center, order:

$F\left(C\right)=p^{*}\left(C\right)C-C\underset{j=1}{\overset{N}{\mathrm{\Σ}}}{\mathrm{\ω}}_j{e}_j{p}_j---\left(5\right)$

By $p^{*}={\stackrel{\‾}{F}}^{-1}\left(\frac{C}{\mathrm{\Λ}}\right),$ Obtain formula (6):

$\frac{\∂F}{\∂C}=p^{*}\left(C\right)-\frac{C}{\mathrm{\Λf}\left(p^{*}\left(C\right)\right)}-\underset{j=1}{\overset{N}{\mathrm{\Σ}}}{\mathrm{\ω}}_j{e}_j{p}_j---\left(6\right)$

Make it equal 0, simultaneous κ=C/ Λ,

can solve κ, and then draw C, p ^*(C).

The 3rd step: according to the route matrix having solved and price p ^*(C), demand function is decomposed to each data center, at this moment need to convert constraint condition.By

price range constraint condition can be changed into

for deferred constraint, by delay condition can be expressed as ${\mathrm{ED}}_j\≈\frac{1}{n_j{\mathrm{\μ}}_j-{\mathrm{\λ}}_j}\≈\frac{1}{{\mathrm{\η}}_j\sqrt{n_j}}\≈\frac{d_j}{\sqrt{n_j}}\≤D_j,$ Simultaneous $(\frac{\stackrel{\‾}{F_j}\left(p^{*}\right)}{f_j\left(p^{*}\right)}\mathrm{\η}-\mathrm{\π})=\mathrm{dq}\≈\frac{q}{\mathrm{\η}}$ :

$\frac{\stackrel{\‾}{F_j}\left(p^{*}\right)}{f_j\left(p^{*}\right)}\frac{1}{D_j\sqrt{n_j}}-q{D}_j\sqrt{n_j}\≤\mathrm{\π}$

Again by $p=p^{*}+\mathrm{\π}/\sqrt{n_j}$ ,

for aggregate demand function.Abbreviation it, the expression formula that postpones condition is:

So constraint condition is simplified a lot, just can obtain the flow that decomposes each data center.The number of servers of Wei Shi data center is minimum, then by trial method determine the value of π, finally obtain optimum price and energy input.

Above-mentioned three step solution procedurees are followed successively by by the approximate routing vector of obtaining load of heavy duty, according to routing vector, and the integrity service ability at optimization data center.Finally, again according to constraint condition, determine the optimal service device quantity at different pieces of information center, finally obtain the optimum solution of energy consumption.In solution procedure, completed the design to mechanism such as load distribution, optimal pricing and energy consumption cost optimizations.In the present invention, affect a lot of because have of server energy consumption, such as server inefficacy, burst flow etc.Data center can arrange the service quality that some standby servers (server in Idle state) guarantee user.

The above is only the preferred embodiment of the present invention; be noted that for those skilled in the art; under the premise without departing from the principles of the invention, can also make some improvements and modifications, these improvements and modifications also should be considered as protection scope of the present invention.

Claims (5)

1. the many data centers Cost Optimization Approach based on Dynamic Pricing strategy, is characterized in that: comprise the steps:

Step 1) based on waiting line theory to majority according to central user be benefited situation and service quality modeling;

Step 2) model of setting up according to step 1), establishes the optimal pricing strategy of each data center;

Step 3), according to determined optimal pricing strategy, is determined the load routing policy between many data centers.

2. the many data centers Cost Optimization Approach based on Dynamic Pricing strategy according to claim 1, it is characterized in that: in described step 1), the foundation of model need to be considered the following condition of data center: energy consumption, service ability, number of servers, power price, the scale upper limit, deferred constraint that per unit service ability consumes.

3. the many data centers Cost Optimization Approach based on Dynamic Pricing strategy according to claim 1, is characterized in that: described step 2) comprise the following steps:

Step 201) determining data center demand function;

Step 202) determine the mapping relations one by one between the same service price of demand function, number of servers;

Step 203) according to 202) in determined mapping relations, solve and obtain optimal pricing strategy.

4. the many data centers Cost Optimization Approach based on Dynamic Pricing strategy according to claim 1, is characterized in that: described step 3) comprises the following steps:

Step 301) solve the load sharing policy of data center, i.e. load routing policy;

Step 302), according to known routing vector, solve the integrity service ability that obtains data center;

Step 303) postponing and serving under the condition of price constraints, solve the optimal service device quantity of each data center, finally obtain optimum energy consumption solution.

5. the many data centers Cost Optimization Approach based on Dynamic Pricing strategy according to claim 1, is characterized in that: described data center is provided with standby server.

Images