CN106201700A - The dispatching method that a kind of virtual machine migrates online - Google Patents

Abstract

The present invention discloses the dispatching method that a kind of virtual machine migrates online, including: step 1, the cpu busy percentage of record per second once every main frame, and the cpu busy percentage of each VM in every main frame；Whether step 2, prediction each main frame of subsequent time transship；Step 3, the VM selecting coefficient of relationship multiple with the rising of host CPU utilization rate maximum on the main frame being judged to overload migrate out；Step 4, finds out low load main frame according to minimum cpu busy percentage method based on threshold value；Step 5, redistributes VM queue to be migrated according to energy sensing optimal descending method.The energy consumption of cloud computing center can be significantly reduced by the method for the present invention, and the most conventional algorithm has lower SLA violation rate, namely has more preferable service quality.

Description

The dispatching method that a kind of virtual machine migrates online

Technical field

The invention belongs to computer architecture memory system construction applications, be specifically related to the scheduling that a kind of virtual machine migrates online Method.

Background technology

Along with pattern rapidly growing of cloud computing " paid i.e. required ", cloud data center task amount and scale are also constantly Expanding, its power consumption is consequently increased.Adding up according to ICTresearch, 2012 China's consumption of data center are up to 664.5 hundred million Degree, accounts for the 1.8% of the most national industrial electricity, is up to 100,000,000,000 degree to China's consumption of data center in 2015.It addition, from American Society of Heating,Refrigerating and Air-Conditioning Engineers (ASHRAE), from the point of view of data analysis, the 75% of whole data center operation cost derives from the energy expenditure of infrastructure.Number High energy consumption according to center will necessarily reduce the profit margin of cloud service provider, increases carbon emission amount.High energy consumption will certainly become system The key factor of the development of about following cloud data center, the most how reducing cloud consumption of data center is that cloud computing system is sustainable Problem demanding prompt solution in evolution.

Under study for action, while reducing the power consumption of cloud data center, it is possible to the decline of service quality can be caused, and then disobey The service-level agreement (Service Level Assignment, SLA) that the back of the body is signed with user, this is complete in actual applications Unacceptable.Therefore research needs to start with in terms of reducing power consumption and ensureing service quality two.

In the data center that IaaS service is provided, it is usually and provides a user with infrastructure money with the form of virtual machine Source, user can complete the request of the access to virtual machine and resource by accessing web interface, and these virtual machines are to be deployed in number According on the server at center.The time-to-live of virtual machine is limited, can unload phase when user is no longer necessary to these resources The virtual machine answered, the deploying virtual machine carried out in data center for a long time along with user, data center can open a lot of server, Unloaded along with virtual machine accordingly, it is the lowest that data center's hardware resource utilization can drop, and by utilizing virtual machine online Migrating technology then can be greatly improved resource utilization and reduce IT cost.And after virtual machine is excessively gathered, data center Response time and availability of service guarantee etc. be likely to be guaranteed, thus violate and service etc. that user signs Level agreement (SLA), the therefore following service quality ensureing data center, or after virtual machine is frequently unloaded, resource utilization Can reduce, at this moment need to detect the resource service condition of data center, carry out the migration of virtual machine if desired.

Generally virtual machine migrates online and is divided into following four steps:

1) overload detection.Detecting whether that main frame transships, if having, then certain virtual machine (vm) migration on this main frame having been gone out.

2) underloading detection.Detect whether main frame underloading, if having, then virtual machine (vm) migrations whole on this main frame have been gone out and incite somebody to action This mian engine changeover to holding state to reduce energy consumption

3) virtual machine is moved out selection.Which move out the virtual machine on this main frame after judging main frame overload.

4) virtual machine reassigns.The virtual machine being needed to be moved out is assigned to other main frames.

The online migration strategy of complete virtual machine all can be optimized at above four aspects.

At present, more existing researchs are devoted to optimize online migration strategy, reduce cloud power consumption of data center.Beloglazov Et al. minimum cpu busy percentage strategy is proposed, when overload, need to go out one or more virtual machine (vm) migrations, it is therefore desirable to On this main frame run numerous virtual in make a choice.Minimum cpu busy percentage strategy (Minimum CPU Utilization, MCU) it is to be applied to this selection course.The core concept of this strategy is to be gone out by the virtual machine (vm) migration that cpu busy percentage is minimum Go to alleviate overload.Research in terms of having a lot of scholar to apply this theory to do virtual machine migration online in recent years.But only lead to Cross cpu busy percentage reckling a little make decision or lose biased, poor effect for the effect of test, it is likely that due to Only migrate out the minimum virtual machine of CPU usage and come back to overload in short time later.Abawajy etc. It is first to select minimum virtual of migration time in all of virtual machine that people proposes MMT (Min Migration Time) strategy Machine, having something of this strategy has taken into full account transport efficiency, and the impact on performance migration brought is preferably minimized.But shortcoming It is also obvious that because the virtual machine (vm) migration raising relation maximum with host resource occupancy not being gone out, therefore it cannot be guaranteed that According to solving the overload problem of this main frame after this policy migration, after migrating, likely main frame is also maintained at transshipping shape State or be on the verge of overload.Anton et al. proposes energy sensing optimal and adapts to descending algorithm (Power Aware Best Fit Decreasing, PABFD), for the popularization in virtual machine placement issue of the BFD algorithm.BFD is a kind of calculation solving bin packing Method, when solving bin packing, its main thought is: " article " i.e. virtual machine first carries out descending, checks all afterwards Non-NULL " chest " i.e. main frame, finds most suitable " chest " being somebody's turn to do " object " and loads in " chest ", without looking for by this object Then " empty van " is opened to such " chest ".This algorithm exclusively for cloud data center low-power consumption Study on Problems out, with this The object of study of problem is consistent, and the research to this problem serves important directive significance.

Summary of the invention

The technical problem to be solved in the present invention is to provide the dispatching method that a kind of virtual machine migrates online, exists at virtual machine Overload detection, underloading that line migrates detect, virtual machine moves out selection and virtual machine reassigns to add in four steps and improves calculation Method, owing to virtual machine migration online itself can cause certain energy consumption, and can cause the reduction of service quality, therefore maximize Migration to be had certain limitations by the energy consumption reduced of being gone out by the virtual machine (vm) migration in underloading main frame simultaneously, it is to avoid " excessively migrating " band The negative effect come, to reach to ensure service quality while reducing energy consumption.By main frame overload detection plan based on prediction Slightly, mobilism and the detection underloading algorithm of restricted migration, the virtual machine that relatively optimizes are moved out selection strategy and reassignment algorithm Optimize the process that virtual machine migrates online.

The dispatching method that a kind of virtual machine migrates online comprises the steps:

Step 1, the cpu busy percentage of record per second once every main frame, and in every main frame, the CPU of each VM utilizes Rate；

Whether step 2, prediction each main frame of subsequent time transship；

Step 3, select to rise, with host CPU utilization rate, the VM that multiple coefficient of relationship is maximum on the main frame being judged to overload Migrate out；

Step 4, finds out low load main frame according to minimum cpu busy percentage method based on threshold value；

Step 5, redistributes VM queue to be migrated according to energy sensing optimal descending method.

As preferably, step 2 comprises the steps:

Step 2.1, calculates the weighted regression curve of current time according to main frame history cpu busy percentage；

Step 2.2, calculates the cpu busy percentage predictive value of subsequent time main frame according to weighted regression curve；

According to 2.2 predictive values calculated, step 2.3, judges whether subsequent time main frame can transship, concrete determination methods As follows:

(1) if predictive value is more than or equal to 0.9, it is determined that main frame subsequent time transships；

(2) if predictive value is less than 0.9, it is determined that main frame subsequent time does not transships.

As preferably, step 3 comprises the steps:

Step 3.1, separately constitutes matrix calculus multiple correlation by the cpu busy percentage of each VM on the main frame of predicted overload Coefficient；

Step 3.2, according to the multiple correlation coefficient of each VM drawn in 3.1, chooses the VM addition that wherein numerical value is maximum and treats Migrate VM queue.

As preferably, step 4 comprises the steps:

Step 4.1, finds out that main frame that cpu busy percentage in cloud computing center is minimum；

Step 4.2, contrasts the cpu busy percentage of this main frame with entering the optimal threshold 0.45 that great many of experiments draws, And judge whether this main frame is in underloading, concrete determination methods is as follows:

(1) if host CPU utilization rate is more than or equal to 0.45, then judge that it is not in light condition；

(2) if host CPU utilization rate is greatly less than 0.45, then judge that it is in light condition.

Step 4.3, adds queue to be migrated by the main frame being judged to underloading.

Compared with prior art, the invention have the advantages that

While maximization virtual machine migrates the energy consumption reduction brought online, it is possible to avoid in underloading decision process as far as possible In because not being any limitation as forming raw " excessively migrating " phenomenon, be therefore compared to the achievement in research of forefathers, the present invention migrates Strategy aspect can have further decline in energy consumption, and is far superior to existing migration strategy in terms of service quality. The energy consumption of cloud computing center can be significantly reduced by the inventive method, and the most conventional algorithm has lower SLA violation rate, Namely there is more preferable service quality.

Accompanying drawing explanation

Fig. 1 is the flow chart of the online moving method of whole virtual machine.

Fig. 2 is the inventive method power consumption and studies optimal algorithm contrast and experiment schematic diagram without algorithm, forefathers；

Fig. 3 is the standard of the service quality of the inventive method: SLA violation rate is shown with forefathers' optimal algorithm contrast and experiment It is intended to.

Detailed description of the invention

For making the purpose of the present invention, technical scheme and advantage are clearer, below in conjunction with accompanying drawing to the present invention Embodiment be described in detail.

Involved in the present invention is the high energy efficiency migration strategy of Virtual machine migration, has 800 physics joints with one As a example by the cloud data center of point, All hosts is modeled as Huawei Fusion Server RH2288H.The simulation program run is CloudSim 3.0.2, the analog data of use is the data of ten days of random acquisition from PlanetLab project, namely altogether Ten groups of workload.Specifically comprise the following steps that

Step 1, the cpu busy percentage of record per second once every main frame, and in every main frame, the CPU of each VM utilizes Rate；

Step 2, utilizes whether local weighted Return Law prediction each main frame of subsequent time transships；

Step 2.1, calculates the local weighted regression curve of current time according to main frame history cpu busy percentageA therein, b are drawn by method of least square (1)

$\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{w}_i\left(x\right){(y_i-a-{\mathrm{bx}}_i)}^2---\left(1\right)$

Wherein n is the history cpu busy percentage quantity of record, x_iFor i-th moment, y_iCPU for this main frame of i-th moment Utilization rate.Weighting function w_iX () is drawn by (2)

$w_i\left(x\right)=T\left(\frac{{\mathrm{\Δ}}_i\left(x_k\right)}{{\mathrm{\Δ}}_1\left(x_k\right)}\right)={(1-{\left(\frac{x_k-x_i}{x_k-x_1}\right)}^3)}^3---\left(2\right)$

Wherein x_kFor k point time, x_iFor i point time, Δ_i(x_k) it is the k point time difference to i point, Δ₁(x_k) it is k point to Just record the time difference of time.

Core T in weighting function is drawn by (3)

$T\left(x\right)\left\{\begin{array}{cc}{(1-{\left|x\right|}^3)}^3,& \left|x\right|<1\\ 0,& otherwise\end{array}\right.---\left(3\right);$

Wherein x is the time.

Step 2.2, according to the weighted regression curve obtained in step 2.1Calculate subsequent time main frame Cpu busy percentage predictive value

According to 2.2 predictive values calculated, step 2.3, judges whether subsequent time main frame can transship, concrete determination methods As follows:

(1) if predictive value is more than or equal to 0.9, it is determined that main frame subsequent time transships；

(2) if predictive value is less than 0.9, it is determined that main frame subsequent time does not transships；

Step 3, by step 2, selects coefficient of relationship multiple with host CPU utilization rate maximum on the main frame being judged to overload VM migrate out；

Step 3.1, separately constitutes matrix calculus multiple correlation by the cpu busy percentage of each VM on the main frame of predicted overload Coefficient.As a example by a VM, the cpu busy percentage history of the same main frame VM in addition to this VM is formed matrix table and is shown as X1, X2 ... Xn, is shown as Y by the cpu busy percentage history lists of this VM, as shown in (4).

Wherein with any one element x in X matrix_a,bAs a example by, wherein a is a platform virtual machine, and b is the b moment.Then x_a,bIt it is a platform virtual machine cpu busy percentage in the b moment.y_nFor evaluation virtual machine the n-th moment CPU utilize Rate.

Then coefficient of multiple correlation R²As shown in (5).

$R^2=\frac{\&lsqb;{\mathrm{\&Sigma;}}_{i=1}^n(y_i-\stackrel{\&OverBar;}{y})({\hat{y}}_i-\stackrel{\&OverBar;}{y})\&rsqb;}{{\mathrm{\&Sigma;}}_{i=1}^n{(y_i-\stackrel{\&OverBar;}{y})}^2{\mathrm{\&Sigma;}}_{i=1}^n{({\hat{y}}_i-\stackrel{\&OverBar;}{y})}^2}=\frac{{\mathrm{\&Sigma;}}_{i=1}^n{({\hat{y}}_i-\stackrel{\&OverBar;}{y})}^2}{{\mathrm{\&Sigma;}}_{i=1}^n{(y_i-\stackrel{\&OverBar;}{y})}^2}---\left(5\right)$

Wherein, y_iFor the cpu busy percentage of the virtual machine at i moment evaluation,CPU for the virtual machine of evaluation The meansigma methods of utilization rate.

WhereinValue as shown in (6)

$\begin{array}{cc}\hat{y}=Xb& b={\left(X^{T}X\right)}^{-1}{X}^{T}y\end{array}---\left(6\right);$

Step 3.2, according to the multiple correlation coefficient of each VM drawn in 3.1, chooses the VM addition that wherein numerical value is maximum and treats Migrate VM queue；

Step 4, finds out low load main frame according to minimum cpu busy percentage method based on threshold value；

Step 4.1, finds out that main frame that cpu busy percentage in cloud computing center is minimum；

Step 4.2, contrasts the cpu busy percentage of this main frame with entering the optimal threshold 0.45 that great many of experiments draws, And judge whether this main frame is in underloading, concrete determination methods is as follows:

(1) if host CPU utilization rate is more than or equal to 0.45, then judge that it is not in light condition；

(2) if host CPU utilization rate is greatly less than 0.45, then judge that it is in light condition；

Step 4.3, adds queue to be migrated by the main frame being judged to underloading.

Step 5, redistributes VM queue to be migrated according to energy sensing optimal descending algorithm, and algorithmic procedure is such as Under:

Concrete analysis is done again below according to experimental result:

The virtual machine online migration strategy main purpose of the present invention is to reduce power consumption and ensure service quality as far as possible, its Middle service quality SLA (Service-Level Agreement, service-level agreement) violation rate is weighed.As shown in Figure 2: Dvfs is a kind of cloud computing center Low Power Strategy not having virtual machine to migrate online, and lr_mmt_mu is best during forefathers study A kind of strategy migrated online based on virtual machine, it can be seen that the merit of the present invention in whole ten groups of Workload Consumption, all well below the power consumption of dvfs, is the most also slightly below lr_mmt_mu.The online migration strategy of virtual machine of the present invention is to service The impact of quality is as shown in Figure 3: the SLA violation rate of whole ten groups of Workload is all well below lr_mmt_mu, it means that this The strategy of invention is also far superior to the research of forefathers on service quality this respect.

Above example is only the exemplary embodiment of the present invention, is not used in the restriction present invention, protection scope of the present invention It is defined by the claims.The present invention can be made respectively in the essence of the present invention and protection domain by those skilled in the art Planting amendment or equivalent, this amendment or equivalent also should be regarded as being within the scope of the present invention.

Claims (4)

1. the dispatching method that a virtual machine migrates online, it is characterised in that comprise the steps:

Step 1, the cpu busy percentage of record per second once every main frame, and the cpu busy percentage of each VM in every main frame；

Whether step 2, prediction each main frame of subsequent time transship；

Step 3, the VM selecting coefficient of relationship multiple with the rising of host CPU utilization rate maximum on the main frame being judged to overload migrate out Go；

Step 4, find out low load main frame according to minimum cpu busy percentage method based on threshold value；

Step 5, VM queue to be migrated is redistributed according to energy sensing optimal descending method.

2. the dispatching method that virtual machine as claimed in claim 1 migrates online, it is characterised in that step 2 comprises the steps:

Step 2.1, according to main frame history cpu busy percentage calculate current time weighted regression curve；

Step 2.2, according to weighted regression curve calculate subsequent time main frame cpu busy percentage predictive value；

Step 2.3, judging whether subsequent time main frame can transship according to 2.2 predictive values calculated, concrete determination methods is as follows:

(1) if predictive value is more than or equal to 0.9, it is determined that main frame subsequent time transships；

(2) if predictive value is less than 0.9, it is determined that main frame subsequent time does not transships.

3. the dispatching method that virtual machine as claimed in claim 1 migrates online, it is characterised in that step 3 comprises the steps:

Step 3.1, the cpu busy percentage of each VM on the main frame of predicted overload is separately constituted matrix calculus complex phase relation Number；

Step 3.2, multiple correlation coefficient according to each VM drawn in 3.1, choose the VM that wherein numerical value is maximum and add to be migrated VM queue.

4. the dispatching method that virtual machine as claimed in claim 1 migrates online, it is characterised in that step 4 comprises the steps:

Step 4.1, find out that main frame that cpu busy percentage in cloud computing center is minimum；

Step 4.2, the cpu busy percentage of this main frame is contrasted with entering the optimal threshold 0.45 that great many of experiments draws, and sentence Whether this main frame disconnected is in underloading, and concrete determination methods is as follows:

(1) if host CPU utilization rate is more than or equal to 0.45, then judge that it is not in light condition；

(2) if host CPU utilization rate is greatly less than 0.45, then judge that it is in light condition；

Step 4.3, will be judged to underloading main frame add queue to be migrated.