CN103036974A - Cloud computing resource scheduling method and system based on hidden markov model - Google Patents

Abstract

The invention relates to related technical field of cloud computing, in particular to a cloud computing resources scheduling method and a system based on a hidden markov model. The cloud computing resources scheduling method comprises a step of model training, and training the hidden markov model to acquire related model parameters; a step of detecting business volume, detecting the business volume of the cloud computing resources to acquire business volume observed value; a step of forecasting the business volume, inputting the acquired business volume observed value to the hidden markov model to carry out state calculation and get a forecasting state, and the hidden markov model adopts the model parameters gained in the step of training the model; and a step of scheduling the resources, scheduling the cloud computing resources according to the forecasting state. Due to the fact that the business volume is forecasted through the hidden markov model (HMM), resource scheduling can be carried out according to the forecast result, a schedule can be carried out in advance, and resources can be utilized effectively.

Description

Cloud computing resource scheduling method and system based on hidden Markov model

Technical field

The present invention relates to the cloud computing correlative technology field, particularly relate to a kind of cloud computing resource scheduling method based on hidden Markov model and system.

Background technology

Scheduling of resource is the key problem of cloud computing: the introducing of the technology such as cloud computing is virtual so that script disperses, idle resource forms unified, standardized virtual resource, becomes possibility so that resource is allocated flexibly; The resource optimization allotment can be satisfied the needs of operation system so that the operation system that is deployed on the cloud computing resource pool can be used resource according to the size of traffic carrying capacity, can avoid the wasting of resources simultaneously.

Present resource regulating method often is conceived to the optimization of virtual machine on physical server and disposes, what mainly solve is how to determine reasonably that virtual machine is to the mapping of node, usually be counted as a bin packing, namely seek the optimum scheme that virtual machine is assigned to node, thereby make the use resource sum of virtual machine in each node be no more than the upper limit that node can provide, and the number of nodes that uses is optimum.The scholars such as Buyya propose based on the economic model resource regulating method, the method has proposed market-oriented cloud computing architecture and market-oriented Resource Distribution and Schedule method, this architecture realizes negotiation between resource user and the resource provider by the SLA resource allocator, realizes that resource optimization distributes.Li Qiangs etc. have proposed to place multiple-objection optimization genetic resources dispatching method etc. with the virtual machine of application service level goal constraint.The people such as Menaud and Van proposes to propose dynamic dispatching method for the management of virtual resource in the cloud computing, take overall scheduling time as target, reshuffle time and virtual machine (vm) migration time by considering, provide a kind of resource convention method Entropy.Wei Guiyi etc. have proposed to solve the cloud computing resources assignment problem based on game theory resource Resources allocation dispatching method, this resource regulating method is optimized for the resource requirement that at first utilizes the integer programming method to single operation system, adopts the evolution resource regulating method to solve the complex optimization problem of a plurality of operation systems again.The people such as Xu Xianghua have proposed a kind of cloud computing resources allocation strategy based on market mechanism, and design one based on the demand in genetic price adjustment resource regulating method processing market and the equilibrium problem of supply.The people such as Lin Weiwei have proposed a kind of cloud computing resource scheduling method based on dynamic reconfiguration virtual resources.The cloud application load information that the method is collected take the cloud application monitor is as foundation, and the load capacity of the virtual resource of then using based on the operation cloud and cloud are used current load and carried out dynamic decision, is that cloud is used dynamic reconfiguration virtual resources according to the result of decision-making.By use the dynamic adjustment that the method for reshuffling virtual resource realizes resource for cloud, do not need dynamically to redistribute physical resource and stop cloud and use execution.Beloglazov etc. propose a kind of resource regulating method based on minimum power consumption, and the energy consumption by the server before and after the computation migration is optimized scheduling.

Above-mentioned resource regulating method major part belongs to static resource regulating method, and namely under the prerequisite that traffic carrying capacity is determined substantially, the resources of virtual machine that the computing service amount is required carries out the binding of resources of virtual machine and physical resource by optimizing resource regulating method.In the production environment of reality, traffic carrying capacity is dynamic change, and these resource regulating methods often under the scene of service based amount maximum, are optimized deploying virtual machine.Therefore, in most of traffic carrying capacitys and little situation, this type of resource regulating method can cause virtual resource idle, can cause certain wasting of resources.The method that the people such as the people such as Lin Weiwei and Li Qiang propose belongs to dynamic resource regulating method, and the former detects by the utilance to present resource node, surpasses set threshold value and carries out the resource increase and decrease; The latter has carried out particular study to the WEB flow, obtains the processing threshold value of WEB flow, and to the monitoring performance of virtual machine, triggers resource mobilization when surpassing threshold value.These two kinds of resource regulating methods all belong to the resource regulating method of passive-type, namely can only by the virtual machine real resource consumption dispatch, when traffic carrying capacity is busy, can cause the scheduling lag behind business demand so that business demand can't in time respond.

The said method overwhelming majority is all dispatched the certainty resource based on current resources situation, and in the cloud computing environment, traffic carrying capacity has certain uncertainty, and the detection traffic carrying capacity is dispatched and tended to cause resource short-term wretched insufficiency or waste.

Summary of the invention

Based on this, be necessary to carry out according to the prediction traffic carrying capacity in cloud computing resources for prior art the computational problem of scheduling of resource, a kind of cloud computing resource scheduling method based on hidden Markov model is provided.

A kind of cloud computing resource scheduling method based on hidden Markov model, use hidden Markov model the traffic carrying capacity situation of change to be described and to predict described method comprises:

The model training step is trained the model parameter that obtains being correlated with to hidden Markov model;

The traffic carrying capacity detection steps is surveyed the traffic carrying capacity of cloud computing resources and to be obtained the traffic carrying capacity measured value;

The traffic carrying capacity prediction steps, the traffic carrying capacity measured value that detection is obtained is inputted described hidden Markov model and is carried out state computation, obtains predicted state, the model parameter that described hidden Markov model adopts the model training step to obtain;

The scheduling of resource step is dispatched cloud computing resources according to described predicted state.

Among embodiment, the traffic carrying capacity of described cloud computing resources is the network traffics parameter therein.

Among embodiment, described traffic carrying capacity prediction steps comprises therein:

As a time slot traffic carrying capacity is carried out record every the t time, every N time slot consists of an observation window, carries out data at each observation window and processes and obtain the traffic carrying capacity measured value, and wherein, t is positive integer more than or equal to 1 greater than 0, N;

Carrying out data at each observation window processes and to obtain the traffic carrying capacity measured value and specifically comprise:

With i observation window T _iJ time slot be designated as t _{I, j}, corresponding traffic log value is x _{I, j}, wherein 1≤j≤N calculates T _iThe mean value of interior all record values

Wherein, 1≤j≤N;

Calculate i observation window T _iRecord value mean value X _iWith i-1 observation window T _I-1Record value mean value X _I-1Difference, as i observation window T _iTraffic carrying capacity measured value the first parameter o _{I, 1}=X _i-X _I-1

Calculate i observation window T _iThe contrast C ON of interior all record values _i=∑ _{M, n}| m-n|p _{I, mn}, as i observation window T _iTraffic carrying capacity measured value the second parameter estimator value o _{I, 2,}Wherein, minx _{I, j}()≤m, n≤max (x _{I, j}), 1≤j≤N, p _{I, mn}I observation window T _iJ record value be that m and j+1 record value situation that is n is at i observation window T _iThe frequency of middle appearance, p _{I, mn}=# (x _{I, j}=m, x _{I, j+1}=n)/(T-1), and wherein, min (x _{I, j})≤m, n≤max (x _{I, j});

To o _{I, 2}And o _{I, 2}Be converted to corresponding traffic carrying capacity measured value.

Therein among embodiment:

Described model parameter comprises: state set, the set of measured value value, state transition probability matrix, measured value produce probability matrix and initial probability distribution matrix;

Described state set S={s ₁, s ₂, s ₃, s wherein ₁The expression traffic carrying capacity increases state, s ₂The fair state of expression traffic carrying capacity, s ₃The expression traffic carrying capacity reduces state;

Described measured value value set V={v ₁, v ₂... v _K, K 〉=1;

Described state transition probability matrix A={a _Mn, a wherein _MnExpression is from state s _mJump to state s _nProbability;

Described measured value produces probability matrix B={b _m(k) }, b wherein _m(k) be illustrated in state s _mLower generation measured value v _kProbability;

Described initial probability distribution matrix π={ π _m, π wherein _mThe expression initial condition is s _mProbability;

Described traffic carrying capacity prediction steps specifically comprises:

Calculate e observation window T _ePredicted state q _e:

Wherein, a _e(i) a _eBe e observation window T _eI forward variable, and: ${\mathrm{\α}}_e\left(i\right)=\left\{\begin{array}{cc}{\mathrm{\π}}_i{b}_i\left(o_e\right),& e=\mathrm{1,1}\≤i\≤M\\ \left[\underset{i=1}{\overset{M}{\mathrm{\Σ}}}{\mathrm{\α}}_{e-1}\left(j\right)a_{\mathrm{ji}}\right]b_i\left(o_e\right),& e\≠\mathrm{1,1}\≤i\≤M\end{array}\right.,$ Wherein, o _eBe e observation window T _eThe traffic carrying capacity predicted value.

Among embodiment, described scheduling of resource step specifically comprises therein:

Reduce state if predicted state is traffic carrying capacity, then reclaim cloud computing resources;

Increase state if predicted state is traffic carrying capacity, then increase cloud computing resources;

If predicted state the is traffic carrying capacity state that maintains an equal level is then kept cloud computing resources.

Wherein, described cloud computing resources can be the resources of virtual machine in the cloud computing system.

A kind of cloud computing resources dispatching patcher based on hidden Markov model comprises:

The model training module is used for hidden Markov model is trained the model parameter that obtains being correlated with;

The traffic carrying capacity detecting module obtains the traffic carrying capacity measured value for the traffic carrying capacity of cloud computing resources is surveyed, and sends to the traffic carrying capacity prediction module;

The traffic carrying capacity prediction module, be used for that the traffic carrying capacity measured value that the detection of traffic carrying capacity detecting module obtains is inputted described hidden Markov model and carry out state computation, obtain predicted state and send to the scheduling of resource module, the model parameter that described hidden Markov model adopts the model training module to obtain;

The scheduling of resource module is used for according to the predicted state that obtains from the traffic carrying capacity prediction module cloud computing resources being dispatched.

Therein among embodiment:

Described traffic carrying capacity detecting module comprises:

Record sub module is used for as a time slot traffic carrying capacity being carried out record every the t time;

Traffic carrying capacity measured value calculating sub module is used for every N time slot and consists of an observation window, carries out the data processing at each observation window and obtains the traffic carrying capacity measured value, specifically is used for:

With i observation window T _iJ time slot be designated as t _{I, j}, corresponding traffic log value is x _{I, j}, wherein 1≤j≤N calculates T _iThe mean value of interior all record values

Wherein, 1≤j≤N;

Calculate i observation window T _iRecord value mean value X _iWith i-1 observation window T _I-1Record value mean value X _I-1Difference, as i observation window T _iTraffic carrying capacity measured value the first parameter o _{I, 1}=X _i-X _I-1

Calculate i observation window T _iThe contrast C ON of interior all record values _i=∑ _{M, n}| m-n|p _{I, mn}, as i observation window T _iTraffic carrying capacity measured value the second parameter estimator value o _{I, 2}, wherein, minx _{I, j}()≤m, n≤max (x _{I, j}), 1≤j≤N, p _{I, mn}I observation window T _iJ record value be that m and j+1 record value situation that is n is at i observation window T _iThe frequency of middle appearance, p _{I, mn}=# (x _{I, j}=m, x _{I, j+1}=n)/(N-1), and wherein, min (x _{I, j})≤m, n≤max (x _{I, j});

To o _{I, 2}And o _{I, 2}Be converted to corresponding traffic carrying capacity measured value and be sent to the traffic carrying capacity prediction module.

Therein among embodiment:

Described model parameter comprises: state set, the set of measured value value, state transition probability matrix, measured value produce probability matrix and initial probability distribution matrix;

Described state set S={s ₁, s ₂, s ₃, s wherein ₁The expression traffic carrying capacity increases state, s ₂The fair state of expression traffic carrying capacity, s ₃The expression traffic carrying capacity reduces state;

Described measured value value set V={v ₁, v ₂... v _K, K 〉=1;

Described state transition probability matrix A={a _Mn, a wherein _MnExpression is from state s _mJump to state s _nProbability;

Described measured value produces probability matrix B={b _m(k) }, b wherein _m(k) be illustrated in state s _mLower generation measured value v _kProbability;

Described initial probability distribution matrix π={ π _m, π wherein _mThe expression initial condition is s _mProbability;

Described traffic carrying capacity prediction module specifically is used for:

Calculate e observation window T _ePredicted state q _e:

Wherein, a _e(i) a _eBe e observation window T _eI forward variable, and: ${\mathrm{\α}}_e\left(i\right)=\left\{\begin{array}{cc}{\mathrm{\π}}_i{b}_i\left(o_e\right),& e=\mathrm{1,1}\≤i\≤M\\ \left[\underset{i=1}{\overset{M}{\mathrm{\Σ}}}{\mathrm{\α}}_{e-1}\left(j\right)a_{\mathrm{ji}}\right]b_i\left(o_e\right),& e\≠\mathrm{1,1}\≤i\≤M\end{array}\right.,$ Wherein, o _eBe e observation window T _eThe traffic carrying capacity predicted value.

Among embodiment, described scheduling of resource module comprises therein:

Traffic carrying capacity reduces the scheduling sublayer module, and being used for predicted state is that traffic carrying capacity reduces state, reclaims cloud computing resources;

Traffic carrying capacity increases the scheduling sublayer module, and being used for predicted state is that traffic carrying capacity increases state, increases cloud computing resources;

The traffic carrying capacity scheduling sublayer module that maintains an equal level is used for predicted state and is the traffic carrying capacity state that maintains an equal level, and keeps cloud computing resources.

Present technology generally adopts dispatches the monitor mode of existing resource, can only accomplish that resource carries out intervention schedule in the not enough situation.Above-mentioned a kind of cloud computing resource scheduling method based on hidden Markov model, because by hidden Markov model (Hidden Markov Model, HMM) traffic carrying capacity is predicted, thereby the result according to prediction carries out the cloud computing resources scheduling, forward scheduling can be carried out, resource can be more effectively utilized.

Description of drawings

Fig. 1 is the flow chart of the cloud computing resource scheduling method based on hidden Markov model of the present invention;

Fig. 2 is the block diagram of the cloud computing resources dispatching patcher based on hidden Markov model of the present invention;

Fig. 3 is that traffic carrying capacity detection time of the present invention is cut apart schematic diagram;

Fig. 4 is that contrast of the present invention is calculated schematic diagram;

Fig. 5 is the flow chart of resource regulating method of the present invention;

Fig. 6 is the code conversion figure that the first quantification measured value of the present invention and second quantizes measured value.

Embodiment

The present invention will be further described in detail below in conjunction with the drawings and specific embodiments.

Show the flow chart of the cloud computing resource scheduling method based on hidden Markov model of the present invention such as Fig. 1.

A kind of cloud computing resource scheduling method based on hidden Markov model, described method comprises:

Step S101 trains the model parameter that obtains being correlated with to hidden Markov model;

Step S102 surveys the traffic carrying capacity of cloud computing resources and to obtain the traffic carrying capacity measured value;

Step S103, the traffic carrying capacity measured value that detection is obtained is inputted described hidden Markov model and is carried out state computation, obtains predicted state, the model parameter that described hidden Markov model adopts the model training step to obtain;

Step S104 dispatches cloud computing resources according to described predicted state.

In the HMM of present embodiment, state set S has 3 states, wherein s ₁The expression traffic carrying capacity increases state, s ₂The fair state of expression traffic carrying capacity, s ₃The expression traffic carrying capacity reduces state.

For the training of HMM, carry out to algorithm before and after adopting, specific as follows:

The front and back of HMM are O (M to the algorithm computation complexity ²T), wherein T is the length of training and forecasting sequence.

Definition forward variable and the following formula 1 of computational methods thereof:

${\mathrm{\α}}_t\left(i\right)\stackrel{\mathrm{\Δ}}{=}P(o_1^t,q_t=s_i|\mathrm{\λ})=\left\{\begin{array}{cc}{\mathrm{\π}}_i{b}_i\left(o_1\right),& t=\mathrm{1,1}\≤i\≤M\\ \left[\underset{i=1}{\overset{M}{\mathrm{\Σ}}}{\mathrm{\α}}_{t-1}\left(j\right)a_{\mathrm{ji}}\right]b_i\left(o_t\right),& t\≠\mathrm{1,1}\≤i\≤M\end{array}\right.,$ Wherein, λ represents "current" model; Backward variable-definition and be calculated as follows formula 2:

${\mathrm{\β}}_t\left(i\right)\stackrel{\mathrm{\Δ}}{=}P(o_{t+1}^T\text{|}{q}_t=s_i,\mathrm{\λ})=\left\{\begin{array}{cc}1,& t=T,1\≤i\≤M\\ \underset{j=1}{\overset{M}{\mathrm{\Σ}}}{a}_{\mathrm{ij}}{\mathrm{\β}}_{t+1}\left(j\right)b_i\left(o_{t+1}\right),& t\≠T,1\≤i\≤M\end{array}\right.;$

Definition intermediate variable γ _t(i) be that t state is s _iProbability, be calculated as follows formula 3:

${\mathrm{\γ}}_t\left(i\right)\stackrel{\mathrm{\Δ}}{=}P(q_t=s_i|o_1^T,\mathrm{\λ})=\frac{{\mathrm{\α}}_t\left(i\right){\mathrm{\β}}_t\left(i\right)}{{\mathrm{\Σ}}_{i=1}^M{\mathrm{\α}}_t\left(i\right){\mathrm{\β}}_t\left(i\right)};$

Definition intermediate variable ξ _t(i, j) is constantly from state s at t _iJump to s _jProbability, be calculated as follows formula 4:

${\mathrm{\ξ}}_t(i,j)\stackrel{\mathrm{\Δ}}{=}P(q_t=s_i,q_{t+1}=s_j|o_1^T,\mathrm{\λ})=\frac{{\mathrm{\α}}_t\left(i\right)q_{\mathrm{ij}}{b}_j\left(o_{t+1}\right){\mathrm{\β}}_{t+1}\left(j\right)}{{\mathrm{\Σ}}_{i=1}^M{\mathrm{\Σ}}_{j=1}^M{\mathrm{\α}}_t\left(i\right)a_{\mathrm{ij}}{b}_j\left(o_{t+1}\right){\mathrm{\β}}_{t+1}\left(j\right)}.$

In the evaluation problem, the probability that is produced given observation sequence by this model solves following formula 5:

$P(O|\λ)={\mathrm{\Σ}}_{i=1}^M{\mathrm{\α}}_T\left(i\right);$

Most probable produces the status switch Q=q of given observation sequence in the decoding problem ₁q ₂Q _TIn each state can obtain by following formula 6:

$q_t=\underset{1\&le;i\&le;M}{\mathrm{<figure-callout\; id="1"\; label="arg\; max"\; filenames="HDA00002578836200022.png,HDA00002578836200023.png"\; state="\{\{state\}\}">arg</figure-callout>}\max }\left[{\mathrm{\&gamma;}}_t\left(i\right)\right],1\&le;t\&le;T.$

With the model parameter that observation sequence mates most, can estimate by the Baum-Welch algorithm, at first use observation sequence

Calculate intermediate variable γ with "current" model parameter Ω=(A, B, π) _t(i) and ξ _t(i, j), the model parameter that obtains upgrading by formula (3.7-9) again

Calculate "current" model

Produce the probability of given observation sequence

If comparing with model does not before have significant change, namely

(ε is threshold value) thinks that then having restrained model parameter Ω=(A, B, π) is the parameter of mating this observation sequence most; Otherwise, order

Repeat said process, until convergence.Model parameter during convergence is calculated as follows:

${\widehat{\mathrm{\&pi;}}}_m=\frac{{\mathrm{\&gamma;}}_1\left(m\right)}{{\mathrm{\&Sigma;}}_m{\mathrm{\&gamma;}}_1\left(m\right)},$ ${\hat{a}}_{\mathrm{ij}}=\frac{\underset{t=1}{\overset{T-1}{\mathrm{\&Sigma;}}}{\mathrm{\&xi;}}_t(i,j)}{\underset{t=1}{\overset{T-1}{\mathrm{\&Sigma;}}}{\mathrm{\&gamma;}}_t\left(i\right)},$ ${\hat{b}}_i\left(k\right)\text{=}\frac{\underset{s.t.o_t=v_k}{\overset{T}{\underset{t=1,}{\mathrm{\&Sigma;}}}}{\mathrm{\&gamma;}}_t\left(i\right)}{\underset{t=1}{\overset{T}{\mathrm{\&Sigma;}}}{\mathrm{\&gamma;}}_t\left(i\right)}.$

In step S102, it is as follows that the traffic carrying capacity of cloud computing resources is surveyed the mode that obtains the traffic carrying capacity measured value: the description of traffic carrying capacity can with network traffics parameter (comprising the parameters such as access bandwidth, request number, linking number), can gather by the load-balancing device on the cloud computing resource pool uniform outlet, firewall box.

As shown in Figure 3, at system's run duration, as a time slot traffic carrying capacity is carried out record every the t time, every N time slot consists of an observation window, carries out the data processing at each observation window and obtains the traffic carrying capacity measured value, wherein, t is positive integer more than or equal to 1 greater than 0, N;

Carrying out data at each observation window processes and to obtain the traffic carrying capacity measured value and specifically comprise:

1) with i observation window T _iJ time slot be designated as t _{I, j}, corresponding traffic log value is x _{I, j},

Wherein 1≤j≤T works as T _iThe record value of all time slots all record complete after, calculate T _iThe mean value of interior all record values

Wherein, 1≤j≤N;

2) calculate i observation window T _iRecord value mean value X _iWith i-1 observation window T _I-1Record value mean value X _I-1Difference (X ₀Be decided to be 0), as i observation window T _iTraffic carrying capacity measured value the first parameter o _{I, 1}=X _i-X _I-1

3) calculate i observation window T _iThe contrast C ON of interior all record values _i=∑ _{M, n}| m-n|p _{I, mn}, as i observation window T _iTraffic carrying capacity measured value the second parameter estimator value o _{I, 2}=CON _T=∑ _{M, n}| m-n|p _{I, mn}, wherein, minx _{I, j}()≤m, n≤max (x _{I, j}), 1≤j≤N, p _{I, mn}I observation window T _iJ record value be that m and j+1 record value situation that is n is at i observation window T _iThe frequency of middle appearance, p _{I, mn}=# (x _{I, j}=m, x _{I, j+1}=n)/(N-1), and wherein, min (x _{I, j})≤m, n≤max (x _{I, j});

Fig. 4 has listed one and has calculated p _{I, mn}Method, once have the observation window T of 10 time slots _iIn, obtain the traffic log value such as the figure left side, then:

$p_{i,\mathrm{10,10}}=\frac{\#(x_j=10,x_{j+1}=10)}{10-1}=\frac{2}{9}=0.22,$ Insert in the table on figure right side.

4) be disposed after, obtain the measured value O of this observation window _i=(o _{I, 1}, o _{I, 2}), record Oi and record value mean value X _i, because measured value O _iBe a two-dimentional variable, need to quantize in the transfer process of measured value at record value, therefore to o _{I, 2}And o _{I, 2}Be converted to corresponding traffic carrying capacity measured value.To o _{I, 2}And o _{I, 2}The mode of changing, those of ordinary skills can adopt various ways conversion.In the present embodiment, adopt following conversion regime: for the first parameter estimator value o _{I, 1}, predetermined minimum is minO ₁(the actual appearance less than minO ₁Measured value think and be equivalent to minO ₁), maximum is maxO ₁(the actual appearance greater than maxO ₁Measured value think and be equivalent to maxO ₁) to interval [minO ₁, maxO ₁] to be equally divided into length be L ₁M ₁Five equilibrium is for dropping on [minO ₁, minO ₁+ L ₁] the first parameter estimator value o _{I, 1}, think that its corresponding first quantification measured value is v _1,1,, for dropping on [minO ₁+ L ₁* (k-1), minO ₁+ L ₁* k] the first parameter estimator value o _{I, 1}, think that its corresponding first quantification measured value is v _{1, k}, 2≤k≤M wherein ₁In like manner, can preset the second parameter estimator value o _{I, 2}Minimum value be minO ₂, maximum is maxO ₂, to interval [minO ₂, maxO ₂] to be equally divided into length be L ₂M ₂Five equilibrium is in the same way to the second parameter estimator value o _{I, 2}Quantize.Carry out as shown in Figure 6 code conversion after the quantification, measured value is shown in the numerical value in Fig. 6 medium square.Can will be the traffic carrying capacity measured value that variable is converted to the one dimension variable by coding.For example: abscissa is the first parameter estimator value o among Fig. 6 _{I, 1}The quantification scale, ordinate is the second parameter estimator value o _{I, 2}The quantification scale, if the first parameter estimator value o then _{I, 1}In (minO1+1*L1, minO1+2*L1), and the second parameter estimator value o _{I, 2}In (minO2+1*L1, minO2+2*L1), then corresponding traffic carrying capacity measured value is M1+2.

The step S102 of present embodiment can also adopt additive method to realize, does not for example carry out time slot dividing, perhaps do not adopt the modes such as watch window to carry out the measured value collection, those skilled in the art after reading technical scheme of the present invention, carry out the limited number of time test after, also can obtain similar method.

And present embodiment has carried out time slot dividing, and adopt that observation window carries out that data calculate mode, because each time slot is smaller, if each time slot once calculates, then data volume is very huge, therefore, it is divided into observation window, in each observation window, calculate, adopt average and contrast to carry out traffic characteristic and describe, can reflect the changes in flow rate general trend, also can embody the severe degree of changes in flow rate in the certain hour, the calculating of average and contrast all can realize by a forward direction iterative manner, has namely guaranteed the accuracy of data, has also avoided googol to calculate according to amount.

In step S103, the HMM that traffic carrying capacity prediction employing has trained and the measured value of traffic carrying capacity detecting module transmission are carried out state computation.According to formula 6, an e observation window T _e, predicted value O _eCorresponding last time point, corresponding predicted state is calculated as follows:

$q_e=\underset{1\&le;i\&le;M}{\mathrm{<figure-callout\; id="1"\; label="arg\; max"\; filenames="HDA00002578836200022.png,HDA00002578836200023.png"\; state="\{\{state\}\}">arg</figure-callout>}\max }\left[{\mathrm{\&gamma;}}_t\left(i\right)\right]=\arg \max \left[\frac{{\mathrm{\&alpha;}}_e\left(i\right){\mathrm{\&beta;}}_e\left(i\right)}{\underset{1\&le;i\&le;M}{{\mathrm{\&Sigma;}}_{i=1}^{\mathrm{<figure-callout\; id="1"\; label="M"\; filenames="HDA00002578836200022.png,HDA00002578836200023.png"\; state="\{\{state\}\}">M</figure-callout>}}}{\mathrm{\&alpha;}}_e\left(i\right){\mathrm{\&beta;}}_e\left(i\right)}\right]=\arg \max \left[\frac{{\mathrm{\&alpha;}}_e\left(i\right)}{\underset{1\&le;i\&le;M}{{\mathrm{\&Sigma;}}_{i=1}^M{\mathrm{\&alpha;}}_e\left(i\right)}}\right].$

Therefore, only needing to calculate forward variable gets final product.The T that starts in system ₀Constantly begin to calculate forward variable, after this each observation window finishes to upgrade by the iterative manner forward variable.The forward variable of e observation window is calculated as follows:

${\mathrm{\&alpha;}}_e\left(i\right)=\left\{\begin{array}{cc}{\mathrm{\&pi;}}_i{b}_i\left(o_e\right),& e=\mathrm{1,1}\&le;i\&le;M\\ \left[\underset{i=1}{\overset{M}{\mathrm{\&Sigma;}}}{\mathrm{\&alpha;}}_{e-1}\left(j\right)a_{\mathrm{ji}}\right]b_i\left(o_e\right),& e\&NotEqual;\mathrm{1,1}\&le;i\&le;M\end{array}\right.,$ Wherein, o _eBe e observation window T _eThe traffic carrying capacity predicted value.

Therefore, by comparing the forward variable value of each state, can obtain the residing state of current operation amount, corresponding is the state that traffic carrying capacity increases, maintains an equal level, reduces.

In step S104, based on the resource regulating method of predicting as shown in Figure 5, comprising:

Step S510, comparison prediction state and current state, process accordingly according to comparative result:

Reduce state (the expression traffic carrying capacity is on a declining curve), execution in step S520 when being in traffic carrying capacity;

When system is in the fair state of traffic carrying capacity (the expression traffic carrying capacity remains unchanged), do not do any processing, execution in step S530;

When being in traffic carrying capacity, system reduces state (the expression traffic carrying capacity is in rising trend), execution in step S540;

Step S520 has judged whether many virtual machines, if having, and execution in step S521 then, otherwise execution in step S530;

Step S521 closes 1 virtual machine, and the recovery section resource finishes scheduling;

Step S530 maintains the statusquo, and finishes scheduling;

Step S540 judges whether existing resource is enough, if enough, execution in step S541, otherwise execution in step S542;

Step S541 starts virtual machine, finishes scheduling;

Step S542 maintains the statusquo, and carries out alarm, finishes scheduling.

Because system adopts HMM to carry out data prediction, therefore can do sth. in advance budget to the trend of resource, thereby by increasing or reduce virtual machine cloud computing resources be dispatched.

Show the block diagram of the cloud computing resources dispatching patcher based on hidden Markov model of the present invention such as Fig. 2.

System adopts the HMM of a three condition that traffic carrying capacity is predicted, three respectively corresponding traffic carrying capacitys of state increase, traffic carrying capacity maintains an equal level and traffic carrying capacity reduces.In the training stage, train by historical data to obtain a stable forecast model, in the traffic forecast stage, the data that gather by existing network generate measured value, judge the present state of operation system according to forecast model, determine whether needs increase and decrease virtual machine according to predicted state.

A kind of cloud computing resources dispatching patcher 200 based on hidden Markov model comprises:

Model training module 210, be used for hidden Markov model is trained the model parameter that obtains being correlated with, this module is according to the historical image data of actual volume, to the model training, obtain stable model, so that model can mate traffic carrying capacity better, the working model training module upgrades model system's free time;

Traffic carrying capacity detecting module 220, for being surveyed, the traffic carrying capacity of cloud computing resources obtains the traffic carrying capacity measured value, and sending to the traffic carrying capacity prediction module, this module is responsible for collecting the traffic carrying capacity between external user 25 and the resource node 26, traffic data is carried out preliminary treatment form measured value;

Traffic carrying capacity prediction module 230, be used for that the traffic carrying capacity measured value that 220 detections of traffic carrying capacity detecting module obtain is inputted described hidden Markov model and carry out state computation, obtain predicted state and send to the scheduling of resource module, the model parameter that described hidden Markov model adopts model training module 210 to obtain, this module is responsible for measured value is analyzed, and according to forecast model the traffic carrying capacity of next stage is predicted;

Scheduling of resource module 240, be used for according to the predicted state that obtains from the traffic carrying capacity prediction module cloud computing resources being dispatched, this module is dispatched with the applicable present situation of resource according to predicting the outcome of traffic carrying capacity prediction module 230, specifically comprise newly-increased virtual machine, reduce virtual machine, and virtual machine (vm) migration etc.

Traffic carrying capacity detecting module 220 comprises:

Record sub module 221 is used for as a time slot traffic carrying capacity being carried out record every the t time;

Traffic carrying capacity measured value calculating sub module 222 is used for every N time slot and consists of an observation window, carries out the data processing at each observation window and obtains the traffic carrying capacity measured value, specifically is used for:

With i observation window T _iJ time slot be designated as t _{I, j}, corresponding traffic log value is x _{I, j}, wherein 1≤j≤N calculates T _iThe mean value of interior all record values

Wherein, 1≤j≤N;

Calculate i observation window T _iRecord value mean value X _iWith i-1 observation window T _I-1Record value mean value X _I-1Difference, as i observation window T _iTraffic carrying capacity measured value the first parameter o _{I, 1}=X _i-X _I-1

Calculate i observation window T _iThe contrast C ON of interior all record values _i=∑ _{M, n}| m-n|p _{I, mn}, as i observation window T _iTraffic carrying capacity measured value the second parameter estimator value o _{I, 2}, wherein, minx _{I, j}()≤m, n≤max (x _{I, j}), 1≤j≤T, p _{I, mn}I observation window T _iJ record value be that m and j+1 record value situation that is n is at i observation window T _iThe frequency of middle appearance, p _{I, mn}=# (x _{I, j}=m, x _{I, j+1}=n)/(T-1), and wherein, min (x _{I, j})≤m, n≤max (x _{I, j});

To o _{I, 2}And o _{I, 2}Be converted to corresponding traffic carrying capacity measured value and be sent to traffic carrying capacity prediction module 230.

Described traffic carrying capacity prediction module 230 specifically is used for:

Calculate e observation window T _ePredicted state q _e:

Wherein, a _e(i) a _eBe e observation window T _eI forward variable, and: ${\mathrm{\&alpha;}}_e\left(i\right)=\left\{\begin{array}{cc}{\mathrm{\&pi;}}_i{b}_i\left(o_e\right),& e=\mathrm{1,1}\&le;i\&le;M\\ \left[\underset{i=1}{\overset{M}{\mathrm{\&Sigma;}}}{\mathrm{\&alpha;}}_{e-1}\left(j\right)a_{\mathrm{ji}}\right]b_i\left(o_e\right),& e\&NotEqual;\mathrm{1,1}\&le;i\&le;M\end{array}\right.,$ Wherein, o _eBe e observation window T _eThe traffic carrying capacity predicted value.

Scheduling of resource module 240 comprises:

Traffic carrying capacity reduces scheduling sublayer module 241, and being used for predicted state is that traffic carrying capacity reduces state, reclaims cloud computing resources;

Traffic carrying capacity increases scheduling sublayer module 242, and being used for predicted state is that traffic carrying capacity increases state, increases cloud computing resources;

The traffic carrying capacity scheduling sublayer module 243 that maintains an equal level is used for predicted state and is the traffic carrying capacity state that maintains an equal level, and keeps cloud computing resources.

The HMM model that utilization of the present invention is comparatively ripe is delineated traffic carrying capacity, and utilizes the trend prediction method of HMM traffic carrying capacity is predicted and to be dispatched, and can carry out forward scheduling, can more effectively utilize resource.

The above embodiment has only expressed several execution mode of the present invention, and it describes comparatively concrete and detailed, but can not therefore be interpreted as the restriction to claim of the present invention.Should be pointed out that for the person of ordinary skill of the art without departing from the inventive concept of the premise, can also make some distortion and improvement, these all belong to protection scope of the present invention.Therefore, the protection range of patent of the present invention should be as the criterion with claims.

Claims (10)

1. the cloud computing resource scheduling method based on hidden Markov model is characterized in that, uses hidden Markov model the traffic carrying capacity situation of change to be described and to predict described method comprises:

Hidden Markov model is trained the model parameter that obtains being correlated with;

The traffic carrying capacity of cloud computing resources surveyed obtain the traffic carrying capacity measured value;

The traffic carrying capacity measured value that detection is obtained is inputted described hidden Markov model and is carried out state computation, obtains predicted state, and described hidden Markov model adopts the model parameter that obtains in the model training step;

According to described predicted state cloud computing resources is dispatched.

2. the cloud computing resource scheduling method based on hidden Markov model according to claim 1 is characterized in that, the traffic carrying capacity of described cloud computing resources is the network traffics parameter.

3. the cloud computing resource scheduling method based on hidden Markov model according to claim 1 is characterized in that, described traffic carrying capacity prediction steps comprises:

As a time slot traffic carrying capacity is carried out record every the t time, every N time slot consists of an observation window, carries out data at each observation window and processes and obtain the traffic carrying capacity measured value, and wherein, t is positive integer more than or equal to 1 greater than 0, N.

4. according to right 3 described cloud computing resource scheduling methods based on hidden Markov model, it is characterized in that, carry out data at each observation window and process and to obtain the traffic carrying capacity measured value and specifically comprise:

Calculate the mean value of all record values in the current observation window as the first mean value;

More than the mean value of all record values in the observation window as the second mean value, calculate the difference of the first mean value and the second mean value, as traffic carrying capacity measured value first parameter of current observation window;

Calculate the contrast of all record values in the current observation window, as traffic carrying capacity measured value second parameter of current observation window, described contrast C ON _i=∑ _{M, n}| m-n|p _Mn, wherein, p _MnThe record value that is current observation window is the frequency that situation that m and a upper record value are n occurs in current observation window; Traffic carrying capacity measured value the first parameter and traffic carrying capacity measured value the second parameter are converted to corresponding traffic carrying capacity measured value.

5. the cloud computing resource scheduling method based on hidden Markov model according to claim 4 is characterized in that:

Described model parameter comprises: state set, the set of measured value value, state transition probability matrix, measured value produce probability matrix and initial probability distribution matrix;

Described state set S={s ₁, s ₂, s ₃, s wherein ₁The expression traffic carrying capacity increases state, s ₂The fair state of expression traffic carrying capacity, s ₃The expression traffic carrying capacity reduces state;

Described measured value value set V={v ₁, v ₂..., v _k..., v _K, K 〉=1, v _kBe k measured value;

Described state transition probability matrix A={a _Mn, a wherein _MnExpression is from state s _mJump to state s _nProbability;

Described measured value produces probability matrix B={b _m(k) }, b wherein _m(k) be illustrated in state s _mLower generation measured value v _kProbability;

Described initial probability distribution matrix π={ π _m, π wherein _mThe expression initial condition is s _mProbability;

Described traffic carrying capacity prediction steps specifically comprises:

Calculate e observation window T _ePredicted state q _e:

Wherein, a _e(i) be e observation window T _eI forward variable, and: ${\mathrm{\&alpha;}}_e\left(i\right)=\left\{\begin{array}{cc}{\mathrm{\&pi;}}_i{b}_i\left(o_e\right),& e=\mathrm{1,1}\&le;i\&le;M\\ \left[\underset{i=1}{\overset{M}{\mathrm{\&Sigma;}}}{\mathrm{\&alpha;}}_{e-1}\left(j\right)a_{\mathrm{ji}}\right]b_i\left(o_e\right),& e\&NotEqual;\mathrm{1,1}\&le;i\&le;M\end{array}\right.,$ Wherein, o _eBe e observation window T _eThe traffic carrying capacity predicted value.

6. the cloud computing resource scheduling method based on hidden Markov model according to claim 5 is characterized in that, described scheduling of resource step specifically comprises:

Reduce state if predicted state is traffic carrying capacity, then reclaim cloud computing resources;

Increase state if predicted state is traffic carrying capacity, then increase cloud computing resources;

If predicted state the is traffic carrying capacity state that maintains an equal level is then kept cloud computing resources.

7. the cloud computing resources dispatching patcher based on hidden Markov model is characterized in that, comprising:

The model training module is used for hidden Markov model is trained the model parameter that obtains being correlated with;

The traffic carrying capacity detecting module obtains the traffic carrying capacity measured value for the traffic carrying capacity of cloud computing resources is surveyed, and sends to the traffic carrying capacity prediction module;

The traffic carrying capacity prediction module, be used for that the traffic carrying capacity measured value that the detection of traffic carrying capacity detecting module obtains is inputted described hidden Markov model and carry out state computation, obtain predicted state and send to the scheduling of resource module, the model parameter that described hidden Markov model adopts the model training module to obtain;

The scheduling of resource module is used for according to the predicted state that obtains from the traffic carrying capacity prediction module cloud computing resources being dispatched.

8. the cloud computing resources dispatching patcher based on hidden Markov model according to claim 7 is characterized in that:

Described traffic carrying capacity detecting module comprises:

Record sub module is used for as a time slot traffic carrying capacity being carried out record every the t time;

Traffic carrying capacity measured value calculating sub module is used for every N time slot and consists of an observation window, carries out the data processing at each observation window and obtains the traffic carrying capacity measured value, specifically is used for:

Calculate the mean value of all record values in the current observation window as the first mean value;

More than the mean value of all record values in the observation window as the second mean value, calculate the difference of the first mean value and the second mean value, as traffic carrying capacity measured value first parameter of current observation window;

Calculate the contrast of all record values in the current observation window, as traffic carrying capacity measured value second parameter of current observation window, described contrast C ON _i=∑ _{M, n}| m-n|p _Mn, p _MnThe record value that is current observation window is the frequency that situation that m and a upper record value are n occurs in current observation window; Traffic carrying capacity measured value the first parameter and traffic carrying capacity measured value the second parameter are converted to corresponding traffic carrying capacity measured value and are sent to the traffic carrying capacity prediction module.

9. the cloud computing resources dispatching patcher based on hidden Markov model according to claim 8 is characterized in that:

Described model parameter comprises: state set, the set of measured value value, state transition probability matrix, measured value produce probability matrix and initial probability distribution matrix;

Described state set S={s ₁, s ₂, s ₃, s wherein ₁The expression traffic carrying capacity increases state, s ₂The fair state of expression traffic carrying capacity, s ₃The expression traffic carrying capacity reduces state;

Described measured value value set V={v ₁, v ₂... v _K, K 〉=1, v _kBe k measured value;

Described state transition probability matrix A={a _Mn, a wherein _MnExpression is from state s _mJump to state s _nProbability;

Described measured value produces probability matrix B={b _m(k) }, b wherein _m(k) be illustrated in state s _mLower generation measured value v _kProbability;

Described initial probability distribution matrix π={ π _m, π wherein _mThe expression initial condition is s _mProbability;

Described traffic carrying capacity prediction module specifically is used for:

Calculate e observation window T _ePredicted state q _e:

Wherein, a _e(i) be e observation window T _eI forward variable, and: ${\mathrm{\&alpha;}}_e\left(i\right)=\left\{\begin{array}{cc}{\mathrm{\&pi;}}_i{b}_i\left(o_e\right),& e=\mathrm{1,1}\&le;i\&le;M\\ \left[\underset{i=1}{\overset{M}{\mathrm{\&Sigma;}}}{\mathrm{\&alpha;}}_{e-1}\left(j\right)a_{\mathrm{ji}}\right]b_i\left(o_e\right),& e\&NotEqual;\mathrm{1,1}\&le;i\&le;M\end{array}\right.,$ Wherein, o _eBe e observation window T _eThe traffic carrying capacity predicted value.

10. the cloud computing resources dispatching patcher based on hidden Markov model according to claim 9 is characterized in that, described scheduling of resource module comprises:

Traffic carrying capacity reduces the scheduling sublayer module, and being used for predicted state is that traffic carrying capacity reduces state, reclaims cloud computing resources;

Traffic carrying capacity increases the scheduling sublayer module, and being used for predicted state is that traffic carrying capacity increases state, increases cloud computing resources;

The traffic carrying capacity scheduling sublayer module that maintains an equal level is used for predicted state and is the traffic carrying capacity state that maintains an equal level, and keeps cloud computing resources.

Images