CN104375621A - Dynamic weighting load assessment method based on self-adaptive threshold values in cloud computing - Google Patents

Abstract

The invention discloses a dynamic weighting load assessment method based on self-adaptive threshold values in cloud computing. The method includes the steps that load assessment threshold values lambda 1 and lambda 2 are calculated according to request parameters of virtual resources; a normalized relative load value L[i] is calculated; according to the value of L[i], load states of the virtual resources are divided into the idle state, the normal state and the overload state by the adoption of the double threshold values lambda 1 and lambda 2; for the virtual resources in the overload state, the overload resources are migrated, and for the virtual resources in the idle state, when the number of times in idle exceeds a migration threshold value coefficient, the virtual resources are released; the load assessment threshold values lambda 1 and lambda 2 are dynamically adjusted in a self-adaptive mode; if a local request average value and the number of the virtual resources change, in the next assessment cycle, the load assessment threshold values lambda 1 and lambda 2 are dynamically adjusted in the self-adaptive mode. Differential and integral operation is not involved, the operation complexity is low, the use ratio of the resources can be increased, and energy loss can be reduced.

Description

Based on the dynamic weighting load evaluation method of adaptive threshold in a kind of cloud computing

Technical field

The present invention relates to a kind of dynamic weighting load evaluation method based on adaptive threshold in cloud computing.

Background technology

The existing research about energy consumption is mainly from two: one is save electric energy by the voltage of dynamic conditioning server or frequency; Two is realize energy-conservation by closing idle server resource.Wherein second point is the focus of research at present, as Kyoto University's HLRS report display: only adopt the strategy of closing idle node aperiodically just to achieve the energy consumption saving of 39%.

In the energy consumption of current field of cloud calculation, the tolerance of energy consumption is a difficult point, the especially energy consumption of virtual resource, and the energy consumption of virtual resource cannot be obtained by hard ware measure.There is this people to propose an indirectly energy consumption of virtual machine measurement mechanism, first follow the tracks of the resource service condition of each hardware component that virtual machine uses, then by a resource energy consumption model, resource utilization is converted to energy consumption utilization rate.

Precision problem in addition about energy consumption measurement is also a challenge, in virtual cloud computation data center, the energy consumption of virtual machine is normally determined by the utilization factor of assessment current virtual machine CPU, but these assessment possibilities also out of true, because the cpu busy percentage measured cannot accurately reflect real cpu usage, it further comprises the internal memory stand-by period, therefore truly can not reflect power consumption situation.Statistics is presented in the server resource for calculating, and CPU energy consumption only accounts for 43%, so it is a bit unilateral only to assess energy consumption with cpu busy percentage.Existing document just carries out simplation verification (energy consumption involved by transition process) to energy consumption from experimental viewpoint.

More research is research resource load situation be combined with energy consumption.In existing research, there is a lot of research for minimum energy losses, achieved the Dynamic Integration problem of the virtual resource of data center by migration.But a lot of the load condition of resource (whether transship, excessively not busy etc.) not to be assessed, just periodically optimize virtual resource allocation.Also some employing certain methods arranges the threshold value passed judgment on host resource and whether transship, and exceedes this threshold value and is judged as overload.But the setting of these static threshold can not meet the demand of resource load dynamic change, load histories analysis was utilized to judge overload by dynamic conditioning threshold value so research and propose again afterwards, and some resource migration integrated strategies are applied to overload resource, energy-conservation to realize.But they only judge the overload situations of resource load, do not divide in more detail resource status.In addition in threshold value is arranged, in some document, there is adjustable factors, but its factor (this factor is lower, and energy consumption is fewer, but SLAV is higher) of just method energy consumption and SLAV being balanced, and how this factor specifically quantizes not explanation.

Researched and proposed a dynamic virtual resource migration strategy, resource is divided into these three states of focus, warm spot and cold spot according to resource utilization situation by it, and according to certain strategy, hot point resource is moved to avoid overload, by cold spot release to save energy consumption.The threshold value of the evaluation and test resource status that this division resource and the method for saving energy consumption adopt is fixing, and does not verify energy consumption situation of saving.

In a word, mainly there is following problem in existing research:

(1) existing appraisal procedure only uses this index of cpu busy percentage as the overall service condition of resource, but in fact cpu busy percentage can not represent the using state of resource completely, it further comprises the internal memory stand-by period etc., so it is accurate not only to assess resource load with cpu busy percentage.

(2) existing appraisal procedure only judges the overload situations of resource load, does not judge for unavailable or slack resources for a long time.

(3) judgement etc. of existing resource assessment to overload all uses fixing threshold value, and this cannot adapt to the feature of cloud resource dynamic change.Though there is an adjustable factors s in indivedual document, it is revised (s is lower, and energy consumption is fewer, but SLAV is higher) the energy consumption of method and performance, and how s specifically quantizes not explanation.

Summary of the invention

The object of the invention is to overcome deficiency of the prior art, provide a kind of dynamic weighting load evaluation method based on adaptive threshold in cloud computing, do not relate to differential, integral operation, computational complexity is low, can improve resource utilization and save energy consumption.

For solving the problems of the technologies described above, the technical solution adopted in the present invention is: based on the dynamic weighting load evaluation method of adaptive threshold in a kind of cloud computing, comprise the steps:

Step one: according to virtual resource required parameter computational load assessment threshold value λ ₁, λ ₂;

Step 2: calculate normalized relative load value L [i];

Step 3: state classification: the value according to L [i] adopts dual threshold λ ₁and λ ₂the load condition of virtual resource is divided into free time, normal, overload three kinds of states;

Step 4: state processing:

For the virtual resource being divided into overload, resource migration will be transshipped, jump to step 5;

For the virtual resource being divided into idle state, the idle number of times of statistics, when idle number of times exceedes mobility threshold coefficient, discharges this virtual resource, jumps to step 5;

For the virtual resource being divided into normal condition, do not deal with, directly jump to step 5;

Step 5: self-adaptation dynamic adjustments load evaluation threshold value λ ₁and λ ₂: if please average in this locality and virtual resource number changes, then in next assessment cycle, self-adaptation dynamic adjustments load evaluation threshold value λ ₁and λ ₂; Otherwise, directly exit.

Load evaluation threshold value λ ₁, λ ₂specific formula for calculation as follows:

λ ₁＝Q-σ (1)

λ ₂＝Q+σ (2)

Wherein Q is the mean value of all resource load intensity of system, and specific formula for calculation is as follows:

$Q=\frac{1}{n}\underset{k=1}{\overset{n}{\mathrm{\Σ}}}{Q}_k---\left(3\right)$

σ is the standard deviation of system load, and specific formula for calculation is as follows:

$\mathrm{\σ}=\sqrt{\frac{1}{n}\underset{k=1}{\overset{n}{\mathrm{\Σ}}}{(x_k-\stackrel{\‾}{x})}^2}---\left(4\right)$

Wherein Q _kfor the mean value of the intensity of load of resource k, n is all resource numbers in system, for the mean value of system load intensity, namely x _kfor the mean value of the intensity of load of resource node k, i.e. x _k=Q _k;

$Q_k=\frac{1}{m}\underset{k=1}{\overset{m}{\mathrm{\Σ}}}{q}_k---\left(5\right)$

Wherein q _kfor resource node U _kthe intensity of load of (1≤k≤n).

The computing formula of L [i] is as follows:

$L\left[i\right]=\left\{\begin{array}{c}1,\mathrm{if}(r_i=R_i\mathrm{or}{h}_i\≥H_i\mathrm{or}{q}_i\≥Q_i)\\ w_1\frac{r_i}{R_i}+w_2\frac{h_i}{H_i}+w_3\frac{q_i}{Q_i},\mathrm{Others}\end{array}\right.---\left(11\right)$

Wherein r _i, R _irepresent resource U respectively _icurrent resource request amount and largest request amount, h _i, H _irepresent resource U respectively _icurrent computing power and max calculation ability, q _iand Q _irepresent U respectively _ipresent load intensity and maximum load intensity; r _i/ R _i, h _i/ H _i, q _i/ Q _ibe respectively r _ito R _i, h _ito H _i, q _ito Q _ibe normalized the value of gained, span is [0,1], w ₁, w ₂, w ₃be respectively r _i/ R _i, h _i/ H _i, q _i/ Q _ito the weighing factor of L [i];

Adopt Dynamic Effect weight, in each assessment cycle, by formula (12) self-adaptation dynamic adjustments weighted value w _j:

w _j＝w ₀+μ(w ₁-w ₀) (12)

Wherein w ₀, w ₁be constant, its scope is [0,0.5], [0,1] respectively, and w ₀>w ₁; μ is the random number distributed in [0,1], and formula (12) makes r _i/ R _i, h _i/ H _i, q _i/ Q _iweighing factor at [w ₀, w ₁] between random variation, and meet formula (13):

$\underset{j=1}{\overset{3}{\mathrm{\Σ}}}{w}_j=1---\left(13\right)$

Described mobility threshold coefficient is dynamic conditioning according to current environment for use, and concrete adjustment mode is:

$\mathrm{\λ}=\left[\frac{\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{r}_i}{n\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{h}_i}\right].$

Compared with prior art, the beneficial effect that the present invention reaches is: utilize dynamic evaluation index, a kind of method using dual threshold to judge is proposed to resource dynamic loading condition, according to resource status self-adaptative adjustment and the dynamic threshold that can specifically quantize, the dynamic of cloud resource load can be more suitable for; Overload, normal and idle three kinds of situations are divided into resource, and the threshold value of evaluation and test is set to self-adaptation dynamic threshold, more accurately can reflect the index of resource using status; According to assessment result by overload resource migration to realize load balancing, improve resource utilization, the free time is exceeded certain threshold value resource unloading to reduce energy consumption; Assessment result can be used for the formulation of the decision-making such as virtual machine (vm) migration, unloading, and as transshipped resource migration, the unloading etc. that the free time exceedes mobility threshold coefficient, can improve resource utilization and save energy consumption.The inventive method does not relate to differential, integral operation, and computing method are comparatively simple, and complexity is relatively low.

Accompanying drawing explanation

Fig. 1 is operational flowchart of the present invention.

Fig. 2 is hardware configuration theory diagram of the present invention.

Embodiment

Below in conjunction with accompanying drawing, the invention will be further described.Following examples only for technical scheme of the present invention is clearly described, and can not limit the scope of the invention with this.

As shown in Figure 2, hardware configuration theory diagram of the present invention, comprise global resource manager and some groups of local resource manager, the corresponding physical node of each local resource manager and several virtual resources (VM), loading condition in local resource manager periodic collection local virtual machine or physical machine, then these resource informations are supplied to global resource manager, global resource manager can adopt dual threshold λ according to the inventive method according to the value of L [i] ₁and λ ₂(λ ₁< λ ₂) load condition of virtual resource is divided into free time, normal, overload three kinds of states.

To the assessment result of resource load, can be formulation associated virtual machine migration strategy and provide Data support, for virtual machine (vm) migration and release provide information.Specific implementation method is the resource release idle number of times c being exceeded mobility threshold coefficient, to save energy consumption; By the resource migration of overload to realize load balancing.Mobility threshold coefficient is dynamic conditioning according to current environment for use, and concrete adjustment mode is:

When local resource manager monitor this locality please average and VM number changes time, in time data notification is transferred to global resource manager, like this in next assessment cycle, just needs self-adaptation dynamic adjustments threshold value λ ₁and λ ₂.And Dynamic weighting values w _jthen carry out renewal adjustment in each assessment cycle according to formula (12).

As shown in Figure 1, be operational flowchart based on the dynamic weighting load evaluation method of adaptive threshold in cloud computing of the present invention.Before carrying out concrete operations, first input resource request parameter, comprising: resource U _icurrent resource request amount r _iwith largest request amount R _i, resource U _icurrent computing power h _iwith max calculation ability H _i, U _ipresent load intensity q _iwith maximum load intensity Q _i.

Concrete operation step of the present invention is as follows:

Step one: according to virtual resource required parameter computational load assessment threshold value λ ₁, λ ₂:

Load evaluation threshold value λ ₁, λ ₂specific formula for calculation as follows:

λ ₁＝Q-σ (1)

λ ₂＝Q+σ (2)

Wherein Q is the mean value of all resource load intensity of system, and specific formula for calculation is as follows:

$Q=\frac{1}{n}\underset{k=1}{\overset{n}{\mathrm{\&Sigma;}}}{Q}_k---\left(3\right)$

σ is the standard deviation of system load, and specific formula for calculation is as follows:

$\mathrm{\&sigma;}=\sqrt{\frac{1}{n}\underset{k=1}{\overset{n}{\mathrm{\&Sigma;}}}{(x_k-\stackrel{\&OverBar;}{x})}^2}---\left(4\right)$

Wherein Q _kfor the mean value of the intensity of load of resource k, n is all resource numbers in system, for the mean value of system load intensity, namely x _kfor the mean value of the intensity of load of resource node k, i.e. x _k=Q _k.

$Q_k=\frac{1}{m}\underset{k=1}{\overset{m}{\mathrm{\&Sigma;}}}{q}_k---\left(5\right)$

Wherein q _kfor resource node U _kthe intensity of load of (1≤k≤n).

Step 2: calculate normalized relative load value L [i]: L [i] can reflect performance difference between resource and potential intensity of load, for realizing the decision-making foundation that load balancing provides best.

The computing formula of L [i] is as follows:

$L\left[i\right]=\left\{\begin{array}{c}1,\mathrm{if}(r_i=R_i\mathrm{or}{h}_i\&GreaterEqual;H_i\mathrm{or}{q}_i\&GreaterEqual;Q_i)\\ w_1\frac{r_i}{R_i}+w_2\frac{h_i}{H_i}+w_3\frac{q_i}{Q_i},\mathrm{Others}\end{array}\right.---\left(11\right)$

R _i/ R _i, h _i/ H _i, q _i/ Q _ibe respectively r _ito R _i, h _ito H _i, q _ito Q _ibe normalized the value of gained, span is [0,1], w ₁, w ₂, w ₃be respectively r _i/ R _i, h _i/ H _i, q _i/ Q _ito the weighing factor of L [i];

Adopt Dynamic Effect weight, in each assessment cycle, by formula (12) self-adaptation dynamic adjustments weighted value w _j:

w _j＝w ₀+μ(w ₁-w ₀) (12)

Wherein w ₀, w ₁be constant, its scope is [0,0.5], [0,1] respectively, and w ₀>w ₁; μ is the random number distributed in [0,1], and formula (12) makes r _i/ R _i, h _i/ H _i, q _i/ Q _iweighing factor at [w ₀, w ₁] between random variation, and meet formula (13):

$\underset{j=1}{\overset{3}{\mathrm{\&Sigma;}}}{w}_j=1---\left(13\right).$

Step 3: state classification: the value according to L [i] adopts dual threshold λ ₁and λ ₂the load condition of virtual resource is divided into free time, normal, overload three kinds of states;

Step 4: state processing:

For the virtual resource being divided into overload, resource migration will be transshipped, jump to step 5;

For the virtual resource being divided into idle state, the idle number of times of statistics, when idle number of times exceedes mobility threshold coefficient, discharges this virtual resource, jumps to step 5;

For the virtual resource being divided into normal condition, do not deal with, directly jump to step 5;

Step 5: self-adaptation dynamic adjustments load evaluation threshold value λ ₁and λ ₂: if please average in this locality and virtual resource number changes, then data notification is transferred to global resource manager, in next assessment cycle, self-adaptation dynamic adjustments load evaluation threshold value λ ₁and λ ₂; Otherwise, directly exit.

Present resource (PMs and the VMs) energy consumption discussed before and after use the present invention.Before integrating, energy consumption is E _before, computing formula is as follows.

E _before＝αP _maxrNT (14)

After integrating, energy consumption is E _after, it is made up of two parts, Current resource energy consumption and transition process energy consumption.Directly be difficult to quantize because moving involved energy consumption, therefore adopt and a kind ofly use resource request amount as calculated amount thus the method for indirect approximate treatment migration amount, after therefore integrating, the computing formula of energy consumption is as follows.

E _after＝αP _maxr(N-b)T+βP _transL _M(15)

$L_M=\underset{k=1}{\overset{a}{\mathrm{\&Sigma;}}}{A}_k---\left(16\right)$

Wherein α and β is regulatory factor, and N is in the number by resource, and T is the averaging time that resource is run, P _maxrfor the maximum energy consumption of each resource, be set to 250w, P _transfor the migration transmitting procedure power consumption values of every computing unit, be set to 2.4W.L _mfor transmission calculated amount.A _kfor overload need the amount of migration, A _k=maximum resource request amount R _i-current resource request amount r _i.

Wherein, a is migration number of times, and b is unloading resource number of times, and wherein the value of b is relevant with mobility threshold coefficient lambda with resources idle number of times c.

From model, the difference E of energy consumption before and after integrating _savecalculating as formula (17)

E _save＝E _before-E _after＝αP _maxrbT-βP _transL _M(17)

Can find out that this difference need the amount of migration A with overload thus _k, migration number of times a is relevant with unloading number of times b.And wherein the value of regulatory factor α and β need the adaptive change according to actual experiment environment.

Below be only the preferred embodiment of the present invention; it should be pointed out that for those skilled in the art, under the prerequisite not departing from the technology of the present invention principle; can also make some improvement and distortion, these improve and distortion also should be considered as protection scope of the present invention.

Claims (4)

1. in cloud computing based on a dynamic weighting load evaluation method for adaptive threshold, it is characterized in that, comprise the steps:

Step one: according to virtual resource required parameter computational load assessment threshold value λ ₁, λ ₂;

Step 2: calculate normalized relative load value L [i];

Step 3: state classification: the value according to L [i] adopts dual threshold λ ₁and λ ₂the load condition of virtual resource is divided into free time, normal, overload three kinds of states;

Step 4: state processing:

For the virtual resource being divided into overload, resource migration will be transshipped, jump to step 5;

For the virtual resource being divided into idle state, the idle number of times of statistics, when idle number of times exceedes mobility threshold coefficient, discharges this virtual resource, jumps to step 5;

For the virtual resource being divided into normal condition, do not deal with, directly jump to step 5;

Step 5: self-adaptation dynamic adjustments load evaluation threshold value λ ₁and λ ₂: if please average in this locality and virtual resource number changes, then in next assessment cycle, self-adaptation dynamic adjustments load evaluation threshold value λ ₁and λ ₂; Otherwise, directly exit.

2. in cloud computing according to claim 1 based on the dynamic weighting load evaluation method of adaptive threshold, it is characterized in that, load evaluation threshold value λ ₁, λ ₂specific formula for calculation as follows:

λ ₁＝Q-σ (1)

λ ₂＝Q+σ (2)

Wherein Q is the mean value of all resource load intensity of system, and specific formula for calculation is as follows:

$Q=\frac{1}{n}\underset{k=1}{\overset{n}{\mathrm{\&Sigma;}}}{Q}_k---\left(3\right)$

σ is the standard deviation of system load, and specific formula for calculation is as follows:

$\mathrm{\&sigma;}=\sqrt{\frac{1}{n}\underset{k=1}{\overset{n}{\mathrm{\&Sigma;}}}{(x_k-\stackrel{\&OverBar;}{x})}^2}---\left(4\right)$

Wherein Q _kfor the mean value of the intensity of load of resource k, n is all resource numbers in system, for the mean value of system load intensity, namely x _kfor the mean value of the intensity of load of resource node k, i.e. x _k=Q _k;

$Q_k=\frac{1}{m}\underset{k=1}{\overset{m}{\mathrm{\&Sigma;}}}{q}_k---\left(5\right)$

Wherein q _kfor resource node U _kthe intensity of load of (1≤k≤n).

3. in cloud computing according to claim 1 based on the dynamic weighting load evaluation method of adaptive threshold, it is characterized in that, the computing formula of L [i] is as follows:

$L\left[i\right]=\left\{\begin{array}{c}1,\mathrm{if}(r_i=R_i\mathrm{or}{h}_i\&GreaterEqual;H_i\mathrm{or}{q}_i\&GreaterEqual;Q_i)\\ w_1\frac{r_i}{R_i}+w_2\frac{h_i}{H_i}+w_3\frac{q_i}{Q_i},\mathrm{Others}\end{array}\right.---\left(11\right)$

Wherein r _i, R _irepresent resource U respectively _icurrent resource request amount and largest request amount, h _i, H _irepresent resource U respectively _icurrent computing power and max calculation ability, q _iand Q _irepresent U respectively _ipresent load intensity and maximum load intensity; r _i/ R _i, h _i/ H _i, q _i/ Q _ibe respectively r _ito R _i, h _ito H _i, q _ito Q _ibe normalized the value of gained, span is [0,1], w ₁, w ₂, w ₃be respectively r _i/ R _i, h _i/ H _i, q _i/ Q _ito the weighing factor of L [i];

Adopt Dynamic Effect weight, in each assessment cycle, by formula (12) self-adaptation dynamic adjustments weighted value w _j:

w _j＝w ₀+μ(w ₁-w ₀) (12)

Wherein w ₀, w ₁be constant, its scope is [0,0.5], [0,1] respectively, and w ₀>w ₁; μ is the random number distributed in [0,1], and formula (12) makes r _i/ R _i, h _i/ H _i, q _i/ Q _iweighing factor at [w ₀, w ₁] between random variation, and meet formula (13):

$\underset{j=1}{\overset{3}{\mathrm{\&Sigma;}}}{w}_j=1---\left(13\right).$

4. in cloud computing according to claim 3 based on the dynamic weighting load evaluation method of adaptive threshold, described mobility threshold coefficient is dynamic conditioning according to current environment for use, and concrete adjustment mode is:

$\mathrm{\&lambda;}=\left[\frac{\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}{r}_i}{n\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}{h}_i}\right].$