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

Abstract

The invention discloses a multi-data center cost optimization method based on a dynamic pricing strategy, which includes the following steps: step 1) modeling the user benefits and service quality of multi-data centers based on queuing theory; step 2) establishing according to step 1) Based on the model, the optimal pricing strategy of each data center is established; Step 3) According to the determined optimal pricing strategy, determine the load routing strategy between multiple data centers. Starting from the interests of service providers, the method of the present invention unifies the load distribution among multiple data centers, service pricing and energy consumption cost optimization; on the basis of satisfying the interests of service providers, the energy consumption of data centers is realized through a load balancing algorithm cost optimization.

Description

A Multi-data Center Cost Optimization Method Based on Dynamic Pricing Strategy

technical field

The present invention relates to the field of cloud computing and data center management, in particular to a multi-data center cost optimization method based on a dynamic pricing strategy.

Background technique

At present, cloud providers generally adopt a flexible pricing scheme to adjust the volume of service requests to maximize their own benefits. For example, Amazon sells cloud resources cheaply when the demand is low, so as to improve the utilization rate of resources and realize the optimization of its business benefits.

Obviously, different price levels have a direct impact on market demand. As shown in Figure 1(a), when the price increases, the quantity demanded decreases, and the relationship between the two is inversely proportional, and this relationship presents certain random fluctuations. Figure 1(b) shows how the quantity demanded changes with time at three different price levels. It can be seen that the price level determines the demand (but not the only determinant), and the demand has a significant impact on the energy management of the data center. At present, many cloud providers, such as Amazon and Google, have established multiple data centers in many places around the world. Deploying data centers in different geographical locations is not only convenient for expanding the cloud service market, but also helps to reduce energy consumption costs (using temperature difference, electricity price difference, etc.). Consider the multi-data center environment shown in Figure 2: Assume that cloud provider P has established multiple data centers in different regions. Due to geographical differences, the electricity prices and user markets are different. For example, at time t ₀ , the service price of the data center is p ₀ , and at time t ₁ , the cloud provider P lowers the service price to p1 in order to stimulate consumption. At this time, the market demand in various places may have different price adjustments reaction. At the same time, electricity prices in various places may also have changed (such as the adoption of tiered electricity prices or time-of-use electricity prices). At this time, if the load can be shifted to the data center with lower demand and electricity price, it will be able to effectively improve the utilization rate of resources and reduce the energy consumption cost of the data center.

Due to the heterogeneity and cross-region characteristics of different data centers, the above optimization strategy faces the following challenges:

1. Load transfer takes a certain amount of time, which may cause a decline in user QoS performance;

2. The service capabilities of different data centers are not the same, and the scale is limited, and the load distribution must consider the heterogeneity of the data center;

3. The purpose of data center price adjustment is to maximize its own interests. The energy consumption optimization strategy should be coordinated with the price adjustment strategy, and there is no conflict between the two;

4. The load flow in the real environment is affected by many factors and has random fluctuations. The energy consumption optimization strategy should be able to effectively adapt to the dynamic changes of the load flow.

Contents of the invention

Purpose of the invention: In order to overcome the deficiencies in the prior art, the present invention provides a multi-data center cost optimization method based on a dynamic pricing strategy. Aiming at the characteristics of differences in electricity price costs and service flows among multiple data centers, a method is proposed. The energy cost optimization mechanism based on the dynamic pricing strategy in a multi-data center environment considers service pricing, load distribution, and server status control mechanisms in a unified way to formulate a coordinated system optimization plan.

Technical scheme: in order to achieve the above object, the technical scheme adopted in the present invention is:

A multi-data center cost optimization method based on a dynamic pricing strategy, which considers service pricing, load distribution, and server status adjustment and other related strategies, and formulates a coordinated system optimization plan to effectively reduce data center energy consumption costs. Purpose. Firstly, the change model of market demand is established, and the relationship between performance indicators such as service price, task response time and service capability is analyzed by using queuing theory, and a reasonable load routing mechanism and energy cost optimization strategy with QoS guarantee are designed accordingly. Specifically include the following steps:

Step 1) Based on queuing theory, model the benefit situation and service quality of multi-data center users;

Step 2) Based on the model established in step 1), establish the optimal pricing strategy for each data center;

Step 3) Determine the load routing strategy among multiple data centers according to the determined optimal pricing strategy.

Specifically, in the step 1), the establishment of the model needs to consider the following conditions of the data center: energy consumption per unit of service capacity, service capacity, number of servers, power price (cost price of the data center), upper limit of scale, delay constraint.

Specifically, the step 2) includes the following steps:

Step 201) Determine the data center demand function;

Step 202) Determine the one-to-one mapping relationship between the demand function, the service price (selling price), and the number of servers;

Step 203) According to the mapping relationship determined in 202), the optimal pricing strategy is obtained by solving.

Specifically, the step 3) includes the following steps:

Step 301) Solve the load distribution strategy of the data center, that is, the load routing strategy;

Step 302) Solve to obtain the overall service capability of the data center according to the known routing vector;

Step 303) Under the constraints of delay and service price, solve for the optimal number of servers in each data center, and finally obtain the optimal energy consumption solution.

The data center is provided with backup servers. Since there are many factors of server energy consumption, such as server failure, burst traffic, etc., setting up backup servers (servers in an idle state) can guarantee the quality of service for users.

Beneficial effects: Compared with the prior art, the multi-data center cost optimization method based on the dynamic pricing strategy provided by the present invention has the following advantages:

1. Among multiple data centers, according to the difference in service demand and electricity price between different data centers, a coordinated system optimization plan has been formulated by considering dynamic service pricing and load distribution strategies; it not only reduces the cost of data centers The overall energy consumption cost, and optimize the data center's revenue;

2. By analyzing the relationship between service pricing and data center energy consumption costs, establish a data center service pricing and load routing model; starting from the interests of service providers, load distribution among multiple data centers, service pricing and energy consumption Unify the energy consumption cost optimization; realize the optimization of the data center energy consumption cost through the load balancing algorithm on the basis of satisfying the interests of service providers;

3. Organically combine data center management issues and pricing issues, providing new ideas for effectively solving cloud computing data center management issues;

4. The data center pricing method and load routing strategy are simple, effective, and highly accurate, and can be applied to large-scale data center environments.

Description of drawings

Figure 1 is a schematic diagram of the relationship between service prices and changes in market demand; where (a) is the relationship between quantity and price, and (b) is the relationship between quantity and time;

FIG. 2 is a schematic diagram of a multi-data center architecture.

Detailed ways

The present invention will be further described below in conjunction with the accompanying drawings.

A multi-data center cost optimization method based on a dynamic pricing strategy, comprising the following steps:

Step 1) Based on queuing theory, model the benefit situation and service quality of multi-data center users;

Step 2) Based on the model established in step 1), establish the optimal pricing strategy for each data center;

Step 3) Determine the load routing strategy among multiple data centers according to the determined optimal pricing strategy.

The data center sets a service price for each user who accesses the service in order to obtain a certain amount of income. Users accessing the data center need to consider not only the service price, but also the service quality. In order to simplify the model, this case uses the delay cost to describe the user's satisfaction with the service quality. The longer the task response time, the higher the delay cost, and the lower the user's service quality satisfaction. Delay cost can be expressed as a function of task response time, and commonly used cost functions include linear function, exponential function, etc. Only when the sum of the service price and the delay cost is less than the user's reserved price (psychological price), the user will choose to visit the data center. For the multi-data center environment shown in Figure 2, the large-scale queuing theory based on overloaded approximation is used to model it as a queuing model. Based on this model, the relationship between different system parameters (service price, delay cost, number of servers, etc.) is studied in a stable system state. In order to clearly express the content of this case, the notation and description of the variables in this case are listed in Table 1. The specific definition of variables will be elaborated in the text.

Symbols of the variables in Table 1

Assume that the service price set by the service provider is p>0, and the task delay is ED. In this case, the linear function q*ED is used to represent the user's delay cost, where q is a constant, which is the coefficient of the user's delay cost. The user's reservation price is a random variable v, which obeys the probability cumulative distribution function F. When p is given, users can decide whether to visit the data center according to their own reservation price v, through the probability P (v≥p+qED). That is, only when the user's reserved price is greater than the sum of the delay cost and the service price, the user will choose to visit the data center. This article mainly analyzes the system behavior when the data center is in a steady state, then the user demand rate should satisfy the following demand relationship:

λ(p)=ψP(v≥p+qED)=ψ(1-F(p+qED))

where ψ is the potential demand in the market. That is, the actual number of visits of users is equal to the potential demand of the market multiplied by the visit probability. The above demand model is commonly used in various economic models. Next, define the concept of demand function elasticity to measure the sensitivity of service demand to service price changes in the market. Only when the demand function satisfies the elastic condition can the relevant conclusions of this case be established. If the market demand does not meet the elastic conditions, that is, when the price decreases or increases, the demand does not change, then the relevant conclusions of this paper are no longer applicable.

Definition 1: Let λ(p) be the demand function of customers, and p be the service price, then the elasticity of λ(p) can be defined as:

$\mathrm{<span\; class="notranslate">\; \&epsiv;</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; p</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span><span\; class="notranslate">\; -</span>\frac{<span\; class="notranslate">\; \&PartialD;</span>\mathrm{<span\; class="notranslate">\; \&lambda;</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; p</span>}<span\; class="notranslate">\; )</span>}{<span\; class="notranslate">\; \&PartialD;</span>\mathrm{<span\; class="notranslate">\; p</span>}}<span\; class="notranslate">\; \&CenterDot;</span>\frac{\mathrm{<span\; class="notranslate">\; p</span>}}{\mathrm{<span\; class="notranslate">\; \&lambda;</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; p</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span>$

If λ(p) satisfies ε(p)>1 on the interval [a,b], then λ(p) is said to be elastic. The intuitive explanation of demand elasticity is: due to the effect of economic laws, there is an inverse relationship between price and demand. That is, when the price rises, the quantity demanded falls, and vice versa. The concept of elasticity aims to illustrate the relationship between the rate of change in price and the rate of change in quantity demanded. This simple yet important relationship between price and demand functions is widely used in various forms of economic models.

i=1，2，…，M（即生成的任务到达强度），p为服务价格。则按照上文分析，

Although data centers are built in different regions with different power prices and service capabilities, the service prices should be consistent, that is, the prices are regionally independent. For example, no matter where the user is, they need the same points to download the data of a certain website. We assume that the demand functions of different markets satisfy the elastic conditions defined above. Users access the data center through agents distributed in various places, so that the distribution functions of user reservation prices in different regional markets are respectively F _i , and the demand function is

i=1, 2, ..., M (that is, the arrival strength of the generated tasks), and p is the service price. According to the above analysis,

make For the total demand of users everywhere, and define the function

$\mathrm{<span\; class="notranslate">\; f</span>}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}^{\mathrm{<span\; class="notranslate">\; m</span>}}{\mathrm{<span\; class="notranslate">\; \&psi;</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}{\mathrm{<span\; class="notranslate">\; f</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; /</span>{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}^{\mathrm{<span\; class="notranslate">\; m</span>}}{\mathrm{<span\; class="notranslate">\; \&psi;</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}$

So it can be deduced as follows:

为任务的路由概率，即Agent i的任务被送到数据中心j的概率为p_ij。因此，[p_ij]构成了一个路由矩阵。令

于是数据中心j的到达强度为

当需求函数满足弹性关系时，有如下定理：make

is the routing probability of the task, that is, the probability that the task of Agent i is sent to data center j is p _ij . Therefore, [p _ij ] constitutes a routing matrix. make

Then the arrival strength of data center j is

When the demand function satisfies the elasticity relation, there is the following theorem:

和数据中心总服务能力 ${\mathrm{\&Sigma;}}_{j=1}^N{n}_j{\mathrm{\&mu;}}_j$ 保持线性关系，即 ${\mathrm{\&Sigma;}}_{j=1}^N{n}_j{\mathrm{\&mu;}}_j=\mathrm{\&kappa;\&psi;},$ 那么，对于任何满足 ${\mathrm{\&Sigma;}}_{i=1}^M{p}_{\mathrm{ij}}=n_j{\mathrm{\&mu;}}_j/\mathrm{\&psi;\&kappa;}$ 的概率路由矩阵P[p_ij]，和服务价格p>0，下列等式成立：Theorem 1: Assume that the agent's potential maximum total demand

and the total service capacity of the data center ${\mathrm{\&Sigma;}}_{j=1}^N{\mathrm{no}}_j{\mathrm{\&mu;}}_j$ maintain a linear relationship, that is, ${\mathrm{\&Sigma;}}_{j=1}^N{\mathrm{no}}_j{\mathrm{\&mu;}}_j=\mathrm{\&kappa;\&psi;},$ Then, for any satisfying ${\mathrm{\&Sigma;}}_{i=1}^m{p}_{\mathrm{ij}}={\mathrm{no}}_j{\mathrm{\&mu;}}_j/\mathrm{\&psi;\&kappa;}$ The probability routing matrix P[p _ij ], and the service price p>0, the following equation holds:

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

The total arrival rate of user requests is:

The service price p has the following structure:

$\mathrm{<span\; class="notranslate">\; p</span>}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; p</span>}}^{<span\; class="notranslate">\; *</span>}<span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}^{\mathrm{<span\; class="notranslate">\; N</span>}}{\mathrm{<span\; class="notranslate">\; p</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}\frac{\mathrm{<span\; class="notranslate">\; \&pi;</span>}}{\sqrt{{\mathrm{<span\; class="notranslate">\; no</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}}}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; o</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; /</span>\sqrt{{\mathrm{<span\; class="notranslate">\; no</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 3</span>}<span\; class="notranslate">\; )</span>$

独立于数据中心的服务器数量，

π是η的函数。in,

independent of the number of servers in the data center,

π is a function of η.

The average latency of data center j is:

${\mathrm{<span\; class="notranslate">\; ED</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; d</span>}}{\sqrt{{\mathrm{<span\; class="notranslate">\; no</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}}}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; o</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; /</span>\sqrt{{\mathrm{<span\; class="notranslate">\; no</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 4</span>}<span\; class="notranslate">\; )</span>$

where (π, η, d) is uniquely determined by ν.

ν为系统的拥塞概率。Below, we consider how to solve this problem in three steps based on the idea of overloaded approximation. First, solve the load distribution strategy of the data center, that is, the load routing strategy, which is essentially the solution of the routing vector. (The reason is that when the service price p is determined, the total demand of the system is also determined. Therefore, different user markets can be merged into one user market, so the routing matrix P[p _ij ] degenerates into a routing vector). For the delay of an M/M/n queuing system, the exact expression should be

ν is the congestion probability of the system.

Under the conditions of delay and price constraints, the optimal number of servers in each data center is solved, and the optimal energy consumption solution is finally obtained. The detailed steps are as follows:

Step 1: When n _j →∞, according to the above analysis, the following equation holds true:

${\mathrm{<span\; class="notranslate">\; D.</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; L</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}{\mathrm{<span\; class="notranslate">\; u</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; \&lambda;</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \&ap;</span>{\mathrm{<span\; class="notranslate">\; D.</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; L</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}{\mathrm{<span\; class="notranslate">\; u</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; \&lambda;</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \&ap;</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; ,</span><span\; class="notranslate">\; \&ForAll;</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; \&le;</span>\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; \&le;</span>\mathrm{<span\; class="notranslate">\; N</span>}$

An approximate routing vector can be found.

Step 2: Let C be the total service capacity of the data center, from p ^* (λ _j )=p ^* (λ _k )=p ^* (C), then:

$\underset{\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; N</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; p</span>}}^{<span\; class="notranslate">\; *</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; \&lambda;</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; \&omega;</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}{\mathrm{<span\; class="notranslate">\; e</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; )</span>{\mathrm{<span\; class="notranslate">\; \&lambda;</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; p</span>}}^{<span\; class="notranslate">\; *</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; C</span>}<span\; class="notranslate">\; )</span>\mathrm{<span\; class="notranslate">\; C</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; C</span>}\underset{\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; N</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{\mathrm{<span\; class="notranslate">\; \&omega;</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}{\mathrm{<span\; class="notranslate">\; e</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}{\mathrm{<span\; class="notranslate">\; p</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}$

where p _j is the probability that service traffic is allocated to data center j, so that:

$\mathrm{<span\; class="notranslate">\; f</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; C</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; p</span>}}^{<span\; class="notranslate">\; *</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; C</span>}<span\; class="notranslate">\; )</span>\mathrm{<span\; class="notranslate">\; C</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; C</span>}\underset{\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; N</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{\mathrm{<span\; class="notranslate">\; \&omega;</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}{\mathrm{<span\; class="notranslate">\; e</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}{\mathrm{<span\; class="notranslate">\; p</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 5</span>}<span\; class="notranslate">\; )</span>$

Depend on $p^{*}={\stackrel{\&OverBar;}{f}}^{-1}\left(\frac{C}{\mathrm{\&Lambda;}}\right),$ Get formula (6):

$\frac{<span\; class="notranslate">\; \&PartialD;</span>\mathrm{<span\; class="notranslate">\; f</span>}}{<span\; class="notranslate">\; \&PartialD;</span>\mathrm{<span\; class="notranslate">\; C</span>}}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; p</span>}}^{<span\; class="notranslate">\; *</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; C</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span>\frac{\mathrm{<span\; class="notranslate">\; C</span>}}{\mathrm{<span\; class="notranslate">\; \&Lambda;f</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; p</span>}}^{<span\; class="notranslate">\; *</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; C</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; -</span>\underset{\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; N</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{\mathrm{<span\; class="notranslate">\; \&omega;</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}{\mathrm{<span\; class="notranslate">\; e</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}{\mathrm{<span\; class="notranslate">\; p</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 6</span>}<span\; class="notranslate">\; )</span>$

即可解出κ，进而得出C，p^*(C)。Let it be equal to 0, and simultaneously κ=C/Λ,

Then κ can be solved, and then C, p ^* (C) can be obtained.

价格区间约束条件可变为

对于延迟约束，由延迟条件可表示为 ${\mathrm{ED}}_j\&ap;\frac{1}{n_j{\mathrm{\&mu;}}_j-{\mathrm{\&lambda;}}_j}\&ap;\frac{1}{{\mathrm{\&eta;}}_j\sqrt{n_j}}\&ap;\frac{d_j}{\sqrt{n_j}}\&le;D_j,$ 联立 $(\frac{\stackrel{\&OverBar;}{F_j}\left(p^{*}\right)}{f_j\left(p^{*}\right)}\mathrm{\&eta;}-\mathrm{\&pi;})=\mathrm{dq}\&ap;\frac{q}{\mathrm{\&eta;}}$ 得：Step 3: According to the solved routing matrix and price p ^* (C), decompose the demand function into each data center, and then need to change the constraints. Depend on

The price range constraint can be changed to

For delay constraints, by The delay condition can be expressed as ${\mathrm{ED}}_j\&ap;\frac{1}{{\mathrm{no}}_j{\mathrm{\&mu;}}_j-{\mathrm{\&lambda;}}_j}\&ap;\frac{1}{{\mathrm{\&eta;}}_j\sqrt{{\mathrm{no}}_j}}\&ap;\frac{d_j}{\sqrt{{\mathrm{no}}_j}}\&le;{\mathrm{D.}}_j,$ Simultaneous $(\frac{\stackrel{\&OverBar;}{f_j}\left(p^{*}\right)}{f_j\left(p^{*}\right)}\mathrm{\&eta;}-\mathrm{\&pi;})=\mathrm{dq}\&ap;\frac{q}{\mathrm{\&eta;}}$ have to:

$\frac{\stackrel{<span\; class="notranslate">\; \&OverBar;</span>}{{\mathrm{<span\; class="notranslate">\; f</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; p</span>}}^{<span\; class="notranslate">\; *</span>}<span\; class="notranslate">\; )</span>}{{\mathrm{<span\; class="notranslate">\; f</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; p</span>}}^{<span\; class="notranslate">\; *</span>}<span\; class="notranslate">\; )</span>}\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{{\mathrm{<span\; class="notranslate">\; D.</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}\sqrt{{\mathrm{<span\; class="notranslate">\; no</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}}}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; q</span>}{\mathrm{<span\; class="notranslate">\; D.</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}\sqrt{{\mathrm{<span\; class="notranslate">\; no</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}}<span\; class="notranslate">\; \&le;</span>\mathrm{<span\; class="notranslate">\; \&pi;</span>}$

为总需求函数。化简之，则延迟条件的表达式为：Then by $p=p^{*}+\mathrm{\&pi;}/\sqrt{{\mathrm{no}}_j}$ have to,

is the aggregate demand function. Simplified, the expression of the delay condition is:

Therefore, the constraint conditions are greatly simplified, and the traffic decomposed to each data center can be obtained. In order to minimize the number of servers in the data center, the heuristic Determine the value of π, and finally get the optimal price and energy consumption.

The above three-step solution process is to approximate the load routing vector by overloading in turn, and optimize the overall service capability of the data center according to the routing vector. Finally, according to the constraints, determine the optimal number of servers in different data centers, and finally obtain the optimal solution for energy consumption. In the process of solving, the design of mechanisms such as load distribution, optimal pricing and energy cost optimization is completed. In the present invention, there are many factors affecting the energy consumption of the server, such as server failure, burst traffic, and the like. The data center will set up some standby servers (servers in an idle state) to ensure the quality of service for users.

The above is only a preferred embodiment of the present invention, it should be pointed out that for those of ordinary skill in the art, without departing from the principle of the present invention, some improvements and modifications can also be made. It should be regarded as the protection scope of the present invention.

Claims (5)

1. A multi-data center cost optimization method based on a dynamic pricing strategy is characterized by comprising the following steps: the method comprises the following steps:

step 1) modeling benefit conditions and service quality of multi-data center users based on queuing theory;

step 2) according to the model established in the step 1), establishing an optimal pricing strategy of each data center;

and 3) determining a load routing strategy among multiple data centers according to the determined optimal pricing strategy.

2. The method for multiple data center cost optimization based on dynamic pricing strategy of claim 1, wherein: in the step 1), the following conditions of the data center need to be considered for establishing the model: energy consumption per unit service capacity, number of servers, price of electricity, upper limit of scale, delay constraints.

3. The method for multiple data center cost optimization based on dynamic pricing strategy of claim 1, wherein: the step 2) comprises the following steps:

step 201) determining a data center demand function;

step 202) determining a one-to-one mapping relation among the demand function, the service price and the number of servers;

step 203) solving to obtain the optimal pricing strategy according to the mapping relation determined in step 202).

4. The method for multiple data center cost optimization based on dynamic pricing strategy of claim 1, wherein: the step 3) comprises the following steps:

step 301) solving a load distribution strategy, namely a load routing strategy, of the data center;

step 302), solving to obtain the overall service capacity of the data center according to the known routing vector;

and 303) solving the optimal number of servers of each data center under the conditions of delay and service price constraint to finally obtain an optimal energy consumption solution.

5. The method for multiple data center cost optimization based on dynamic pricing strategy of claim 1, wherein: the data center is provided with a standby server.

Images