CN106529989A - Server adaptive pricing strategy for green data center - Google Patents

Abstract

The invention relates to a server adaptive pricing strategy for a green data center, taking into account the factors such as the supply and demand of the server, the production of renewable resources, the real-time electricity price and the battery power. The server adaptive pricing strategy is based on the Lyapunov theory, which can dynamically adjust the server pricing according to the surrounding environmental condition and maximize the profit of the cloud provider.

Description

A kind of server self adaptation pricing strategy of green data center

Technical field

The invention belongs to data center server Dynamic Pricing technical field, more particularly to a kind of clothes of green data center Business device self adaptation pricing strategy.

Background technology

Cloud computing brings the new opportunities for reconstructing enterprises computing basic facility with cheap cost for modern enterprise.Amazon With the public cloud provider such as Microsoft, start (IaaS, the Infrastructure as a that service that provide infrastructures to cloud user Service) to hire out their computing resource, save the infrastructure investment that cloud user buys hardware facility.Additionally, cloud user will be numerous Cloud provider is given in trivial and complicated hardware maintenance work, focuses more on the design business logic of oneself and then improves work Efficiency.

The basic problem that cloud provider faces, is how to formulate an efficient server to hire out pricing strategy, with Maximize its profit.So far, cloud computing industry experienced various different pricing methods.Early stage, static price occupy leading Status.Although the characteristics of static pricing method has simple, as the resource left unused does not bring profit, or even can be due to demoting It is worth and increases extra expense, therefore this pricing strategy can not fully meets the demand of cloud provider profit maximization.Dynamic Pricing method causes the great interest of industrial quarters and academia as a kind of alternative pricing method.Dynamic Pricing Basic thought, is to would sit idle for resource to hire out with relatively low expense, to improve resource utilization.If resource requirement increases sharply, lifted Price is obtaining more profits.

The content of the invention

In order to overcome the shortcoming of above-mentioned prior art, it is an object of the invention to provide a kind of service of green data center Device self adaptation pricing strategy, based on the supply-demand relationship of instantaneous server, maximizes the lease of cloud service provider server as far as possible The profit of business；The pricing strategy be additionally contemplates that simultaneously regenerative resource, Spot Price and accumulator electric-quantity etc. it is uncertain because Element, and during pricing algorithm operation, it is not necessary to understand the statistical information of these uncertain factors, realize easy.

To achieve these goals, the technical solution used in the present invention is：

A kind of server self adaptation pricing strategy of green data center, describes server Dynamic Pricing by equation below Problem：

Constraints is：

d_t ^b+d_t ^g=[ρ m_t-r_t]⁺,

[r_t-ρm_t]⁺≥c_t ^r>=0,

B_t+1=B_t+η_c(c_t ^g+c_t ^r)-η_dd_t ^b,

0≤B_t≤B_max<∞,

Server Dynamic Pricing and energy operation are realized maximizing time average profit by above formula；

Wherein, P={ P_t, t=1,2 ..., T, it is a sequence of decisions, P_tIt is a decision vector, is defined as It is the instantaneous profit of the time slot t under P, d_t ^bThere is provided by the operation of accumulator back-level server Electric energy,It is the back-level server electric energy that provides of operation by electrical network, ρ is the power of a server, m_tFor t service Device resource aggregate demand, r_tFor t production of renewable energy resources amount,The electric energy of accumulator, B are filled with for regenerative resource_tFor electric power storage Pond electricity, η_cAnd η_dIt is charge efficiency and discharging efficiency respectively,The electric energy of accumulator, B are filled with for electrical network_maxFor accumulator Limiting electricity quantity, c_t ^gIt is electric energy that electrical network is provided for accumulator, C is maximum charge efficiency, and G is the maximum that can be bought from electrical network Electric energy, δ are maximum discharging efficiency, p_tFor the spot price of t server.

Director server demand is obtained by setting up data center model, it is in data center model, in time slot t, false If the load of cloud user i is λ_t,i, server price is p_t, then number of servers m leased_t,iDetermined by cost function is minimized It is fixed：

m_t,i=argmin { α_iD_i(m_t,i,λ_t,i)+p_tm_t,i}

Wherein D_i(m_t,i,λ_t,i) be cloud user disutility function, be defined as one with regard to m_t,iConvex function, its physics It is meant that the loss of income produced when the resource rented can not meet demand, α_iIt is that the disutility extent of damage is converted into into The weights of this loss, make server aggregate demand be m_t=∑_im_t,i, show that director server demand is：

λ_t,iObey arbitrary stochastic process, it is assumed that λ_t,iIt is bounded, i.e.,Wherein λ_max,iIt is have Limit constant；

In data center model, it is assumed that market of the data center in alterable electricity price is carried out, electricity price is with real-time supply-demand relationship Change, cloud provider do not have the statistical information of electricity price, do not know the electricity price in future, however, it can be opened in each time slot yet Current electricity price is observed during the beginningIt is constant in current decision time groove, and be less than certain constant, i.e.,

The constraints is drawn by setting up regenerative resource and battery model：

Assume that regenerative resource is that, without any cost, data center preferentially meets server fortune using regenerative resource Capable energy demand, in moment t, makes regenerative resource be r_t, it is assumed that

Equipped with accumulator in data center, the electric energy for wherein storing is available for future usage, carries out when electricity price is low Charge, the electric discharge when electricity price is high meets the demand of server electricity consumption, if ρ is the power of individual server, as ρ m_t-r_t>0, i.e., always When electrical energy demands exceed the electric energy provided by regenerative resource, then the extra acquisition electric energy d from electrical network is needed_t ^g, or battery discharge d_t ^bGo to meet electrical energy demands, i.e.,：

d_t ^b+d_t ^g=[ρ m_t-r_t]⁺

Otherwise, regenerative resource can charge for accumulator, i.e.,

[r_t-ρm_t]⁺≥c_t ^r≥0

As charging and discharging is operated, accumulator electric-quantity can change over time, and the dynamic changing process of battery is as follows：

B_t+1=B_t+η_c(c_t ^g+c_t ^r)-η_dd_t ^b

Under practical situation, the electricity of accumulator be it is limited, i.e.,

Following restriction is done finally for each control variable：

Dynamic state server pricing problem is solved using Liapunov optimum theory, liapunov function L is defined respectively_t With Liapunov drift △_tFor：

△_t=L_t+1=L_t,

Wherein θ>0 is a constant；

Under the constraints, Liapunov optimum theory minimizes formula △_t-VR_t, to obtain dynamic state server The optimal solution of pricing problem：

The right for minimizing above formula can be solved to problem.

By formula △_t-VR_tIt is decomposed into following two subproblems to solve：

s.t.:m_t=∑_i(D'_i)^-1(p_t,λ_t,i)

ρm_t-r_t≥0

With

s.t.:m_t=∑_i(D'_i)^-1(p_t,λ_t,i)

ρm_t-r_t≤0

In P1, it is assumed that regenerative resource electric energy is not enough to the energy demand of support data center, that is, constrain ρ m_t-r_t>=0 meets, Therefore haveDo not have unnecessary regenerative resource to charge a battery；In P2, it is assumed that have enough regenerative resources Meet data center's energy demand, that is, constrain ρ m_t-r_t≤ 0 meets；NowWith0 is only have to be equal to, that is, does not need electrical network to power And battery discharge, if ρ is m_t=r_t, then the target function value of P1 is identical with P2, and

In each decision point, two simple convex optimization problems P1 and P2, P1 are solved using dynamic state server pricing algorithm 4 and 3 control variable are only included respectively with P2, and problem scale is unrelated with cloud user and intensity of load.

The dynamic state server pricing algorithm is comprised the following steps that：

Algorithm idea is, in each decision-making time t, to solve two optimization problems of P1 and P2 respectively, and optimal solution is corresponding O is arrived in object function storage₁And O₂In；Then, method comparison O₁And O₂, and a wherein less corresponding solution is set to entirely Office's optimal solution S；Finally, charge and discharge is carried out to system battery according to decision-making electrically operated.

Compared with prior art, the present invention has considered the supply-demand relationship of server, in real time Renewable resource yield, electricity The factor such as valency and accumulator electric-quantity.The algorithm is based on Lyapunov's theory, server can be fixed a price according to ambient environmental conditions Dynamic regulation is carried out, the profit of cloud provider is maximized.

Description of the drawings

Fig. 1 is system model schematic diagram.

Specific embodiment

Describe embodiments of the present invention with reference to the accompanying drawings and examples in detail.

1. system structure：

As shown in Figure 1, cloud provider operation one has the data center of many cloud users to the system structure of the present invention.Data Central interior lays a large amount of servers.Electric power energy needed for server work is local from two：Electrical network and renewable energy Source (solar energy or wind energy).As regenerative resource has the defect of interruption, therefore between cloud provider is using accumulator equipment or not Cut-off electric system (UPS, Uninterruptible Power System), to tackle the situation of regenerative resource shortage.

2. symbol description：

Symbol description in the present invention is as shown in table 1.

1 symbol description of table

3. data center model：

A. resource requirement description：The target of cloud user, is to determine number of servers to be leased, to minimize rental charge With.More precisely, in time slot t, it is assumed that the load of cloud user i is λ_t,i, server price is p_t, then the service leased Device quantity m_t,iCan be determined by cost function is minimized：

m_t,i=argmin { α_iD_i(m_t,i,λ_t,i)+p_tm_t,i} (1)

Wherein D_i(m_t,i,λ_t,i) be cloud user disutility function, be defined as one with regard to m_t,iConvex function, its physics It is meant that the loss of income produced when the resource rented can not meet demand.Parameter alpha_iIt is that the disutility extent of damage is converted For the weights of cost allowance.In order to obtain the solution of formula (1), to m_t,iDerivation and make which be equal to 0, obtain α_iD'_i+p_t=0.Therefore HaveServer aggregate demand is made to be m_t=∑_im_t,i, it can be deduced that director server demand is：

For Resources requirement model (2), λ_t,iArbitrary stochastic process may be obeyed.λ is only assumed that now_t,iIt is bounded, i.e.,Wherein λ_max,iIt is limited constant.

B. Spot Price：Assume market of the data center in alterable electricity price is carried out, electricity price changes with real-time supply-demand relationship. Cloud provider does not have the statistical information of electricity price, does not know the electricity price in future yet.However, it can be seen when each time slot starts Measure current electricity priceIt is constant in current decision time groove, and be less than certain constant, i.e.,

4. regenerative resource and battery model：

Assume that regenerative resource is without any cost.In order to save power consumption, data center preferentially can use can be again Give birth to the energy to meet the energy demand of server operation.In moment t, regenerative resource is made to be r_t, it is assumed that

Equipped with accumulator in data center, the electric energy for wherein storing is available for future usage.Using accumulator another The Electricity price fluctuation of advantage Shi Shi data center is smoothened：It is charged when electricity price is low, the electric discharge when electricity price is high meets The demand of server electricity consumption.If ρ is the power of individual server, as ρ m_t-r_t>0, i.e., total electrical energy demands exceed regenerative resource institute During the electric energy of offer, then the extra acquisition electric energy d from electrical network is needed_t ^g, or battery discharge d_t ^bGo to meet electrical energy demands, i.e.,：

d_t ^b+d_t ^g=[ρ m_t-r_t]⁺ (3)

Otherwise, regenerative resource can charge for accumulator, i.e.,

[r_t-ρm_t]⁺≥c_t ^r≥0 (4)

Wherein c_t ^rIt is electric energy that regenerative resource is provided for accumulator.

As charging and discharging is operated, accumulator electric-quantity can change over time.The dynamic changing process of battery is as follows：

B_t+1=B_t+η_c(c_t ^g+c_t ^r)-η_dd_t ^b (5)

Wherein η_cAnd η_dIt is charge efficiency and discharging efficiency respectively.c_t ^gIt is electric energy that electrical network is provided for accumulator.Actual feelings Under condition, the electricity of accumulator be it is limited, i.e.,

Following restriction is done finally for each control variable：

Formula (7) describes the charge efficiency of maximum and is limited by a constant C.Formula (8) is described can be from electrical network The maximum power of middle purchase is G.Formula (9) represents that maximum discharging efficiency is δ.

5. dynamic state server pricing problem description：

The target of dynamic state server pricing strategy, is to make control decision in each time slot t To maximize its profit.The profit of cloud provider can be stated with following formula：

Wherein preceding paragraph is with price p_tRent m_tIncome obtained by platform server, consequent is produced by purchase electrical network electric energy Cost.

Make P={ p_t, t=1,2 ..., T is a sequence of decisions, andIt is the instantaneous profit of the time slot t under P.It is logical Cross following equation and describe dynamic state server pricing problem：

Constraints be (3), (4), (5), (6), (7), (8), (9), (10),

Dynamic state server pricing problem is by maximizing time average profit to server Dynamic Pricing and energy operation.

Specifically, the present invention solves dynamic state server pricing problem using Liapunov optimum theory：

A. liapunov function L is defined respectively_tWith Liapunov drift △_tFor：

△_t=L_t+1=L_t,

Wherein θ>0 is a constant.

B. Liapunov optimum theory minimizes formula (13), to obtain the optimal solution of dynamic state server pricing problem.

△_t-VR_t (14)

(3) are constrained to, (4), (5), (6), (7), (8), (9), (10).

Have by battery dynamic formula (5)

To above formula both sides square and arrange obtain

Wherein B is constant and is defined as follows

Both sides addition object function has

Therefore, problem is solved by the right for only needing minimum (17).

Algorithm design is as follows：

A. due to (3), the presence of (4) formula constraint directly can not be minimized on the right of (15) formula.The problem can be broken down into Following two subproblems solve：

s.t.:m_t=∑_i(D'_i)^-1(p_t,λ_t,i)

ρm_t-r_t≥0

With

s.t.:m_t=∑_i(D'_i)^-1(p_t,λ_t,i)

ρm_t-r_t≤0

Observation can be pinpointed the problems the solution of (15), must be the solution of P1 or P2.In P1, it is assumed that regenerative resource electric energy is not enough to prop up Hold data center's energy demand (constraint ρ m_t-r_t>=0 meets), therefore haveThere is no unnecessary regenerative resource to storage Battery charges.In P2, it is assumed that have enough regenerative resources to meet data center's energy demand (constraint ρ m_t-r_t≤ 0 meets). NowWith0 is only have to be equal to, that is, does not need electrical network to power and battery discharge.If ρ is m_t=r_t, then the target function value of P1 It is identical with P2, and

B. dynamic state server pricing algorithm.As shown in algorithm 1, in each decision point, algorithm 1 solves two simply to algorithm Convex optimization problem P1 and P2.P1 and P2 only includes 4 and 3 control variable respectively, and problem scale is strong with cloud user and load Degree is unrelated.Therefore, the algorithm can be used in large-scale data center.

Algorithm 1:Dynamic state server pricing algorithm, concrete steps：

。

Claims (7)

1. the server self adaptation pricing strategy of a kind of green data center, it is characterised in that serviced by equation below description Device Dynamic Pricing problem：

$\underset{P}{\max }\underset{T\&RightArrow;\mathrm{\&infin;}}{\lim }\frac{1}{T}\underset{t=0}{\overset{T-1}{\&Sigma;}}{R}_t^p,$

Constraints is：

d_t ^b+d_t ^g=[ρ m_t-r_t]⁺,

[r_t-ρm_t]⁺≥c_t ^r>=0,

B_t+1=B_t+η_c(c_t ^g+c_t ^r)-η_dd_t ^b,

0≤B_t≤B_max<∞,

$c_t^g+c_t^r\&le;C,$

$d_t^g+c_t^g\&le;G,$

$d_t^b\&le;\mathrm{\&delta;},$

$c_t^g,c_t^r,d_t^g,d_t^b,p_t\&GreaterEqual;0,\&ForAll;t;$

Server Dynamic Pricing and energy operation are realized maximizing time average profit by above formula；

Wherein, P={ P_t, t=1,2 ..., T, it is a sequence of decisions, P_tIt is a decision vector, is defined as It is the instantaneous profit of the time slot t under P, d_t ^bThere is provided by the operation of accumulator back-level server Electric energy,It is the back-level server electric energy that provides of operation by electrical network, ρ is the power of a server, m_tFor t service Device resource aggregate demand, r_tFor t production of renewable energy resources amount,The electric energy of accumulator, B are filled with for regenerative resource_tFor electric power storage Pond electricity, η_cAnd η_dIt is charge efficiency and discharging efficiency respectively,The electric energy of accumulator, B are filled with for electrical network_maxFor the pole of accumulator The amount of rationing the power supply, c_t ^gIt is electric energy that electrical network is provided for accumulator, C is maximum charge efficiency, and G is the maximum electricity that can be bought from electrical network Can, δ is maximum discharging efficiency, p_tFor the spot price of t server.

2. the server self adaptation pricing strategy of green data center according to claim 1, it is characterised in that by setting up Data center model obtains director server demand, in data center model, in time slot t, it is assumed that the load of cloud user i is λ_t,i, server price is p_t, then number of servers m leased_tiDetermined by cost function is minimized：

m_t,i=argmin { α_iD_i(m_t,i,λ_t,i)+p_tm_t,i}

Wherein D_i(m_t,i,λ_t,i) be cloud user disutility function, be defined as one with regard to m_t,iConvex function, its physical meaning is The loss of income produced when the resource rented can not meet demand, α_iIt is that the disutility extent of damage is converted into into cost allowance Weights, make server aggregate demand be m_t=∑_im_t,i, show that director server demand is：

$m_t=\underset{i}{\&Sigma;}{\left({D^{\&prime;}}_i\right)}^{-1}(\frac{p_t}{{\mathrm{\&alpha;}}_i},{\mathrm{\&lambda;}}_{t,i})$

λ_{T, i}Obey arbitrary stochastic process, it is assumed that λ_tiIt is bounded, i.e.,Wherein λ_max,iIt is limited normal Number；

In data center model, it is assumed that market of the data center in alterable electricity price is carried out, electricity price change with real-time supply-demand relationship, Cloud provider does not have the statistical information of electricity price, does not know the electricity price in future, however, it can be seen when each time slot starts yet Measure current electricity priceIt is constant in current decision time groove, and be less than certain constant, i.e.,

3. the server self adaptation pricing strategy of green data center according to claim 1, it is characterised in that the constraint Condition is drawn by setting up regenerative resource and battery model：

Assume that regenerative resource is that, without any cost, data center preferentially meets server operation using regenerative resource Energy demand, in moment t, makes regenerative resource be r_t, it is assumed that

$0\&le;r_t\&le;r_{max},\&ForAll;t$

Equipped with accumulator in data center, the electric energy for wherein storing is available for future usage, is charged when electricity price is low, When electricity price is high, electric discharge meets the demand of server electricity consumption, if ρ is the power of individual server, as ρ m_t-r_t>0, i.e., total electric energy is needed When the electric energy provided more than regenerative resource is provided, then need extra to obtain electric energy d from electrical network_t ^g, or battery discharge d_t ^bGo full Sufficient electrical energy demands, i.e.,：

d_t ^b+d_t ^g=[ρ m_t-r_t]⁺

Otherwise, regenerative resource can charge for accumulator, i.e.,

[r_t-ρm_t]⁺≥c_t ^r≥0

As charging and discharging is operated, accumulator electric-quantity can change over time, and the dynamic changing process of battery is as follows：

B_t+1=B_t+η_c(c_t ^g+c_t ^r)-η_dd_t ^b

Under practical situation, the electricity of accumulator be it is limited, i.e.,

$0\&le;B_t\&le;B_{max}<\mathrm{\&infin;},\&ForAll;t$

Following restriction is done finally for each control variable：

$c_t^g+c_t^r\&le;C$

$d_t^g+c_t^g\&le;G$

$d_t^b\&le;\mathrm{\&delta;}$

$c_t^g,c_t^r,d_t^g,d_t^b,p_t\&GreaterEqual;0.$

4. the server self adaptation pricing strategy of green data center according to claim 1, it is characterised in that utilize Li Ya Pu Nuofu optimum theories solve dynamic state server pricing problem, define liapunov function L respectively_tDrift about with Liapunov △_tFor：

$L_t=\frac{1}{2}{(B_t-\mathrm{\&theta;})}^2,$

△_t=L_t+1=L_t,

Wherein θ>0 is a constant；

Under the constraints, Liapunov optimum theory minimizes formula △_t-VR_t, to obtain dynamic state server price The optimal solution of problem：

$\begin{array}{c}{\mathrm{\&Delta;}}_t-{\mathrm{VR}}_t\\ \&le;B+(B_t-\mathrm{\&theta;})\left({\mathrm{\&eta;}}_c\right(c_t^q+c_t^r)-{\mathrm{\&eta;}}_d{d}_t^b)-V(m_t{p}_t-(d_t^g+c_t^g\left)c_t^p\right)\\ \&le;B+\left(\right(B_t-\mathrm{\&theta;}){\mathrm{\&eta;}}_c+{\mathrm{Vc}}_t^p)c_t^g+(B_t-\mathrm{\&theta;}){\mathrm{\&eta;}}_c{c}_t^r-(B_t-\mathrm{\&theta;}){\mathrm{\&eta;}}_d{d}_t^b\\ -{\mathrm{Vm}}_t{p}_t+{\mathrm{Vc}}_t^p{d}_t^g\end{array}.$

The right for minimizing above formula can be solved to problem.

5. the server self adaptation pricing strategy of green data center according to claim 4, it is characterised in that by formula △_t-VR_tIt is decomposed into following two subproblems to solve：

$\begin{array}{c}P1:min\left(\right(B_t-\mathrm{\&theta;}){\mathrm{\&eta;}}_c+{\mathrm{Vc}}_t^p)c_t^q-(B_t-\mathrm{\&theta;}){\mathrm{\&eta;}}_d{d}_t^b\\ -{\mathrm{Vm}}_t{p}_t+{\mathrm{Vc}}_t^p{d}_t^g\end{array}$

s.t.:m_t=∑_i(D'_i)^-1(p_t,λ_t,i)

ρm_t-r_t≥0

$d_t^g+d_t^b={\mathrm{\&rho;m}}_t-r_t$

$c_t^g\&le;C$

$d_t^g+c_t^g\&le;G$

$d_t^b\&le;\mathrm{\&delta;}$

$c_t^g,d_t^g,d_t^b,p_t\&GreaterEqual;0$

With

$P2:min\left(\right(B_t-\mathrm{\&theta;}){\mathrm{\&eta;}}_c+{\mathrm{Vc}}_t^p)c_t^g+(B_t-\mathrm{\&theta;}){\mathrm{\&eta;}}_c{c}_t^r-{\mathrm{Vm}}_t{p}_t$

s.t.:m_t=∑_i(D'_i)^-1(p_t,λ_t,i)

ρm_t-r_t≤0

$r_t-{\mathrm{\&rho;m}}_t\&GreaterEqual;c_t^r\&GreaterEqual;0$

$c_t^g+c_t^r\&le;C$

$c_t^g\&le;G$

$c_t^g,c_t^r,p_t\&GreaterEqual;0$

In P1, it is assumed that regenerative resource electric energy is not enough to the energy demand of support data center, that is, constrain ρ m_t-r_t>=0 meets, therefore HaveDo not have unnecessary regenerative resource to charge a battery；In P2, it is assumed that have enough regenerative resources to meet Data center's energy demand, that is, constrain ρ m_t-r_t≤ 0 meets；NowWith0 is only have to be equal to, that is, does not need electrical network to power and electricity Tank discharge, if ρ is m_t=r_t, then the target function value of P1 is identical with P2, and

6. the server self adaptation pricing strategy of green data center according to claim 5, it is characterised in that determine at each Plan point, solves two simple convex optimization problems P1 and P2, P1 using dynamic state server pricing algorithm and P2 only includes 4 respectively With 3 control variable, and problem scale is unrelated with cloud user and intensity of load.

7. the server self adaptation pricing strategy of green data center according to claim 6, it is characterised in that the dynamic Server pricing algorithm is comprised the following steps that：

Algorithm idea is, in each decision-making time t, to solve two optimization problems of P1 and P2 respectively, and by optimal solution corresponding target O is arrived in function storage₁And O₂In；Then, method comparison O₁And O₂, and a wherein less corresponding solution is set to into the overall situation most Excellent solution S；Finally, charge and discharge is carried out to system battery according to decision-making electrically operated.