CN105141697A - Multi-QoS constrained cloud computing task scheduling method - Google Patents

Abstract

The invention belongs to the cloud computing technical field and relates to a multi-QoS constrained cloud computing task scheduling method. According to the method, required QoS descriptions of all tasks submitted by users are given, and QoS service providing ability descriptions of all resources when the resources are distributed are given; and a scheduling algorithm tries to search a QoS optimal task scheduling scheme according to the QoS descriptions, so that the QoS optimal task scheduling scheme can be obtained. With the multi-QoS constrained cloud computing task scheduling method adopted, problems existing in multi-QoS constrained cloud computing task scheduling can be solved. The method can be applied to cloud computing task scheduling.

Description

A kind of cloud computing method for scheduling task of multi-QoS constraint

Technical field

The present invention relates to field of cloud computer technology, especially a kind of cloud computing method for scheduling task of multi-QoS constraint.

Background technology

Cloud computing, as a kind of new computation schema and business model, is just being subject to the extensive attention of academia and industrial circle.Cloud computing technology is Distributed Calculation, the further developing of parallel computation and grid computing, Intel Virtualization Technology is utilized to turn to dynamic telescopic virtual resources pond virtual to computational resource, storage resources and bandwidth resources, take the Internet as carrier, be supplied to different users in the form of services as required and use.Data center under cloud computing environment can reduce the difficulty of operation, maintenance and management effectively, also improves the safety and reliability of data simultaneously.By adopting the mechanism of paying as required, user there is no need to buy extra hardware resource on network peak, not only saves the expense buying hardware like this, also eliminates the problem that maintenance and management brings simultaneously.

Task scheduling is very crucial in cloud computing technology and the problem of complexity, and the quality of this Resolving probiems not only has influence on the satisfaction of user, also greatly have impact on the overall performance of system.Therefore, under cloud environment, how scheduling is efficiently carried out rationally to task, improve the satisfaction of user, become the Focal point and difficult point of cloud computing research.Generally, task scheduling is a np complete problem.

At present a lot of scholars, expert and enterprise has been had to drop into a large amount of human and material resources and financial resources are studied task scheduling strategy wherein.Cloud computing produces on the basis of Distributed Calculation, effectiveness calculating, grid computing, and have a lot of task scheduling algorithms to be all find out on the basis of grid computing in cloud computing.And be exactly make all task runs complete required minimal time in a main target of Job Scheduling for Grid Computing algorithm, most task scheduling algorithm is all optimized task scheduling with this target.But in cloud computing model, the cost required for tasks carrying, the time expended and load balancing are all the resources of the focus instantly studied, different computing capability, and its use cost is also different.For the application of time-sensitive, provide the resource of stronger disposal ability, make task run complete the required time shorter; User for cost sensitivity applies, and provide lower to manage the resource of cost, make task run complete required cost lower, and cloud computing task scheduling algorithm is used by the just needs that meet of these different demands.

Having in cloud computing task scheduling manyly has problem to be solved, such as load balancing, minimum completion time etc.The minimum completion time of consideration task is a kind of strategy making full use of system resource, the resource in cloud resource pool has plenty of isomery, has plenty of isomorphism.In heterogeneous distributed cloud resource pool, by considering that minimum completion time factor can make the fast machine of the speed of service better be utilized, also make the slow machine of the speed of service to be unlikely to distribute too much task simultaneously thus cause tasks carrying time-out, so be necessary to the consideration of this factor of minimum completion time.

Load balancing is the method often used in cloud computing task scheduling, has following two kinds of forms: the first, in static equilibrium, utilize the information such as number of tasks and time of implementation, distributed by the method in mathematics between task and resource.The shortcoming of this strategy to cause resource utilization not high, and for the change of virtual machine information, this kind of balance policy cannot dynamically change; The second, in dynamic equalization, generally all realized by prediction.The foundation that algorithm carries out predicting is the current of virtual machine and historical information, then carries out task scheduling according to prediction result out.For this algorithm, rational standard is very important.The HDFS system of the product E C2 of Amazon and the product improvement of Yahoo also has in the product G FS of Google has all carried out actual research and apply to load balancing, so have great importance to the better cloud computing task scheduling algorithm of proposition.

In cloud computing task scheduling, also have another one to need key issue and the service quality (QoS, QualityofService) of solution badly.QoS is a kind of method solving network stabilization performance, and the delayed and congestion problems in network is also solved by service quality.And be applied to now in cloud computing, QoS is exactly some demands that reflection user proposes when using cloud computing system.QoS comprises much different parameters, and we compare and common are CPU, cost, access rate, distance, resource stability, memory capacity etc.Qos parameters much is more so had, so cloud computing system develops can provide different platform of serving to meet different user institute and putting forward the different demand of task owing to using in the demand of the user of cloud computing system.These platforms can meet the dissimilar demand of user, such as ensure the reliability of system, consume less cost, provide higher message transmission rate etc.The so many realization of QoS demand to cloud computing task scheduling algorithm are huge challenges.Even if consider the QoS demand of user to task for the current task scheduling based on QoS proposed, but also just consider one dimension QoS demand simply, namely single time or cost.Therefore, the task scheduling strategy proposing to have multi-QoS constraint is necessary for the development of cloud computing.

Summary of the invention

The technical problem that the present invention solves is the cloud computing method for scheduling task providing a kind of multi-QoS constraint; To propose the task scheduling strategy with multi-QoS constraint.

The technical scheme that the present invention solves the problems of the technologies described above is:

All users describe the QoS that submitting to of task provides its demand, and all resources also provide it when issuing and provide QoS service ability description; Dispatching algorithm describes according to these QoS, attempts search and obtains the optimized task scheduling approach of QoS.

Described method concrete steps are:

Obtain task-set T and cloud resource set R, and extract the QoS of task-set and resource;

Standardization is carried out to task-set and resource QoS, obtains standardization task QoS matrix and resource QoS matrix;

Calculating is respectively the QoS weights of resource;

Calculation task concentrates the comprehensive QoS demand of each task, and according to QoS demand by task descending, obtains new task-set TT;

To the task t1 of first in new task-set TT, calculate the QoS satisfaction that it arrives each resource; Obtain the resource set RR of QoS Maximum Satisfaction;

If only have a resource in maximum resource set RR, distribute t1 and perform to this resource; Otherwise, calculate the QoS distance that t1 to RR concentrates each resource, distribute t1 to QoS and perform apart from minimum resource;

If there is multiple task, then the task of Resources allocation is deleted, then, carry out according to the method for back the distribution performing resource; Until whole task matching execution resources;

After tasks carrying Resourse Distribute completes, whether polling tasks is finished;

As task is not finished, then check whether available free resource; When available free resource situation, judge further in the highest QoS satisfaction task sequence in this resource with or without not being finished of task; Not during available free resource, then turn back to the poll that whether complete tasks carrying is;

As complete in QoS satisfaction tasks carrying the highest in idling-resource, then by the task scheduling that is not finished to the idling-resource of this highest QoS satisfaction until tasks carrying is complete;

As tasks carrying is complete, then terminate.

Described set of tasks is expressed as T, T={t1, t2 ..., tn}, n=|T| are user task quantity, and ti represents i-th task of user, i ∈ [1, n]; Ti={tID, tLen, tQoS, tSta};

TID: the unique identification representing task;

TLen: the length representing task, unit: MI (MillionInstruction);

TQoS:tQoS={QoS ₁, QoS ₂..., QoS _krepresenting the multi-QoS demand of task, k represents QoS dimension;

TSta:tSta={tAlloc, tExecu, tSucc}, represent the state of user task, and divide three kinds: tAlloc to represent that task treats dispatch state, tExecu represents execution status of task, and tSucc represents tasks carrying completion status;

Being P, P={p1, p2..., p1}, l=|P| by the physical host set table method of resource, is physical host quantity; Ph={pID, pType, pSta}.

PID: the unique identification representing physical host;

Ptype: the type representing physical host, as work station, large-scale computer or microcomputer etc.;

Psta: represent the state of physical host, Psta={pFree, pRun}, pFre represent that physical host is in idle condition, does not namely dispose virtual machine or virtual machine is not executed the task, and pRun represents that physical host is in running order;

Represent resource with virtual machine form, resource collection is expressed as R, R={r1, r2 ..., rm}, m=|R|, be resource (virtual machine) quantity that cloud data center provides, rj represents a jth resource, j ∈ [1, m]; Rj={rID, rCap, rQoS, rSta, rLoc};

RID: the unique identification representing resource;

RCap: the computing capability representing resource, unit: MIPS (MillionInstructinPerSecond);

RQoS:rQoS={QoS ₁, QoS ₂..., QoS _krepresenting the multi-QoS service ability of resource, k represents QoS dimension.

RSta:rSta={rRun, rFre} represent the state of resource, and divide two kinds: rRun to represent that resource is in the state of executing the task, rFre represents that resource is in idle condition.

RLoc: the physical host representing resource place.

Described QoS standardization is:

The QoS dimension making resource provide is k, and resource quantity is m, then the QoS that m resource provides is m × k matrix, is expressed as:

${\mathrm{QoS}}_{m,k}^R=\left[\begin{array}{c}{\mathrm{qos}}_{1,1}^R,{\mathrm{qos}}_{1,2}^R,...,{\mathrm{qos}}_{1,k}^R\\ {\mathrm{qos}}_{2,1}^R,{\mathrm{qos}}_{2,2}^R,...,{\mathrm{qos}}_{2,k}^R\\ \begin{array}{cccc}...& ...& ...& ...\end{array}\\ {\mathrm{qos}}_{m,1}^R,{\mathrm{qos}}_{m,2}^R,...,{\mathrm{qos}}_{m,k}^R\end{array}\right]$

Make number of tasks be n, then the demand of n task on k dimension QoS can be expressed as n × k matrix, is expressed as:

${\mathrm{QoS}}_{n,k}^T=\left[\begin{array}{c}{\mathrm{qos}}_{1,1}^T,{\mathrm{qos}}_{1,2}^T,...,{\mathrm{qos}}_{1,k}^T\\ {\mathrm{qos}}_{2,1}^T,{\mathrm{qos}}_{2,2}^T,...,{\mathrm{qos}}_{2,k}^T\\ \begin{array}{cccc}...& ...& ...& ...\end{array}\\ {\mathrm{qos}}_{n,1}^T,{\mathrm{qos}}_{n,2}^T,...,{\mathrm{qos}}_{n,k}^T\end{array}\right]$

Then the mode of clicking carries out standardization;

First simultaneous two matrixes, obtain simultaneous matrix,

${\mathrm{QoS}}_{m+n,k}=\left[\begin{array}{c}{\mathrm{qos}}_{1,1}^R,{\mathrm{qos}}_{1,2}^R,...,{\mathrm{qos}}_{1,k}^R\\ {\mathrm{qos}}_{2,1}^R,{\mathrm{qos}}_{2,2}^R,...,{\mathrm{qos}}_{2,k}^R\\ \begin{array}{cccc}...& ...& ...& ...\end{array}\\ {\mathrm{qos}}_{m,1}^R,{\mathrm{qos}}_{m,2}^R,...,{\mathrm{qos}}_{m,k}^R\\ {\mathrm{qos}}_{1,1}^T,{\mathrm{qos}}_{1,2}^T,...,{\mathrm{qos}}_{1,k}^T\\ {\mathrm{qos}}_{2,1}^T,{\mathrm{qos}}_{2,2,}^T,...,{\mathrm{qos}}_{2,k}^T\\ \begin{array}{cccc}...& ...& ...& ...\end{array}\\ {\mathrm{qos}}_{n,1}^T,{\mathrm{qos}}_{n,2}^T,...,{\mathrm{qos}}_{n,k}^T\end{array}\right]$

During standardization, QoS is divided into actively tolerance and negative metric; Positive metric is higher shows that service quality is higher, as reliability, fail safe, stability; Consume that metric is lower shows that service quality is higher, as service fee; Positive tolerance carries out standardization with formula 1, consumes tolerance and carries out standardization with formula 2;

${\mathrm{qos}}_{i,j}^S=\left\{\begin{array}{cc}1,& {\mathrm{qos}}_j^{\max }-{\mathrm{qos}}_j^{\min }=0\\ \frac{{\mathrm{qos}}_{i,j}-{\mathrm{qos}}_j^{\min }}{{\mathrm{qos}}_j^{\max }-{\mathrm{qos}}_j^{\min }},& {\mathrm{qos}}_j^{\max }-{\mathrm{qos}}_j^{\min }\≠0\end{array}\right.$ Formula 1

${\mathrm{qos}}_{i,j}^S=\left\{\begin{array}{cc}1,& {\mathrm{qos}}_j^{\max }-{\mathrm{qos}}_j^{\min }=0\\ \frac{{\mathrm{qos}}_j^{\max }-{\mathrm{qos}}_{i,j}}{{\mathrm{qos}}_j^{\max }-{\mathrm{qos}}_j^{\min }},& {\mathrm{qos}}_j^{\max }-{\mathrm{qos}}_j^{\min }\≠0\end{array}\right.$ Formula 2

Wherein, qos _{i, j}representing matrix QoS _{m+n, k}in the i-th row jth row qos value, 1 <=i <=m+n, 1 <=j <=k, be illustrated respectively in matrix QoS _{m+n, k}the minimum value of middle jth row qos parameter value and maximum, represent the qos value of the i-th row jth row after standardization, normalized matrix is expressed as:

Separation criterion matrix obtain standardization resource QoS matrix and task QoS matrix, be expressed as with

${\mathrm{QoS}}_{m,k}^{SR}=\left[\begin{array}{c}{\mathrm{qos}}_{1,1}^{SR},{\mathrm{qos}}_{1,2}^{SR},...,{\mathrm{qos}}_{1,k}^{SR}\\ {\mathrm{qos}}_{2,1}^{SR},{\mathrm{qos}}_{2,2}^{SR},...,{\mathrm{qos}}_{2,k}^{SR}\\ \begin{array}{cccc}...& ...& ...& ...\end{array}\\ {\mathrm{qos}}_{m,1}^{SR},{\mathrm{qos}}_{m,2}^{SR},...,{\mathrm{qos}}_{m,k}^{SR}\end{array}\right],$ ${\mathrm{QoS}}_{n,k}^{ST}=\left[\begin{array}{c}{\mathrm{qos}}_{1,1}^{ST},{\mathrm{qos}}_{1,2}^{ST},...,{\mathrm{qos}}_{1,k}^{ST}\\ {\mathrm{qos}}_{2,1}^{ST},{\mathrm{qos}}_{2,2}^{ST},...,{\mathrm{qos}}_{2,k}^{ST}\\ \begin{array}{cccc}...& ...& ...& ...\end{array}\\ {\mathrm{qos}}_{n,1}^{ST},{\mathrm{qos}}_{n,2}^{ST},...,{\mathrm{qos}}_{n,k}^{ST}\end{array}\right].$

The calculating of described resource QoS satisfaction is:

For positive tolerance, calculate with formula 1 satisfaction that user obtains on this dimension QoS, for negative metric, calculate satisfaction with formula 2;

${\mathrm{QoS}}_{i,j}^{Deg}=\left\{\begin{array}{cc}\frac{{\mathrm{qos}}_{i,j}^{SR}}{{\mathrm{qos}}_{i,j}^{ST}},& {\mathrm{qos}}_{i,j}^{SR}<{\mathrm{qos}}_{i,j}^{ST}\\ 1,& {\mathrm{qos}}_{i,j}^{SR}\&GreaterEqual;{\mathrm{qos}}_{i,j}^{ST}\end{array}\right.$ Formula 1

${\mathrm{QoS}}_{i,j}^{Deg}=\left\{\begin{array}{cc}\frac{{\mathrm{qos}}_{i,j}^{ST}}{{\mathrm{qos}}_{i,j}^{SR}},& {\mathrm{qos}}_{i,j}^{ST}<{\mathrm{qos}}_{i,j}^{SR}\\ 1,& {\mathrm{qos}}_{i,j}^{ST}\&GreaterEqual;{\mathrm{qos}}_{i,j}^{SR}\end{array}\right.$ Formula 2

Then user task t _icomprehensive satisfaction in performed resource on each dimension QoS is:

${\mathrm{QoS}}_i^{Deg}=\frac{{\&Sigma;}_{j=1}^k{\mathrm{QoS}}_{i,j}^{Deg}}{k}$

The average QoS satisfaction of all user tasks is:

${\mathrm{QoS}}^{Deg}=\frac{{\&Sigma;}_{i=1}^n{\mathrm{QoS}}_i^{Deg}}{n}$

In task scheduling process, select the highest resource of satisfaction to distribute as far as possible.

Described QoS distance calculates and is:

Utilize Weighted distance to the QoS distance between the task of measuring and resource,

$D=\sqrt{{\&Sigma;}_{j=1}^k{w}_j{({\mathrm{qos}}_{i,j}^{ST}-{\mathrm{qos}}_{i,j}^{SR})}^2}$

$\begin{array}{cc}s.t.& w_j\&GreaterEqual;0,j=1,2,...,k,{\&Sigma;}_{j=1}^k{w}_j^2=1\end{array}$

Wherein, w _jrepresent jth dimension QoS shared weight when distance calculates; w _jmaximizing deviation method is used to determine size; The capacity of water gap that resource provides on jth dimension QoS is larger, illustrates that the impact of this parameter when measuring distance is larger, corresponding w _jlarger, otherwise, w _jless; Consider that the capacity of water that provide of all resources on jth dimension QoS is equal, then this parameter affects when measuring distance is 0, now should w _jbe decided to be 0; For jth dimension QoS, use D _{i, j}w () represents resource ri and other resources deviation in this QoS service ability, then

$D_{i,j}\left(w\right)={\&Sigma;}_{i^{\&prime;}=1}^m|{\mathrm{rs}}_{i,j}{w}_j-{\mathrm{rs}}_{i^{\&prime;},j}{w}_j|,i=1,2,...,m,j=1,2,...,k$

Order

$D_j={\&Sigma;}_{i=1}^m{D}_{i,j}\left(w\right)={\&Sigma;}_{i=1}^m{\&Sigma;}_{i^{\&prime;}=1}^m|{\mathrm{rs}}_{i,j}{w}_j-{\mathrm{rs}}_{i^{\&prime;},j}{w}_j|,j=1,2,...,k$

Dj represents the service ability of each resource and total deviation of other resource service abilities on jth dimension QoS;

The value of resource QoS comprehensive service capability and user QoS integration requirement after qos parameter standardization is calculated with following formula;

$U_{qos}\left(w\right)={\&Sigma;}_{j=1}^k{\mathrm{qos}}_{i,j}{w}_j$

For each resource, U _qosw () is larger, its comprehensive QoS service ability is better; Construct target function thus,

$\max D\left(w\right)={\&Sigma;}_{j=1}^m{D}_j\left(w\right)={\&Sigma;}_{j=1}^k{\&Sigma;}_{i=1}^m{\&Sigma;}_{i^{\&prime;}=1}^m|{\mathrm{rs}}_{i,j}-{\mathrm{rs}}_{i^{\&prime;},j}|\&times;w_j$

Be equal to

$\max D\left(w\right)={\&Sigma;}_{j=1}^k{\&Sigma;}_{i=1}^m{\&Sigma;}_{i^{\&prime;}=1}^m|{\mathrm{rs}}_{i,j}-{\mathrm{rs}}_{i^{\&prime;},j}|\&times;w_j$

$\begin{array}{cc}s.t.& w_j\&GreaterEqual;0,j=1,2,...,k,{\&Sigma;}_{j=1}^k{w}_j^2=1\end{array}$

By Lagrangian method can try to achieve target function maximum time,

$w_j=\frac{{\&Sigma;}_{i=1}^m{\&Sigma;}_{i^{\&prime;}=1}^m\left|{\mathrm{rs}}_{i,j}-{\mathrm{rs}}_{i^{\&prime;},j}\right|}{{\&Sigma;}_{j=1}^k{\&Sigma;}_{i=1}^m{\&Sigma;}_{i^{\&prime;}=1}^m\left|{\mathrm{rs}}_{i,j}-{\mathrm{rs}}_{i^{\&prime;},j}\right|},j=1,2,...,k.$

Scheduling user task problem different to the requirement of QoS goal constraint under the present invention is directed to cloud computing environment, proposes a kind of cloud computing method for scheduling task of multi-QoS constraint.The method considers based on maximum QoS satisfaction and minimum QoS distance between task and resource, by the target function of structure reflection QoS service ability, and solves target function with Lagrangian method, obtains the solution of resource selection and task scheduling.The present invention, under the prerequisite meeting the maximum satisfaction of user task QoS, selects the resource of minimum QoS distance to map, and not only can guarantee the task scheduling efficiency during QoS constraint of multidimensional, reduce average task execution time, can also ensure resource utilization.

Accompanying drawing explanation

Below in conjunction with accompanying drawing, the present invention is further described:

Fig. 1 is cloud computing Task Scheduling Model of the present invention;

Fig. 2 is algorithm flow chart of the present invention.

Embodiment

As shown in Figure 2, task scheduling concrete steps of the present invention are:

Obtain task-set T and cloud resource set R, and extract the QoS of task-set and resource;

Standardization is carried out to task-set and resource QoS, obtains standardization task QoS matrix and resource QoS matrix;

Calculating is respectively the QoS weights of resource;

Calculation task concentrates the comprehensive QoS demand of each task, and according to QoS demand by task descending, obtains new task-set TT;

To the task t1 of first in new task-set TT, calculate the QoS satisfaction that it arrives each resource; Obtain the resource set RR of QoS Maximum Satisfaction;

If only have a resource in maximum resource set RR, distribute t1 and perform to this resource; Otherwise, calculate the QoS distance that t1 to RR concentrates each resource, distribute t1 to QoS and perform apart from minimum resource;

If there is multiple task, then the task of Resources allocation is deleted, then, carry out according to the method for back the distribution performing resource; Until whole task matching execution resources;

After tasks carrying Resourse Distribute completes, whether polling tasks is finished;

As task is not finished, then check whether available free resource; When available free resource situation, judge further in the highest QoS satisfaction task sequence in this resource with or without not being finished of task; Not during available free resource, then turn back to the poll that whether complete tasks carrying is;

As complete in QoS satisfaction tasks carrying the highest in idling-resource, then by the task scheduling that is not finished to the idling-resource of this highest QoS satisfaction until tasks carrying is complete;

As tasks carrying is complete, then terminate.

Below, be described in detail from the aspect such as multi-QoS constraint, problem description, multi-QoS constrained procedure of the QoS of cloud task scheduling, cloud computing.

1, the QoS of cloud task scheduling

In cloud computing Task Scheduling Model based on Qos constraint, the task that all users submits to needs the QoS providing its demand to describe, and all resources also need to provide it when issuing and provide QoS service ability description.Dispatching algorithm describes according to these QoS, attempts search and obtains the optimized task scheduling approach of QoS.The QoS that task and resource provide describes can some classifications, describes for each class QoS, for quantizing the income that when it is met in various degree, system obtains, draws the concept of benefit function.

QoS as module describes and comprises:

Timeliness describes: the attribute description of time correlation.The timeliness QoS of task describes and comprises total deadline, time started, latest finishing time etc.The timeliness QoS of resource describes the computing capability etc. comprising resource.Without loss of generality, the latest finishing time considering task and the running time of task in resource drawn by Resource Calculation ability.

Reliability describes: long playing task may the failure due to resource failed.Re-execute task and will cause the resource consumption of repetition, cause systematic function to reduce.The reliability of task by its requirement is dispatched, the generation of this situation can be reduced.Such of task describes to comprise and is minimumly successfully completed rate etc.; Such description of resource comprises unit interval failure rate etc.

Priority: the relative importance describing task.The higher task of priority needs comparatively early to be performed, and the higher resource of priority will preferentially be used when benefit is identical.

QoS as strategy describes and is mainly service class, and it acts on above all kinds of descriptions, comprising:

Rigid level: the QoS with this rank describes has the strongest restraining force.Such description of task must be met, otherwise is exactly invalid to its scheduling.

Soft level: having some QoS to describe does not need so strong constraint.This type of of task describes as being satisfied, then obtain greatest benefit, but if do not met, scheduling is not considered to invalid yet, and just benefit is affected.

As possible level: this rank is used for showing some and is not concerned or inessential QoS describes.This type of description of task and resource can be met as far as possible or be realized.

2, the multi-QoS constraint of cloud computing

Traditional cloud task scheduling target is generally the guarantee of one-dimensional QoS, as the shortest task completion time.In the cloud computing environment of reality, the interests desired by cloud user are multiple, are not single demand.Be described as example with timeliness QoS, some users not only pursue task and obtain the shortest deadline, also require to finish the work within the time limit sometime.Cloud environment relates to the distributed environment of multiple entity, and from the angle of cloud user subject and cloud resource entity, its QoS target in administrative mechanism, security strategy and expense etc. is all not quite similar.As, user not only expects also to require the least cost of tasks carrying by the shortest time that task completes.And cloud resource entity is more it is considered that self carry out the income of tasks carrying, pay close attention to the throughput-maximized of whole system simultaneously.

The present invention will pay close attention to the QoS of following Four types, and makes quantification treatment:

1) fail safe

Safety QoS is defined as:

${\mathrm{QoS}}_{Security}=\frac{{\&Sigma;}_{i=1}^n{t}_i{s}_i}{{\&Sigma;}_{i=1}^n{t}_i}$

Wherein, n represents security parameter number, and ti represents the weight of i-th security parameter, and si=1 represents that i-th security parameter is met, otherwise is 0.

2) QoS is stablized

${\mathrm{QoS}}_{Stality}=\frac{RunTime}{RunTime+FailureTime}$

Wherein, RunTime represents the resource uptime, and FailureTime represents the resource failed time.

3) success rate QoS

${\mathrm{QoS}}_{Success}=\frac{N_{Success}}{N_{Total}}$

Wherein, Nsuccess represents the number of tasks of successful execution in resource, and Ntotal represents and is assigned to number of tasks total in resource.

4) expense QoS

QoS _cost＝V×T

Wherein, V represents utilization of resources unit price, and T represents the time that task is run in resource.

3, problem describes

About cloud computing task, first make the following assumptions: the cloud task type of consideration is mainly calculation type Meta task, do not have dependence between task, do not consider performing across resource in implementation when task performs in resource, the Meta task that namely user submits to is the minimum unit of task scheduling.

Set of tasks is expressed as T, T={t1, t2 ..., tn}, n=|T| are user task quantity, and ti represents i-th task of user, i ∈ [1, n].ti＝{tID，tLen，tQoS，tSta}。

TID: the unique identification representing task;

TLen: the length representing task, unit: MI (MillionInstruction);

TQoS:tQoS={QoS ₁, QoS ₂..., QoS _krepresenting the multi-QoS demand of task, k represents QoS dimension.

TSta:tSta={tAlloc, tExecu, tSucc}, represent the state of user task, and divide three kinds: tAlloc to represent that task treats dispatch state, tExecu represents execution status of task, and tSucc represents tasks carrying completion status.

Cloud computing provides the mode of resource virtual for the physical host multiple stage virtual machine that turns to be carried out resource and provides in the data in the heart, and physical host set table method is P, P={p1, p2..., p1}, l=|P|, is physical host quantity.Ph＝{pID，pType，pSta}。

PID: the unique identification representing physical host;

Ptype: the type representing physical host, as work station, large-scale computer or microcomputer etc.;

Psta: represent the state of physical host, Psta={pFree, pRun}, pFre represent that physical host is in idle condition, does not namely dispose virtual machine or virtual machine is not executed the task, and pRun represents that physical host is in running order.

Because task is finally perform on a virtual machine, represent resource with virtual machine form, resource collection is expressed as R, R={r1, r2 ..., rm}, m=|R|, for resource (virtual machine) quantity that cloud data center provides, rj represents a jth resource, j ∈ [1, m].rj＝{rID，rCap，rQoS，rSta，rLoc}。

RID: the unique identification representing resource;

RCap: the computing capability representing resource, unit: MIPS (MillionInstructinPerSecond);

RQoS:rQoS={QoS ₁, QoS ₂..., QoS _krepresenting the multi-QoS service ability of resource, k represents QoS dimension.

RSta:rSta={rRun, rFre} represent the state of resource, and divide two kinds: rRun to represent that resource is in the state of executing the task, rFre represents that resource is in idle condition.

RLoc: the physical host representing resource place.

Based on above three class entities, cloud computing Task Scheduling Model as shown in Figure 1.

4, multi-QoS constrained procedure

1) multi-QoS standardization

The QoS dimension making resource provide is k, and resource quantity is m, then the QoS that m resource provides is m × k matrix, is expressed as:

${\mathrm{QoS}}_{m,k}^R=\left[\begin{array}{c}{\mathrm{qos}}_{1,1}^R,{\mathrm{qos}}_{1,2}^R,...,{\mathrm{qos}}_{1,k}^R\\ {\mathrm{qos}}_{2,1}^R,{\mathrm{qos}}_{2,2}^R,...,{\mathrm{qos}}_{2,k}^R\\ \begin{array}{cccc}...& ...& ...& ...\end{array}\\ {\mathrm{qos}}_{m,1}^R,{\mathrm{qos}}_{m,2}^R,...,{\mathrm{qos}}_{m,k}^R\end{array}\right]$

Make number of tasks be n, then the demand of n task on k dimension QoS can be expressed as n × k matrix, is expressed as:

${\mathrm{QoS}}_{n,k}^T=\left[\begin{array}{c}{\mathrm{qos}}_{1,1}^T,{\mathrm{qos}}_{1,2}^T,...,{\mathrm{qos}}_{1,k}^T\\ {\mathrm{qos}}_{2,1}^T,{\mathrm{qos}}_{2,2}^T,...,{\mathrm{qos}}_{2,k}^T\\ \begin{array}{cccc}...& ...& ...& ...\end{array}\\ {\mathrm{qos}}_{n,1}^T,{\mathrm{qos}}_{n,2}^T,...,{\mathrm{qos}}_{n,k}^T\end{array}\right]$

QoS due to resource provides the otherness with the QoS demand of task, and for the ease of evaluating the QoS service ability of user's QoS demand and resource, then two matrixes carry out standardization below.

First simultaneous two matrixes, obtain simultaneous matrix,

${\mathrm{QoS}}_{m+n,k}=\left[\begin{array}{c}{\mathrm{qos}}_{1,1}^R,{\mathrm{qos}}_{1,2}^R,...,{\mathrm{qos}}_{1,k}^R\\ {\mathrm{qos}}_{2,1}^R,{\mathrm{qos}}_{2,2}^R,...,{\mathrm{qos}}_{2,k}^R\\ \begin{array}{cccc}...& ...& ...& ...\end{array}\\ {\mathrm{qos}}_{m,1}^R,{\mathrm{qos}}_{m,2}^R,...,{\mathrm{qos}}_{m,k}^R\\ {\mathrm{qos}}_{1,1}^T,{\mathrm{qos}}_{1,2}^T,...,{\mathrm{qos}}_{1,k}^T\\ {\mathrm{qos}}_{2,1}^T,{\mathrm{qos}}_{2,2,}^T,...,{\mathrm{qos}}_{2,k}^T\\ \begin{array}{cccc}...& ...& ...& ...\end{array}\\ {\mathrm{qos}}_{n,1}^T,{\mathrm{qos}}_{n,2}^T,...,{\mathrm{qos}}_{n,k}^T\end{array}\right]$

During standardization, QoS is divided into actively tolerance and negative metric.Positive metric is higher shows that service quality is higher, as reliability, fail safe, stability.Consume that metric is lower shows that service quality is higher, as service fee.Positive tolerance carries out standardization with formula 1, consumes tolerance and carries out standardization with formula 2,

${\mathrm{qos}}_{i,j}^S=\left\{\begin{array}{cc}1,& {\mathrm{qos}}_j^{\max }-{\mathrm{qos}}_j^{\min }=0\\ \frac{{\mathrm{qos}}_{i,j}-{\mathrm{qos}}_j^{\min }}{{\mathrm{qos}}_j^{\max }-{\mathrm{qos}}_j^{\min }},& {\mathrm{qos}}_j^{\max }-{\mathrm{qos}}_j^{\min }\&NotEqual;0\end{array}\right.$

${\mathrm{qos}}_{i,j}^S=\left\{\begin{array}{cc}1,& {\mathrm{qos}}_j^{\max }-{\mathrm{qos}}_j^{\min }=0\\ \frac{{\mathrm{qos}}_j^{\max }-{\mathrm{qos}}_{i,j}}{{\mathrm{qos}}_j^{\max }-{\mathrm{qos}}_j^{\min }},& {\mathrm{qos}}_j^{\max }-{\mathrm{qos}}_j^{\min }\&NotEqual;0\end{array}\right.$

Wherein, qos _{i, j}representing matrix QoS _{m+n, k}in the i-th row jth row qos value, 1 <=i <=m+n, 1 <=j <=k, be illustrated respectively in matrix QoS _{m+n, k}the minimum value of middle jth row qos parameter value and maximum, represent the qos value of the i-th row jth row after standardization, normalized matrix is expressed as:

${\mathrm{QoS}}_{m+n,k}^S=\left[\begin{array}{c}{\mathrm{qos}}_{1,1}^S,{\mathrm{qos}}_{1,2}^S,...,{\mathrm{qos}}_{1,k}^S\\ {\mathrm{qos}}_{2,1}^S,{\mathrm{qos}}_{2,2}^S,...,{\mathrm{qos}}_{2,k}^S\\ \begin{array}{cccc}...& ...& ...& ...\end{array}\\ {\mathrm{qos}}_{m,1}^S,{\mathrm{qos}}_{m,2}^S,...,{\mathrm{qos}}_{m,k}^S\\ {\mathrm{qos}}_{m+1,1}^S,{\mathrm{qos}}_{m+1,2,}^S,...,{\mathrm{qos}}_{m+1,k}^S\\ \begin{array}{cccc}...& ...& ...& ...\end{array}\\ {\mathrm{qos}}_{m+n,1}^S,{\mathrm{qos}}_{m+n,2}^S,...,{\mathrm{qos}}_{m+n,k}^S\end{array}\right],{\mathrm{qos}}_{i,j}^S\&Element;\&lsqb;0,1\&rsqb;,1\&le;i\&le;m+n,1\&le;j\&le;k$

Separation criterion matrix obtain standardization resource QoS matrix and task QoS matrix, be expressed as with

${\mathrm{QoS}}_{m,k}^{SR}=\left[\begin{array}{c}{\mathrm{qos}}_{1,1}^{SR},{\mathrm{qos}}_{1,2}^{SR},...,{\mathrm{qos}}_{1,k}^{SR}\\ {\mathrm{qos}}_{2,1}^{SR},{\mathrm{qos}}_{2,2}^{SR},...,{\mathrm{qos}}_{2,k}^{SR}\\ \begin{array}{cccc}...& ...& ...& ...\end{array}\\ {\mathrm{qos}}_{m,1}^{SR},{\mathrm{qos}}_{m,2}^{SR},...,{\mathrm{qos}}_{m,k}^{SR}\end{array}\right],$ ${\mathrm{QoS}}_{n,k}^{ST}=\left[\begin{array}{c}{\mathrm{qos}}_{1,1}^{ST},{\mathrm{qos}}_{1,2}^{ST},...,{\mathrm{qos}}_{1,k}^{ST}\\ {\mathrm{qos}}_{2,1}^{ST},{\mathrm{qos}}_{2,2}^{ST},...,{\mathrm{qos}}_{2,k}^{ST}\\ \begin{array}{cccc}...& ...& ...& ...\end{array}\\ {\mathrm{qos}}_{n,1}^{ST},{\mathrm{qos}}_{n,2}^{ST},...,{\mathrm{qos}}_{n,k}^{ST}\end{array}\right].$

2) the QoS satisfaction of user

Task scheduling should the QoS demand of satisfied users substantially.For positive tolerance, calculate with formula 1 satisfaction that user obtains on this dimension QoS, for negative metric, calculate satisfaction with formula 2.

${\mathrm{QoS}}_{i,j}^{Deg}=\left\{\begin{array}{cc}\frac{{\mathrm{qos}}_{i,j}^{SR}}{{\mathrm{qos}}_{i,j}^{ST}},& {\mathrm{qos}}_{i,j}^{SR}<{\mathrm{qos}}_{i,j}^{ST}\\ 1,& {\mathrm{qos}}_{i,j}^{SR}\&GreaterEqual;{\mathrm{qos}}_{i,j}^{ST}\end{array}\right.$

${\mathrm{QoS}}_{i,j}^{Deg}=\left\{\begin{array}{cc}\frac{{\mathrm{qos}}_{i,j}^{ST}}{{\mathrm{qos}}_{i,j}^{SR}},& {\mathrm{qos}}_{i,j}^{ST}<{\mathrm{qos}}_{i,j}^{SR}\\ 1,& {\mathrm{qos}}_{i,j}^{ST}\&GreaterEqual;{\mathrm{qos}}_{i,j}^{SR}\end{array}\right.$

Then user task t _icomprehensive satisfaction in performed resource on each dimension QoS is:

${\mathrm{QoS}}_i^{Deg}=\frac{{\&Sigma;}_{j=1}^k{\mathrm{QoS}}_{i,j}^{Deg}}{k}$

The average QoS satisfaction of all user tasks is:

${\mathrm{QoS}}^{Deg}=\frac{{\&Sigma;}_{i=1}^n{\mathrm{QoS}}_i^{Deg}}{n}$

In task scheduling process, distribute selecting the highest resource of satisfaction as far as possible.

3) QoS distance metric

In order to the resource not making the task of low QoS demand take high QoS, affect the tasks carrying of other users, cause total time of implementation to increase, task matching should be made as far as possible to perform on the resource similar with oneself QoS demand.Utilize Weighted distance to the QoS distance between the task of measuring and resource,

$D=\sqrt{{\&Sigma;}_{j=1}^k{w}_j{({\mathrm{qos}}_{i,j}^{ST}-{\mathrm{qos}}_{i,j}^{SR})}^2}$

$\begin{array}{cc}s.t.& w_j\&GreaterEqual;0,j=1,2,...,k,{\&Sigma;}_{j=1}^k{w}_j^2=1\end{array}$

Wherein, w _jrepresent jth dimension QoS shared weight when distance calculates.The capacity of water gap provided on each dimension QoS due to resource is different, in order to carry out range measurement better, uses maximizing deviation method to determine w _jsize.The capacity of water gap that resource provides on jth dimension QoS is larger, illustrates that the impact of this parameter when measuring distance is larger, corresponding w _jlarger, otherwise, w _jless.Consider that the capacity of water that provide of all resources on jth dimension QoS is equal, then this parameter affects when measuring distance is 0, now should w _jbe decided to be 0.For jth dimension QoS, use D _{i, j}w () represents resource r _iwith the deviation of other resources in this QoS service ability, then

$D_{i,j}\left(w\right)={\&Sigma;}_{i^{\&prime;}=1}^m|{\mathrm{rs}}_{i,j}{w}_j-{\mathrm{rs}}_{i^{\&prime;},j}{w}_j|,i=1,2,...,m,j=1,2,...,k$

Order

$D_j={\&Sigma;}_{i=1}^m{D}_{i,j}\left(w\right)={\&Sigma;}_{i=1}^m{\&Sigma;}_{i^{\&prime;}=1}^m|{\mathrm{rs}}_{i,j}{w}_j-{\mathrm{rs}}_{i^{\&prime;},j}{w}_j|,j=1,2,...,k$

Dj represents the service ability of each resource and total deviation of other resource service abilities on jth dimension QoS.

The value of resource QoS comprehensive service capability and user QoS integration requirement after qos parameter standardization is calculated with following formula.

$U_{qos}\left(w\right)={\&Sigma;}_{j=1}^k{\mathrm{qos}}_{i,j}{w}_j$

Obviously, for each resource, U _qosw () is larger, its comprehensive QoS service ability is better.Construct target function thus,

$\max D\left(w\right)={\&Sigma;}_{j=1}^m{D}_j\left(w\right)={\&Sigma;}_{j=1}^k{\&Sigma;}_{i=1}^m{\&Sigma;}_{i^{\&prime;}=1}^m|{\mathrm{rs}}_{i,j}-{\mathrm{rs}}_{i^{\&prime;},j}|\&times;w_j$

Be equal to

$\max D\left(w\right)={\&Sigma;}_{j=1}^k{\&Sigma;}_{i=1}^m{\&Sigma;}_{i^{\&prime;}=1}^m|{\mathrm{rs}}_{i,j}-{\mathrm{rs}}_{i^{\&prime;},j}|\&times;w_j$

$\begin{array}{cc}s.t.& w_j\&GreaterEqual;0,j=1,2,...,k,{\&Sigma;}_{j=1}^k{w}_j^2=1\end{array}$

By Lagrangian method can try to achieve target function maximum time,

$w_j=\frac{{\&Sigma;}_{i=1}^m{\&Sigma;}_{i^{\&prime;}=1}^m\left|{\mathrm{rs}}_{i,j}-{\mathrm{rs}}_{i^{\&prime;},j}\right|}{{\&Sigma;}_{j=1}^k{\&Sigma;}_{i=1}^m{\&Sigma;}_{i^{\&prime;}=1}^m\left|{\mathrm{rs}}_{i,j}-{\mathrm{rs}}_{i^{\&prime;},j}\right|},j=1,2,...,k.$

Claims (8)

1. a cloud computing method for scheduling task for multi-QoS constraint, is characterized in that: all users describe the QoS that submitting to of task provides its demand, and all resources also provide it when issuing and provide QoS service ability description; Dispatching algorithm describes according to these QoS, attempts search and obtains the optimized task scheduling approach of QoS.

2. the cloud computing method for scheduling task of multi-QoS constraint according to claim 1, is characterized in that: described method concrete steps are:

Obtain task-set T and cloud resource set R, and extract the QoS of task-set and resource;

Standardization is carried out to task-set and resource QoS, obtains standardization task QoS matrix and resource QoS matrix;

Calculating is respectively the QoS weights of resource;

Calculation task concentrates the comprehensive QoS demand of each task, and according to QoS demand by task descending, obtains new task-set TT;

To the task t1 of first in new task-set TT, calculate the QoS satisfaction that it arrives each resource; Obtain the resource set RR of QoS Maximum Satisfaction;

If only have a resource in maximum resource set RR, distribute t1 and perform to this resource; Otherwise, calculate the QoS distance that t1 to RR concentrates each resource, distribute t1 to QoS and perform apart from minimum resource;

If there is multiple task, then the task of Resources allocation is deleted, then, carry out according to the method for back the distribution performing resource; Until whole task matching execution resources;

After tasks carrying Resourse Distribute completes, whether polling tasks is finished;

As task is not finished, then check whether available free resource; When available free resource situation, judge further in the highest QoS satisfaction task sequence in this resource with or without not being finished of task; Not during available free resource, then turn back to the poll that whether complete tasks carrying is;

As complete in QoS satisfaction tasks carrying the highest in idling-resource, then by the task scheduling that is not finished to the idling-resource of this highest QoS satisfaction until tasks carrying is complete;

As tasks carrying is complete, then terminate.

3. the cloud computing method for scheduling task of multi-QoS constraint according to claim 2, is characterized in that: described set of tasks is expressed as T, T={t1, t2, ..., tn}, n=|T| are user task quantity, ti represents i-th task of user, i ∈ [1, n]; Ti={tID, tLen, tQoS, tSta};

TID: the unique identification representing task;

TLen: the length representing task, unit: MI (MillionInstruction);

TQoS:tQoS={QoS ₁, QoS ₂..., QoS _krepresenting the multi-QoS demand of task, k represents QoS dimension;

TSta:tSta={tAlloc, tExecu, tSucc}, represent the state of user task, and divide three kinds: tAlloc to represent that task treats dispatch state, tExecu represents execution status of task, and tSucc represents tasks carrying completion status;

Being P, P={p1, p2..., p1}, l=|P| by the physical host set table method of resource, is physical host quantity; Ph={pID, pType, pSta}.

PID: the unique identification representing physical host;

Ptype: the type representing physical host, as work station, large-scale computer or microcomputer etc.;

Psta: represent the state of physical host, Psta={pFree, pRun}, pFre represent that physical host is in idle condition, does not namely dispose virtual machine or virtual machine is not executed the task, and pRun represents that physical host is in running order;

Represent resource with virtual machine form, resource collection is expressed as R, R={r1, r2 ..., rm}, m=|R|, be resource (virtual machine) quantity that cloud data center provides, rj represents a jth resource, j ∈ [1, m]; Rj={rID, rCap, rQoS, rSta, rLoc};

RID: the unique identification representing resource;

RCap: the computing capability representing resource, unit: MIPS (MillionInstmctinPerSecond);

RQoS:rQoS={QoS ₁, QoS ₂..., QoS _krepresenting the multi-QoS service ability of resource, k represents QoS dimension.

RSta:rSta={rRun, rFre} represent the state of resource, and divide two kinds: rRun to represent that resource is in the state of executing the task, rFre represents that resource is in idle condition.

RLoc: the physical host representing resource place.

4. the cloud computing method for scheduling task of multi-QoS constraint according to claim 2, is characterized in that: described QoS standardization is:

The QoS dimension making resource provide is k, and resource quantity is m, then the QoS that m resource provides is m × k matrix, is expressed as:

Make number of tasks be n, then the demand of n task on k dimension QoS can be expressed as n × k matrix, is expressed as:

Then the mode of clicking carries out standardization;

First simultaneous two matrixes, obtain simultaneous matrix,

During standardization, QoS is divided into actively tolerance and negative metric; Positive metric is higher shows that service quality is higher, as reliability, fail safe, stability; Consume that metric is lower shows that service quality is higher, as service fee; Positive tolerance carries out standardization with formula 1, consumes tolerance and carries out standardization with formula 2;

${\mathrm{qos}}_{i,j}^S=\left\{\begin{array}{cc}1,& {\mathrm{qos}}_j^{\max }-{\mathrm{qos}}_j^{\min }=0\\ \frac{{\mathrm{qos}}_{i,j}-{\mathrm{qos}}_j^{\min }}{{\mathrm{qos}}_j^{\max }-{\mathrm{qos}}_j^{\min }},& {\mathrm{qos}}_j^{\max }-{\mathrm{qos}}_j^{\min }\&NotEqual;0\end{array}\right.$ Formula 1

${\mathrm{qos}}_{i,j}^S=\left\{\begin{array}{cc}1,& {\mathrm{qos}}_j^{\max }-{\mathrm{qos}}_j^{\min }=0\\ \frac{{\mathrm{qos}}_j^{\max }-{\mathrm{qos}}_{i,j}}{{\mathrm{qos}}_j^{\max }-{\mathrm{qos}}_j^{\min }},& {\mathrm{qos}}_j^{\max }-{\mathrm{qos}}_j^{\min }\&NotEqual;0\end{array}\right.$ Formula 2

Wherein, qos _{i, j}representing matrix QoS _{m+n, k}in the i-th row jth row qos value, 1 <=i <=m+n, 1 <=j <=k, be illustrated respectively in matrix QoS _{m+n, k}the minimum value of middle jth row qos parameter value and maximum, represent the qos value of the i-th row jth row after standardization, normalized matrix is expressed as:

Separation criterion matrix obtain standardization resource QoS matrix and task QoS matrix, be expressed as with

5. the cloud computing method for scheduling task of multi-QoS constraint according to claim 3, is characterized in that: described QoS standardization is:

The QoS dimension making resource provide is k, and resource quantity is m, then the QoS that m resource provides is m × k matrix, is expressed as:

Make number of tasks be n, then the demand of n task on k dimension QoS can be expressed as n × k matrix, is expressed as:

Then the mode of clicking carries out standardization;

First simultaneous two matrixes, obtain simultaneous matrix,

During standardization, QoS is divided into actively tolerance and negative metric; Positive metric is higher shows that service quality is higher, as reliability, fail safe, stability; Consume that metric is lower shows that service quality is higher, as service fee; Positive tolerance carries out standardization with formula 1, consumes tolerance and carries out standardization with formula 2;

${\mathrm{qos}}_{i,j}^S=\left\{\begin{array}{cc}1,& {\mathrm{qos}}_j^{\max }-{\mathrm{qos}}_j^{\min }=0\\ \frac{{\mathrm{qos}}_{i,j}-{\mathrm{qos}}_j^{\min }}{{\mathrm{qos}}_j^{\max }-{\mathrm{qos}}_j^{\min }},& {\mathrm{qos}}_j^{\max }-{\mathrm{qos}}_j^{\min }\&NotEqual;0\end{array}\right.$ Formula 1

${\mathrm{qos}}_{i,j}^S=\left\{\begin{array}{cc}1,& {\mathrm{qos}}_j^{\max }-{\mathrm{qos}}_j^{\min }=0\\ \frac{{\mathrm{qos}}_j^{\max }-{\mathrm{qos}}_{i,j}}{{\mathrm{qos}}_j^{\max }-{\mathrm{qos}}_j^{\min }},& {\mathrm{qos}}_j^{\max }-{\mathrm{qos}}_j^{\min }\&NotEqual;0\end{array}\right.$ Formula 2

Wherein, qos _{i, j}representing matrix QoS _{m+n, k}in the i-th row jth row qos value, 1 <=i <=m+n, 1 <=j <=k, be illustrated respectively in matrix QoS _{m+n, k}the minimum value of middle jth row qos parameter value and maximum, represent the qos value of the i-th row jth row after standardization, normalized matrix is expressed as:

Separation criterion matrix obtain standardization resource QoS matrix and task QoS matrix, be expressed as with

6. the cloud computing method for scheduling task of the multi-QoS constraint according to any one of claim 2 to 5, is characterized in that: the calculating of described resource QoS satisfaction is:

For positive tolerance, calculate with formula 1 satisfaction that user obtains on this dimension QoS, for negative metric, calculate satisfaction with formula 2;

${\mathrm{QoS}}_{i,j}^{Deg}=\left\{\begin{array}{cc}\frac{{\mathrm{qos}}_{i,j}^{SR}}{{\mathrm{qos}}_{i,j}^{ST}},& {\mathrm{qos}}_{i,j}^{SR}<{\mathrm{qos}}_{i,j}^{ST}\\ 1,& {\mathrm{qos}}_{i,j}^{SR}\&GreaterEqual;{\mathrm{qos}}_{i,j}^{ST}\end{array}\right.$ Formula 1

${\mathrm{QoS}}_{i,j}^{Deg}=\left\{\begin{array}{cc}\frac{{\mathrm{qos}}_{i,j}^{ST}}{{\mathrm{qos}}_{i,j}^{ST}},& {\mathrm{qos}}_{i,j}^{ST}<{\mathrm{qos}}_{i,j}^{SR}\\ 1,& {\mathrm{qos}}_{i,j}^{ST}\&GreaterEqual;{\mathrm{qos}}_{i,j}^{SR}\end{array}\right.$ Formula 2

Then user task t _icomprehensive satisfaction in performed resource on each dimension QoS is:

${\mathrm{QoS}}_i^{Deg}=\frac{{\mathrm{\&Sigma;}}_{j=1}^k{\mathrm{QoS}}_{i,j}^{Deg}}{k}$

The average QoS satisfaction of all user tasks is:

${\mathrm{QoS}}^{Deg}=\frac{{\mathrm{\&Sigma;}}_{i=1}^n{\mathrm{QoS}}_i^{Deg}}{n}$

In task scheduling process, select the highest resource of satisfaction to distribute as far as possible.

7. the cloud computing method for scheduling task of the multi-QoS constraint according to any one of claim 2 to 5, is characterized in that: described QoS distance calculates and is:

Utilize Weighted distance to the QoS distance between the task of measuring and resource,

$D=\sqrt{{\mathrm{\&Sigma;}}_{j=1}^k{w}_j{({\mathrm{qos}}_{i,j}^{ST}-{\mathrm{qos}}_{i,j}^{SR})}^2}$

$\begin{array}{cc}s.t.& w_j\&GreaterEqual;0,j=1,2,...,k,{\mathrm{\&Sigma;}}_{j=1}^k{w}_j^2=1\end{array}$

Wherein, w _jrepresent jth dimension QoS shared weight when distance calculates; w _jmaximizing deviation method is used to determine size; The capacity of water gap that resource provides on jth dimension QoS is larger, illustrates that the impact of this parameter when measuring distance is larger, corresponding w _jlarger, otherwise, w _jless; Consider that the capacity of water that provide of all resources on jth dimension QoS is equal, then this parameter affects when measuring distance is 0, now should w _jbe decided to be 0; For jth dimension QoS, use D _{i, j}w () represents resource ri and other resources deviation in this QoS service ability, then

$D_{i,j}\left(w\right)={\mathrm{\&Sigma;}}_{i^{\&prime;}=1}^m|{\mathrm{rs}}_{i,j}{w}_j-{\mathrm{rs}}_{i^{\&prime;},j}{w}_j|,i=1,2,...,m,j=1,2,...,k$

Order

$D_j={\mathrm{\&Sigma;}}_{i=1}^m{D}_{i,j}\left(w\right)={\mathrm{\&Sigma;}}_{i=1}^m{\mathrm{\&Sigma;}}_{i^{\&prime;}=1}^m|{\mathrm{rs}}_{i,j}{w}_j-{\mathrm{rs}}_{i^{\&prime;},j}{w}_j|,j=1,2,...,k$

Dj represents the service ability of each resource and total deviation of other resource service abilities on jth dimension QoS;

The value of resource QoS comprehensive service capability and user QoS integration requirement after qos parameter standardization is calculated with following formula;

$U_{qos}\left(w\right)={\mathrm{\&Sigma;}}_{j=1}^k{\mathrm{qos}}_{i,j}{w}_j$

For each resource, U _qosw () is larger, its comprehensive QoS service ability is better; Construct target function thus,

$\max D\left(w\right)={\mathrm{\&Sigma;}}_{j=1}^m{D}_j\left(w\right)={\mathrm{\&Sigma;}}_{j=1}^k{\mathrm{\&Sigma;}}_{i=1}^m{\mathrm{\&Sigma;}}_{i^{\&prime;}=1}^m|{\mathrm{rs}}_{i,j}-{\mathrm{rs}}_{i^{\&prime;},j}|\&times;w_j$

Be equal to

$\max D\left(w\right)={\mathrm{\&Sigma;}}_{j=1}^k{\mathrm{\&Sigma;}}_{i=1}^m{\mathrm{\&Sigma;}}_{i^{\&prime;}=1}^m|{\mathrm{rs}}_{i,j}-{\mathrm{rs}}_{i^{\&prime;},j}|\&times;w_j$

$\begin{array}{cc}s.t.& w_j\&GreaterEqual;0,j=1,2,...,k,{\mathrm{\&Sigma;}}_{j=1}^k{w}_j^2=1\end{array}$

By Lagrangian method can try to achieve target function maximum time,

$w_j=\frac{{\mathrm{\&Sigma;}}_{i=1}^m{\mathrm{\&Sigma;}}_{i^{\&prime;}=1}^m|{\mathrm{rs}}_{i,j}-{\mathrm{rs}}_{i^{\&prime;},j}|}{{\mathrm{\&Sigma;}}_{j=1}^k{\mathrm{\&Sigma;}}_{i=1}^m{\mathrm{\&Sigma;}}_{i^{\&prime;}=1}^m|{\mathrm{rs}}_{i,j}-{\mathrm{rs}}_{i^{\&prime;},j}|},j=1,2,...,k.$

8. the cloud computing method for scheduling task of multi-QoS constraint according to claim 6, is characterized in that: described QoS distance calculates and is:

Utilize Weighted distance to the QoS distance between the task of measuring and resource,

$D=\sqrt{{\mathrm{\&Sigma;}}_{j=1}^k{w}_j{({\mathrm{qos}}_{i,j}^{ST}-{\mathrm{qos}}_{i,j}^{SR})}^2}$

$\begin{array}{cc}s.t.& w_j\&GreaterEqual;0,j=1,2,...,k,{\mathrm{\&Sigma;}}_{j=1}^k{w}_j^2=1\end{array}$

Wherein, w _jrepresent jth dimension QoS shared weight when distance calculates; w _jmaximizing deviation method is used to determine size; The capacity of water gap that resource provides on jth dimension QoS is larger, illustrates that the impact of this parameter when measuring distance is larger, corresponding w _jlarger, otherwise, w _jless; Consider that the capacity of water that provide of all resources on jth dimension QoS is equal, then this parameter affects when measuring distance is 0, now should w _jbe decided to be 0; For jth dimension QoS, use D _{i, j}w () represents resource ri and other resources deviation in this QoS service ability, then

$D_{i,j}\left(w\right)={\mathrm{\&Sigma;}}_{i^{\&prime;}=1}^m|{\mathrm{rs}}_{i,j}{w}_j-{\mathrm{rs}}_{i^{\&prime;},j}{w}_j|,i=1,2,...,m,j=1,2,...,k$

Order

$D_j={\mathrm{\&Sigma;}}_{i=1}^m{D}_{i,j}\left(w\right)={\mathrm{\&Sigma;}}_{i=1}^m{\mathrm{\&Sigma;}}_{i^{\&prime;}=1}^m|{\mathrm{rs}}_{i,j}{w}_j-{\mathrm{rs}}_{i^{\&prime;},j}{w}_j|,j=1,2,...,k$

Dj represents the service ability of each resource and total deviation of other resource service abilities on jth dimension QoS;

The value of resource QoS comprehensive service capability and user QoS integration requirement after qos parameter standardization is calculated with following formula;

$U_{qos}\left(w\right)={\mathrm{\&Sigma;}}_{j=1}^k{\mathrm{qos}}_{i,j}{w}_j$

For each resource, U _qosw () is larger, its comprehensive QoS service ability is better; Construct target function thus,

$\max D\left(w\right)={\mathrm{\&Sigma;}}_{j=1}^m{D}_j\left(w\right)={\mathrm{\&Sigma;}}_{j=1}^k{\mathrm{\&Sigma;}}_{i=1}^m{\mathrm{\&Sigma;}}_{i^{\&prime;}=1}^m|{\mathrm{rs}}_{i,j}-{\mathrm{rs}}_{i^{\&prime;},j}|\&times;w_j$

Be equal to

$\max D\left(w\right)={\mathrm{\&Sigma;}}_{j=1}^k{\mathrm{\&Sigma;}}_{i=1}^m{\mathrm{\&Sigma;}}_{i^{\&prime;}=1}^m|{\mathrm{rs}}_{i,j}-{\mathrm{rs}}_{i^{\&prime;},j}|\&times;w_j$ $\begin{array}{cc}s.t.& w_j\&GreaterEqual;0,j=1,2,...,k,{\mathrm{\&Sigma;}}_{j=1}^k{w}_j^2=1\end{array}$

By Lagrangian method can try to achieve target function maximum time,

$w_j=\frac{{\mathrm{\&Sigma;}}_{i=1}^m{\mathrm{\&Sigma;}}_{i^{\&prime;}=1}^m|{\mathrm{rs}}_{i,j}-{\mathrm{rs}}_{i^{\&prime;},j}|}{{\mathrm{\&Sigma;}}_{j=1}^k{\mathrm{\&Sigma;}}_{i=1}^m{\mathrm{\&Sigma;}}_{i^{\&prime;}=1}^m|{\mathrm{rs}}_{i,j}-{\mathrm{rs}}_{i^{\&prime;},j}|},j=1,2,...,k.$