CN101488098A - Multi-core computing resource management system based on virtual computing technology - Google Patents

Abstract

The invention relates to a multi-core system computer resource management system based on a virtual computer technology. The system comprises a plurality of virtual machines, a virtual machine monitor and a virtual machine manager. The virtual machine monitor monitors the load condition and the operation state of the virtual machine at real time. The virtual machine manager is a bond for communication between a virtual machine and a physical host. The virtual machine operates on the virtual machine manager and provides the user with a virtual platform. At the same time, the invention divides the virtual machine into three general categories, and different resource adjusting strategies are adopted for each category of the virtual machine respectively. The invention provides a practical and feasible way for the dynamic adjusting and distributing problem of multi-core computer resource and realizes the maximization of energy saving and resource utilization.

Description

Multinuclear managing computing resources system based on virtual computing technique

Technical field

The invention belongs to computer virtualized technical field, be specifically related to a kind of multinuclear managing computing resources system based on virtual computing technique, this system adopts virtual computing technique that the multinuclear computational resource is effectively managed.

Background technology

The appearance of multi-core technology be the Computer Systems Organization field, and earth-shaking variation has been brought in the parallel computation field.On the one hand, computing machine is just pursued more " soon " no longer as before, and dominant frequency constantly raises, but crosswise development has had the more computational resource of " many "; On the other hand, application software must improve the degree of parallelism of self, thereby reaches the target that improves self speed, rather than can only wait CPU as before and become faster.

Has the Computer Systems Organization research field that is changed to that these are a series of proposed new challenge: computational resource how to manage more " many "? in traditional computation model (comprising that emerging cluster calculates), because the computational resource of single node is less relatively, the general operating system that adopts is managed, computational resource in the method by time-sharing multiplex between multitask, shared system.Yet, the multinuclear and following crowd assess the development of calculation technology, brought the computational resource of more and more and become increasingly complex (as heterogeneous polynuclear), under this background, adopt the single operation system that system is managed, some collision problems between the application will be brought, and the waste and the unreasonable distribution of computational resource may be caused.

The recovery of virtual computing technique is the solution of this problem, and a key is provided.This technology can be moved a plurality of virtual machines (each virtual machine moves a kind of operating system) on a main frame, hold more applications, thereby improves the resource utilization of system.Because virtual computing technique possesses These characteristics, this technology is adopted by a lot of enterprise data centers at present, and be proved to be able to greatly improve the utilization factor of server resource in the data center, the quantity of minimizing server, thereby reach the purpose of reducining the construction costs with energy savings.These characteristics of virtual calculating can provide support for managing multiple nucleus system better technically.Yet as recovery technology soon, virtual computing technique is in the work of being done aspect the multinuclear resource management and few, and is also ripe far away.

The present invention is based on the Xen virtual machine manager, propose a kind of multinuclear managing computing resources system, can dynamically be optimized configuration, reach the purpose of energy savings and utilization of resources according to the demand of virtual machine to cpu resource based on virtual computing technique.Wherein, Xen is a outstanding virtual machine manager of increasing income, and supervision that existing Xen system provides and management tool such as Xenmon, Virt-Manager etc. can't carry out perception to the demand of each virtual machine inherence, more can't realize dynamic allocation of resources.

Summary of the invention

The object of the present invention is to provide a kind of multinuclear managing computing resources system based on virtual computing technique, this system can obtain the resource requirement of virtual machine, has realized the dynamic adjustment and the assignment problem of multinuclear computational resource.

Multiple nucleus system managing computing resources system based on virtual computing technique provided by the invention, it is characterized in that: this system comprises monitor of virtual machine, virtual machine manager, physical host and several virtual machines from top to bottom, is respectively arranged with Network Performance Monitor in the client operating system of each virtual machine;

Virtual machine provides virtual platform for the user, and client operating system is for supporting hot plug function operations system, and Network Performance Monitor is used for monitoring the load variations situation of application program, the loading condition of analyzing virtual machine, and analysis result passed to virtual machine manager;

Monitor of virtual machine is monitored the loading condition and the running status of virtual machine in real time, the result is in time shown feed back to the user; The user can pass through the computational resource of monitor of virtual machine manual adjustment virtual machine, and with adjusted mapping relations notice virtual machine manager;

Virtual machine manager is used for virtual machine is managed, and the accessing communication between virtual machine and the physical host is provided.

Above-mentioned Network Performance Monitor and virtual machine manager can adopt following optimal way to realize:

For the crucial virtual machine of using of operation, when knowing the cpu resource amount that needs in advance, then Network Performance Monitor directly sends its resources requirement to virtual machine manager, otherwise, Network Performance Monitor adopts multiplication to increase the linear strategy that reduces and obtains resources requirement, and it is sent virtual machine manager; Virtual machine manager gives operation the crucial virtual machine of using resources allocation according to above-mentioned resources requirement;

For the virtual machine that operation service is used, virtual machine manager is according to service specified rank agreement, and the virtual machine of distributing to the operation service application can satisfy the minimum stock number of service request;

The idling-resource that virtual machine manager will satisfy outside above-mentioned two kinds of application is distributed to the virtual machine that moves background application.

The present invention is intended to adopt virtual computing technique, by to the judgement of upper layer application behavior with to the dynamic adjustment of bottom physical computing resource, solves the dynamic adjustment and the assignment problem of multinuclear computational resource.The present invention dynamically adjusts the corresponding relation of virtual processor in this virtual machine (VCPU) and real physical cpu center by to using the judgement of behavior in the virtual machine.By adjustment, improved the performance that virtual machine is used, the configuration of optimizing computer entire system, and the target of reduction system energy consumption to the computational resource that virtual machine had.The present invention has following technique effect:

1. the present invention has realized the optimum management of computational resource, and self-defined configuration interface is provided.By real-time performance monitoring to client operating system, analyze the resource requirement of virtual machine, dynamically physical resource is carried out reasonable distribution.

2. the present invention is divided into 3 big classes according to the demand degree of using cpu resource with the virtual machine application: crucial application, service are used, background application, and different cpu resource adjustment strategies is adopted in different virtual machine application.Further strengthen resource management capacity, the granularity that dynamic resource is adjusted is thinner, and resource adjusting strategies has more specific aim.Improve the cpu busy percentage and the service quality of entire system, more can improve the loss of the usage factor of system resource and the minimizing energy substantially.

3. the invention provides the efficient real-time virtual machine performance monitoring of keeper, the virtual machine Network Performance Monitor that provides can be monitored the running status of a plurality of virtual machines in real time, specifically can also monitor utilization of resources situation, VCPU and the CPU mapping relations etc. of virtual machine VCPU.

4. last, the present invention is a platform with the unit multinuclear, and the support to the ACPI power management is provided.Can dynamically close or open the physical cpu resource according to the resource consumption state of virtual machine platform.

Description of drawings

Fig. 1 is the architecture that the present invention is based on the multiple nucleus system managing computing resources system of virtual computing technique;

Fig. 2 is computational resource mapping relations figure under the default situation;

Fig. 3 enters computational resource mapping relations figure after the dormancy for virtual machine among Fig. 2;

Fig. 4 is the computational resource mapping relations figure of virtual machine after busy among Fig. 3;

Fig. 5 operates in computational resource mapping relations figure on the same virtual platform for the multiclass virtual machine.

Embodiment

Below by will describing the present invention more in detail by following examples, and following examples only are illustrative, and the present invention is not subjected to the restriction of these embodiment.

Hardware of the present invention has Intel VT or AMD Pacifica technology, adopt the XEN3.2 version to build virtual platform, all kinds of virtual machines operate in this virtual platform, by different classes of virtual machine is adopted different resource adjusting strategies, realize utilizing substantially resource.

As shown in Figure 1, from the level of architecture, the applied computer system of the present invention comprises virtual machine 1.1 from top to bottom, 1.2 ..., 1.n, monitor of virtual machine 2, virtual machine manager 3 and physical host 4, the present invention can hold a plurality of virtual machines, and n is a positive integer.Each virtual machine 1.1,1.2 ..., the client operating system 6.1,6.2 of 1.n ..., be respectively arranged with Network Performance Monitor 7.1,7.2 among the 6.n ..., 7.n.For explaining conveniently, hereinafter with virtual machine 1.1,1.2 ..., 1.n, client operating system 6.1,6.2 ..., 6.n and Network Performance Monitor 7.1,7.2 ..., 7.n is referred to as virtual machine 1, client operating system 6 and Network Performance Monitor 7 respectively.

Virtual machine 1 includes Network Performance Monitor 7 for the user provides virtual platform in its client operating system 6.Client operating system 6 can increase or reduce its CPU number for supporting hot plug function operations system.

Network Performance Monitor 7 operates in the client operating system 6, is used for monitoring the load variations situation of application program, as CPU usage, memory usage etc.Judging whether need be more () computational resources perhaps still less, CPU user mode in the analyzing virtual machine 1, whether whether CPU at full capacity or idle need to increase or reduce the cpu resource of distributing to virtual machine 1 thereby draw, and go out the cpu resource demand by corresponding algorithm predicts.This demand is under the situation that guarantees the operation of virtual machine 1 normal service, minimum cpu resource consumption, and judged result passed to the virtual cpu in the virtual machine manager 3 and the mapping block 32 of physical cpu.

As above-mentioned, for more refinement ground realization resources allocation better, the present invention uses virtual machine and is divided into 3 big classes: crucial application, service application, background application.

For crucial application virtual machine,, so just directly give mapping block 32 its resources requirements of transmission of virtual cpu and physical cpu if can know in advance and need how many cpu resources; If how many cpu resources can not know in advance needs, adopt multiplication to increase the linear strategy that reduces, doubly increase cpu resource, reduce cpu resource successively.Concrete prediction algorithm is as described below:

Set the reasonable CPU usage scope N1～N2 of the crucial virtual machine of using of operation, the threshold value of the crucial resource request of using is w.By the cpu busy percentage of Network Performance Monitor 7 real-time monitoring virtual machines, if the resources occupation rate of such virtual machine is very high and a plurality of " hungering and thirst " process is arranged at the same CPU of competition, this just needs generation to increase the signal of cpu resource.With N1=50%, N2=70%, w=5 are example, and the generation of the increase of cpu resource and minimizing signal is to produce according to following process:

1, calculate interior cpu busy percentage and degree of parallelism of pre-designed sampling time, the calculating of degree of parallelism is to draw according to the process number that is in run mode of current competition computational resource and the number of CPU;

If 2 resource utilizations occur w time greater than 1 continuously greater than N2 and degree of parallelism, then change step 5;

If 3 resource utilizations occur w time less than 1 continuously less than N1 and degree of parallelism, change step 6;

If 4 resource utilizations between N1 and N2, are changeed step 1;

5, producing resource requirement increases signal, and sends to virtual machine manager 3, changes step 1;

6, produce resource requirement and reduce signal, and send to virtual machine manager 3, change step 1.

Consume certain resource owing to dynamically adjust the resources allocation meeting,, will influence the service performance of system if the resource adjustment is too frequent.Therefore to have set a threshold value for the crucial resource request of using be 5 to this example, when the resource request number of times equals 5, then produces resource and adjust signal.

The service application virtual machine mainly provides service to use to the user, comprising: network application, database application or the like.For the virtual machine of this class, need guarantee service quality, the response time is one of leading indicator of estimating its service quality.At the resource adjustment of service application virtual machine, its main thought is under service specified rank agreement (SLA), distributes to the minimum resource of virtual machine and satisfies service request.

Web service is an I/O intensive task, is the cpu resource demand that example is studied service virtual machine with the Web server based on virtual machine among the present invention.At first introduce the parameter of using always in calculating below:

T: the time of expression whole test process.

K: the number of resource in the expression system.

B _i: in whole observation process T, the busy time of resource i.

A _i: in the process of observation T, the request quantity of resource i.

C _i: observe among the process T quantity of finishing of resource i.C represents the request number that total system is finished.

By above several known quantities, can obtain several more useful parameters:

λ _i: arrival rate λ _iThe quantity of work request in the representation unit time.

${\mathrm{\λ}}_i=\frac{A_i}{T}---\left(1.1\right)$

U _i: the utilization factor of resource i;

$U_i=\frac{B_i}{T}---\left(1.2\right)$

X _i: the throughput of resource i, the request quantity that the representation unit time finishes;

$X_i=\frac{C_i}{T}---\left(1.3\right)$

D _iThe time of a kind of request consumption of natural resource i of given type is handled in expression.By defining,

$D_i=\frac{B_i}{C}=\frac{U_i\×T}{C}=\frac{U_i}{C\÷T}=\frac{U_i}{X}---\left(1.4\right)$

In the system of turnover balance,

X＝λ (1.5)

Therefore,

$D_i=\frac{U_i}{\mathrm{\λ}}---\left(1.6\right)$

Because each resource utilization can not surpass 100%, get by following formula,

U _i＝λ×D _i≤1 (1.7)

Therefore,

$\mathrm{\λ}\≤\frac{1}{D_i}---\left(1.8\right)$

So have

$\mathrm{\λ}\≤\frac{1}{{\max }_{i=1}^K{D}_i}---\left(1.9\right)$

Response time is an important performance measurement index.Below two formula provided the calculated response time method.

$R=\underset{i=1}{\overset{K}{\mathrm{\Σ}}}{R}_i---\left(1.10\right)$

$R_i=\frac{D_i}{1-U_i}---\left(1.11\right)$

Deceleration parameter S of this example definition _v, when being illustrated in same operation of virtual machine and local operation system execution, the ratio of busy time of cpu resource:

$S_v=\frac{B_{\mathrm{CPU}}^{\mathrm{HVM}}}{B_{\mathrm{CPU}}^{\mathrm{Linux}}}---\left(1.12\right)$

At virtual machine and local system test duration Duan Douwei T, get by formula (1.2) and formula (1.12),

$S_v=\frac{B_{\mathrm{CPU}}^{\mathrm{HVM}}}{B_{\mathrm{CPU}}^{\mathrm{Linux}}}=\frac{U_{\mathrm{CPU}}^{\mathrm{HVM}}\×T}{U_{\mathrm{CPU}}^{\mathrm{Linux}}\×T}=\frac{U_{\mathrm{CPU}}^{\mathrm{HVM}}}{U_{\mathrm{CPU}}^{\mathrm{Linux}}}---\left(1.13\right)$

If in virtual machine and local operation system, test with identical arrival rate, can push away by formula (1.13) and formula (1.6),

$S_v=\frac{U_{\mathrm{CPU}}^{\mathrm{HVM}}}{U_{\mathrm{CPU}}^{\mathrm{Linux}}}=\frac{D_{\mathrm{CPU}}^{\mathrm{HVM}}\×\mathrm{\λ}}{D_{\mathrm{CPU}}^{\mathrm{Linux}}\×\mathrm{\λ}}=\frac{D_{\mathrm{CPU}}^{\mathrm{HVM}}}{D_{\mathrm{CPU}}^{\mathrm{Linux}}}---\left(1.14\right)$

In virtual machine, handle affairs identical with the local operation system, not only virtual machine needs consumption of natural resource, Dom0 also wants consumption of natural resource simultaneously, and wherein Dom0 represents the virtual machine of a privilege in Xen, and the I/O operation of all virtual machines all needs to handle by Dom0.In order to calculate the performance cost of Dom0, define a factor

${\mathrm{Cost}}_{\mathrm{HVM}}^{\mathrm{Dom}0}=\frac{B_{\mathrm{CPU}}^{\mathrm{Dom}0}}{B_{\mathrm{CPU}}^{\mathrm{HVM}}}---\left(1.15\right)$

Dom0 has identical test duration section T with virtual machine in test, then can push away by formula (1.2) and formula (1.15),

${\mathrm{Cost}}_{\mathrm{HVM}}^{\mathrm{Dom}0}=\frac{B_{\mathrm{CPU}}^{\mathrm{Dom}0}}{B_{\mathrm{CPU}}^{\mathrm{HVM}}}=\frac{U_{\mathrm{CPU}}^{\mathrm{Dom}0}\×T}{U_{\mathrm{CPU}}^{\mathrm{HVM}}\×T}=\frac{U_{\mathrm{CPU}}^{\mathrm{Dom}0}}{U_{\mathrm{CPU}}^{\mathrm{HVM}}}---\left(1.16\right)$

Can push away by formula (1.16) and formula (1.6),

${\mathrm{Cost}}_{\mathrm{HVM}}^{\mathrm{Dom}0}=\frac{U_{\mathrm{CPU}}^{\mathrm{Dom}0}}{U_{\mathrm{CPU}}^{\mathrm{HVM}}}=\frac{D_{\mathrm{CPU}}^{\mathrm{Dom}0}\×\mathrm{\λ}}{D_{\mathrm{CPU}}^{\mathrm{HVM}}\×\mathrm{\λ}}=\frac{D_{\mathrm{CPU}}^{\mathrm{Dom}0}}{D_{\mathrm{CPU}}^{\mathrm{HVM}}}---\left(1.17\right)$

For the maximum throughput rate of calculating based on the service application virtual machine needs following parameter:

With the local (SuSE) Linux OS of virtual machine same hardware configuration in, handle the CPU service needs of each HTTP request.This parameter is by testing and can calculate at local linux system.

S _v: linux system is to the deceleration parameter of hardware virtual machine.Calculate this parameter and not only need test in linux system, also need carry out identical test at the identical hardware virtual machine of configuration, through type (1.13) calculates this parameter.

According to the test data under the hardware virtual machine, utilize formula (1.16) can calculate this parameter.

Above Several Parameters has been arranged, and use formula (1.14) and formula (1.17) can draw

With

Virtual machine maximum cpu resource that can use under the intensive load of I/O is provided by parameters C eil, therefore can obtain:

${\mathrm{\λ}}_{\max }\≈\min (\frac{1}{D_{\mathrm{cpu}}^{\mathrm{Dom}0}},\frac{\mathrm{Ceil}}{D_{\mathrm{cpu}}^{\mathrm{vrn}}})---\left(2.18\right)$

Limit the maximum throughput rate that can change Web server by cpu resource to virtual machine.When distribute cpu resource quantity for virtual machine is that U (during U＜Ceil), then has

${\mathrm{\λ}}_{\max }^U\≈\min (\frac{1}{D_{\mathrm{CPU}}^{\mathrm{Dom}0}},\frac{U}{D_{\mathrm{CPU}}^{\mathrm{HVM}}})---\left(2.19\right)$

Arrival rate λ in the virtual machine server services request of appointment _NeedCan calculate the required cpu resource quantity of hardware virtual machine by formula (2.2) down:

$U_{\mathrm{need}}=D_{\mathrm{CPU}}^{\mathrm{HVM}}\×{\mathrm{\λ}}_{\mathrm{need}}---\left(2.20\right)$

Formula (2.18) has provided the computing method of maximum throughput under the assignment of allocation cpu resource, and formula (2.20) provides the method for specifying arrival rate to calculate needed cpu resource quantity.For example the arrival rate of given server needs is λ _Need, can calculate needed cpu resource quantity by (2.20).But cpu busy percentage is not invariable when the server actual motion, and it fluctuates at average utilization.Therefore be in course of adjustment, the resource of distributing to hardware virtual machine need be slightly larger than the needed cpu resource quantity of hardware virtual machine.If only the cpu resource number that goes out for the hardware virtual machine Distribution Calculation will appear at when moving, sometimes the situation of cpu resource deficiency.This can cause the unstable properties of server, thereby normal service can not be provided.Simultaneously by formula (1.11), as can be seen cpu busy percentage near in limited time, the response time can sharply rise, and especially surpasses after 80%, the response time can be doubled and redoubled.Therefore when Resources allocation, be the resource of hardware virtual machine overabsorption 20%, so both can have guaranteed the stable of server, can also guarantee that the response time is in desirable scope simultaneously.Computing method are as follows:

$U_{\mathrm{allocate}}={\mathrm{\λ}}_{\mathrm{need}}{D}_{\mathrm{cpu}}^{\mathrm{vm}}\×120\%---\left(2.21\right)$

Wherein

For with the virtual machine of virtual machine same hardware configuration in, CPU handles each HTTP request needed time of service.

For the background application virtual machine, the present invention does not carry out the resource requirement forecast analysis for it, because what move on the background application virtual machine mainly is batch processing job, background task etc., virtual machine for this class, when the task of its operation and operation are finished not is very important, therefore the present invention, just distributes to the background application virtual machine and uses as long as idle resource is arranged not for it carries out the cpu resource demand forecast.

Monitor of virtual machine 2 and virtual machine 1 the same operating on the virtual machine manager 3.Monitor of virtual machine 2 is not only monitored the performance state of virtual machine in real time, and graphical demonstration and user control interface are provided.Monitor of virtual machine 2 is the loading condition and the running status of monitoring virtual machine 1 in real time, the result is graphically shown in time feed back to the user.The user can pass through the computational resource of monitor of virtual machine 2 manual adjustment virtual machines 1, and adjusted mapping relations is notified the mapping block 32 of virtual cpu and physical cpu.

Under virtual machine 1 a lot of situations is to communicate by virtual machine manager 3 and physical host 4, decides its visit to all resources on the physical host 4 by virtual machine manager 3.Virtual machine manager 3 comprises the mapping block 32 and the scheduler 33 of analog power administration module 31, virtual cpu and physical cpu.

Whether analog power administration module 31 perception virtual machine automatically enters dormancy, and these information is in time sent to the mapping block 32 of virtual cpu and physical cpu.If analog power administration module 31 is learnt virtual machine 1 and is entered the information of dormancy, then this dormancy information is sent to the mapping block 32 of virtual cpu and physical cpu, by the latter computational resource of virtual machine 1 is adjusted, reached the purpose of energy savings and the loss of minimizing computing machine.The mapping block 32 of virtual cpu and physical cpu is safeguarded mapping table adjustment to it, the corresponding corresponding relation of the CPU that deletion need be closed, VCPU according to the dormancy information that receives.

The mapping block 32 of virtual cpu and physical cpu is safeguarded the mapping table of virtual cpu and physical cpu, the mapping block 32 of virtual cpu and physical cpu can change the mapping relations of virtual computing unit and physical computing unit automatically by the information of collecting, as the above, the present invention uses virtual machine and is divided into 3 classes according to the cpu demand degree, different resource adjusting strategies is adopted in different classes of application, and adjusted mapping relations is notified to the scheduler 33 and monitor of virtual machine 2 of actual execution mapping.

Crucial use generally responsively, weigh its performance by the deadline, therefore need a large amount of computational resources time ratio.The resource adjusting strategies that this class virtual machine is used is fairly simple, in order to realize the approaching of execution time on performance and the real machine, the present invention uses for such virtual machine its desired cpu resource is provided as much as possible, satisfies the crucial cpu resource demand of using to greatest extent.Resource conversion and priority policy are adopted in such virtual application, distribute cpu resource promptly preferentially for such virtual machine, and carry out virtual machine cpu resource and physical cpu resource corresponding one by one.If remaining cpu resource can not satisfy the requirement of such application, to carry out conversion and deprive the formula strategy, the cpu resource of other virtual machines has been distributed in conversion, thereby guarantees the cpu resource that such virtual machine application is asked to greatest extent.

Because the main resource requirement I/O resource of service application virtual machine needs to consume a large amount of computational resources unlike key is used, the present invention has adopted the adjustment strategy of resource sharing to the service application virtual machine.In order to utilize cpu resource better, the present invention adopts fine-grained computational resource to adjust mode to service virtual machine.The present invention is divided into ten parts with the single cpu resource, realizes by dividing CPU time.By in a period of time, restriction service application virtual machine consumes CPU and realizes service time, has guaranteed that the service application virtual machine under the prerequisite that guarantees cpu resource, limits its consumption to cpu resource.For example draw, specifying arrival rate λ by the resources algorithm _NeedDown, the required cpu resource quantity of virtual machine is 2.4.Fraction part is wherein represented the share that it can use the cpu resource of distribution, and promptly at a period of time t in second, this virtual machine consumes the 2.4t CPU time of second at most.Can better more preferably realize utilization of resources maximization by fine-grained resource adjustment.

The present invention has adopted resources shared to adjust strategy to the background application virtual machine, the priority of its Resources allocation is minimum, normally provide on the basis of service at the preferential front two class virtual machines that guarantee,, just distribute to such virtual machine and use as long as in the system unnecessary cpu resource is arranged.

Scheduler 33 is realized the computational resource and the physical computing resource mapping of virtual machine 1.Scheduler 33 reads the mapping table of the mapping block 32 of virtual cpu and physical cpu, carries out the adjustment of computational resource according to mapping table.Scheduler will be according to mapping table, and virtual cpu is dispatched to the operation queue of corresponding physical cpu, waits for the scheduled for executing of physical cpu.

As shown in Figure 2, started two virtual machines on virtual manager 3, only for illustrating, native system can start a plurality of virtual machines simultaneously herein, this figure is when adopting virtual computing technique, the default corresponding relation of Virtual Processing Unit and physical processing unit in two virtual machines.Wherein, physical host has 4 processing units, and virtual machine 1.1 has two Virtual Processing Units, all be activated, and corresponding separately physical processing unit; Virtual machine 1.2 has 4 Virtual Processing Units, but because own load heavy (can be obtained by the Network Performance Monitor of virtual machine 1.2), so only activated wherein 2 and corresponding respectively physical processing unit.

In order to realize the computational resource mapping under this default situation, the present invention arrives the mapping relations of physical computing unit (mapping block 32 by module virtual cpu and physical cpu is realized) by realize the virtual computing unit of Fig. 2 in virtual machine manager 3, and the part Virtual Processing Unit (being realized by the client operating system 2 of supporting the hot plug function) of closing in the virtual machine 1.2 reaches the purpose that cpu resource is distributed rationally.

As shown in Figure 3, after system operation a period of time, virtual machine 1.1 is not owing to there is task for a long time, and entered dormant state, for the consideration of saving energy consumption, needs to discharge two physical processing units that take before it.At this moment, because not variation of the load of virtual machine 1.2, and the physical computing resource that virtual machine 1.1 is discharged is idle, thereby closed (or reducing dominant frequency) by virtual machine manager 3, thus save electric power resource.

For realizing the mapping relations among Fig. 3, the present invention need realize following function:

1. the state of real time monitoring virtual machine obtains the information that virtual machine 1.1 enters dormancy by analog power administration module 31;

2. analog power administration module 31 passes to the mapping block 32 of virtual cpu and physical cpu with the information of virtual machine 1.1 dormancy, is dynamically adjusted the corresponding relation of virtual cpu and physical cpu by the latter;

3. the adjustment result of the mapping block 32 of scheduler 33 read module virtual cpus and physical cpu closes corresponding processing unit in the physical host by it.

As shown in Figure 4.On the basis of Fig. 3, if the load in the virtual machine 1.2 increases, need more computational resource, the present invention activates the Virtual Processing Unit of the dormancy in this execution environment, and the virtual cpu in the notice virtual machine manager 3 and the mapping block 32 of physical cpu, allow the latter that the CPU or the nuclear of closing on the physical host are opened, and the virtual computing unit orientation that will activate (or scheduling) is to CPU that newly opens or nuclear.

In order to realize this function, realization flow of the present invention is as follows:

1. by the loading condition in the module monitor of virtual machine 2 real-time monitoring virtual machines 1.2, need more computational resource to determine this environment;

2. the mapping block 32 dynamic mapping relations that realize of virtual cpu and physical cpu are adjusted automatically;

3. the Network Performance Monitor of virtual machine 1.2 activates the virtual computing unit in this environment.

4. open physical computing unit on the main frame by scheduler, and virtual computing unit is mapped on the physical computing unit of newly opening.

For more refinement ground realization resources allocation better, the present invention uses virtual machine and is divided into 3 big classes: crucial application, service application, background application.Use for different virtual machines, adopt different resource adjusting strategies, realize the flexible adjustment of resource.

As shown in Figure 5, have 4 CPU on the physical host, moved 3 virtual machines on the virtual machine manager 3, on virtual machine 1.1, moving crucial the application, operation service is used on the virtual machine 1.2, has moved background application on the virtual machine 1.3, and the multiclass virtual machine operates on the same virtual platform jointly.The crucial application of regulation of the present invention has the highest priority of resource allocation, and its CPU allocation strategy is corresponding conversion formula one by one.As shown in Figure 5, in order to guarantee the performance of virtual machine 1.1, need to distribute the computational resource of 2 CPU.According to resource adjusting strategies, distribute two virtual cpus and corresponding one by one with physical cpu respectively for virtual machine 1.1, just illustrate herein, if along with the task of virtual machine 1.1 need to increase the weight of more cpu resource, the resource that can seize virtual machine 1.2 and virtual machine 1.3 by force guarantees its performance, if task alleviates, need to discharge cpu resource, the cpu resource that is discharged can be distributed to virtual machine 1.2 and 1.3 and use.

The priority that all the other two classes are used is lower, the priority of service application virtual machine will be higher than the background application virtual machine, the two all adopts the resource sharing strategy on resource allocation policy, be that both share the use cpu resource, pay the utmost attention to the resource request of virtual machine 1.2, satisfy in maximization under the request of virtual machine 1.2, just unnecessary cpu resource is distributed to virtual machine 1.3 and used.

The present invention not only is confined to above-mentioned embodiment; persons skilled in the art are according to content disclosed by the invention; can adopt other multiple embodiment to implement the present invention; therefore; every employing project organization of the present invention and thinking; do some simple designs that change or change, all fall into the scope of protection of the invention.

Claims (5)

1, a kind of multinuclear managing computing resources system based on virtual computing technique, it is characterized in that: this system comprises monitor of virtual machine (2), virtual machine manager (3), physical host (4) and several virtual machines (1) from top to bottom, is respectively arranged with Network Performance Monitor (7) in the client operating system (6) of each virtual machine (1);

Virtual machine (1) provides virtual platform for the user, client operating system (6) is for supporting hot plug function operations system, Network Performance Monitor (7) is used for monitoring the load variations situation of application program, the loading condition of analyzing virtual machine (1), and analysis result passed to virtual machine manager (3);

Monitor of virtual machine (2) is monitored the loading condition and the running status of virtual machine (1) in real time, the result is in time shown feed back to the user; The user can pass through the computational resource of monitor of virtual machine (2) manual adjustment virtual machine (1), and with adjusted mapping relations notice virtual machine manager (3);

Virtual machine manager (3) is used for virtual machine (1) is managed, and the accessing communication between virtual machine (1) and the physical host (4) is provided.

2, multinuclear managing computing resources according to claim 1 system is characterized in that:

For the crucial virtual machine of using of operation, when knowing the cpu resource amount that needs in advance, then Network Performance Monitor (7) sends its resources requirement directly for virtual machine manager (3), otherwise, Network Performance Monitor (7) adopts multiplication to increase the linear strategy that reduces and obtains resources requirement, and it is sent virtual machine manager (3); Virtual machine manager (3) gives operation the crucial virtual machine of using resources allocation according to above-mentioned resources requirement;

For the virtual machine that operation service is used, virtual machine manager (3) is according to service specified rank agreement, and the virtual machine of distributing to the operation service application can satisfy the minimum stock number of service request;

The idling-resource that virtual machine manager (3) will satisfy outside above-mentioned two kinds of application is distributed to the virtual machine that moves background application.

3, multinuclear managing computing resources according to claim 1 and 2 system, it is characterized in that: virtual machine manager (3) comprises the mapping block (32) and the scheduler (33) of analog power administration module (31), virtual cpu and physical cpu;

Analog power administration module (31) is used for automatic perception virtual machine and whether enters dormancy, and these information is in time sent to the mapping block (32) of virtual cpu and physical cpu;

The mapping block of virtual cpu and physical cpu (32) is used to safeguard the mapping table of virtual cpu and physical cpu, changes the mapping relations of virtual computing unit and physical computing unit automatically by the information of collecting; It also is notified to adjusted mapping relations the scheduler (33) and monitor of virtual machine (2) of actual execution mapping;

Scheduler (33) is used to realize the computational resource and the physical computing resource mapping of virtual machine (1), and scheduler (33) reads the mapping table of the mapping block (32) of virtual cpu and physical cpu, carries out the adjustment of computational resource according to mapping table.

4, multinuclear managing computing resources according to claim 2 system is characterized in that: the reasonable CPU usage scope of setting the crucial virtual machine of using of operation is N1～N2, and the threshold value of the crucial resource request of using is w; Multiplication increases the linear strategy that reduces and comprises the steps:

1., cpu busy percentage and degree of parallelism in the calculating sampling time, wherein, degree of parallelism calculates according to the process number that is in run mode of current competition computational resource and the number of CPU;

If 2. resource utilization occurs w time greater than 1 continuously greater than N2 and degree of parallelism, then change step 3., if resource utilization occurs w time less than 1 continuously less than N1 and degree of parallelism, change step 4., otherwise, change step over to 1.;

3., produce resource requirement and increase signal, and send to virtual machine manager (3), change step 1.;

4., produce resource requirement and reduce signal, and send to virtual machine manager (3), change step 1..

5, according to claim 2 or 4 described multinuclear managing computing resources systems, it is characterized in that: for Web server based on virtual machine, the stock number U that the virtual machine that virtual machine manager (3) is used to operation service distributes _AllocateComputing formula be:

$U_{\mathrm{allocate}}={\mathrm{\λ}}_{\mathrm{need}}{D}_{\mathrm{cpu}}^{\mathrm{vm}}\×120\%$

Wherein, λ _NeedBe the arrival rate of the virtual machine server services request of appointment,

For with the virtual machine of virtual machine same hardware configuration in, CPU handles each HTTP request needed time of service.

Images