CN104811491A - Cloud computing resource scheduling method based on genetic algorithm - Google Patents

Abstract

The invention discloses a cloud computing resource scheduling method based on a genetic algorithm. A scheduling scheme in resources in cloud computation is obtained by using the genetic algorithm; each gene in chromosomes of the genetic algorithm is a sub-task of a cloud task; site sequence numbers of the chromosomes represent numbers distributed to a virtual machine by each sub-task. According to the method, predicated Qos is set according to the cloud task and the chromosomes meeting the predicated Qos are remained in a population iteration updating process and the chromosomes which do not meet the predicated Qos are discarded, and the current chromosomes with a high population fitness value are directly copied to the next generation of population according to the selected possibility; then, crossed and mutation operation is carried out on the residual chromosomes of the current population and the quality of the chromosomes in the population is guaranteed; the resources of the cloud computation are used according to the requirements and elastically expanded; the cloud computing resource scheduling arrangement is done in real time and optimal resource construction and adjustment strategies are decided; under the condition that the service quality of a cloud computing network is guaranteed, the effectiveness ratio of the cloud computation is improved.

Description

A kind of cloud computing resource scheduling method based on genetic algorithm

Technical field

The invention belongs to field of cloud calculation, particularly a kind of cloud computing resource scheduling method based on genetic algorithm.

Background technology

In recent years, cloud computing, as a kind of novel information processing manner, has been deep in the every field of our life and work.Cloud computing take Intel Virtualization Technology as technical support, the most basic theory is exactly use as required, and in the process, affect the critical problem of cloud computing performance because the energy consumption of cloud data center and resource thereof provide efficiency to become, how effective and reasonable the resource of use cloud computing also become a key point.

Along with Internet era information and date quick growth, science, engineering and business calculate field to be needed to process data that are extensive, magnanimity, to the demand of the computing capability computing capability far beyond self IT architecture, at this moment just need to continue to increase the extensibility that system hardware drops into the system that realizes.In addition, due to the limitation of Traditional parallel programming model application, the new multiple programming framework of a kind of easy study, use, deployment is objectively required.In this case, in order to save cost and realize the scalability of system, the concept of cloud computing is carried out.Cloud computing is Distributed Calculation, the further developing of parallel processing and grid computing, and it is the calculating based on the Internet, can provide the system of hardware service, infrastructure services, platform service, software service, stores service to various internet, applications.The mechanism that usual cloud system is had by third party provides service, and user is only concerned about the service that cloud provides.At present about the definition that cloud computing system is ununified, cloud computing supplier releases relevant cloud computing strategy according to oneself business event.USA National Institute of Standard and Technology (NIST) gives the cloud computing definition of authority at present: (1) cloud computing is a kind of pattern can being accessed a configurable computing resource sharing pond (comprising network, server, storage, application and service etc.) by network in mode easily, as required, this resource-sharing pond with minimum administration overhead and minimum mutual with supplier, can configure rapidly, provides or releasing resource; (2) cloud computing mode has 5 essential characteristics: Self-Service as required, widely access to netwoks, shared resource pool, fast elasticity capacity, measurable service, also comprise 3 kinds of service modes: namely software serve (SaaS), namely platform serves (PaaS), namely infrastructure serve (IaaS), and 4 kinds of deployment way: privately owned cloud, community's cloud, publicly-owned cloud, mixed cloud.

The customer group of the service orientation that cloud computing provides is huge, therefore in " cloud ", task quantity is huge, the personage of system all out magnanimity all the time, so resource management is a critical problem of cloud computing, its scheduling strategy and algorithm directly affect performance and the cost of cloud system.Along with user is to the continuous growth of cloud computing demand, cloud data center scale is day by day huge, and its energy consumption problem is also more and more serious.According to reported literature, global data center added about 56% at 2005 ~ 2010, and the energy consumption at U.S. Data center adds 36%.The electric energy that the data center of 50,000 nodes consumes every year is more than 100,000,000 kilowatts, and energy consumption account in data center's O&M cost 40%.At home, the data center of common carrier is power consumption rich and influential family, data center of China Telecom year power consumption 11.2 hundred million kilowatt hour, data center of CHINAUNICOM be 9.9 hundred million kilowatt hours to the year two thousand twenty, the energy consumption of the main cloud computing operator in the whole world will close to 2,000,000,000,000 kilowatt hours.Therefore, the resource that how effectively to utilize studied under cloud computing environment has urgent demand.

Summary of the invention

The object of the invention is to overcome the shortcoming of prior art and deficiency, a kind of cloud computing resource scheduling method based on genetic algorithm is provided.The method can make the scheduling arrangement of cloud computing resources in real time, and decision-making goes out best resource construction and adjustable strategies, under the prerequisite ensureing system for cloud computing service quality, improves the effectiveness ratio of cloud computing.

Object of the present invention is achieved through the following technical solutions: a kind of cloud computing resource scheduling method based on genetic algorithm, and step is as follows:

S1, by user submit to cloud task be cut into several subtasks, and according to cloud task arrange one expection Qos (Quality of Service, service quality); When the Qos of expection refers to that user obtains cloud computing resources, a requirement of institute's spended time and expense;

S2, initialization: initialization population scale, and population maximum iteration time is set;

S3, generation initial population: according to coding rule stochastic generation chromosome, by abstract for cloud task be chromosomal coding input, each gene in chromosome is the subtask in this cloud task, and chromosomal position sequence number represents that the representative of each subtask is assigned to the number of virtual machine; Then judge whether chromosome meets the Qos of expection, if meet, is then joined in initial population by this chromosome, if do not meet, then abandons this chromosome, until reach the scale of initial population;

S4, calculate the chromosomal fitness value of every bar in current population according to fitness function;

S5, selection, crossover and mutation operation: select the chromosome that fitness value is high, by these chromosome replications to new population of future generation according to select probability; And carry out crossover and mutation operation for chromosome remaining in current population;

Whether the chromosome in S6, determining step S5 after crossover and mutation operation meets the Qos of expection, if meet, then joins in new population of future generation, if do not meet, then abandons, until new population reaches population scale, and and population iterations adds 1;

S7, judge whether population iterations reaches maximum iteration time, if not, then return step S4, if so, then enter step S8;

S8, using chromosome the highest for fitness value in the new population that finally obtains as optimal solution, and carry out for this chromosome the number that decode operation obtains virtual machine, using the optimal solution of this chromosome as scheduling of resource.

Preferably, the coded system based on path is adopted to carry out chromosome coding in described step S3.

Preferably, each chromosomal fitness in population is calculated by following auto-adaptive function in described step S4:

f(X _k)＝1/(lg(EC _total+1))；

Wherein EC _totalfor chromosome x _kin the total energy consumption of resource scheduling scheme, f (X _k) be chromosome x _kfitness.

Further, described chromosome x _kin the total energy consumption EC of resource scheduling scheme _totalfor:

${\mathrm{EC}}_{\mathrm{total}}=\underset{j=1}{\overset{N}{\mathrm{\Σ}}}({\mathrm{DEC}}_j+{\mathrm{IEC}}_j+{\mathrm{WEC}}_j);$

Wherein WEC _jfor the operating power consumption of host j in one-period, m _jfor the subtask number that host j runs; DEC _jfor the energy consumption of host j in one-period during resting state; IEC _jfor the energy consumption during host j free time; N is the number of host;

${\mathrm{WEC}}_j=\underset{i=1}{\overset{m_j}{\mathrm{\Σ}}}({\mathrm{CEC}}_{\mathrm{ij}}+{\mathrm{SEC}}_{\mathrm{ij}});$

Wherein CEC _ijfor:

CEC _ij＝CT _ij×PC _j＝CI _i/CS _j×PC _j；

CT _ijrepresent the power consumption of subtask i on host j, CI _irepresent the fill order number of subtask i on host j, CS _jrepresent the CPU processing speed of host j, PC _jrepresent power consumption during host j work;

Wherein SEC _ijfor:

SEC _ij＝SD _i/SS _j×PS _j；

SEC _ijrepresent the storage energy consumption of subtask i on host j, SD _irepresent the data volume of i required read-write in subtask on host j, SS _jrepresent host j disk reading rate, PS _jrepresent that host j stores power consumption.

Further, in the selection operation in described step S5, select probability adopts wheel disc operator, and in population, each chromosomal select probability is:

$P_k=f\left(X_k\right)/\underset{X_k=1}{\overset{\mathrm{popSize}}{\mathrm{\Σ}}}f\left(X_k\right);$

Wherein popSize is the scale of population.

Preferably, the mutation operation in described step S5, adopts and replaces variation mode, first from parent chromosome, selects a sub-bit string, and then in remaining bit string, selects a position at random, and inserts this sub-bit string.

Preferably, the mutation operation in described step S5 adopts semi-match method, and Stochastic choice two crosspoints, the position between two points will intersect, and other positions are copied.

Preferably, in the Qos of the expection arranged in described step S1, in each host, the time of implementation of each subtask meets the following conditions:

TE _ij＝CT _ij+ST _ij；

TE _ijrepresent the operation total time of subtask i on host j, CT _ijrepresent the task execution time of subtask i on host j, ST _ijrepresent the time that host j reads and writes data;

The time of implementation T of all subtasks in host j _jmeet the following conditions:

$T_j=\underset{i=1}{\overset{m_j}{\mathrm{\Σ}}}{\mathrm{TE}}_{\mathrm{ij}};$

M _jfor the quantity of subtask in host j;

The time T of cloud tasks carrying _totalmeet the following conditions:

$T_{\mathrm{total}}=\underset{j=1}{\overset{N}{\mathrm{\Σ}}}{T}_j;$

N is the number of host;

In the Qos of the expection arranged in described step S1, expense meets the following conditions:

${\cos t}_{\mathrm{ij}}=(\underset{i=1}{\overset{m_j}{\mathrm{\Σ}}}\mathrm{sRe}{q}_i\×s{\mathrm{Price}}_j)\×{\mathrm{TE}}_{\mathrm{ij}};$

Wherein cost _ijrepresent expense needed for subtasking i in host j, s Req _iwhat represent is the resource that host j subtask i is corresponding, s Price _jwhat represent is expense needed for resource that in host j, subtask i is corresponding.

Preferably, in described step S1, adopt the MapReduce programming model in cloud computing, the cloud task that user submits to is cut into several subtasks.

The present invention has following advantage and effect relative to prior art:

(1) the inventive method utilizes genetic algorithm to get resource scheduling scheme optimum in cloud computing, the Qos of expection is set according to cloud task in the inventive method, wherein this expection Qos for user obtain cloud computing resources time, a requirement of institute's spended time and expense; In population iteration renewal process, the chromosome of the Qos of satisfied expection is stayed, the chromosome of the Qos not meeting expection is abandoned, and according to select probability by chromosomal inheritance high for fitness value in population of future generation, ensure that chromosomal quality in population, the resource achieving cloud computing uses as required, the feature of resilient expansion, the scheduling arrangement of cloud computing resources can be made in real time, decision-making goes out best resource construction and adjustable strategies, under the prerequisite ensureing system for cloud computing service quality, improve the effectiveness ratio of cloud computing.

(2) chromosomal fitness is calculated by the auto-adaptive function relevant to the resource scheduling scheme total energy consumption in chromosome in the inventive method, this fitness function can evaluate chromosomal good and bad state, its fitness value is larger, represent that chromosomal fitness is better, otherwise it is poorer, can select by this fitness function the chromosome that realistic optimum scheduling of resource requires more accurately, further ensuring the resource scheduling scheme that genetic algorithm finally obtains is optimal solution.

Accompanying drawing explanation

Fig. 1 is the inventive method flow chart.

Embodiment

Below in conjunction with embodiment and accompanying drawing, the present invention is described in further detail, but embodiments of the present invention are not limited thereto.

Embodiment

As shown in Figure 1, present embodiment discloses a kind of cloud computing resource scheduling method based on genetic algorithm, step is as follows:

The MapReduce programming model of S1, employing cloud computing, is cut into several subtasks by the cloud task that user submits to, and arranges the Qos of an expection according to cloud task; When the Qos of expection refers to that user obtains cloud computing resources, a requirement of institute's spended time and expense.

S2, initialization: initialization population scale popSize, and population maximum iteration time Tmax is set.

S3, generation initial population: according to coding rule stochastic generation chromosome, by abstract for cloud task be chromosomal coding input, each gene in chromosome is the subtask in this cloud task, and chromosomal position sequence number represents that each subtask is assigned to the number of virtual machine; Then judge whether chromosome meets the Qos of expection, if meet, is then joined in initial population by this chromosome, if do not meet, then abandons this chromosome, until reach the scale of initial population.The coded system that have employed in the present embodiment based on path carries out chromosomal coding, can certainly adopt based on binary system, based on matrix, based on adjacent, based on the coded system of index etc.

S4, calculate the chromosomal fitness of every bar in current population according to fitness function; Wherein calculate each chromosomal fitness f (X in population by following auto-adaptive function in this step _k):

f(X _k)＝1/(lg(EC _total+1))；

Wherein EC _totalfor chromosome x _kin the total energy consumption of resource scheduling scheme; Fitness function have rated chromosomal quality, and its fitness value is larger, represents that chromosomal fitness is better, otherwise poorer.EC in this step _totalless, fitness value is larger.

Chromosome x _kin the total energy consumption EC of resource scheduling scheme _totalfor:

${\mathrm{EC}}_{\mathrm{total}}=\underset{j=1}{\overset{N}{\mathrm{\Σ}}}({\mathrm{DEC}}_j+{\mathrm{IEC}}_j+{\mathrm{WEC}}_j);$

Wherein WEC _jfor the operating power consumption of host j in one-period, m _jfor the subtask number that host j runs; DEC _jfor the energy consumption of host j in one-period during resting state; IEC _jfor the energy consumption during host j free time; N is the number of host;

${\mathrm{WEC}}_j=\underset{i=1}{\overset{m_j}{\mathrm{\Σ}}}({\mathrm{CEC}}_{\mathrm{ij}}+{\mathrm{SEC}}_{\mathrm{ij}});$

Wherein CEC _ijfor:

CEC _ij＝CT _ij×PC _j＝CI _i/CS _j×PC _j；

CT _ijrepresent the power consumption of subtask i on host j, CI _irepresent the fill order number of subtask i on host j, CS _jrepresent the CPU processing speed of host j, PC _jrepresent power consumption during host j work;

Wherein SEC _ijfor:

SEC _ij＝SD _i/SS _j×PS _j；

SEC _ijrepresent the storage energy consumption of subtask i on host j, SD _irepresent the data volume of i required read-write in subtask on host j, SS _jrepresent host j disk reading rate, PS _jrepresent that host j stores power consumption.

S5, selection, crossover and mutation operation: select the chromosome that fitness value is high, by chromosome replication high for these fitness to new population of future generation according to select probability; And carry out crossover and mutation operation for chromosome remaining in current population.

Select select probability in operation to adopt wheel disc operator in this step, the probability that in population, chromosome is selected is directly proportional to its fitness; In population, each chromosomal select probability is:

$P_k=f\left(X_k\right)/\underset{X_k=1}{\overset{\mathrm{popSize}}{\mathrm{\Σ}}}f\left(X_k\right);$

Wherein popSize is the scale of population, f (X _k) be chromosome x _kfitness.

Mutation operation in this step, adopts and replaces variation mode, first from parent chromosome, selects a sub-bit string, and then in remaining bit string, selects a position at random, and inserts this sub-bit string.In certain the present embodiment, mutation operation also can adopt exchange mutation (EM), insertion variation (IM), simple inversion variation (SIM), and inversion makes a variation (IVM), fight for the variation modes such as variation (SM).

In this step, mutation operation adopts semi-match method, and Stochastic choice two crosspoints, the position between two points will intersect, and other positions are copied.In certain the present embodiment, interlace operation also can adopt semi-match method (PMX), recycling cross method (CX), order interior extrapolation method (OX), location-based interior extrapolation method (POS) etc.

Whether the chromosome in S6, determining step S5 after crossover and mutation operation meets the Qos of expection, if meet, then join in new population of future generation, if do not meet, then abandon, until new population reaches population scale popSize, and population iterations T adds 1.

S7, judge whether population iterations reaches maximum iteration time Tmax, if not, then return step S4, if so, then enter step S8.

S8, using chromosome the highest for fitness value in the new population that finally obtains as optimal solution, and carry out for this chromosome the number that decode operation obtains virtual machine, using the optimal solution of this chromosome as scheduling of resource.

The Qos of cloud computing refers to, when consumer obtains cloud computing resources, wish the time, the indexs such as cost can meet the re-set target of consumer.When consumer relies on resource provider to meet its computation requirement, in order to mate their requirement, user task multidimensional Qos demand (time, cost) must be guaranteed.

The energy consumption of material resources equals the product of power consumption and time.Resource power consumption in varied situations is also different, and physical host is divided into dormancy usually, idle and work three states.In the Qos of the expection arranged in the present embodiment above-mentioned steps S1, in each host, the time of implementation of each subtask meets the following conditions:

TE _ij＝CT _ij+ST _ij；

TE _ijrepresent the operation total time of subtask i on host j, CT _ijrepresent the task execution time of subtask i on host j, ST _ijrepresent the time that host j reads and writes data;

The time of implementation T of all subtasks in host j _jmeet the following conditions:

$T_j=\underset{i=1}{\overset{m_j}{\mathrm{\Σ}}}{\mathrm{TE}}_{\mathrm{ij}};$

M _jfor the quantity of subtask in host j;

The time T of cloud tasks carrying _totalmeet the following conditions:

$T_{\mathrm{total}}=\underset{j=1}{\overset{N}{\mathrm{\Σ}}}{T}_j;$

N is the number of host;

In the Qos of the expection arranged in described step S1, expense meets the following conditions:

${\cos t}_{\mathrm{ij}}=(\underset{i=1}{\overset{m_j}{\mathrm{\Σ}}}\mathrm{sRe}{q}_i\×s{\mathrm{Price}}_j)\×{\mathrm{TE}}_{\mathrm{ij}};$

Wherein cost _ijrepresent expense needed for subtasking i in host j, s Req _iwhat represent is the resource that host j subtask i is corresponding, s Price _jwhat represent is expense needed for resource that in host j, subtask i is corresponding.

Resource scheduling scheme is exactly a balance of time and expense, and the present embodiment draws a time of expecting and the expense Qos as cloud task according to user at the cloud task execution time and expense that use above-mentioned demand fulfillment in resource process.

Above-described embodiment is the present invention's preferably execution mode; but embodiments of the present invention are not restricted to the described embodiments; change, the modification done under other any does not deviate from Spirit Essence of the present invention and principle, substitute, combine, simplify; all should be the substitute mode of equivalence, be included within protection scope of the present invention.

Claims (9)

1. based on a cloud computing resource scheduling method for genetic algorithm, it is characterized in that, step is as follows:

S1, by user submit to cloud task be cut into several subtasks, and according to cloud task arrange one expection Qos; When the Qos of expection refers to that user obtains cloud computing resources, a requirement of institute's spended time and expense;

S2, initialization: initialization population scale, and population maximum iteration time is set;

S3, generation initial population: according to coding rule stochastic generation chromosome, by abstract for cloud task be chromosomal coding input, each gene in chromosome is the subtask in this cloud task, and chromosomal position sequence number represents that the representative of each subtask is assigned to the number of virtual machine; Then judge whether chromosome meets the Qos of expection, if meet, is then joined in initial population by this chromosome, if do not meet, then abandons this chromosome, until reach the scale of initial population;

S4, calculate the chromosomal fitness value of every bar in current population according to fitness function;

S5, selection, crossover and mutation operation: select the chromosome that fitness value is high, by these chromosome replications to new population of future generation according to select probability; And carry out crossover and mutation operation for chromosome remaining in current population;

Whether the chromosome in S6, determining step S5 after crossover and mutation operation meets the Qos of expection, if meet, then joins in new population of future generation, if do not meet, then abandons, until new population reaches population scale, and and population iterations adds 1;

S7, judge whether population iterations reaches maximum iteration time, if not, then return step S4, if so, then enter step S8;

S8, using chromosome the highest for fitness value in the new population that finally obtains as optimal solution, and carry out for this chromosome the number that decode operation obtains virtual machine, using the optimal solution of this chromosome as scheduling of resource.

2. the cloud computing resource scheduling method based on genetic algorithm according to claim 1, is characterized in that, adopts the coded system based on path to carry out chromosome coding in described step S3.

3. the cloud computing resource scheduling method based on genetic algorithm according to claim 1, is characterized in that, calculates each chromosomal fitness in population in described step S4 by following auto-adaptive function:

$f\left(X_k\right)=1/\left(\lg ({\mathrm{EC}}_{\mathrm{total}}+1)\right);$

Wherein EC _totalfor chromosome x _kin the total energy consumption of resource scheduling scheme, f (X _k) be chromosome x _kfitness.

4. the cloud computing resource scheduling method based on genetic algorithm according to claim 3, is characterized in that, described chromosome x _kin the total energy consumption EC of resource scheduling scheme _totalfor:

${\mathrm{EC}}_{\mathrm{total}}=\underset{j=1}{\overset{N}{\mathrm{\Σ}}}({\mathrm{DEC}}_j+{\mathrm{IEC}}_j+{\mathrm{WEC}}_j);$

Wherein WEC _jfor the operating power consumption of host j in one-period, m _jfor the subtask number that host j runs; DEC _jfor the energy consumption of host j in one-period during resting state; IEC _jfor the energy consumption during host j free time; N is the number of host;

${\mathrm{WEC}}_j=\underset{i=1}{\overset{m_j}{\mathrm{\Σ}}}({\mathrm{CEC}}_{\mathrm{ij}}+{\mathrm{SEC}}_{\mathrm{ij}});$

Wherein CEC _ijfor:

CEC _ij＝CT _ij×PC _j＝CI _i/CS _j×PC _j；

CT _ijrepresent the power consumption of subtask i on host j, CI _irepresent the fill order number of subtask i on host j, CS _jrepresent the CPU processing speed of host j, PC _jrepresent power consumption during host j work;

Wherein SEC _ijfor:

SEC _ij＝SD _i/SS _j×PS _j；

SEC _ijrepresent the storage energy consumption of subtask i on host j, SD _irepresent the data volume of i required read-write in subtask on host j, SS _jrepresent host j disk reading rate, PS _jrepresent that host j stores power consumption.

5. the cloud computing resource scheduling method based on genetic algorithm according to claim 3, is characterized in that, in the selection operation in described step S5, select probability adopts wheel disc operator, and in population, each chromosomal select probability is:

$P_k=f\left(X_k\right)/\underset{X_k=1}{\overset{\mathrm{popSize}}{\mathrm{\Σ}}}f\left(X_k\right);$

Wherein popSize is the scale of population.

6. the cloud computing resource scheduling method based on genetic algorithm according to claim 1, it is characterized in that, mutation operation in described step S5, adopt and replace variation mode, first from parent chromosome, select a sub-bit string, and then in remaining bit string, select a position at random, and insert this sub-bit string.

7. the cloud computing resource scheduling method based on genetic algorithm according to claim 1, it is characterized in that, the mutation operation in described step S5 adopts semi-match method, Stochastic choice two crosspoints, position between two points will intersect, and other positions are copied.

8. the cloud computing resource scheduling method based on genetic algorithm according to claim 1, is characterized in that, in the Qos of the expection arranged in described step S1, in each host, the time of implementation of each subtask meets the following conditions:

TE _ij＝CT _ij+ST _ij；

TE _ijrepresent the operation total time of subtask i on host j, CT _ijrepresent the task execution time of subtask i on host j, ST _ijrepresent the time that host j reads and writes data;

The time of implementation T of all subtasks in host j _jmeet the following conditions:

$T_j=\underset{i=1}{\overset{m_j}{\mathrm{\Σ}}}{\mathrm{TE}}_{\mathrm{ij}};$

M _jfor the quantity of subtask in host j;

The time T of cloud tasks carrying _totalmeet the following conditions:

$T_{\mathrm{total}}=\underset{j=1}{\overset{N}{\mathrm{\Σ}}}{T}_j;$

N is the number of host;

In the Qos of the expection arranged in described step S1, expense meets the following conditions:

${\cos t}_{\mathrm{ij}}=(\underset{i=1}{\overset{m_j}{\mathrm{\Σ}}}\mathrm{sRe}{q}_i\×{\mathrm{sPrice}}_j)\×{\mathrm{TE}}_{\mathrm{ij}};$

Wherein cost _ijrepresent expense needed for subtasking i in host j, sReq _iwhat represent is the resource that host j subtask i is corresponding, sPrice _jwhat represent is expense needed for resource that in host j, subtask i is corresponding.

9. the cloud computing resource scheduling method based on genetic algorithm according to claim 1, is characterized in that, adopts the MapReduce programming model in cloud computing in described step S1, and the cloud task that user submits to is cut into several subtasks.