CN104102544A - Multi QoS (quality of service)-constrained parallel task scheduling cost optimizing method under mixed cloud environment - Google Patents

Abstract

The invention relates to a multi QoS (quality of service)-constrained parallel task scheduling cost optimizing method under a mixed cloud environment. The method comprises the steps of task scheduling of a private cloud, task rescheduling and the minimizing of the leasing cost of public resources. In the task scheduling of the private cloud, resources are distributed for the tasks according to the improved maximal and minimum strategies, and a fast heuristic algorithm-TSOPR is provided. The task rescheduling step is finished by an RSD algorithm and comprises steps of deciding the task type distributed to a public cloud, judging whether the public cloud can meet the final time constraint and the budget constraint or not; generating a scheduling table for the task in the submitted operation by a system if all constraint conditions are met. The leasing cost of the public cloud resources can be minimized under the premise that the budget control constraint is met.

Description

The Parallel Task Scheduling Cost Optimization Approach of multi-QoS constraint under mixed cloud environment

Technical field

The present invention relates to the method for scheduling task under a kind of mixed cloud environment, particularly the Parallel Task Scheduling Cost Optimization Approach of multi-QoS constraint under the mixed cloud environment of deadline date constraint and budget control constraint.

Background technology

Cloud computing in recent years becomes a research topic becoming more and more important.Many existing cloud service platforms, cloud computing platform as blue in Google cloud computing platform, IBM, Amazon elasticity calculate cloud and Microsoft cloud computing platform verified they success and to the public, provide the i.e. mode of use of a kind of payable at sight.Cloud environment can be divided into privately owned cloud and publicly-owned cloud simply.Move a privately owned cloud data center, mechanism of data center will buy, safeguards, manages and operate all software and hardware infrastructure, but still is faced with the risk that supply falls short of demand.There is the continuous 46Tian dynamic monitoring of the researchist U.S. second largest Online Video sharing website Yahoo video workload hourly.Peak load is far longer than mean value, but peak load is of short duration and uncertain.If institute's Constrained of operating load is attempted to meet in a private data center, peak load will force owner in privately owned cloud, to invest more hardware resource.In this case, it can cause most of the time hardware resource be waste.The payable at sight i.e. publicly-owned cloud of use can help us to process these unpredictable workload peak values, and be only used in that publicly-owned cloud processes overload task need during this period of time to pay extra-pay, and in privately owned cloud without any unnecessary resource.Therefore,, if privately owned cloud exists, the method for building mixed cloud can be avoided the waste of lower deployment cost and operation cost.And Parallel Task Scheduling is one of main challenge of mixed cloud.

Different scheduling strategies may change resource utilization, response time, reliability, operating cost and maintenance cost.In addition, observe under the prerequisite of QoS constraint in system, user prefers cost-effective mode Gains resources conventionally.When submitting task to for user, mixed cloud select different service levels that dirigibility is provided.Because publicly-owned cloud is the i.e. use of payable at sight, when privately owned cloud can not be met consumers' demand, find the publicly-owned cloud resource that meets QoS constraint and cost benefit optimum to become an important subject of mixed cloud.Researchist distributes with scheduling problem and has done a large amount of research task both at home and abroad, and a lot of heuristic driving methods have been proposed, yet these methods are all for grid or the isomerous environment such as distributed, can not be applied in mixed cloud environment, and the mixed cloud task scheduling Cost Optimization Approach proposing is at present less.

Therefore, be necessary to provide the Parallel Task Scheduling Cost Optimization Approach of multi-QoS constraint under mixed cloud environment to overcome above-mentioned the deficiencies in the prior art.

Summary of the invention

Parallel Task Scheduling Cost Optimization Approach (the Multi-QoS Constraints Cost Optimal Algorithm for Parallel Task Scheduling in Hybrid Cloud that the object of the invention provides multi-QoS constraint under a kind of mixed cloud environment of the present invention with regard to being to overcome above-mentioned the deficiencies in the prior art, hereinafter to be referred as MQCOHC), the method can minimize the hiring cost of publicly-owned cloud resource under the prerequisite that meets budget control constraint.

The technical scheme that realizes the object of the invention employing is the Parallel Task Scheduling Cost Optimization Approach of multi-QoS constraint under a kind of mixed cloud environment, and the method comprises privately owned cloud task scheduling and publicly-owned cloud task scheduling,

Described privately owned cloud task scheduling comprises: in mixed cloud environment, according to operation, need to distribute for the operation of each submission privately owned cloud resource slot, if privately owned cloud resource slot can not meet the deadline date constraint of operation, according to the deadline date constraint of operation, determine that the task of which operation should be dispatched to publicly-owned cloud, privately owned cloud also generates two dispatch lists of this operation, one for one of privately owned cloud scheduling for publicly-owned cloud scheduling;

Described publicly-owned cloud task scheduling comprises that task reschedules with publicly-owned resource and minimizes hiring cost, wherein,

Described task reschedules and comprises: determine that what task should be arranged into publicly-owned cloud; Judge whether publicly-owned cloud can meet deadline date constraint and budgetary restraints; If constraint condition can meet, system is that the task in submit job generates a dispatch list;

Described publicly-owned resource minimizes hiring cost resource slot according to resource quality, CPU T.T., always charge by storage space and wastage in bulk or weight bandwidth, minimizes the hiring cost of publicly-owned cloud resource under the prerequisite that meets budget control constraint.

In technique scheme, described privately owned cloud task scheduling comprises the following steps:

(1) definition J _i, PrR _iand RT{1 ... m}; J wherein _irepresent to need the operation of submission, this operation comprises n task, PrR _irepresent to distribute privately owned cloud resource slot, and this privately owned resource slot quantity is m, RT{1 ... privately owned resource slot of m} record becomes the available shortest time from current;

(2) according to data volume size DS _i,joperation J _iall tasks according to descending sort;

(3) calculate each task T _i,jestimation execution time Eet[i, j, k];

(4) give task T _i,jresources allocation groove, and record mapping;

(5) if all RT are less than deadline date constraint, return to a privately owned cloud schedule of tasks; Otherwise, utilize QoS comprehensive estimation method to select and operation J _ithe publicly-owned cloud resource that matches of safety and reliability.

In technique scheme, determine that what task should be arranged into publicly-owned cloud and comprise the following steps:

(1) definition Z _i, PrR _i, RT[1 ... m], JR; Z wherein _irepresent operation J _ibe arranged in the dispatch list moving on privately owned cloud, PrR _irepresent privately owned cloud resource slot, its resource slot quantity is m, RT[1 ... m] represent that privately owned resource slot of record becomes the available shortest time from current, JR is the Output rusults of QoS comprehensive estimation method;

(2) task-set that need to be assigned on publicly-owned cloud is put sky, meter TPPU=φ;

(3) task is carried out to ascending sort according to the size of estimating the execution time and obtain TPPR;

(4) when the execution time of the upper task of privately owned resource slot k, be greater than the final coutoff time limit, inquiry is distributed in the task-set of resource slot k, adds this task-set to TPPU, and front n the task in privately owned cloud resource slot that be about to moves on to TPPU from TPPR;

(5) deadline of calculating publicly-owned cloud task is while being less than the final coutoff limit, the more individual task of n ' is moved on to TPPR from TPPU;

(6) output two tuple <Z ' _i, TP _i>, wherein Z ' _ioperation J _idispatch list on publicly-owned cloud, TP _iit is the task-set of transferring on publicly-owned cloud.

The hiring cost that minimizes publicly-owned cloud resource in technique scheme under the prerequisite that meets budget control constraint comprises the following steps:

(1) definition of T P _i, PRT, t _init; TP wherein _ioperation J _ioperate in the task-set on publicly-owned cloud, PRT is the set of types of publicly-owned cloud resource slot, t _initit is the time initial value of publicly-owned cloud resource slot;

(2) initializing variable, NR _i=φ, Z _i=φ, TotalCost=0, wherein NR _irepresent one group of publicly-owned cloud resource slot, Z _ithe dispatch list that represents task on publicly-owned cloud, TotalCost represents the hiring cost of publicly-owned cloud;

(3) on publicly-owned cloud for task is found suitable resource type, it is minimum that this resource meets under the condition of deadline date constraint price;

(4), if best resource type is found, system can on publicly-owned cloud, create an example types and allocating task arrives this publicly-owned cloud resource example, i.e. allocating task T _i,jto resource k and at Z _imiddle record mapping;

(5) calculate the T moving on k _i,jcost and join TotalCost;

(6) return to the dispatch list of publicly-owned cloud and correspondingly dispatch publicly-owned cloud resource slot for the task in operation.

The present invention utilizes execution time evaluation method and fast dispatch method, to retrain the deadline date with budget control constraint and be incorporated in method for scheduling task, meeting under the prerequisite of deadline date constraint and budget control constraint, minimizing the cost of the publicly-owned cloud resource of lease.

Accompanying drawing explanation

Fig. 1 is parallel task adaptive scheduling model under mixed cloud environment.

Fig. 2 is the process flow diagram of the Parallel Task Scheduling Cost Optimization Approach of multi-QoS constraint under mixed cloud environment of the present invention.

Embodiment

Below in conjunction with the drawings and specific embodiments, the present invention is described in further detail.

Mixed cloud can be controlled cost by its inner base facility and publicly-owned cloud resource, meets the real needs of data security, security, performance and delay aspect.The IT platform of each enterprise has network, server and the storage hardware of oneself, so just can form a privately owned cloud.From the viewpoint of the work for the treatment of load peak problem of cost, privately owned cloud resource can dynamically be added a publicly-owned cloud resource to and be formed a mixed cloud environment.

Cloud alliance is intended to focus on cost benefit and resource optimization, and under isomerous environment, all clouds coact to obtain unlimited computational resource, therefore becomes new business opportunity.When the privately owned cloud of any one tissue reaches a specific operating load threshold value, once demand, can select a publicly-owned cloud resource and become mixed cloud environment.

Parallel task adaptive scheduling model under model of the present invention mixed cloud environment as shown in Figure 1, user is by an interface Gains resources, and this interface allows user to set the budget of application program capacity, deadline date and the publicly-owned cloud of use.When a new operation (task) arrives, dispatcher components will be triggered, and then scheduler program obtains information and the resource of task, as size of data, load etc.Dissimilar scheduler program may have different sub-components, and the scheduling mechanism of proposition contains four sub-components: execution time algorithm for estimating, cost function, dynamic programming module and scheduling selection algorithm.Estimate that execution time and cost function are by the information acquisition of task and resource.Wherein, algorithm for estimating is estimated respectively execution time, transmission time and deadline; Cost function generates its value at cost to different publicly-owned cloud resource slots.Result based on execution time algorithm for estimating and cost function, the scheduling mechanism of proposition adopts dynamic programming technology from cost value and completes time limit aspect and calculates best resource configuration.Scheduling selection algorithm, by carrying out assigned tasks by the result of dynamic programming assembly, guarantees that the operation of all submissions can also in the end complete before the time limit with least cost.These tasks are specifically dispatched to privately owned cloud or publicly-owned cloud, and this will depend on the result of method for scheduling and selecting.

Under mixed cloud environment of the present invention in the Parallel Task Scheduling Cost Optimization Approach of multi-QoS constraint, Mission Scheduling under QoS constraint is converted to a kind of multidimensional multiple-choice knapsack problem of variation, utilize execution time evaluation method and fast dispatch method, task execution time is dropped to minimum, and the expense of renting publicly-owned cloud resource can be controlled and is a part of the budget.Be the utilization factor that MQCOHC maximizes privately owned cloud, minimize the hiring cost of publicly-owned cloud.

Under the mixed cloud environment of above-mentioned proposition, parallel task adaptive scheduling model parameter is carried out as given a definition:

Under the mixed cloud environment of above-mentioned proposition, in parallel task adaptive scheduling model, user submits to mixed cloud by operation, and each operation comprises n task.A plurality of tasks can be processed by a resource slot, and a resource slot is configured to a virtual machine conventionally.Define operation and task below:

Definition 1 (operation J _iwith deadline D _i) operation i is defined as J _i, in model, an operation i may be by n task { T _{i, 1}, T _{i, 2}..., T _i,j..., T _i,nform T _i,jrepresent j task and 1≤j≤n in operation i.Each operation i has a deadline D _i, this is user-defined, is the maximum execution time of operation i.

Definition 2 (task T _i,j) T _i,joperation J _ia task, be the base unit that user asks, task of a resource slot single treatment.Task is four-tuple T _i,j={ D _i, SC _i,j, SD _i,j, M _i, wherein:

(1) D _ibe deadline, represent that user is to operation J _ithe deadline date of definition.Each operation J _ithere is a deadline D _i, this is user-defined, is operation J _imaximum execution time.Operation J _ieach task in the end carry out completely before the time limit and result returned to user.If the operation deadline surpasses the time limit of appointment, just violated QoS constraint.

(2) SC _i,jbe operating load, represent task T _i,jthe size of workload, be task T _i,jthe time of being carried out by a standard resource groove.

(3) SD _i,jbe data volume size, represent task T _i,jthe size of data.It affects the time of data transmission.Data volume is weighed with MBs.

(4) M _ibe executory cost, be illustrated in and in publicly-owned cloud, carry out operation J _iexpense.

In the mixed cloud scheduling model proposing, physical machine is the resource kernel as much slot count object resource that can have with CPU.Although the more slot of CPU core that physical machine can distribution ratio physical machine, in the situation that surpassing supply, the efficiency of resource slot will significantly decline.Resource slot has different computing powers according to the mode of the computing power of resource and resource sharing.According to existing cloud system, the publicly-owned cloud resource with different prices may have different performances.Define resource slot below:

Definition 3 (resource slot) resource slot k is expressed as P _k, P _kby privately owned cloud or publicly-owned cloud, created.Seven tuple P for resource slot _k={ pr μ _k, x _k, y _k, dti _k, dto _k, NB, L _krepresent.

(1) pr μ _kbe the computing power of privately owned cloud resource slot k, represent the computing power of resource slot k million instructions per second.

(2) x _kbe assessing the cost of publicly-owned cloud resource slot k, represent the executory cost of every 1,000,000 instructions in publicly-owned cloud.

(3) y _kbe the carrying cost of publicly-owned cloud resource slot k, be illustrated in the cost of the every MB of save data in publicly-owned cloud.

(4) dti _kbe the cost to resource slot k input data, represent that input data are to the cost of resource slot.

(5) dto _kbe the cost of resource slot k output data, represent from the cost of resource slot output data.

(6) NB is the network bandwidth between privately owned cloud and publicly-owned cloud, and it affects the transmission time of data.

(7) L _kthe buffer memory task in resource slot k, a plurality of copies that it comprises task.When an operation is sent to privately owned cloud, the task of operation by automatic deployment to privately owned cloud resource slot.

In the mixed cloud Task Scheduling Model proposing, private data center operations and maintenance cost are thought low-down so negligible, are set to x _k=0, y _k=0, dti _k=0 and dto _k=0.Yet the publicly-owned cloud resource slot of lease has various prices.This is because there is different pricing strategies in different publicly-owned cloud provider.If only consider operation and maintenance cost, the resource of common publicly-owned cloud is more than the resource costliness of privately owned cloud.

Service quality (QoS) is an overall target, different priority, user, data stream can be provided different application programs or guarantee the data stream of certain performance.The mixed cloud scheduling model proposing focuses on execution time (deadline date) and the price (value at cost) of QoS standard.Next define the parameter of required QoS:

Definition 4 (estimated time to completion Est _k) Est _krepresent the time that resource slot k has estimated.It is determined by the residue workload size operating on resource slot k.Based on this, estimate, when having new task to be used to this resource, we can predict estimated time to completion.

Definition 5 (data transmission period Dtt) data transmission occurs in ought not have task T by a resource slot k _ijdata time.If desired these data can be transferred to resource slot.Transmission time is depended on network bandwidth NB.Data transmission period Dtt is as given a definition:

$\mathrm{Dtt}[i,j,k]=\frac{{\mathrm{SD}}_{i,j}}{\mathrm{NB}}---\left(1\right)$

Definition 6 (estimating execution time EEt) estimates that the execution time equals workload size SC _i,jcomputing power p divided by resource slot k _kadd data transmission period Dtt.Estimation task was necessary in the execution time of different resource groove, Resources allocation that like this can be appropriate for the operation with specific QoS standard.Estimate that execution time EEt is as given a definition:

$\mathrm{EEt}[i,j,k]=\frac{{\mathrm{SC}}_{i,j}}{P_k}+\mathrm{Dtt}[i,j,k]---\left(2\right)$

The general publicly-owned cloud service of definition 7 (cost function CostF) provider, chargeable service has three aspects: calculating, storage and data transmission.Therefore, cost function can pass through calculation task T _i,jworkload size SC _i,jcarry out computing cost, calculate the price xk that rents a resource slot, data volume size SD _i,jstorage overhead, rent the storage overhead y of stores service _k, size of data SD _i,jtransport overhead, data inputs expense dti _kwith data output expense dto _k.In privately owned cloud, the expense that we arrange any resource slot equals 0.

CostF[i,j,k]＝SC _i,j×x _k+SD _i,j×y _k+SD _i,j×(dti _k+dto _k) (3)

Definition 8 (deadline date constraint) given operation J _i={ T _{i, 1}, T _{i, 2}..., T _i,n, k resource slot, the deadline date is D _i, operation J _ideadline be the time that last resource slot is finished the work to be less than or equal to D _i.Because the deadline date only relates to the resource slot for Activity Calculation, we arrange two choice variable a _{i, j, k}and b _k.A _{i, j, k}expression task T _i,jwhether distribute to resource slot k, work as a _{i, j, k}=1 represents task T _i,jdistribute to resource slot k, work as a _{i, j, k}=0, otherwise.B _kexpression resource slot, whether being used, is worked as b _k=1 represents that resource slot k, being used, works as b _k=0, otherwise.Deadline date constraint must meet formula below:

$\underset{k=1~m}{\max }((\underset{j=1}{\overset{n}{\mathrm{\&Sigma;}}}\left(\mathrm{EEt}\right[i,j,k]\&times;a_{i,j,k})+{\mathrm{Est}}_k)\&times;b_k)\&le;D_i---\left(4\right)$

Definition 9 (budget control constraint) given operation J _i, budget M _ibe a user-defined variable, represent operation J _ithe expense of carrying out in publicly-owned cloud.In other words, workload size is SC _i,jwith data volume size be SD _i,jtask T _i,jexecution is at public resource groove q ∈ { PuR ₁, PuR ₂..., PuR _mexpense is lower than the x that assesses the cost _qwith carrying cost y _qand, i.e. budget M _i.Budget control constraint is as given a definition:

$\underset{k=1}{\overset{m}{\mathrm{\&Sigma;}}}\underset{j=1}{\overset{n}{\mathrm{\&Sigma;}}}\left(\mathrm{CostF}\right[i,j,k]\&times;a_{i,j,k})\&le;M_i---\left(5\right)$

Multidimensional multiple-choice knapsack problem DO-MMKP (the Dual-Objective Multi-dimension Multi-choice Knapsack Problem)

Because this patent need to be each task Resources allocation groove, minimize the executory cost on publicly-owned cloud and minimize total execution time.Therefore, partner selection can be summed up as Bi-objective multidimensional multiple-choice knapsack problem, a given operation J _icomprise n task, had m available resources groove (to be defined as resource slot D _i>Est _k), Bi-objective multidimensional more options problem can be as given a definition:

Define 10 Bi-objective multidimensional multiple-choice knapsack problem TDO-MMKP (The Dual-Objective Multi-dimension Multi-choice Knapsack Problem)

function)

Make to estimate that the execution time is less than the deadline date

The present invention need to be each task Resources allocation groove, and mixed cloud scheduling mechanism must construct dispatch list and to meet user's QoS constraint and the privately owned cloud of maximum using, minimize the cost of the publicly-owned cloud of use.Partner selection is mapped as the multidimensional multiple-choice knapsack problem of a variation: 1., in TDO-MMKP, each task of operation is mapped to a resource slot, and each resource slot is mapped to an object in a group; 2. in TDO-MMKP, the QoS of each task constraint is mapped to the required resource of object.3. the first profit that cost function is mapped to object must be optimized; 4. use the second profit that least estimated deadline of resource slot is mapped to object to optimize; 5. deadline date constraint and budgetary restraints are looked at as the restriction of available resources in knapsack; 6. in TDO-MMKP, with choice variable, represent that whether a certain object in a group is selected; 7. the solution that has better the first profit is good solution, if but two schemes have identical the first profit, and the scheme that has better the second profit is better.

On m resource slot for optimum solution is found in the scheduling with n task of constrained, we list likely in situation, by select an object (resource slot) from each group of objects (task), then calculate its corresponding profit.This solution is only feasible when n and m are smaller, and time complexity is O (n ^m).Unfortunately, the resource slot quantity of publicly-owned cloud resource is very large.Therefore, under a mixed cloud environment, it may find optimum solution hardly.Must use heuristic to solve this problem, at this, we adopt improved Max-Min algorithm (specifically seeing 5.1 joints), and the time complexity of Max-Min algorithm is O (n ²m), so just greatly reduce the time of finding optimum solution, can realize the optimization of Bi-objective.

Under mixed cloud environment of the present invention, the Parallel Task Scheduling Cost Optimization Approach of multi-QoS constraint specifically comprises the following steps:

S100, privately owned cloud task scheduling TSOPR (Task Scheduling On Private)

In mixed cloud environment, according to operation, need to distribute for the operation of each submission privately owned cloud resource slot, if privately owned cloud resource slot can not meet the deadline date constraint of operation, according to the deadline date constraint of operation, determine that the task of which operation should be dispatched to publicly-owned cloud, privately owned cloud also generates two dispatch lists of this operation, one for one of privately owned cloud scheduling for publicly-owned cloud scheduling.

When operation is distributed to one group of privately owned cloud resource slot and started to carry out, task dispatch will be to each task Resources allocation groove.Based on above-mentioned definition 10, it is cost that first object task scheduling optimizes to use resource.Suppose that the privately owned cloud resource slot using is that expense is 0, if therefore privately owned cloud can be processed the operation of submission, scheduler program can directly concentrate on the second target execution time and optimize.

The present invention adopts the strategy of improved Max-Min, the maximal workload of this policy selection task and the minimum completion time of task.The present invention is by a Fast Heuristic Algorithm-TSOPR, calculates a good solution and do not need to consume the too many time.

The step of TSOPR method

(1) definition J _i, PrR _iand RT{1 ... m}; J wherein _irepresent to need the operation of submission, this operation comprises n task, PrR _irepresent to distribute privately owned cloud resource slot, and this privately owned resource slot quantity is m, RT{1 ... privately owned resource slot of m} record becomes the available shortest time from current;

(2) according to data volume size DS _i,joperation J _iall tasks according to descending sort;

(3) calculate each task T _i,jestimation execution time Eet[i, j, k];

(4) give task T _i,jresources allocation groove, and record mapping;

(5) if all RT are less than deadline date constraint, return to a privately owned cloud schedule of tasks; Otherwise, utilize QoS comprehensive estimation method to select and operation J _ithe publicly-owned cloud resource that matches of safety and reliability, enter RSD method step.

S200, publicly-owned cloud task scheduling

Publicly-owned cloud task scheduling comprises that task reschedules with publicly-owned resource and minimizes hiring cost, specifically comprises the following steps:

S201, task reschedule RSD (Rescheduling Decision)

When privately owned resource slot is occupied, be finished, the operation of new incoming may be dispatched to publicly-owned cloud by needs in order to meet its time limit constraint.Problem is that we need to be dispatch list of Activity Calculation in mixed cloud.Scheduling arrangement need to meet deadline date constraint, rents public resource groove and minimizes cost and minimize the operation deadline.Here we adopt task to reschedule technology and solve this problem.When an operation fails to be arranged to privately owned cloud, this system should be assigned to some tasks publicly-owned cloud, and each task in operation meets deadline date constraint like this.The scheduling mechanism proposing completes this target by four steps: 1. system must determine that what task should be arranged into publicly-owned cloud; 2. system need to judge whether publicly-owned cloud can meet deadline date constraint and budgetary restraints; If 3. constraint condition can meet, system should generate a dispatch list for the task in submit job; 4. minimize the hiring cost of publicly-owned cloud.

When an operation has arrived privately owned cloud, scheduler program should distribute privately owned resource and use TSOPR algorithm for dispatch list of Activity Calculation for this operation, can determine so whether operation can move on privately owned cloud.If of course, based on calculating scheduling, scheduler program directly sends task and their data arrive corresponding privately owned cloud resource slot.Otherwise, calculating scheduling and in the end under time limit constraint condition, fulfil assignment, scheduler program will call RSD algorithm, determine that what task should be arranged into publicly-owned cloud.

RSD algorithm need to be from three parameters of TSOPR arithmetic result.

The step of RSD method

(1) definition Z _i, PrR _i, RT[1 ... m], JR; Z wherein _irepresent operation J _ibe arranged in the dispatch list moving on privately owned cloud, PrR _irepresent privately owned cloud resource slot, its resource slot quantity is m, RT[1 ... m] represent that privately owned resource slot of record becomes the available shortest time from current, JR is the Output rusults of QoS comprehensive estimation method;

(2) task-set that need to be assigned on publicly-owned cloud is put sky, TPPU=φ;

(3) task is carried out to ascending sort according to the size of estimating the execution time and obtain TPPR;

(4) when the execution time of the upper task of privately owned resource slot k, be greater than the final coutoff time limit, inquiry is distributed in the task-set of resource slot k, adds this task-set to TPPU, and front n the task in privately owned cloud resource slot that be about to moves on to TPPU from TPPR;

(5) deadline of calculating publicly-owned cloud task is while being less than the final coutoff limit, the more individual task of n ' is moved on to TPPR from TPPU;

(6) output two tuple <Z ' _i, TP _i>, wherein Z ' _ioperation J _idispatch list on publicly-owned cloud, TP _iit is the task-set of transferring on publicly-owned cloud.

S202, publicly-owned resource minimize hiring cost MRCPR (Minimizing Renting Cost of Public Resource)

After allocating task arrives publicly-owned cloud, next step is exactly to minimize hiring cost and form the schedule of tasks on a publicly-owned cloud for the scheduling mechanism of proposition.Problem is that the scheduler task at publicly-owned cloud relates to publicly-owned cloud service model.User utilizes the existing mode with spot payment can rent the resource slot of any type.Resource slot is according to resource quality, CPU T.T., always charge by storage space and wastage in bulk or weight bandwidth.We adopt MRCPR algorithm, minimize the hiring cost of publicly-owned cloud resource under the prerequisite that meets budget control constraint.Three parameters that input algorithm needs, parameter TP _ifrom the result of RSD and other two parameters from scheduling mechanism data maintenance.

MRCPR method comprises the following steps:

(1) the method need to be inputted three parameters, one of them TP _iparameter is from the result and the maintenance of two other parameter from scheduling mechanism metadata of RSD method.Definition of T P _i, PRT, t _init; TP wherein _ioperation J _ioperate in the task-set on publicly-owned cloud, PRT is the set of types of publicly-owned cloud resource slot, t _initit is the time initial value of publicly-owned cloud resource slot;

(2) initializing variable, NR _i=φ, Z _i=φ, TotalCost=0, wherein NR _irepresent one group of publicly-owned cloud resource slot, Z _ithe dispatch list that represents task on publicly-owned cloud, TotalCost represents the hiring cost of publicly-owned cloud;

(3) on publicly-owned cloud for task is found suitable resource type, it is minimum that this resource meets under the condition of deadline date constraint price;

(4), if best resource type is found, system can on publicly-owned cloud, create an example types and allocating task arrives this publicly-owned cloud resource example, i.e. allocating task T _i,jto resource k and at Z _imiddle record mapping;

(5) calculate the T moving on k _i,jcost and join TotalCost;

(6) return to the dispatch list of publicly-owned cloud and correspondingly dispatch publicly-owned cloud resource slot for the task in operation.

Claims (4)

1. a Parallel Task Scheduling Cost Optimization Approach for multi-QoS constraint under mixed cloud environment, is characterized in that comprising privately owned cloud task scheduling and publicly-owned cloud task scheduling,

Described privately owned cloud task scheduling comprises: in mixed cloud environment, according to operation, need to distribute for the operation of each submission privately owned cloud resource slot, if privately owned cloud resource slot can not meet the deadline date constraint of operation, according to the deadline date constraint of operation, determine that the task of which operation should be dispatched to publicly-owned cloud, privately owned cloud also generates two dispatch lists of this operation, one for one of privately owned cloud scheduling for publicly-owned cloud scheduling;

Described publicly-owned cloud task scheduling comprises that task reschedules with publicly-owned resource and minimizes hiring cost, wherein,

Described task reschedules and comprises: determine that what task should be arranged into publicly-owned cloud; Judge whether publicly-owned cloud can meet deadline date constraint and budgetary restraints; If constraint condition can meet, system is that the task in submit job generates a dispatch list;

Described publicly-owned resource minimizes hiring cost resource slot according to resource quality, CPU T.T., always charge by storage space and wastage in bulk or weight bandwidth, minimizes the hiring cost of publicly-owned cloud resource under the prerequisite that meets budget control constraint.

2. the Parallel Task Scheduling Cost Optimization Approach of multi-QoS constraint under mixed cloud environment according to claim 1, is characterized in that described privately owned cloud task scheduling comprises the following steps:

(1) definition J _i, PrR _iand RT{1 ... m}, wherein J _irepresent to need the operation of submission, this operation comprises n task, PrR _irepresent to distribute privately owned cloud resource slot, and this privately owned resource slot quantity is m, RT{1 ... privately owned resource slot of m} record becomes the available shortest time from current;

(2) according to data volume size DS _i,joperation J _iall tasks according to descending sort;

(3) calculate each task T _i,jestimation execution time Eet[i, j, k];

(4) give task T _i,jresources allocation groove, and record mapping;

(5) if all RT are less than deadline date constraint, return to a privately owned cloud schedule of tasks; Otherwise, utilize QoS comprehensive estimation method to select and operation J _ithe publicly-owned cloud resource that matches of safety and reliability.

3. the Parallel Task Scheduling Cost Optimization Approach of multi-QoS constraint under mixed cloud environment according to claim 1, is characterized in that determining that what task should be arranged into publicly-owned cloud and comprise the following steps:

(1) definition Z _i, PrR _i, RT[1 ... m], JR; Z wherein _irepresent operation J _ibe arranged in the dispatch list moving on privately owned cloud, PrR _irepresent privately owned cloud resource slot, its resource slot quantity is m, RT[1 ... m] represent that privately owned resource slot of record becomes the available shortest time from current, JR is the Output rusults of QoS comprehensive estimation method;

(2) task-set that need to be assigned on publicly-owned cloud is put sky, meter TPPU=φ;

(3) task is carried out to ascending sort according to the size of estimating the execution time and obtain TPPR;

(4) when the execution time of the upper task of privately owned resource slot k, be greater than the final coutoff time limit, inquiry is distributed in the task-set of resource slot k, adds this task-set to TPPU, and front n the task in privately owned cloud resource slot that be about to moves on to TPPU from TPPR;

(5) deadline of calculating publicly-owned cloud task is while being less than the final coutoff limit, the more individual task of n ' is moved on to TPPR from TPPU;

(6) output two tuple <Z ' _i, TP _i>, wherein Z ' _ioperation J _idispatch list on publicly-owned cloud, TP _iit is the task-set of transferring on publicly-owned cloud.

4. the Parallel Task Scheduling Cost Optimization Approach of multi-QoS constraint under mixed cloud environment according to claim 1, is characterized in that the hiring cost that minimizes publicly-owned cloud resource under the prerequisite that meets budget control constraint comprises the following steps:

(1) definition of T P _i, PRT, t _init; TP wherein _ioperation J _ioperate in the task-set on publicly-owned cloud, PRT is the set of types of publicly-owned cloud resource slot, t _initit is the time initial value of publicly-owned cloud resource slot;

(2) initializing variable, NR _i=φ, Z _i=φ, TotalCost=0, wherein NR _irepresent one group of publicly-owned cloud resource slot, Z _ithe dispatch list that represents task on publicly-owned cloud, TotalCost represents the hiring cost of publicly-owned cloud;

(3) on publicly-owned cloud for task is found suitable resource type, it is minimum that this resource meets under the condition of deadline date constraint price;

(4), if best resource type is found, system can on publicly-owned cloud, create an example types and allocating task arrives this publicly-owned cloud resource example, i.e. allocating task T _i,jto resource k and at Z _imiddle record mapping;

(5) calculate the T moving on k _i,jcost and join TotalCost;

(6) return to the dispatch list of publicly-owned cloud and correspondingly dispatch publicly-owned cloud resource slot for the task in operation.