CN104657215A - Virtualization energy-saving system in Cloud computing - Google Patents

Abstract

The invention provides a virtualization energy-saving system in Cloud computing, comprising a Cloud computing resource management prototype system and a backstage data processing system; the backstage data processing system comprises a physical resource pool, a virtual machine layer, a local module, a user's application layer and a global resource manager; the Cloud computing resource management prototype system comprises a physical machine management module, a virtual machine management module, a virtual machine dispatching strategy module, a monitoring module, a mirror image management module and a Web module; the resources are integrated to the least physical machine nodes by transferring the resources of the virtual machine, and the idle resources are closed, so that the consumption of the resources is reduced. The virtualization energy conservation in the Cloud computing is designed and realized in the system by using a VM (Virtual Machine) transfer cost model and an SLA (Service Level Agreement) measurement model as the energy consumption estimation models of a data center.

Description

Virtual energy conserving system in cloud computing

Technical field

The present invention relates to field of cloud calculation, particularly relate to virtual energy conserving system in cloud computing.

Background technology

Cloud computing is a kind of account form based on internet, in this way, the software and hardware resources shared and information can be supplied to computing machine and other equipment by demand, mainly based on the increase of the related service of internet, use and delivery mode, be usually directed to provide dynamically easily expansion by internet and be often virtualized resource.Cloud is the one metaphor saying of network, internet.Past often represents telecommunications network with cloud in the drawings, is also used for afterwards representing the abstract of internet and underlying infrastructure.Narrow sense cloud computing refers to payment and the using forestland of IT infrastructure, refers to obtain resource requirement by network in the mode as required, easily expanded; Broad sense cloud computing refers to payment and the using forestland of service, refers to obtain required service by network in the mode as required, easily expanded.It is relevant with software, internet that this service can be IT, may also be other services.It means that computing power also be can be used as a kind of commodity and circulated by internet.

At present, due to the fast development of Intel Virtualization Technology, cloud computation data center have also been obtained to be used widely.Energy consumption problem in cloud computing platform is day by day serious, judges that whether a data center is energy-conservation, and the size of energy consumption is one of major criterion weighed, and therefore how to calculate energy consumption and seems extremely important.In current data center, the resource consumption of server is relatively more serious, causes energy-conserving and environment-protective performance poor, in use there is certain defect.

In sum, for the defect that cloud computing platform energy consumption is larger, virtual energy conserving system in cloud computing is proposed, to solve the deficiencies in the prior art.

Summary of the invention

The object of this invention is to provide virtual energy conserving system in cloud computing, Cost Model and the SLA measurement model energy consumption assessment model as data center is moved by VM, virtual energy-conservation in design and implimentation cloud computing, solve the phenomenon that energy consumption is too high in cloud computing process.

The technical scheme that the present invention adopts for its technical matters of solution is,

Virtual energy conserving system in cloud computing, includes cloud computing resources management prototype system and back-end data disposal system;

Back-end data disposal system includes physical resource pond, virtual machine layer, local module, user application layer, global resource manager; Physical resource pond includes physical machine, network server, storage, network;

When creating virtual machine: user receives this request of application virtual machine by the administration module of the overall situation, then this request is analyzed, finally by the physical machine that selection one is suitable, physical machine is by Hypervisor software creation virtual machine, finally the virtual machine of establishment is returned to user, the application program needed for user can install at user application layer;

Energy-conservation in order to reach cloud computation data center, take into account the target of customer sla simultaneously, between virtual machine layer and user application layer, with the addition of energy consumption monitoring module, SLA judge module, requirement analysis module, request scheduling module respectively;

Cloud computing resources management prototype system includes physical machine administration module, Virtual Machine Manager module, virtual machine scheduling policy module, monitoring module, mirror image administration module, Web module;

Physical machine administration module is responsible for the management of physical machine, comprises registration physical machine, adds physical machine, physical machine performance monitoring is safeguarded with, essential information;

The management of Virtual Machine Manager module in charge virtual machine state, comprises and creates virtual machine, virtual machine life cycle management, many VMM support and the maintenance of virtual machine essential information;

Virtual machine scheduling policy module in charge adds virtual machine Placement and virtual machine selection algorithm;

Monitoring module comprises physical machine monitoring and virtual machine monitoring two submodules, mainly collects cpu load information;

Mirror image administration module comprises the uploading of mirror image, the retrieval of mirror image and the deletion of mirror image;

Web module comprises several submodules such as monitor message displaying, the management of physical machine Virtual Machine Manager, mirror policy, user management and log management, user and keeper by Web module operation whole cloud computing resources management prototype system, also can understand the information in whole resource pool by Web module.

Further, described Hypervisor software comprises Xen, KVM, physical machine can fictionalize the mutually isolated virtual machine of multiple stage on a single server, the user that virtual machine uses is be transparent mutually between coming, they can not perceive this physical server how many virtual machines, can not judge two virtual machines whether in same physical machine.

Further, described requirement analysis module is used for tackling and analyzes user's request, makes whether accepting this user request, simultaneously before submission user request, by the information obtained from energy consumption monitoring module and Virtual Machine Manager module, requirement analysis module makes the decision whether accepted request;

Request scheduling is the virtual machine of distributing user request, determines when to open and close virtual machine to meet user's request;

SLA judge module is the QoS in order to Deterministic service between user and cloud computing service provider, all can ensure in a kind of mode of agreement, the handling capacity of such as serving, corresponding time etc.;

The open and close deciding physical machine of energy consumption monitoring module, when physical machine is idle, close the energy consumption that physical machine can save free physical machine.

Further, described global administration's module is positioned on the master control node of whole data center, the main effect of global administration's module is exactly collect the information uploaded from each local management module collection, obtain the resource utilization situation of whole data center, and the resource at service data center uses, when global administration's module judges that the utilization of resources of data center is unreasonable, the order just by optimizing adjusts the state of the virtual machine in data center, makes it reach a rational target.

Further, described local management module is positioned in each physical machine, the cpu busy percentage of the local real-time specific physical machine of monitoring, further, according to the loading condition of CPU, the service condition of physical machine resource is judged, determine which platform virtual machine needs migration, when move.

Further, described physical machine Registering modules primary responsibility adds management transfer system to physical machine, physical machine is by sending request to overhead control end, overhead control end determines whether allow this physical machine to be added in the middle of system, whether overhead control end constantly detects by the mode of heartbeat the physical machine having joined system normal, if abnormal, from database information, make corresponding amendment, and send corresponding warning message to keeper, the establishment of virtual machine can change the existing available resources of physical machine, and update module is responsible for the information upgrading physical machine;

Physical machine administration module adopts C/S framework, when a physical machine wants the physical resource pond added in cloud computing environment, client in physical machine is first responsible for collection and the inspection of physical machine information, then these information are sent to long-range cloud computing control end server, control end network in charge adds to physical machine in physical resource pond, finally adds corresponding information in a database;

Physical machine in cloud computing environment is divided into some groups in the mode of physics unit, some physical machine are had in each physics unit, each physical machine there are again some virtual machines, the essential information of physical machine and current state, comprise physical machine name, IP address, maximum memory, state, internal memory, free memory, the information such as available CPU number.

Further, described Virtual Machine Manager module is after local management module determines which virtual machine needs migration, the migration task of virtual machine has been responsible for by Virtual Machine Manager module, after migration completes, the power consumption of adjustment source physical machine and target physical machine, after entering adjustment, some idle physical machine just can close to reach energy-conservation target, some physical machine is in dormant state, all these physical machine of not closing be because judge according to demand, when asking more, the physical machine being in dormant state can be transformed into running status fast to meet the request of user, the high concurrent request in rush hour can be tackled like this,

The administration module of virtual machine comprises unlatching, the operation such as closes, moves, suspends, destroys and restart.

Further, described monitoring module adopts Nagios to come physical machine in supervisory system and virtual machine.The information spinner of monitoring will comprise the load of CPU, the information such as internal memory and network, and the information of the physical machine that monitoring obtains and virtual machine is stored in MySQL database, and dispatching algorithm can do the adjustment of some virtual machines by these monitor messages; The monitor message obtained by Nagios has all been deposited in MySQL database, the scheduling strategy of such virtual machine just can obtain the information of the current and history of physical machine and virtual machine by the information in database, carry out the virtual machine in the whole cloud platform of dynamic conditioning by these information.

Further, virtual machine is placed and adjusting module is mainly used to the placement and the adjustable strategies that add concrete virtual machine, system provides the interface of this module, to place and adjustable strategies only needs to realize according to the interface of system so new virtual machine will be added, under the file that the system that uploads to after having realized is specified, once have new strategy to upload to system will set out corresponding method to load this strategy, it is here main that adopt is the User Defined ClassLoader of Java, the Class file beyond running environment dynamically can be added by this ClassLoader, so just can realize the extensibility of system, virtual machine scheduling policy module is the nucleus module of cloud computing resources management prototype system, other module (monitoring modules, physical machine administration module, Virtual Machine Manager module, Web module etc.) all in order to virtual machine scheduling policy module service, the data that virtual machine scheduling policy module needs need to obtain from monitoring module, the action needs such as the establishment migration of virtual machine have been come by Virtual Machine Manager module.

The invention has the advantages that, this system passes through migrating technology by virtual machine, can operate between different physical machine nodes, when the inadequate resource that the physical machine running virtual machine provides or virtual machine need more resource, virtual machine just can adjust dynamically or move in other physical machine, carries out the adjustment of resource.By moving resources of virtual machine, resource consolidation being closed idle resource to minimum physical machine node, the consumption of resource can be made to reduce.Native system moves Cost Model and the SLA measurement model energy consumption assessment model as data center by VM, virtual energy-conservation in design and implimentation cloud computing.

Accompanying drawing explanation

The present invention is described in detail below in conjunction with the drawings and specific embodiments:

Fig. 1 is configuration diagram of the present invention;

Fig. 2 is module relationship schematic diagram of the present invention;

Fig. 3 is cloud computing resources of the present invention management prototype system block architecture diagram;

Fig. 4 is physical machine module architectures block diagram of the present invention;

Fig. 5 is virtual machine module block architecture diagram of the present invention;

Fig. 6 is virtual machine operations communication process figure of the present invention;

Fig. 7 is monitoring module block architecture diagram of the present invention;

Fig. 8 is virtual machine scheduling policy module architectures block diagram of the present invention;

Fig. 9 is the broken line graph between target cpu load of the present invention and power;

Figure 10 is the broken line graph between CPU actual loading of the present invention and power;

Figure 11 is MMT algorithm flow chart of the present invention;

Figure 12 is the consumption figure of electric energy under different CPU threshold value of the present invention;

Figure 13 is the probability graph violating SLA under different CPU threshold value of the present invention;

Figure 14 is the power consumption figure of the different physical machine selection algorithm of the present invention;

Figure 15 is the SLA situation map of the different physical machine selection algorithm of the present invention;

Figure 16 is the migration number of times figure of the different physical machine selection algorithm of the present invention;

Embodiment

The technological means realized to make the present invention, creation characteristic, reaching object and effect is easy to understand, below in conjunction with diagram and specific embodiment, setting forth the present invention further.

As shown in Figure 1, Figure 2, shown in Fig. 3, Fig. 4, Fig. 5, Fig. 6, Fig. 7, Fig. 8, virtual energy conserving system in the cloud computing that the present invention proposes, includes cloud computing resources management prototype system and back-end data disposal system;

Back-end data disposal system includes physical resource pond, virtual machine layer, local module, user application layer, global resource manager; Physical resource pond includes physical machine, network server, storage, network;

When creating virtual machine: user receives this request of application virtual machine by the administration module of the overall situation, then this request is analyzed, finally by the physical machine that selection one is suitable, physical machine is by Hypervisor software creation virtual machine, finally the virtual machine of establishment is returned to user, the application program needed for user can install at user application layer;

Energy-conservation in order to reach cloud computation data center, take into account the target of customer sla simultaneously, between virtual machine layer and user application layer, with the addition of energy consumption monitoring module, SLA judge module, requirement analysis module, request scheduling module respectively;

Cloud computing resources management prototype system includes physical machine administration module, Virtual Machine Manager module, virtual machine scheduling policy module, monitoring module, mirror image administration module, Web module;

Physical machine administration module is responsible for the management of physical machine, comprises registration physical machine, adds physical machine, physical machine performance monitoring is safeguarded with, essential information;

The management of Virtual Machine Manager module in charge virtual machine state, comprises and creates virtual machine, virtual machine life cycle management, many VMM support and the maintenance of virtual machine essential information;

Virtual machine scheduling policy module in charge adds virtual machine Placement and virtual machine selection algorithm;

Monitoring module comprises physical machine monitoring and virtual machine monitoring two submodules, mainly collects cpu load information;

Mirror image administration module comprises the uploading of mirror image, the retrieval of mirror image and the deletion of mirror image;

Web module comprises several submodules such as monitor message displaying, the management of physical machine Virtual Machine Manager, mirror policy, user management and log management, user and keeper by Web module operation whole cloud computing resources management prototype system, also can understand the information in whole resource pool by Web module.

In addition, Hypervisor software comprises Xen, KVM, physical machine can fictionalize the mutually isolated virtual machine of multiple stage on a single server, the user that virtual machine uses is be transparent mutually between coming, they can not perceive this physical server how many virtual machines, can not judge two virtual machines whether in same physical machine.In order to meet effective utilization of resource in cloud computation data center, the reduction of energy consumption, each physical server loading onto monitoring module, then being carried out the dynamic conditioning of whole cloud platform by one of cloud platform overall control center.

Further, described requirement analysis module is used for tackling and analyzes user's request, makes whether accepting this user request, simultaneously before submission user request, by the information obtained from energy consumption monitoring module and Virtual Machine Manager module, requirement analysis module makes the decision whether accepted request;

Request scheduling is the virtual machine of distributing user request, determines when to open and close virtual machine to meet user's request;

SLA judge module is the QoS in order to Deterministic service between user and cloud computing service provider, all can ensure in a kind of mode of agreement, the handling capacity of such as serving, corresponding time etc.;

The open and close deciding physical machine of energy consumption monitoring module, when physical machine is idle, close the energy consumption that physical machine can save free physical machine.

Further, described global administration's module is positioned on the master control node of whole data center, the main effect of global administration's module is exactly collect the information uploaded from each local management module collection, obtain the resource utilization situation of whole data center, and the resource at service data center uses, when global administration's module judges that the utilization of resources of data center is unreasonable, the order just by optimizing adjusts the state of the virtual machine in data center, makes it reach a rational target.

Further, described local management module is positioned in each physical machine, the cpu busy percentage of the local real-time specific physical machine of monitoring, further, according to the loading condition of CPU, the service condition of physical machine resource is judged, determine which platform virtual machine needs migration, when move.

Further, described physical machine Registering modules primary responsibility adds management transfer system to physical machine, physical machine is by sending request to overhead control end, overhead control end determines whether allow this physical machine to be added in the middle of system, whether overhead control end constantly detects by the mode of heartbeat the physical machine having joined system normal, if abnormal, from database information, make corresponding amendment, and send corresponding warning message to keeper, the establishment of virtual machine can change the existing available resources of physical machine, and update module is responsible for the information upgrading physical machine;

Physical machine administration module adopts C/S framework, when a physical machine wants the physical resource pond added in cloud computing environment, client in physical machine is first responsible for collection and the inspection of physical machine information, then these information are sent to long-range cloud computing control end server, control end network in charge adds to physical machine in physical resource pond, finally adds corresponding information in a database;

Physical machine in cloud computing environment is divided into some groups in the mode of physics unit, some physical machine are had in each physics unit, each physical machine there are again some virtual machines, the essential information of physical machine and current state, comprise physical machine name, IP address, maximum memory, state, internal memory, free memory, the information such as available CPU number.

Further, described Virtual Machine Manager module is after local management module determines which virtual machine needs migration, the migration task of virtual machine has been responsible for by Virtual Machine Manager module, after migration completes, the power consumption of adjustment source physical machine and target physical machine, after entering adjustment, some idle physical machine just can close to reach energy-conservation target, some physical machine is in dormant state, all these physical machine of not closing be because judge according to demand, when asking more, the physical machine being in dormant state can be transformed into running status fast to meet the request of user, the high concurrent request in rush hour can be tackled like this,

The administration module of virtual machine comprises unlatching, the operation such as closes, moves, suspends, destroys and restart,

Virtual Machine Manager module, comprises and creates virtual machine, virtual machine life cycle management, many VMM support and the maintenance of virtual machine essential information.The Intel Virtualization Technology of current main flow comprises Xen and KVM two kinds, is encapsulated these two kinds of technology by Libvirt, makes the interface unification of calling.The life cycle management of virtual machine comprise virtual machine establishment, time-out, migration, destroy and to restart etc. several.

Libvirt is a Virtual Machine Manager API independent of Hypervisor, and it can interact to the virtualization capability of most of operating system.Libvirt provides that one common, general and stable layer carries out safe management to the virtual machine in node.Libvirt can create virtual machine, revise, monitor, control, move and the operation such as stopping.

The main Java EE adopted of Web end develops, and the Libvirt that call control end server is to manage whole cloud platform, and what adopt here is the communication that SOAP Web Service realizes between both.Axis2 is a kind of open source technology framework realizing SOA Web Service, and he has two language version Axis2/C and Axis2/Java, and these two versions can be used in combination.Due to the Java language that Web end in our prototype system adopts, and Libvirt is that what to adopt is C language, so the different language version of Axis2 two kinds is well suited for realizing in our system.The communication process of virtual machine operations as illustrated in figs. 2-7.The Virtual Machine Manager interface of Web end receives user's request, then the Axis2/Java module asking to send to Web to hold, this module complexity sends complicated Web Service information to control end, the Axis2/C of control end receives Web and holds the Web Service sent to ask, then call the Libvirt of bottom, Libvrit determines specifically to call the Hypervisor in which platform physical machine again.When creating successfully, then information is sent to Web module end.

Further, described monitoring module adopts Nagios to come physical machine in supervisory system and virtual machine.The information spinner of monitoring will comprise the load of CPU, the information such as internal memory and network, and the information of the physical machine that monitoring obtains and virtual machine is stored in MySQL database, and dispatching algorithm can do the adjustment of some virtual machines by these monitor messages; The monitor message obtained by Nagios has all been deposited in MySQL database, the scheduling strategy of such virtual machine just can obtain the information of the current and history of physical machine and virtual machine by the information in database, carry out the virtual machine in the whole cloud platform of dynamic conditioning by these information.

Further, virtual machine is placed and adjusting module is mainly used to the placement and the adjustable strategies that add concrete virtual machine, system provides the interface of this module, to place and adjustable strategies only needs to realize according to the interface of system so new virtual machine will be added, under the file that the system that uploads to after having realized is specified, once have new strategy to upload to system will set out corresponding method to load this strategy, it is here main that adopt is the User Defined ClassLoader of Java, the Class file beyond running environment dynamically can be added by this ClassLoader, so just can realize the extensibility of system, virtual machine scheduling policy module is the nucleus module of cloud computing resources management prototype system, other module (monitoring modules, physical machine administration module, Virtual Machine Manager module, Web module etc.) all in order to virtual machine scheduling policy module service, the data that virtual machine scheduling policy module needs need to obtain from monitoring module, the action needs such as the establishment migration of virtual machine have been come by Virtual Machine Manager module.

Meanwhile, cloud computation data center is when running, and need corresponding electric power resource supply, energy consumption is the total amount of system at a period of time internal consumption electric power resource.Model is a kind of expression-form of studied system, process, things or concept, usually can use mathematical relation, sequence of algorithms and logical relation to represent.Therefore, energy consumption model is exactly use mathematical relation to represent the total amount of system at a period of time internal consumption electric power resource.In cloud computing environment, accurately estimate by corresponding energy consumption model the energy consumption that system produces within certain hour, judge that whether the energy consumption management method of cloud computing platform is energy-conservation with this.

At present there are two kinds of main methods for energy consumption modeling: a kind of method carries out modeling from the angle of hardware; Another kind method is then carry out modeling according to the dissimilar of task of resource different conditions when in use and service.Modeling based on power consumption is also the direction that research is more popular at present, but power consumption is different from the concept of energy consumption, and energy consumption represents the summation of power consumption, and power consumption represents the speed of computing machine power consumption resource, and energy consumption equals the integration of power consumption to the time.Usually, system energy consumption low in energy consumption is not necessarily low, but the size of power consumption can from the change of certain angle reflection energy consumption.

For the angle modeling from hardware, setting up energy consumption model to server, is generally carry out computing system energy consumption according to the utilization factor of CPU and the check figure of CPU.The electric energy that one station server consumes, be normally made up of the consumption of CPU, internal memory, disk and network interface card, wherein CPU consumes most energy.Therefore, most of energy consumption model is all carry out modeling based on the utilization factor of CPU.Use CPU as the factor of modeling, be applicable to the system of computation-intensive; For the energy consumption model considering the equipment such as internal memory, network interface card, be then applicable to data-intensive and application that is communications-intensive.The memory device of data center also can produce a large amount of energy consumptions, and the modeling for energy consumption also can be considered from server and memory device two aspects.

For the angle modeling used from resource, the state of system can be in conversion, work and idle three kinds of states, and under these three states, the summation of energy consumption is the energy consumption of system.When system is in idle condition, the consumption for energy consumption is minimum, and meanwhile, when physical resource is in different states, the total energy consumption of consumption is not identical yet.

As can be seen from said method, just carry out modeling based on cpu busy percentage based on energy consumption model major part at present, current energy consumption model does not take into full account the SLA of data center.From the angle of user, SLA is also the problem that they the most directly pay close attention to, and the quality of SLA directly represent the quality of the service quality that cloud computing service provider provides.Native system, mainly from the angle that server is energy-conservation, utilizes the dynamic conditioning of virtual machine to reach the energy-conservation of cloud computation data center, takes into account the SLA between user and cloud service provider simultaneously, makes data center not reduce again the standard of SLA while energy-conservation.

The energy consumption of data center server mainly comprises CPU energy consumption, internal memory energy consumption, network adapter energy consumption, and hard disk energy consumption etc., wherein CPU energy consumption account for the overwhelming majority, and native system represents the energy consumption of whole server with CPU energy consumption, calculates the energy consumption of data center with this.

At present, the CPU of server has dynamic voltage/frequency adjustment technology (DVFS).DVFS dynamic technique be the application program run according to chip to the different needs of computing power, the running frequency of dynamic adjustments chip and voltage (for same chip, frequency is higher, and required voltage is also higher), thus reach energy-conservation object.DVFS technology is mainly for the adjustable system unit of the voltage/frequencies such as CPU, according to the funtcional relationship between energy consumption and frequency, there are some researches show, when completing identical workload, when the frequency stabilization of CPU is in alap situation, its energy consumption is minimum.For application program, not always need to perform with the fastest speed, if the performance requirement of task can be met with lower cpu frequency, then can reduce system energy consumption or reach the battery life of expectation.Key to the issue is the demand of correct Prediction operating load to CPU, rationally processes the quota of CPU simultaneously.

SPEC(The Standard Performance Evaluation Corporation) standard performance evaluation and test mechanism, be the authoritative organization in the world system application performance being carried out to standard evaluation and test, it is intended to establish, revise and assert the standard that a series of server application performance is assessed.From the relation between the load and power of the Proliant ML110G5 server of the Hewlett-Packard that SPEC announces as shown in table 3-1.

Relation between the load of table 3-1Proliant ML110G5 server and power

Targeted loads	Actual loading	Power (unit: W)
			100%	99.5%	135
90%	91.1%	133
			80%	80.5%	129
70%	69.9%	125
			60%	60.2%	121
50%	49.8%	116
			40%	39.4%	110
30%	30.0%	105
			20%	19.8%	101
10%	10.1%	97.0
			Idle condition	Idle condition	93.7

Broken line graph is drawn as according to the data in table 3-1, as shown in Figure 9, Figure 10, the as can be seen from the figure power of the CPU linear relationship that becomes to be similar to load.CPU in an idle state load also reaches 69.4% under cpu load full load condition, so under server is in idle condition always, and is not in dormant state or closedown is the waste easily causing electric energy.

According to the data that SPEC provides, native system is mapped to linear relation the utilization rate of the power consumption of CPU and CPU.So the energy consumption model of server can represent according to formula (1):

Pu＝ Pmin+1·kk* Pmin*u (1)

In formula (1) represent the CPU power consumption under CPU idle condition, the power consumption of what k represented is server CPU in an idle state accounts for the ratio of the power consumption under CPU full load condition, the utilization factor of the CPU that u represents.

Server energy consumption is in an idle state about 70% under full load condition, and namely the value of k approximates 70%.From the relation between the load and power of the Hewlett-Packard Proliant ML110G3 server of SPEC announcement, we can find out that K value is 69.4%, and this also demonstrates the correctness that k value approximates 70%.

Native system just equals 70% to simplify the power consumption model of server with K, so just can release formula (2) from formula (1).

Pu＝Pmin+ 0.30.7* Pmin*u＝ Pmin*1+37*u (2)

Because the utilization factor u of server is along with time variations, so the cpu busy percentage of certain time point can represent with u (t).

Therefore, the power consumption in server a period of time just can be represented at an integration of this time period by P (u), as shown in Equation (3).

E＝ tPut(3)

According to above server energy consumption model, the power consumption of server is weighed by the utilization factor of CPU.So in order to reduce the energy consumption of data center, native system reaches energy-conservation effect mainly through the cpu busy percentage of server in adjustment data center.The estimation of consumption of data center can pass through following equation expression.

$\mathrm{EYIDCY}=\underset{i=0}{\overset{n}{\mathrm{\Σ}}}\mathrm{Ei}$

Wherein EYIDCY represents the energy consumption of whole data center, and Ei is the performance consumption of a physical server in data center, and whole data center runs n platform physical server, and consumption of data center is approximately equal to the energy consumption summation of Servers-all.

Dynamic migration of virtual machine technology can make a virtual machine move to another physical machine from a physical machine, realize the dynamic migration of virtual machine, the following condition of demand fulfillment:

(1) the minimal disruption time: virtual machine will ensure that stop time is minimum, because any operation service thereon all cannot perform when virtual machine is shut down time when moving.

(2) consistance: virtual machine is when migration, and the state of 2 virtual machines can change, the stability of meeting influential system when virtual machine is inconsistent.

(3) least interference: need when migration to ensure not have other virtual machines to fight for the resources such as such as CPU, internal memory and the network bandwidth, in order to avoid the service that interference is movable.

(4) transparency: in the whole process of migration should be completely transparent for user, and network connects and the state of application program should not be affected.

In the process of virtual machine (vm) migration, the performance of virtual machine can be affected, and the application run on a virtual machine also all can be affected.The impact of virtual machine (vm) migration on application depends mainly on the degree relied on page when application performs, to traditional web application, virtual machine (vm) migration is roughly 10% of cpu busy percentage on the impact of performance, the meaning between the lines is exactly, virtual machine (vm) migration can impact SLA, so should reduce the number of times of migration as far as possible.

Virtual machine moves, and first will produce the blank of a virtual machine on target machine, then the memory content of original virtual machine in the mode of shared storage, by network, these contents are transmitted in the past.After internal storage data copies, then data stream is transferred to target machine from original machine gets on.After data stream, internal memory are all consistent, just can close original virtual machine, open new virtual machine.A basic point of dynamic migration must be share to store exactly, and operating system must be public.

Virtual machine (vm) migration needs the network bandwidth between the internal memory of how long main and this virtual machine and physical machine relevant.Because now in cloud computation data center, virtual machine is all the shared storage adopted, so a virtual machine moves in another physical machine from a physical machine, as long as internal memory is moved to another physical machine from a physical machine.

So virtual machine (vm) migration can represent with formula (4) the impact of virtual machine performance

$T_{\mathrm{mj}}=\frac{M_j}{B_j}$

The impact that what Udj represented is on performance in whole virtual machine (vm) migration process.The start time of the virtual machine (vm) migration that T0 represents, the time of what Tmj represented is whole transition process needs, he is the memory size Mj that occupied by this virtual machine except the bandwidth B j of upper virtual machine represents.The cpu busy percentage of what Uj represented is exactly virtual machine.

Cloud computing service, as a kind of commodity, has the qos requirement of oneself, and parameter comprises time, cost, reliability and trust coefficient etc.Current You Duojia IT company provides cloud computing service, and in an environment relatively competed, the lifting of service quality is extremely urgent.Service quality is a dynamic concept, can change along with the change of time, for the user using cloud computing service, the concept introducing SLA can ensure the QoS of cloud service effectively, be conducive to the configuration optimizing cloud resource, thus the interests coordinated between cloud user and cloud supplier, reach the effect of a doulbe-sides' victory.

Service-level agreement (Service Level Agreement, SLA) is the agreement provided about service guarantees signed by user and cloud service provider.SLA carries out simply abstract to resource, the index can measured by some, ensures the promise of cloud service provider to user service type and service quality.SLA generally includes two aspects: technological layer and legal perspective.From technological layer, SLA normally carrys out integrating representation by some parameters, as the bandwidth etc. that the time delay of service, the shake in service process, service provide, by setting up an interactively agreement between cloud service provider and user, can ensure cloud service provider provide to user needed for the grade service that reaches, need effectively to safeguard all kinds of services provided and the management of resource simultaneously, and then good QoS can be provided to ensure.From legal perspective, when cloud service provider can not meet the SLA of signing, the content that user can sign according to actual conditions and SLA is compensated accordingly; Conversely, cloud service provider also can be served according to the SLA of different brackets, provides different services to different clients, collects different expenses.

Present each large cloud service provider all using the bright spot of QoS as oneself cloud products propaganda, attracts various user with high-quality QoS.And QoS is showed as a kind of agreement is exactly SLA, user by and cloud service provider sign SLA, quantize the service what quality is cloud service provider should provide, user also can supervise by SLA the service quality that cloud service provider provides.What define in general SLA is all the maximum response time of serving, minimum throughout etc.

Because the request of application huge change can occur along with the change of time, there will be crest situation at some time point, this be if now virtual machine can not meet request, just there will be the situation that can not meet SLA.In order to the correctness of system testing needs and verification system, need to make a model quantized to SLA.The ability of the execution instruction of CPU can with MIPS(Million Instructions Per Second) represent, namely can perform how many 1,000,000 instructions p.s..

User applies for virtual machine at the beginning, a virtual CPU can be required to cloud service provider, the executive capability of this CPU is determined, but huge change may can be there is along with the change of time in the application of user, also can change to the requirement of CPU, this CPU ability of applying at the beginning likely can not meet the request of user's application under some condition, and Here it is violates the situation that SLA occurs.In order to indicate the ratio that this situation occurs, system testing formula (5) represents this ratio.

SLA＝j＝lMtUrjt-Uajtdtj＝lMtUrjtdt (5)

The MIPS of application request on certain time point virtual machine that in formula (5), Urj represents, that Uaj represents is the actual MIPS being supplied to the CPU of virtual machine of certain time point physical server.Here the MIPS of the CPU of the request of whole virtual machine is deducted the MIPS of actual all CPU of virtual machine, and then the integration of MIPS than the CPU of upper whole virtual machine request, represent the quality of the SLA of whole cloud computing service by this value how.

For the code of the BFD algorithm of energy optimization, when there being request to create virtual machine, method findHostForVm just detects physical machine successively, can hold again physical machine the returning as function of lower virtual machine while that a final selection load being the highest.Cpu load dual threshold setting code is whether two physical machine loads are more than the CPU upper limit and the function lower than CPU lower limit.

Virtual machine scheduling policy is a np hard problem in fact, needs to use auxiliary heuritic approach.A physical server can be regarded as a container, traditional heuritic approach comprises MAX-MIN algorithm, MIN-MIN algorithm, genetic algorithm and ant group algorithm etc.Native system is in order to reach the solution of problem fast, so adopt BFD algorithm, this algorithm can reach target fast, also can reach one and preferably separate.Quick relative users request is needed, so be not suitable for native system as the algorithm that the complexity such as genetic algorithm, ant group algorithm is larger in virtual machine Placement Strategy.And genetic algorithm, ant group algorithm etc. are all random algorithms, these algorithms need repeatedly computing, the poor reliability of result, can not obtain stable solution.

BFD algorithm has been proved to be and can be no more than (11/9*OPT+1) chest, deposits all knapsack (what OPT represented is under optimum solution, the chest number of needs).In system realizes, first the virtual machine of all requests is sorted from high to low according to the utilization factor of CPU, from this queue, then once select a virtual machine to be placed in physical machine.The selection of physical machine is, first the Current resource of this physical machine can meet the requirement of virtual machine, then therefrom selects a cpu busy percentage peak physical machine.More space can be flowed out like this to subsequent user application virtual machine.Native system is owing to mainly considering the power consumption of server, and the power consumption that server is similar to can calculate from cpu load, so the BFD algorithm in native system is mainly using the cpu load of server as overriding concern factor.

Can know from discussion above, virtual machine Placement Problems is a knapsack problem, when virtual machine the least possible be placed in less physical machine time, due to the decline of physical server usage quantity, the electric energy of cloud computation data center also can decline; But this can cause another one problem, too much virtual machine concentrates on a physical server, and virtual machine performance can be caused to can not be guaranteed, especially in peak period.The QoS of the service that such cloud computing service provider provides just can not be guaranteed, and the SLA that user signs also can not be guaranteed.So physical server should leave certain idling-resource to ensure in peak period, the SLA of user is guaranteed.

In order to make the resource of physical server have certain reservation, arrange a threshold value can to the resource of physical server, when placing virtual machine, the use of resource can not exceed this threshold value, instead of whole available resources of physical server.Simultaneously, in order to energy-conservation also needs resets a threshold value, when the resource utilization of physical server lower than this threshold value time, the just whole virtual machines of migration on it, then this physical server is in dormant state or closedown, to reach energy-conservation object.

The adjustment process of virtual machine, the cpu busy percentage of present physical machine is monitored exactly by local management module, when the CPU SC service ceiling of physical server cpu busy percentage higher than physical server, just notice global administration module, global administration's module in charge determines which platform virtual machine of migration, when move, move to any platform physical server.When physical server cpu busy percentage lower than CPU use lower in limited time, also be notify global administration's module by local management module, the whole virtual machines of global administration's module in charge migration on this physical server, and determine to move on which platform or which physical server, when move.

When occurring that the load of the whole virtual machines on certain physical server is greater than the CPU upper limit on this physical server, just need to move on it one or polymorphic virtual machine on other more idle physical server, at this moment which virtual machine how just there will be one to select by the problem of moving, after virtual machine is moved, make the load of this physical server will lower than the upper limit of cpu load, the cost of moving will be considered simultaneously, move less virtual machine as far as possible, the virtual machine that migration cost is little.

How to select there is multiple choices algorithm by the virtual machine moved, propose two kinds of virtual machine selection algorithms with different targets here.The main target of minimum transition time algorithm (Minimum Migration Time, MMT) is each virtual machine selecting always to select transit time minimum when needing the virtual machine of migration.The main target of maximum utilization rate migration algorithm (Maximum Utilization, MU) is after making virtual machine (vm) migration, and the difference between the cpu load of physical server and the CPU upper limit is little as far as possible.

The thinking of the minimum transition time algorithm (MMT) that we propose, when the CPU usage of certain physical server is more than the upper limit of CPU, just need migration one or a part of virtual machine in fact, first to whole virtual machines according to the utilization rate of CPU according to descending sort, then the size delta of actual CPU usage more than the CPU upper limit of present physical server is calculated, the virtual machine sorted, whether exceed Δ according to the cpu busy percentage of current virtual machine and be divided into two parts, if represent the queue not empty exceeding Δ part, just represent that migration any virtual machine wherein just can make the CPU of this physical server use the upper limit dropping to CPU, the object as migration of the virtual machine then selecting committed memory minimum from this queue, the time of virtual machine (vm) migration is main relevant with the memory size that virtual machine takies.If this queue being greater than Δ is for empty, then represent that needing to move polymorphic virtual machine can be just that the cpu busy percentage of physical server drops to and meets this target of the CPU upper limit.If this situation, just use according to the CPU of virtual machine and move from high to low, till the CPU usage always moving to whole physical server drops to the CPU upper limit, as shown in figure 11.

After choosing the virtual machine needing migration, BFD algorithm is just adopted to come virtual machine (vm) migration on suitable physical server.

The main target of maximum utilization rate migration algorithm (MU) is after making virtual machine (vm) migration, and the difference between the cpu load of physical server and the upper limit of cpu load is minimum.Effectively can utilize the resource of physical server like this, reduce the waste of resource.

Flow process (MU) and the minimum transition time algorithm of the maximum utilization rate migration algorithm of our proposition are close, main thought is as follows, first virtual machine is sorted according to cpu load, then successively from high capacity to low load detecting, make the difference between the cpu load of the physical server after migration and the upper limit of cpu load minimum.

In the process selecting the virtual machine needing migration, a document mistake! Do not find Reference source.In propose a kind of algorithm (Minimization of Migrations) of minimum transition number of times.The virtual machine that this algorithm selects load maximum from the physical server more than the CPU upper limit, as migrating objects, makes the CPU of physical server be reduced to below the threshold value of CPU.The time of cost and the migration of moving is not considered in this algorithm.So consider the time of migration cost and migration in previous algorithm, as the place of improving.The roughly thought of MM algorithm is such: if the cpu load of physical server is more than the upper limit of CPU, then need to move certain virtual machine from this physical server, this algorithm moves cpu load virtual machine the earliest successively, until the cpu load of this physical server is reduced to the upper limit of cpu load.

In order to contrast the quality of minimum transition number of times, the shortest transit time algorithm, some other virtual machine selection algorithms are needed to contrast.The MM algorithm introduced above as the algorithm of a contrast, can also introduce some other algorithms.

Stochastic selection algorithm, when local management module detects the upper limit of cpu load beyond CPU of this physical server, then loosen information to global administration's module, the virtual machine on global administration's this physical server of module complexity migration is reduced under the upper limit of CPU to make the cpu load of this physical server.Stochastic selection algorithm, selection random from virtual machine list virtual machine is as the object of migration, the situation of the current CPU load on this physical server is judged again after migration, if be also above the upper limit of cpu load, then continue random migration virtual machine, if cpu load is low than the upper limit of cpu load, then algorithm terminates.

The same as already mentioned previously, in order to ensure the request leaving certain idling-resource reply peak time of physical server, CPU is provided with a upper limit threshold, and the idling-resource exceeding this threshold value is just used to tackle peak time.The setting of this threshold value and the energy-saving effect of whole data center and SLA Relationship Comparison large, if it is higher that this threshold value is arranged, then there is more material resources can hold more virtual machine, electric energy can be saved like this, but the ensureing of QoS just likely can not get good guarantee in peak thing because idling-resource stay insufficient.If it is low that this upper limit threshold is arranged, then QoS can well be ensured, but the virtual machine that same physical server can hold is just few, and the effect of saving electric energy is just not obvious.

One is exactly the setting of cpu load lower limit in addition, when the cpu load of physical server is lower than this threshold value, just show that the load of this physical server is now very low, need the whole virtual machines of migration on this physical server, then allow this physical server be in the state of dormancy or closedown.

In order to reach energy-conservation effect, guaranteeing service quality simultaneously, being provided with two threshold values to the CPU of data center's physical server, these two threshold values need to be determined by actual conditions, and global administration's module did one-time detection every 5 minutes to the cpu load of server.By adopting the threshold value of six groups of different cpu loads, and setting different CPU threshold values, observing the change between the energy consumption of whole data center and SLA, the change of these six groups of different CPU threshold values is as shown in table 4-1.

Table 4-1 threshold range

Numbering	Most high threshold	Lowest threshold
			1	0.7	0.31
2	0.75	0.35
			3	0.8	0.4
4	0.85	0.45
			5	0.9	0.5
6	0.95	0.55

When the scope of CPU threshold value setting is different, different to the consumption of electric energy, data below adopt MMT algorithm, and carry out energy consumption analysis to the CPU scope of six groups of different threshold values, test result is as follows:

The different threshold range power consumption of table 4-2

Numbering	Threshold range	Power consumption (KW/h)
			1	0.3——0.7	11.5
2	0.35——0.75	12.1
			3	0.4——0.8	13.2
4	0.45——0.85	13.6
			5	0.5——0.9	15.2
6	0.55——0.95	16.1

When the scope of the threshold value setting of CPU is different, the requirement that SLA reaches is not identical yet, and test below adopts the MMT algorithm of chapter 3, and observe when using different CPU threshold ranges, the situation of change of SLA, test result is as follows:

The different threshold range SLA of table 4-3

Numbering	Threshold range	SLA
			1	0.3——0.7	5.2
2	0.35——0.75	3.3
			3	0.4——0.8	2.4
4	0.45——0.85	2.3
			5	0.5——0.9	1.1
6	0.55——0.95	0.5

As can be seen from Figure 12, Figure 13, along with the continuous increase of CPU threshold value, the power consumption of data center reduces gradually, but SLA but increases gradually.This mainly because when CPU threshold value arrange larger time, more virtual machine can be had to concentrate on a physical server, and physical machine idle like this will be many, and the physical server of closing these free time just can reach the target of reduction electric energy.Meanwhile, more virtual machine concentrates on a station server, and the probability that the situation violating SLA occurs also will increase, accordingly so the value of SLA can become large.

It can also be seen that from test, along with the continuous increase of cpu load, in data center, the consumption of server electric energy reduces gradually, and trend comparison is gentle, but when cpu load changes to 0.55 ~ 0.95 from 0.5 ~ 0.9 time, the trend that SLA increases is apparently higher than change above.So native system thinks that the threshold value of load of data center CPU answers most recent value to be set in 0.5 ~ 0.9.

Next, contrast under different virtual machine selection algorithms, the consumption of electric energy, the situation of SLA and the number of times of virtual machine (vm) migration.Analyzed by the energy consumption of five kinds of algorithms below test comparison:

Table 4-4 testing algorithm

Algorithm	Describe
		DVFS	Without Placement
MMT	Minimum transition time algorithm
		MU	Peak use rate algorithm
MM	Minimum transition number of times algorithm
		Random	Stochastic selection algorithm

As can be seen from Figure 14, by arranging two threshold values of CPU, and adopt dynamic migration technology, the power consumption of whole server is about 70% when not adopting dynamic migration technology.

As can be seen from Figure 15, Figure 16, minimum transition time algorithm (MMT) is because the virtual machine selecting transit time minimum is as migrating objects at every turn, so the situation of the violation SLA caused due to migration is minimum.

The maximum utilization rate migration algorithm (MU) that we propose, owing to considering the difference between the load of the rear physical machine of migration and the CPU upper limit, so can use physical machine resource to greatest extent, is the most energy-conservation in several virtual machine selection algorithm.The virtual machine that minimum migration number of times algorithm (MM) selects load the highest at every turn from physical machine, as migrating objects, can ensure the least number of times of moving like this.Because native system is mainly from energy-conservation angle, it is best that maximum utilization rate migration algorithm (MU) is embodied in energy-conservation.

Test analysis is carried out in this test altogether use 10 physical machine, and the load situation of change before and after algorithms of different adjustment is as follows:

Loading condition before and after the adjustment of table 4-5DFVS server

	PM1	PM2	PM3	PM4	PM5	PM6	PM7	PM8	PM9	PM10
											Load before adjustment	40.58	37.12	58.39	54.94	51.21	32.36	44.03	0	44.31	0
Adjustment back loading	0	90.12	77.93	75.26	73.03	0	0	0	0	0

Loading condition before and after the adjustment of table 4-6MMT arithmetic server

	PM1	PM2	PM3	PM4	PM5	PM6	PM7	PM8	PM9	PM10
											Load before adjustment	40.58	37.12	58.39	54.94	51.21	32.36	44.03	0	44.31	0
Adjustment back loading	69.88	0	75.93	65.26	70.03	0	48.49	0	45.34	0

Loading condition before and after the adjustment of table 4-7MU arithmetic server

	PM1	PM2	PM3	PM4	PM5	PM6	PM7	PM8	PM9	PM10
											Load before adjustment	40.58	37.12	58.39	54.94	51.21	32.36	44.03	0	44.31	0
Adjustment back loading	86.88	0	85.93	87.26	90.03	0	0	0	0	0

Loading condition before and after the adjustment of table 4-8MM arithmetic server

	PM1	PM2	PM3	PM4	PM5	PM6	PM7	PM8	PM9	PM10
											Load before adjustment	40.58	37.12	58.39	54.94	51.21	32.36	44.03	0	44.31	0
Adjustment back loading	0	45.89	65.93	67.26	70.03	45.49	0	0	78.28	0

Loading condition before and after the adjustment of table 4-9Random arithmetic server

	PM1	PM2	PM3	PM4	PM5	PM6	PM7	PM8	PM9	PM10
											Load before adjustment	40.58	37.12	58.39	54.94	51.21	32.36	44.03	0	44.31	0
Adjustment back loading	42.89	36.45	55.93	62.26	46.03	40.25	45.49	0	30.18	0

Loading condition as can be seen from the above table before and after server adjustment, occupy 8 physical servers before adjustment, and the load major part of server is in about 50%, the utilization rate of server is not high.After adjustment, use MMT, MU and MM algorithm can reduce the use amount of physical server, several physical servers can be closed more, reduce the consumption of electric energy.DVFS algorithm and Random algorithm just adjust the distribution condition of resource, well can not reach energy-saving effect.Because the threshold value being provided with cpu load is at the beginning 50%-90%, after adjustment, the loading range of CPU is all in this interval substantially, but because some server can not hold more virtual machine, so the virtual machine on this server cannot move away, the load of CPU can a little more than threshold value.In general, after adjustment, the quantity of physical machine reduces, and can close corresponding physical machine and carry out saves energy.

Based on above-mentioned, advantage of the present invention is: this system passes through migrating technology by virtual machine, can operate between different physical machine nodes, when the inadequate resource that the physical machine running virtual machine provides or virtual machine need more resource, virtual machine just can adjust dynamically or move in other physical machine, carries out the adjustment of resource.By moving resources of virtual machine, resource consolidation being closed idle resource to minimum physical machine node, the consumption of resource can be made to reduce.Native system moves Cost Model and the SLA measurement model energy consumption assessment model as data center by VM, virtual energy-conservation in design and implimentation cloud computing.

More than show and describe ultimate principle of the present invention, principal character and advantage of the present invention.The technician of the industry should understand; the present invention is not restricted to the described embodiments; what describe in above-described embodiment and instructions just illustrates principle of the present invention; the present invention also has various changes and modifications without departing from the spirit and scope of the present invention, and these changes and improvements all fall in the claimed scope of the invention.Application claims protection domain is defined by appending claims and equivalent thereof.

Claims (9)

1. virtual energy conserving system in cloud computing, is characterized in that, includes cloud computing resources management prototype system and back-end data disposal system;

Back-end data disposal system includes physical resource pond, virtual machine layer, local module, user application layer, global resource manager; Physical resource pond includes physical machine, network server, storage, network;

When creating virtual machine: user receives this request of application virtual machine by the administration module of the overall situation, then this request is analyzed, finally by the physical machine that selection one is suitable, physical machine is by Hypervisor software creation virtual machine, finally the virtual machine of establishment is returned to user, the application program needed for user can install at user application layer;

Energy-conservation in order to reach cloud computation data center, take into account the target of customer sla simultaneously, between virtual machine layer and user application layer, with the addition of energy consumption monitoring module, SLA judge module, requirement analysis module, request scheduling module respectively;

Cloud computing resources management prototype system includes physical machine administration module, Virtual Machine Manager module, virtual machine scheduling policy module, monitoring module, mirror image administration module, Web module;

Physical machine administration module is responsible for the management of physical machine, comprises registration physical machine, adds physical machine, physical machine performance monitoring is safeguarded with, essential information;

The management of Virtual Machine Manager module in charge virtual machine state, comprises and creates virtual machine, virtual machine life cycle management, many VMM support and the maintenance of virtual machine essential information;

Virtual machine scheduling policy module in charge adds virtual machine Placement and virtual machine selection algorithm;

Monitoring module comprises physical machine monitoring and virtual machine monitoring two submodules, mainly collects cpu load information;

Mirror image administration module comprises the uploading of mirror image, the retrieval of mirror image and the deletion of mirror image;

Web module comprises several submodules such as monitor message displaying, the management of physical machine Virtual Machine Manager, mirror policy, user management and log management, user and keeper by Web module operation whole cloud computing resources management prototype system, also can understand the information in whole resource pool by Web module.

2. virtual energy conserving system in cloud computing according to claim 1, it is characterized in that, described Hypervisor software comprises Xen, KVM, physical machine can fictionalize the mutually isolated virtual machine of multiple stage on a single server, the user that virtual machine uses is be transparent mutually between coming, and they can not perceive this physical server how many virtual machines, can not judge two virtual machines whether in same physical machine.

3. virtual energy conserving system in cloud computing according to claim 1, it is characterized in that, described requirement analysis module is used for tackling and analyzes user's request, before submission user request, make whether accepting this user request, simultaneously, by the information obtained from energy consumption monitoring module and Virtual Machine Manager module, requirement analysis module makes the decision whether accepted request;

Request scheduling is the virtual machine of distributing user request, determines when to open and close virtual machine to meet user's request;

SLA judge module is the QoS in order to Deterministic service between user and cloud computing service provider, all can ensure in a kind of mode of agreement, the handling capacity of such as serving, corresponding time etc.;

The open and close deciding physical machine of energy consumption monitoring module, when physical machine is idle, close the energy consumption that physical machine can save free physical machine.

4. virtual energy conserving system in cloud computing according to claim 1, it is characterized in that, described global administration's module is positioned on the master control node of whole data center, the main effect of global administration's module is exactly collect the information uploaded from each local management module collection, obtain the resource utilization situation of whole data center, and the resource at service data center uses, when global administration's module judges that the utilization of resources of data center is unreasonable, order just by optimizing adjusts the state of the virtual machine in data center, it is made to reach a rational target.

5. virtual energy conserving system in cloud computing according to claim 1, it is characterized in that, described local management module is positioned in each physical machine, the cpu busy percentage of the local real-time specific physical machine of monitoring, further, according to the loading condition of CPU, the service condition of physical machine resource is judged, determine which platform virtual machine needs migration, when move.

6. virtual energy conserving system in cloud computing according to claim 1, it is characterized in that, described physical machine Registering modules primary responsibility adds management transfer system to physical machine, physical machine is by sending request to overhead control end, overhead control end determines whether allow this physical machine to be added in the middle of system, whether overhead control end constantly detects by the mode of heartbeat the physical machine having joined system normal, if abnormal, from database information, make corresponding amendment, and send corresponding warning message to keeper, the establishment of virtual machine can change the existing available resources of physical machine, update module is responsible for the information upgrading physical machine,

Physical machine administration module adopts C/S framework, when a physical machine wants the physical resource pond added in cloud computing environment, client in physical machine is first responsible for collection and the inspection of physical machine information, then these information are sent to long-range cloud computing control end server, control end network in charge adds to physical machine in physical resource pond, finally adds corresponding information in a database;

Physical machine in cloud computing environment is divided into some groups in the mode of physics unit, some physical machine are had in each physics unit, each physical machine there are again some virtual machines, the essential information of physical machine and current state, comprise physical machine name, IP address, maximum memory, state, internal memory, free memory, the information such as available CPU number.

7. virtual energy conserving system in cloud computing according to claim 1, it is characterized in that, described Virtual Machine Manager module is after local management module determines which virtual machine needs migration, the migration task of virtual machine has been responsible for by Virtual Machine Manager module, after migration completes, the power consumption of adjustment source physical machine and target physical machine, after entering adjustment, some idle physical machine just can close to reach energy-conservation target, some physical machine is in dormant state, all these physical machine of not closing be because judge according to demand, when asking more, the physical machine being in dormant state can be transformed into running status fast to meet the request of user, the high concurrent request in rush hour can be tackled like this,

The administration module of virtual machine comprises unlatching, the operation such as closes, moves, suspends, destroys and restart.

8. virtual energy conserving system in cloud computing according to claim 1, is characterized in that, described monitoring module adopts Nagios to come physical machine in supervisory system and virtual machine.The information spinner of monitoring will comprise the load of CPU, the information such as internal memory and network, and the information of the physical machine that monitoring obtains and virtual machine is stored in MySQL database, and dispatching algorithm can do the adjustment of some virtual machines by these monitor messages; The monitor message obtained by Nagios has all been deposited in MySQL database, the scheduling strategy of such virtual machine just can obtain the information of the current and history of physical machine and virtual machine by the information in database, carry out the virtual machine in the whole cloud platform of dynamic conditioning by these information.

9. virtual energy conserving system in cloud computing according to claim 1, it is characterized in that, virtual machine is placed and adjusting module is mainly used to the placement and the adjustable strategies that add concrete virtual machine, system provides the interface of this module, to place and adjustable strategies only needs to realize according to the interface of system so new virtual machine will be added, under the file that the system that uploads to after having realized is specified, once have new strategy to upload to system will set out corresponding method to load this strategy, it is here main that adopt is the User Defined ClassLoader of Java, the Class file beyond running environment dynamically can be added by this ClassLoader, so just can realize the extensibility of system, virtual machine scheduling policy module is the nucleus module of cloud computing resources management prototype system, other module (monitoring modules, physical machine administration module, Virtual Machine Manager module, Web module etc.) all in order to virtual machine scheduling policy module service, the data that virtual machine scheduling policy module needs need to obtain from monitoring module, the action needs such as the establishment migration of virtual machine have been come by Virtual Machine Manager module.