CN103677957B - Cloud data center high energy efficiency based on multiple resource virtual machine placement method - Google Patents

Abstract

The invention discloses a kind of cloud data center high energy efficiency based on multiple resource virtual machine placement method, including step: set up the energy efficiency model of multiple resource；Using first-fit algorithm to generate particle, the degree of belief of definition particle individual optimal solution and globally optimal solution instructs particle evolution；Define the fitness function of population according to multiple resource energy efficiency model and evaluate particle with this.The inventive method can effectively reduce the energy consumption of the placement of the virtual machine of cloud computation data center, and system resource obtains an equitable breakdown.

Description

Cloud data center high energy efficiency based on multiple resource virtual machine placement method

Technical field

The present invention relates to field of cloud calculation, specifically a kind of based on combining particle cluster algorithm in cloud data center with energy-conservation be The virtual machine placement method of target.

Background technology

High energy consumption is that cloud data center resource manages a major challenge faced, along with the continuous expansion of data center's scale, High energy consumption problem is more prominent.Such as, the energy consumption that Google data center produces is equivalent to the total energy consumption of a micropolis.Number High energy consumption according to center not only causes the waste of electric energy, the instability that system is run, and environment also results in bad impact simultaneously. The main cause causing cloud data center high energy consumption has two aspects: be on the one hand as the increase of number of users, data center Infrastructure construction increases considerably, and is on the other hand that system resource allocation is unreasonable.Study efficient system resource allocation side Method, can reduce the system energy consumption of cloud data center, and make the utilization of resources develop towards sustainable direction.

The laying method of research virtual machine, the mapping relations between definition virtual machine and physical nodes are in cloud data center The reasonable distribution of resource has vital effect, also produces important to energy consumption, performance and the resource utilization of system simultaneously Impact.Virtual machine Placement Problems can be described as vector bin packing: the article of loading are operating virtual machines, and chest is Physical nodes, the resource requirement of virtual machine and the resource quantity of physical nodes represent article and size of a case respectively.Virtual machine CPU, internal memory, bandwidth, disk etc. are included, the kind of the resource i.e. dimension of vector with the resource of physical nodes.For M physics knot Point, the cloud data center of N number of virtual machine, is M by the solution space of deploying virtual machine to physical nodes^N, belong to NP-hard problem. To this problem owing to not having multinomial optimal solution algorithm, heuritic approach based on greedy algorithm is generally used to obtain optimum or secondary Excellent solution.

At present with energy-conservation for target place virtual machine time, the energy efficiency model of use is based on CPU system resource mostly. Some documents being pointed out, server energy consumption and cpu busy percentage are linear approximate relationship, also can when server is in idle condition Consume its electric power of about 70% when loading that is at the highest notch.Beloglazov etc. propose a kind of high energy efficiency based on this energy efficiency model Resource allocation algorithm, placement and migration by virtual machine significantly reduce system energy consumption and ensure that the property of task simultaneously Energy.Liu Zhi wafts etc. and to propose a kind of Energy-aware virtual machine based on discrete particle cluster method based on this energy efficiency model and place intelligence Optimization method.But, the combining of the resource such as the CPU in cloud data center physical nodes, internal memory, bandwidth, disk in actual applications Close service condition and the efficiency of system is had important impact.Srikantaiah etc. have studied system resource to energy consumption, the shadow of performance Ring, conclude that for busy physical nodes by reality measurement, when cpu busy percentage is 70% and disk utilization is When 50%, the energy consumption of physical nodes is minimum and the performance of task can be effectively ensured.

The placement of cloud data center virtual machine is a bin packing, i.e. finds virtual machine the reflecting to physical nodes of optimum Penetrate relation, make placement result reach optimum.Bin packing belongs to NP-hard problem, typically uses heuritic approach, and mostly opens Hairdo algorithm is based on greedy algorithm, and uses some simple rules, as suboptimum coordinate, profile error and best fit etc., at present Virtual machine Placement, be mostly based on tradition heuritic approach innovatory algorithm.Energy efficiency model design virtual machine is used to put Also use the innovatory algorithm of tradition heuritic approaches when putting algorithm more.Beloglazov etc. are based on BFD(Best Fit Decrease) algorithm, it is proposed that MBFD(Modified Best Fit Decrease) algorithm carries out virtual machine placement.MBFD calculates First virtual machine is carried out descending sort according to CPU resource utilization by method, then according to virtual machine is put by virtual machine request order Put in the physical nodes of energy consumption increment minimum.Srikantaiah etc. are by BFH(Best Fit Heuristics) heuristic calculation Method improves, and uses MBFH(Modified Best Fit Heuristics) algorithm carries out the placement of virtual machine, Ji Jiangxu Plan machine is placed in the physical nodes that the total Euclidean distance of busy physical nodes is minimum.The heuristic calculation of tradition based on greedy algorithm Method can optimize the placement of virtual machine, but tradition heuritic approach generally uses single point search strategy, is easily trapped into local optimum, Overall effect of placing can not be reached optimum, still need to optimize further.

Summary of the invention

For above deficiency of the prior art, it is an object of the invention to provide a kind of effectively reduction in cloud computing data The energy consumption of the placement of the virtual machine of the heart, the cloud data center high energy efficiency based on multiple resource that system resource obtains an equitable breakdown is virtual Machine laying method.Technical scheme is as follows: high energy efficiency based on multiple resource virtual machine placement side of a kind of cloud data center Method, it comprises the following steps:

101, multiple resource kind quantity d in cloud data center is obtained, quantity n of cloud data center physical node and actual profit By rateAnd according to Practical Calculation environment set efficiency optimum utilizationWhereinRepresent jth kind in physical nodes i The practical efficiency of resource,Represent the efficiency optimum utilization of jth kind resource in physical nodes i, set up multiple resource efficiency Model, as the formula (1):

$\mathrm{\δ}=\underset{i=1}{\overset{n}{\mathrm{\Σ}}}\sqrt{\underset{j=1}{\mathrm{\Σ}}{(u_j^i-{\mathrm{ubest}}_j^i)}^2}---\left(1\right);$

102, stochastic generation N number of virtual machine request sequence, uses adaptation First Fit algorithm first to obtain N kind virtual machine Initial placement sequence, the most N number of particle, thus constitute initial population；

103, according to the multiple resource energy efficiency model obtained in step 101,Build grain The fitness function f (δ) of son, and calculate fitness f (δ), locally optimal solution and global optimum according to particle swarm optimization algorithm Solve, draw the new position of particle；

104, according to the new position of the particle obtained in step 103, it may be judged whether meetAnd $\underset{j}{\mathrm{\Σ}}{r}_j^{\mathrm{CPU}}*x_h^j\≤c_h^{\mathrm{CPU}},\underset{j}{\mathrm{\Σ}}{r}_j^{\mathrm{RAM}}*x_h^j\≤c_h^{\mathrm{RAM}},\underset{j}{\mathrm{\Σ}}{r}_j^{\mathrm{BW}}*x_h^j\≤c_h^{\mathrm{BW}},\underset{j}{\mathrm{\Σ}}{r}_j^{\mathrm{DISK}}*x_h^j\≤c_h^{\mathrm{DISK}}---\left(5\right)$

WhereinRepresent whether virtual machine j is placed on physical nodes h, when virtual machine j is placed in physical nodes h,For 1, it is otherwise 0；Represent that each virtual machine can be only placed in a physical nodes；Formula (5) represents that multiple virtual machine is placed on physics During node h, resources of virtual machine not can exceed that the total resources of physical nodes h, wherein,Represent respectively The size of CPU, internal memory, bandwidth and disk needed for virtual machine j, Represent physical nodes h respectively CPU, internal memory, bandwidth, the capacity of disk；

If 105 meet the formula (4) in step 104 and formula (5), the position of the most more new particle, iterations adds 1；If it is discontented Foot formula (4) and formula (5) then particle position are constant, and iterations adds 1；When iterations >=maximum iteration time N3, iteration is tied Restraint and export globally optimal solution, virtual machine being set according to the physical node location corresponding to this globally optimal solution, completes virtual machine Placement.

Further, the multiple resource kind of step 101 medium cloud data center includes CPU, internal memory, bandwidth and disk.

Further, CPU and the optimal efficiency utilization rate of disk in step 101It is respectively 0.7 and 0.5.

Advantages of the present invention and having the beneficial effect that:

(1) the inventive method uses multiple resource energy efficiency model, it is possible to adapts to cloud data center and comprises the virtual of multiple resources The Placement Problems of machine, considers the different resource impact for system energy consumption, and is more than the shadow considering CPU for energy consumption Ring.

(2) particle cluster algorithm is used to process virtual machine Placement Problems, it is possible to avoid tradition heuritic approach to carry out single-point and search Hitch fruit is easily trapped into the problem of local optimum, effectively reduces the system energy consumption that virtual machine is placed, and system resource is rationally divided Join.

(3) the method energy efficiency model based on multiple resource and particle cluster algorithm, it is proposed that the virtual machine placement side of high energy efficiency Method, has considered many factors such as energy consumption, performance, resource utilization, it is ensured that virtual machine places effect.

Accompanying drawing explanation

Fig. 1 is preferred embodiment of the present invention cloud data center high energy efficiency based on multiple resource virtual machine placement method flow process Figure.

Detailed description of the invention

The invention will be further elaborated to provide the embodiment of an indefiniteness below in conjunction with the accompanying drawings.

(1) multiple resource energy efficiency model is set up

The comprehensive service condition of the resources such as the CPU of data center's physical nodes, internal memory, bandwidth, disk is to physical nodes Efficiency has important impact.Existing document has reported by experimentation the energy consumption of physical nodes, performance and various system resources Between relation.Controlled k application program service of client by 4 physical nodes, each physical nodes connects a survey Energy meter surely and the tracker of a monitoring resource utilization, by the various resource utilizations of physical nodes from 10%～ 90% changes with the increment of 10% respectively, measures performance and the energy consumption of physical nodes of application program under different utilization rate.Through too much Secondary experiment, measurement result shows, when the cpu busy percentage of busy physical nodes is 70% and disk utilization is 50%, energy Consumption is minimum and mission performance can be effectively ensured.The resource utilization best combination of busy physical nodes after using virtual machine to place The Euclidean distance of point evaluates the quality of physical nodes efficiency, uses total Euclidean distance of all busy physical nodes to evaluate system The quality of overall efficiency.Represent with formula (1):

$\mathrm{\δ}=\underset{i=1}{\overset{n}{\mathrm{\Σ}}}\sqrt{\underset{j=1}{\mathrm{\Σ}}{(u_j^i-{\mathrm{ubest}}_j^i)}^2}---\left(1\right);$

The quantity of resource category during wherein d represents model, resource can be CPU, disk, internal memory and the network bandwidth etc., n table The quantity of Shi Yun data center physical node,Represent the practical efficiency of i jth kind resource in physical nodes,Represent The utilization rate that in physical nodes i, the efficiency of jth kind resource is optimal.The resource utilization of all busy physical nodes of data center With the Euclidean distance sum of best joint, i.e. represent the departure degree of current system and optimum state.

The threshold values of resource utilization when this energy efficiency model determines efficiency optimum according to actual measured results, closer to really Cloud data center environment, the emulation to cloud data center has bigger reference value.

For simplified model, this considers two spike-type cultivars CPU and disk in describing, according to having been reported CPU and disk Optimal efficiency utilization rate is respectively 0.7,0.5.

(2) definition of virtual machine Placement Problems

Virtual machine Placement Problems is a combinatorial optimization problem, can be defined as multi-C vector bin packing.Each The available resources (chest) of physical nodes are a d dimensional vector, every one-dimensional representation one system resource (CPU, internal memory, bandwidth, magnetic Dish etc.).Virtual machine (article) is also a corresponding d dimensional vector.Target be make virtual machine place after system close to efficiency Good state, i.e. δ are the least, as shown in expression formula (2).Virtual machine Placement Problems can be described as follows:

Target: min δ (2)

Constraint: $x_h^j\∈\left\{\mathrm{0,1}\right\}---\left(3\right)$

$\underset{h}{\mathrm{\Σ}}{x}_h^j=1,\∀j---\left(4\right)$

In expression formula (3),Represent whether virtual machine j is placed on physical nodes h, when virtual machine j is placed on physical nodes h On,It is 1, is otherwise 0；Expression formula (4) represents that each virtual machine can be only placed in a physical nodes.

$\underset{j}{\mathrm{\Σ}}{r}_j^{\mathrm{CPU}}*x_h^j\≤c_h^{\mathrm{CPU}},\underset{j}{\mathrm{\Σ}}{r}_j^{\mathrm{RAM}}*x_h^j\≤c_h^{\mathrm{RAM}},\underset{j}{\mathrm{\Σ}}{r}_j^{\mathrm{BW}}*x_h^j\≤c_h^{\mathrm{BW}},\underset{j}{\mathrm{\Σ}}{r}_j^{\mathrm{DISK}}*x_h^j\≤c_h^{\mathrm{DISK}}---\left(5\right)$

Formula (4) and formula (5) citing

For example, it is assumed that cloud computing center has 4 physical nodes, 4 virtual machines；The configuration of reference preferred server is successively Creating physical nodes, as shown in table 1, virtual machine is as shown in table 2 in physical nodes configuration.

Table 1 physical node configures

Table 2 virtual machine configures

Formula (4) represents that a virtual machine can only be placed on a physical node, it may be assumed that save at physics when virtual machine 1 is placed on During point 1,AndIt is 0, then has(h represents 1,2,3,4), this population meets formula (4)；

Formula (5) represents that, when multiple virtual machine is placed on physical nodes h, resources of virtual machine not can exceed that the money of physical nodes h Source total amount.Such as virtual machine 1,2,3,4 when being all placed on physical node 4,RemainingIt is 0, then $\underset{j}{\mathrm{\Σ}}{r}_j^{\mathrm{CPU}}*x_4^j=60+60+60=240\≤c_4^{\mathrm{CPU}}\left(400\right),$ Have equally $\underset{j}{\mathrm{\Σ}}{r}_j^{\mathrm{RAM}}*x_h^j\≤c_h^{\mathrm{RAM}},\underset{j}{\mathrm{\Σ}}{r}_j^{\mathrm{BW}}*x_h^j\≤c_h^{\mathrm{BW}},$ But $\underset{j}{\mathrm{\Σ}}{r}_j^{\mathrm{DISK}}*x_4^j=100+200+100=400\≤c_4^{\mathrm{CPU}}\left(250\right),$ It is unsatisfactory for $\underset{j}{\mathrm{\Σ}}{r}_j^{\mathrm{DISK}}*x_h^j\≤c_h^{\mathrm{DISK}},$ Now, this population is unsatisfactory for the condition of formula (5).

Expression formula (5) represents that, when multiple virtual machine is placed on physical nodes h, resources of virtual machine not can exceed that physical nodes h Total resources.Wherein,Represent the CPU needed for virtual machine j, internal memory, bandwidth respectively Size with disk.Represent the CPU of physical nodes h, internal memory, bandwidth, magnetic respectively The capacity of dish.

(3) operator and the position updating process of particle cluster algorithm are improved

Particle swarm optimization algorithm (Particle Swarm Optimizer, PSO) be by Kenney and Eberhart in Nineteen ninety-five proposes, for solving the optimization problem of continuous space.Compared with other colony's evolution algorithmics, particle cluster algorithm due to Parameter is few, fast convergence rate and easily realizing, and has obtained the concern of more and more people, has achieved notable achievement, but virtual The discrete combination optimization problems such as machine placement need to study further.Owing to virtual machine Placement Problems is discrete Combinatorial Optimization Problem, needs to redefine the operator of particle cluster algorithm and renewal process.Assume that there are m virtual machine, n physics in cloud data center Node.

1) definition of particle position

Particle position vectorRepresent the l kind feasible program that virtual machine is placed, whereinRepresent During in t generation, updates, the sequence number of the physical nodes at virtual machine j place.Assume the l kind feasible program that virtual machine is placedIt is respectively (1,2,1), represents that virtual machine 1,2,3 is individually positioned in 1,2, No. 1 physical nodes.To particle When position is updated, the position vector of particleIt is converted into 0,1 location matrix

${\mathrm{XX}}_l^t=\left[\begin{array}{cccccc}{s}_{11}^t& s_{12}^t& \·\·\·& s_{1h}^t& \·\·\·& s_{1n}^t\\ s_{21}^t& s_{22}^t& \·\·\·& s_{2h}^t& \·\·\·& s_{2n}^t\\ \·& \·& & \·& & \·\\ \·& \·& & \·& & \·\\ \·& \·& & \·& & \·\\ s_{j1}^t& s_{j2}^t& \·\·\·& s_{\mathrm{jh}}^t& \·\·\·& s_{\mathrm{jn}}^t\\ \·& \·& & \·& & \·\\ \·& \·& & \·& & \·\\ \·& \·& & \·& & \·\\ s_{m1}^t& s_{m2}^t& \·\·\·& s_{\mathrm{mh}}^t& \·\·\·& s_{\mathrm{mn}}^t\end{array}\right]---\left(6\right)$

In formula (6),Represent whether virtual machine j is placed in physical nodes h, if virtual machine j is put in t generation updates In physical nodes h, thenOtherwise Value according toJth dimension value depending on.Such as, ifFor (1,2,1), First dimensionIt is that 1 expression virtual machine has been placed in the 1st physical nodes, thenThe first row inIt is 1, in like mannerIt is 1,For 1.Owing to a virtual machine can be only placed in a physical nodes, so $\underset{h=1}{\overset{n}{\mathrm{\Σ}}}{s}_{\mathrm{jh}}^t=1,\∀j\∈\{\mathrm{1,2},...,m\}.$

2) renewal process of position

Herein according to the individual optimal solution of each particle and the globally optimal solution of all particles, particle position is updated,Represent the l particle individual optimal solution after t iteration, whereinTable Show the position coordinates that particle l individual optimal solution jth is tieed up,Represent that all particles are through t Globally optimal solution after secondary iteration,Represent the position coordinates of the jth dimension of all particle globally optimal solutions.Gbestt Also corresponding location matrix is will convert into, in formula (7)～(10) during particle updatesIt is i.e. right Answer the value that the jth row of location matrix, h arrange.

Assuming herein under random manner, during the t+1 time location updating, virtual machine j is put or is not placed in physical nodes h Probability be 0.5, i.e.Use p_p、p_gRepresent individual optimal solution and the letter of globally optimal solution respectively Ren Du, i.e. finds the probability of particle final optimal solution, judges the position of particle with this.In order to describe simplicity, introduce two herein Parameter k₁、k₂.Obtained by Bayesian formula:

$\begin{array}{c}P({s_{\mathrm{jh}}}^{t+1}=1|{\mathrm{pbest}}_{\mathrm{jh}}^t=1,\mathrm{gbes}{t}_{\mathrm{jh}}^t=1)\\ =\frac{p_p*p_g}{p_p*p_g+(1-p_p)(1-p_g)}=k_1\end{array}---\left(7\right)$

$\begin{array}{c}P({s_{\mathrm{jh}}}^{t+1}=0|{\mathrm{pbest}}_{\mathrm{jh}}^t=1,{\mathrm{gbest}}_{\mathrm{jh}}^t=1)\\ =\frac{(1-p_p)*(1-p_g)}{p_p*p_g+(1-p_p)(1-p_g)}=1-k_1\end{array}---\left(8\right)$

$\begin{array}{c}P({s_{\mathrm{jh}}}^{t+1}=1|{\mathrm{pbest}}_{\mathrm{jh}}^t=1,\mathrm{gbes}{t}_{\mathrm{jh}}^t=0)\\ =\frac{p_p*(1-p_g)}{p_p*(1-p_g)+p_p*(1-p_g)}=k_2\end{array}---\left(9\right)$

$\begin{array}{c}P({s_{\mathrm{jh}}}^{t+1}=0|{\mathrm{pbest}}_{\mathrm{jh}}^t=1,\mathrm{gbes}{t}_{\mathrm{jh}}^t=0)\\ =\frac{p_p*(1-p_g)}{p_p*(1-p_g)+p_g*(1-p_p)}={1-k}_2\end{array}---\left(10\right)$

Formula (7)～(10) represent particle position coordinate according to individual optimal solution and globally optimal solution take 0 or 1 general Rate.Owing to individual optimal solution and globally optimal solution are the history optimal solutions that iterative process produces, the probability finding optimal solution should More than meansigma methods, i.e. 0.5 < p_p< 1,0.5 < p_g< 1, in order to make particle carry out scatter searching, it is to avoid particle is absorbed in local optimum, should Make p_g<p_p, i.e. 0.5 < p_g<p_p<1.Take p herein_p=0.7, p_g=0.8, corresponding k₁=0.9, k₂=0.7。

The false code of position updating process is as follows:

(4) algorithm flow describes

1) population is initialized

Stochastic generation N number of virtual machine request sequence, to each request sequence, calculates according to adapting to (First Fit) first Method, is placed on virtual machine in first physical nodes meeting resource, obtains N kind placement schemes, the most N number of particle, thus structure Becoming initial population, its time complexity is o (nlogn).

2) according to fitness function, calculate the fitness of initial population, obtain particle individual optimal solution and globally optimal solution.

3) according to 4.2.1(2) calculate particle new position.

4) judge whether new position meets requirement, then iterations is added 1.Judge each virtual of each particle successively Whether the new position of machine meets formula (4), the constraints of (5), if all virtual machines all meet constraints, then particle is more New is new position, and otherwise particle keeps original value.The method whether judgment formula (4), the constraints of (5) meet is such as Under: for jth virtual machine, if 1.I.e. virtual machine j is placed in multiple physical nodes, then record position is 1 The sequence number of physical nodes, is then arrangedThe order being incremented by according to physical nodes sequence number, takes recorded physics successively Node, is placed on virtual machine j in first physical nodes of satisfied (5), and its position is put 1, then judges particle other Virtual machine.If the physical nodes of record is all unsatisfactory for (5), then the new position of this virtual machine is unsatisfactory for condition, and particle does not updates.② IfI.e. virtual machine is placed in certain physical nodes, it may be judged whether meet constraint (5), if it is satisfied, then virtual machine New position is eligible, and particle is updated to new position, and otherwise particle does not updates.If 3.After i.e. updating, virtual machine does not has Distributing to any main frame, the new position of particle is unsatisfactory for condition, and particle does not updates.

5) determine whether maximum iteration time, the most then terminate, export globally optimal solution, if it is not, then forward (2) to.

Context of methods improves on the basis of binary system discrete particle cluster algorithm, and no longer Negotiation speed is converted to grain The position of son, but directly determine the new position of particle by probability, search for more direct.Algorithms T-cbmplexity is o (mn).With biography System heuritic approach is compared, and context of methods compares evolution by particle and obtains the overall placement result placing effect optimum, can have Effect is avoided being absorbed in locally optimal solution.

The above embodiment is interpreted as being merely to illustrate the present invention rather than limiting the scope of the invention.? After the content of the record having read the present invention, the present invention can be made various changes or modifications by technical staff, and these equivalences become Change and modify and fall into the inventive method claim limited range equally.

Claims (3)

1. Zhong Yun data center high energy efficiency based on a multiple resource virtual machine placement method, it is characterised in that comprise the following steps:

101, multiple resource kind quantity d in cloud data center, quantity n of cloud data center physical node and practical efficiency are obtainedAnd set efficiency optimum utilizationWhereinRepresent the practical efficiency of jth kind resource in physical nodes i, Represent the efficiency optimum utilization of jth kind resource in physical nodes i, set up multiple resource energy efficiency model, as shown in formula (1):

$\mathrm{\&delta;}=\underset{i=1}{\overset{n}{\&Sigma;}}\sqrt{\underset{j=1}{\overset{d}{\&Sigma;}}{(u_j^i-{\mathrm{ubest}}_j^i)}^2}---\left(1\right);$

102, stochastic generation N number of virtual machine request sequence, uses adaptation First Fit algorithm first to obtain at the beginning of N kind virtual machine Begin to place sequence, the most N number of particle, thus constitute initial population；

103, according to the multiple resource energy efficiency model obtained in step 101,Build particle Fitness function f (δ), and calculate fitness, locally optimal solution and globally optimal solution according to particle swarm optimization algorithm, draw grain The new position of son；

104, according to the new position of the particle obtained in step 103, it may be judged whether meetAnd

$\underset{j}{\&Sigma;}{r}_j^{CPU}*x_h^j\&le;c_h^{CPU},\underset{j}{\&Sigma;}{r}_j^{RAM}*x_h^j\&le;c_h^{RAM},\underset{j}{\&Sigma;}{r}_j^{BW}*x_j^j\&le;c_h^{EW},\underset{j}{\&Sigma;}{r}_j^{DISK}*x_h^j\&le;c_h^{DISK}---\left(5\right)$

In formula (4)Represent whether virtual machine j is placed on physical nodes h, when virtual machine j is placed in physical nodes h, Representing that each virtual machine can be only placed in a physical nodes, virtual machine j is not placed in physical nodes h thenFormula (5) table Showing when multiple virtual machine is placed on physical nodes h, resources of virtual machine not can exceed that the total resources of physical nodes h, wherein,The capacity of CPU, internal memory, bandwidth and the disk needed for expression virtual machine j respectively, Represent the CPU of physical nodes h, internal memory, bandwidth, the capacity of disk respectively；

If 105 meet the formula (4) in step 104 and formula (5), the position of the most more new particle, iterations adds 1；If the formula of being unsatisfactory for (4) and formula (5), then particle position is constant, and iterations adds 1；When iterations is more than or equal to maximum iteration time N3, iteration Terminate and export globally optimal solution, virtual machine being set according to the physical node location corresponding to this globally optimal solution, completes virtual The placement of machine.

Cloud data center the most according to claim 1 high energy efficiency based on multiple resource virtual machine placement method, its feature exists In: the multiple resource kind of step 101 medium cloud data center includes CPU, internal memory, bandwidth and disk.

Cloud data center the most according to claim 1 high energy efficiency based on multiple resource virtual machine placement method, its feature exists In: the efficiency optimum utilization of CPU and disk in step 101It is respectively 0.7 and 0.5.