A kind of resource allocation method of Oriented Green cloud computing
Technical field
The present invention relates to a kind of resource allocation method of Oriented Green cloud computing.
Background technology
Cloud computing(Cloud Computing)It is the focus of current computer realm research, calculating task is dispatched to by it
By on a large amount of resource pools calculating and constituting with storage resource node, enable users to obtain computing power, memory space and information on demand
Service.Cloud computing is energy-conservation in itself, for example, pass through Intel Virtualization Technology, effectively reallocate resources, and improves resource utilization;Pass through
Closing/dormant technology, reduces idle energy consumption, realizes the reduction of energy consumption.Develop rapidly so that traditional number with cloud computing technology
Change according to center, create the data center of a new generation, referred to as cloud data center(Cloud Data Center).Cloud
Data center comprises a large amount of servers, and the quantity of these servers all increases continuous every year.Even if cloud computing is section
Can, cloud data center is all consuming huge energy daily.
According to US Gov Env Protection Agency(Environmental Protection Agency, EPA)As shown by data,
The energy consumption of data center has exceeded the 3% of american energy total amount consumed, accounts for the 1.5%-2% of global energy consumption, and annual with
12% speed increases.《Data center's efficiency test and appraisal guide》Display, data center of China total power consumption has reached 700 in 2011
Hundred million kWh, account for the 1.5% of national power consumption total amount then, are equivalent to the annual total electricity consumption in Tianjin in 2011.In cloud data
The high energy consumption issues of the heart not only cause the waste of electric energy, system operation unstable, also environment is had undesirable effect simultaneously.Beautiful
State's federal agency points out that the high energy consumption issues of cloud computing will be to the side such as air quality, national security, climate change, electric network reliability
Face causes to have a strong impact on.Reduce cloud data center energy resource consumption, realize high-effect green cloud computing become impact low-carbon energy-saving,
The importance of sustainable development.Reduce cloud data center links energy resource consumption, improve the correlation of resource utilization
Research has been subject to the extensive concern of industrial quarters and academia and has achieved certain achievement in research.
At present, the green calculating field in cloud data center relates generally to 3 kinds of power-saving technologies:Dynamic voltage scaling technology
(Dynamic Voltage Scaling, abbreviation DVS), closing/dormant technology(Resource Hibernation)And virtualization
Technology(Virtualization).
DVS technology passes through the dynamic running frequency changing CPU and running voltage in calculate node running, reaches fall
The purpose of low system power dissipation.But when DVS is applied to cloud computing system, problems with can be run into:(a) task reach system when
Between be uncertain, so reach task type be difficult to prediction;Even if b () can predict the type of task, it is suitable for this task
Processor voltage also be difficult to determine;C () DVS is mainly used to reduce the energy consumption of host-processor, but be used for optimizing whole main frame
Or the energy consumption of whole cloud computing system just compares limitation.Some research worker mainly study the strategy based on DVS.This tactful basis
The implementation status of current bag task is predicted it is ensured that completing task before user-defined Deadline, and dynamic adjustment is processed
The working voltage of device, reduces the energy consumption of system.
Closing/dormant technology to reduce idle energy consumption by way of closing or dormancy idle node.Its shortcoming is to work as to need
When the node wanted is unsatisfactory for demand, reset node takes long enough, this can lead to system response time elongated, affect user
Experience.Some research worker adopt the node of dormancy light load, reduce the energy consumption of system.This method is the focus wound of research
Idle node is transferred to from dynamic node in new property ground, reduces energy consumption by the idle node of dormancy.This strategy supposes dormancy
The energy consumption of posterior nodal point is 0, and does not consider the copy of dormancy node storage, but must take into this two in actual applications and ask
Topic.Closing/dormant technology typically sets in advance or predicts the opportunity needing closing/dormancy main frame or critical component.So, for
For having the cloud system of a large amount of computing resources, a closing/dormant technology difficult problem to be solved is in known unit time task
On the premise of amount of reach, determining needs to close how many main frames, and the problems such as close which main frame.
Intel Virtualization Technology can realize the different virtual machine that multiple tasks operate in same physical host(Virtual
Machine, VM)On, by improving the resource utilization of physical host, to reduce the quantity of required main frame, thus reducing energy consumption.
Some research worker propose an energy module administrative model, including two assemblies:Host-level subsystem and virtual machine-level
Subsystem.Host-level subsystem responsible regulates and controls the energy consumption of whole system, reasonably distributes all of hardware money according to application request
Source, the energy consumption of each virtual machine not can exceed that respective threshold so that system has the ability to carry out fine-grained energy for application-specific
Source control.Virtual machine-level subsystem redistributes the hardware resource of virtual machine it is ensured that the energy that each task consumes is less than phase
Answer threshold values.Intel Virtualization Technology is a kind of mode realizing cloud computing energy-conservation, by by abstract for the physical resource side for virtual resource
Formula, improves the utilization rate of resource.But, virtualization itself will pay higher efficiency cost, and virtualized level is deeper
Cost is higher.Because Intel Virtualization Technology be in layer carry out virtualized(From the hardware of lowermost layer to top application),
Each layer of the virtual cost that will pay efficiency.
Content of the invention
For above-mentioned technical problem, the technical problem to be solved is that offer one kind can be in deploying virtual machine mistake
Effectively carry out resource distribution in journey, reduce cloud computing system energy consumption, it is to avoid the resource of the Oriented Green cloud computing of the wasting of resources
Collocation method.
The present invention employs the following technical solutions to solve above-mentioned technical problem:The present invention devises a kind of Oriented Green cloud
The resource allocation method calculating, including cloud computing system, system, while processing task in current period, walks including following
Suddenly:
The actual request amount according to history cycle task for the step 01. and the actual request amount of current period task, under prediction
The predictions request amount of a cycle task;And the predictions request amount according to next cycle task, obtain next cycle and wait
The prediction task amount of system execution;
The prediction task amount that step 02. executes according to next cycle waiting system, obtains virtual needed for next cycle
The pre- quantitation of machine;
The actual startup quantity according to virtual machine in current period for the step 03. and the prediction of virtual machine needed for next cycle
Quantity, obtains the quantity that in next cycle, virtual machine is turned on and off;
The quantity that step 04. is turned on and off according to virtual machine in next cycle, allocated in advance and controls in next cycle
Virtual machine quantity on each physical host.
As a preferred technical solution of the present invention, described step 01 comprises the steps:
Step 0101. sets k-th cycle as current period, for all tasks in+1 cycle of kth, according to as follows
Formula(1), each task in prediction+1 cycle of kth respectivelyiThe predictions request amount x ' of type tasksi(k+1), k=1 ..., i=
1st ..., m, m are the type sum of task;
x′i(k+1)=a 'i(k)+b′i(k)+c′i(k) (1)
Wherein, a 'i(k)、b′i(k)、c′iK () is expressed as follows respectively:
Wherein,For a smooth value,For secondary smooth value,For three smooth values, represent respectively
As follows:
Wherein, xiK () is task in k-th cycleiThe actual request amount of type tasks, α is smoothing factor, interval
For(0,1);
Step 0102. is according to equation below(8)With(9), obtain each task in+1 cycle of kth respectivelyiType tasks etc.
Treat the prediction task amount d ' of system executioni(k+1);
d′imin(k+1)≤d′i(k+1)≤d′imax(k+1)
Wherein, the type sum of task is equal with the type sum of virtual machine, all types of virtual machines alignment processing phase respectively
Answer the task of type, gijRepresent host in current periodjVM is started on physical hostiThe actual quantity of type Virtual machine, j=
1st ..., n, n are the quantity having turned on physical host in system, diK () represents the task that in current k-th cycle, system is processediClass
The actual task amount of type, δiRepresent taskiType tasks predict the preset value of fluctuating error, and T is Cycle Length,Table
Show that in current period, system is to taskiThe disposal ability of type tasks, uijRepresenttask i The task of type is in hostjUpper execution
Mean Speed.
As a preferred technical solution of the present invention, described step 02 includes following process:
Predetermined system real-time response than γ, according to each task in+1 cycle of kthiIt is pre- that type tasks waiting system executes
Survey task amount d 'iAnd equation below (k+1)(10):
v′i(k+1)=γ d 'i(k+1) (10)
Respectively obtain+1 cycle of kth in each taskiThe corresponding required each VM of type tasksiThe prediction of type Virtual machine
Quantity v 'i(k+1).
As a preferred technical solution of the present invention, described step 03 includes following process:
By VM each in current periodiThe actual startup quantity of type Virtual machineRequired with described+1 cycle of kth
Each VMiThe pre- quantitation v ' of type Virtual machinei(k+1) correspondence is compared, and according to the following rules, obtains virtual in+1 cycle of kth
The pre- quantitation that machine is turned on and off:
IfThen maximize and openIndividual VMiType is empty
Plan machine;
IfThen minimize and closeIndividual
VMiType Virtual machine;
IfThen maintain each VM in current periodiType Virtual machine starts
Actual quantity
Wherein, Z represents VM in system current periodiThe idle count of type Virtual machine.
As a preferred technical solution of the present invention, described step 04 comprises the steps:
Virtual machine amount threshold q on step 0401. predetermined physical main framethreshold, judged according to the following rules, respectively
To each hostjPhysical host is divided:
If qj>qthreshold, then by this hostjPhysical host is divided into high applicable main frame, constitutes high applicable host complexes;
If 0<qj<qthreshold, then by this hostjPhysical host is divided into poorly rated main frame, constitutes poorly rated host complexes;
If qj=0, then by this hostjPhysical host is divided into dormancy main frame, constitutes dormancy host complexes;
Step 0402. is directed to the pre- quantitation that in described+1 cycle of kth, virtual machine is turned on and off, and presses such as lower section respectively
Formula, allocates and controls the virtual machine quantity on each physical host in+1 cycle of kth in advance:
For the VM needing closingiType Virtual machine, if there are poorly rated host complexes, preferentially from poorly rated host set
The virtual machine of free time is closed in conjunction;If there are not poorly rated host complexes, it is suitable for, from height, the void closing the free time host complexes
Plan machine;
For the VM needing unlatchingiType Virtual machine, if there are high applicable host complexes, is preferentially suitable for host set from high
Corresponding virtual machine is opened in conjunction;If there are not high applicable host complexes, open empty accordingly from poorly rated host complexes
Plan machine;If there are not high applicable host complexes, there are not poorly rated host complexes yet, then open phase from dormancy host complexes
The virtual machine answered.
As a preferred technical solution of the present invention, described step 0402 comprises the steps:
Step 040201. is respectively directed to each VM in all virtual machines to be allocatediType Virtual machine, selects as follows
Select primary carrying physics host complexes:
Step a)According toObtain high applicable main frame in current period
Each of set physical host can carry VMiThe ability of type Virtual machineIfThen current period
The middle high host complexes that are suitable for are primary carrying physics host complexes, jump to step 040202, otherwise enter next step;
Wherein,Represent hostjThe memory size of physical host,Represent hostjPhysical host
Cpu performance, memiRepresent VMiThe memory size of type Virtual machine, mipsiRepresent VMiThe disposal ability of type Virtual machine;
Step b)According toObtain poorly rated main frame in current period
Each of set physical host can carry VMiThe ability of type Virtual machineIfThen currently all
Interim poorly rated host complexes are primary carrying physics host complexes, jump to step 040202, otherwise enter next step;
Step c)According toObtain dormancy host set in current period
Each of conjunction physical host can carry VMiThe ability of type Virtual machineThen dormancy host complexes in current period
Carry physics host complexes for primary, enter next step;
Step 040202. basisObtain each host in primary carrying physics host complexesjPhysical host
With each VMiMatching degree MR of type Virtual machineij, wherein,μ is constant coefficient;
Step 040203. searches primary carrying in physics host complexesPhysical host, according to
Obtain this hostjPhysical host is to each VM to be allocatediThe matching probability of type Virtual machine
Pij, and according to PijPhysical host is selected to carry VM to be allocated in physics host complexes in primary carryingiType Virtual machine;
The resource of all physical hosts in step 040204. more new system, if the primary physics host complexes that carry are low suitable
With host complexes, then will choose carrying VM to be allocated in step 040203iThe physical host of type Virtual machine is classified as high being suitable for and leads
Machine set;If the primary physics host complexes that carry are dormancy host complexes, carry choose in step 040203 of VM to be allocatedi
The physical host of type Virtual machine is classified as poorly rated host complexes.
As a preferred technical solution of the present invention, also include step 05 after described step 04 as follows:
Step 05. is directed to virtual machine to be allocated in+1 cycle of kth, obtains cloud computing system energy in+1 cycle of kth
Consumption, including physical host energy consumption operating in next cycle, operating energy consumption of virtual machine and switch virtual machine, physics
The control energy consumption of main frame, and feed back to cloud computing system.
A kind of resource allocation method of Oriented Green cloud computing of the present invention adopts above technical scheme and prior art
Compare, there is following technique effect:
(1)The resource allocation method of the Oriented Green cloud computing of present invention design, using Intel Virtualization Technology, by task scheduling
Problem abstract for deploying virtual machine problem;The task amount of user's request is predicted, and combines current system conditions and resource
Distribution, takes conservative control strategy, in advance system resource is allocated, controls, and reduces cloud computing system energy consumption, it is to avoid resource
Waste;
(2)In the resource allocation method of Oriented Green cloud computing of present invention design, real using Three-exponential Smoothing method
The now prediction to task requests amount, and predicting the outcome according to task requests amount, are allocated to system resource in advance, control,
Can effectively solve the problem that resource distribution lags behind the problem of user's request, improve the response speed of system process task request;
(3)In the resource allocation method of Oriented Green cloud computing of present invention design, introduce on the basis of Forecasting Methodology
Conservative control strategy, has exchanged stable in systematic function and the task response ratio that improve for the increase of suitable energy consumption,
Avoid and make forecast error big in the case that real-time task fluctuation ratio is larger, lead to system response ratio low, stability difference
Problem;
(4)The resource allocation method of the Oriented Green cloud computing of present invention design so that resource is pre-configured more rationalizes,
Activate less physical host number, improve the resource utilization of system, reduce system energy consumption, and pre-configured in resource
During, the feasible solution of optimization can be quickly found, reduce the energy consumption of system to a certain extent, and this method takes
Few, can effectively adapt to the pre-configured demand of short cycle resource.
Brief description
Fig. 1 is the schematic flow sheet of the resource allocation method that the present invention designs Oriented Green cloud computing.
Specific embodiment
With reference to Figure of description, the specific embodiment of the present invention is described in further detail.
The present invention designs in the resource allocation method of Oriented Green cloud computing, in order to improve the utilization of resources of cloud computing system
Rate, reduces system energy consumption, needs each physical host resource in system is configured, complete the scheduling of virtual machine.Here mistake
Cheng Zhong, can only access present physical main frame and virtual machine process task by monitoring physical host and virtual machine real-time status
Situation, resource distribution process but lags behind the task requests of user all the time.Therefore, the present invention builds forecast model, takes conservative
Control strategy, carries out cyclic forecast and control to the request amount of task dissimilar in cloud computing system, thus realizing rationally
Resource pre-configured, it is to avoid resource distribution lags behind a difficult problem for user's request.
In order that the result of resource distribution can persistently meet the demand to resource for each generic task of user's submission, to next
In the individual cycle, the request amount of each generic task is predicted.The present invention predicts respective type task based on Three-exponential Smoothing method
Request amount size.The size of predetermined period should depend on execution time of task, method takes, physical host and virtual machine
Switching on and shutting down take etc. factor.If the predetermined period chosen is too short, large effect can be produced to the stability of system, can simultaneously
Increase the expense that system energy consumption controls.System can be according to the size of different situation reasonable selection predetermined period.
During actual prediction, no matter adopted which kind of Forecasting Methodology, predictive value is likely to there is more or less error
Fluctuation, leads to predictive value and the larger situation about not being inconsistent of real time load value, and system real time requires to ensure;Except this it
Outward, the frequent fluctuation of predictive value, leads to system stability extreme difference it is impossible to meet actual demand.In order to solve the above problems, this
Bright take conservative control strategy in prediction it is therefore an objective to the response ratio of effective lift system and stability.
The present invention by abstract for the scheduling problem of task for " by as far as possible optimum in the way of, be possible to meet user's need
The multiple scheduling virtual machines asked are on the cloud system cluster for N for the scale ";The purport that the present invention calculates from low energy consumption green,
Propose the resource allocation method based on probability matching;Resource allocation method based on probability matching substantially belongs to heuristic side
Method, for the purpose of reducing system energy consumption, by the coupling between the division of suitability mainframe cluster and supply and demand resource, finds and optimizes
Feasible solution.
During resource distribution, need to be suitable for the maximum of the equilibrium of load and resource between host complexes in view of height
Change and utilize.In order to equalize the CPU of physical host and the use of memory source, during preventing deploying virtual machine, physical host occurs
Excessive " wooden pail effect ", that is, low memory but CPU computing capability is superfluous, or internal memory surplus but cpu busy percentage is not enough, make
Become the situation that the wasting of resources, virtual machine are not normally functioning.
As shown in figure 1, a kind of resource allocation method of Oriented Green cloud computing of present invention design, including cloud computing system
System, system, while processing task in current period, comprises the steps:
The actual request amount according to history cycle task for the step 01. and the actual request amount of current period task, under prediction
The predictions request amount of a cycle task;And the predictions request amount according to next cycle task, obtain next cycle and wait
The prediction task amount of system execution, wherein detailed process comprises the steps:
Step 0101. sets k-th cycle as current period, for all tasks in+1 cycle of kth, according to as follows
Formula(1), each task in prediction+1 cycle of kth respectivelyiThe predictions request amount x ' of type tasksi(k+1), k=1 ..., i=
1st ..., m, m are the type sum of task;
x′i(k+1)=a 'i(k)+b′i(k)+c′i(k) (1)
Wherein, a 'i(k)、b′i(k)、c′iK () is expressed as follows respectively:
Wherein,For a smooth value,For secondary smooth value,For three smooth values, represent respectively
As follows:
Wherein, xiK () is task in k-th cycleiThe actual request amount of type tasks, α is smoothing factor, interval
For(0,1);
Step 0102. is according to equation below(8)With(9), obtain each task in+1 cycle of kth respectivelyiType tasks etc.
Treat the prediction task amount d ' of system executioni(k+1);
d′imin(k+1)≤d′i(k+1)≤d′imax(k+1)
Wherein, the type sum of task is equal with the type sum of virtual machine, all types of virtual machines alignment processing phase respectively
Answer the task of type, gijRepresent host in current periodjVM is started on physical hostiThe actual quantity of type Virtual machine, j=
1st ..., n, n are the quantity having turned on physical host in system, diK () represents the task that in current k-th cycle, system is processediClass
The actual task amount of type, δiRepresent taskiType tasks predict the preset value of fluctuating error, and T is Cycle Length,Table
Show that in current period, system is to taskiThe disposal ability of type tasks, uijRepresenttask i The task of type is in hostjUpper execution
Mean Speed.
The prediction task amount that step 02. executes according to next cycle waiting system, obtains virtual needed for next cycle
The pre- quantitation of machine, specifically includes following process:
Predetermined system real-time response than γ, according to each task in+1 cycle of kthiIt is pre- that type tasks waiting system executes
Survey task amount d 'iAnd equation below (k+1)(10):
v′i(k+1)=γ d 'i(k+1) (10)
Respectively obtain+1 cycle of kth in each taskiThe corresponding required each VM of type tasksiThe prediction of type Virtual machine
Quantity v 'i(k+1).
The actual startup quantity according to virtual machine in current period for the step 03. and the prediction of virtual machine needed for next cycle
Quantity, obtains the quantity that in next cycle, virtual machine is turned on and off, specifically includes following process:
By VM each in current periodiThe actual startup quantity of type Virtual machineRequired with described+1 cycle of kth
Each VMiThe pre- quantitation v ' of type Virtual machinei(k+1) correspondence is compared, and according to the following rules, obtains virtual in+1 cycle of kth
The pre- quantitation that machine is turned on and off:
IfThen maximize and openIndividual VMiType is empty
Plan machine;
IfThen minimize and closeIndividual
VMiType Virtual machine;
IfThen maintain each VM in current periodiType Virtual machine starts
Actual quantity
Wherein, Z represents VM in system current periodiThe idle count of type Virtual machine.
The quantity that step 04. is turned on and off according to virtual machine in next cycle, allocated in advance and controls in next cycle
Virtual machine quantity on each physical host, specifically includes following steps:
Virtual machine amount threshold q on step 0401. predetermined physical main framethreshold, judged according to the following rules, respectively
To each hostjPhysical host is divided:
If qj>qthreshold, then by this hostjPhysical host is divided into high applicable main frame, constitutes high applicable host complexes;
If 0<qj<qthreshold, then by this hostjPhysical host is divided into poorly rated main frame, constitutes poorly rated host complexes;
If qj=0, then by this hostjPhysical host is divided into dormancy main frame, constitutes dormancy host complexes;
Step 0402. is directed to the pre- quantitation that in described+1 cycle of kth, virtual machine is turned on and off, and presses such as lower section respectively
Formula, allocates and controls the virtual machine quantity on each physical host in+1 cycle of kth in advance:
For the VM needing closingiType Virtual machine, if there are poorly rated host complexes, preferentially from poorly rated host set
The virtual machine of free time is closed in conjunction;If there are not poorly rated host complexes, it is suitable for, from height, the void closing the free time host complexes
Plan machine;
For the VM needing unlatchingiType Virtual machine, if there are high applicable host complexes, is preferentially suitable for host set from high
Corresponding virtual machine is opened in conjunction;If there are not high applicable host complexes, open empty accordingly from poorly rated host complexes
Plan machine;If there are not high applicable host complexes, there are not poorly rated host complexes yet, then open phase from dormancy host complexes
The virtual machine answered.
Wherein, described step 0402 specifically includes following steps:
Step 040201. is respectively directed to each VM in all virtual machines to be allocatediType Virtual machine, selects as follows
Select primary carrying physics host complexes:
Step a)According toObtain high applicable main frame in current period
Each of set physical host can carry VMiThe ability of type Virtual machineIfThen current period
The middle high host complexes that are suitable for are primary carrying physics host complexes, jump to step 040202, otherwise enter next step;
Wherein,Represent hostjThe memory size of physical host,Represent hostjPhysical host
Cpu performance, memiRepresent VMiThe memory size of type Virtual machine, mipsiRepresent VMiThe disposal ability of type Virtual machine;
Step b)According toObtain poorly rated main frame in current period
Each of set physical host can carry VMiThe ability of type Virtual machineIfThen current period
In poorly rated host complexes be primary carry physics host complexes, jump to step 040202, otherwise enter next step;
Step c)According toObtain dormancy host set in current period
Each of conjunction physical host can carry VMiThe ability of type Virtual machineThen dormancy host complexes in current period
Carry physics host complexes for primary, enter next step;
Step 040202. basisObtain each host in primary carrying physics host complexesjPhysical host
With each VMiMatching degree MR of type Virtual machineij, wherein,μ is constant coefficient;
Step 040203. searches primary carrying in physics host complexesHostjPhysical host, according toObtain this hostjPhysical host is to each VM to be allocatediThe matching probability P of type Virtual machineij, and according to
According to PijPhysical host is selected to carry VM to be allocated in physics host complexes in primary carryingiType Virtual machine;
The resource of all physical hosts in step 040204. more new system, if the primary physics host complexes that carry are low suitable
With host complexes, then will choose carrying VM to be allocated in step 040203iThe physical host of type Virtual machine is classified as high being suitable for and leads
Machine set;If the primary physics host complexes that carry are dormancy host complexes, carry choose in step 040203 of VM to be allocatedi
The physical host of type Virtual machine is classified as poorly rated host complexes.
Step 05. is directed to virtual machine to be allocated in+1 cycle of kth, obtains cloud computing system energy in+1 cycle of kth
Consumption, including physical host energy consumption operating in next cycle, operating energy consumption of virtual machine and switch virtual machine, physics
The control energy consumption of main frame, and feed back to cloud computing system.
Wherein, E (n, aij) representing the energy consumption of operating physical host and virtual machine, E (Δ V, Δ H) represents that switch is virtual
The control energy consumption of machine and physical host;The resource requirement of the resource space of physical host and virtual machine is all represented with bivector,
I.e.:hostj:And VMi:(mipsi,memi),Represent hostjThe internal memory of physical host is big
It is little,Represent hostjThe cpu performance of physical host, memiRepresent VMiThe memory size of type Virtual machine, mipsiTable
Show VMiThe disposal ability of type Virtual machine.The present invention establishes multi-objective constrained optimization model for problem to be studied, its
Mathematical form is as follows:
Wherein, aijRepresent host in current periodjVM is started on physical hostiThe number of type Virtual machine, qjRepresent current
Host in cyclejThe quantity of the virtual machine starting on physical host,WithRepresent host respectivelyjThe power consumption of physical host and
Open the power consumption that each virtual machine needs to increase,It is to open in virtual machine process to produce
Control energy consumption,It is to close the extra control energy consumption producing in virtual machine process,It is to open/close the control energy consumption producing in host process,WithIt is illustrated respectively in hostjThe instantaneous power of virtual machine is opened on physical host, opens
The time opening virtual machine, the instantaneous power closing virtual machine and the time closing virtual machine, and
WithRepresent respectively and open hostjThe instantaneous power of physical host, opening time, closing hostjPhysical host instantaneous
Power and shut-in time;Represent that the virtual machine sum of the respective type of unlatching is equal to measuring and calculating and needs unlatching
Virtual machine number;Represent that the virtual machine sum of the respective type of closing is equal to measuring and calculating and needs the virtual of closing
Machine number;WithPhysical host in expression system
The available resources of CPU and internal memory are to the constraint opening virtual machine.
To sum up, the resource allocation method of the Oriented Green cloud computing of present invention design, using Intel Virtualization Technology, task is adjusted
Degree problem abstract for deploying virtual machine problem;The task amount of user's request is predicted, and combines current system conditions and money
Source distribution, takes conservative control strategy, in advance system resource is allocated, controls, and reduces cloud computing system energy consumption, it is to avoid money
The waste in source;The prediction to task requests amount for the Three-exponential Smoothing method realization is employed during prediction, and please according to task
Predicting the outcome of the amount of asking, is allocated to system resource in advance, controls, and can effectively solve the problem that resource distribution lags behind user's request
Problem, improve system process task request response speed;And introduce conservative control on the basis of Forecasting Methodology
Strategy, exchanged stable in systematic function and the task response ratio that improve for the increase of suitable energy consumption, it is to avoid in reality
When task fluctuation ratio larger in the case of make forecast error big, lead to system response ratio low, the problem of stability difference;Make cloud
In computing system, resource is pre-configured more rationalizes, and activates less physical host number, improves the resource utilization of system, fall
Low system energy consumption, and the feasible solution of optimization during resource is pre-configured, can be quickly found, to a certain extent
Reduce the energy consumption of system, and this method takes less, can effectively adapt to the pre-configured demand of short cycle resource.
Above in conjunction with accompanying drawing, embodiments of the present invention are explained in detail, but the present invention is not limited to above-mentioned enforcement
Mode, in the ken that those of ordinary skill in the art possess, can also be on the premise of without departing from present inventive concept
Make a variety of changes.