CN104102544B - Mix the Parallel Task Scheduling Cost Optimization Approach of multi-QoS constraint under cloud environment - Google Patents
Mix the Parallel Task Scheduling Cost Optimization Approach of multi-QoS constraint under cloud environment Download PDFInfo
- Publication number
- CN104102544B CN104102544B CN201410309665.0A CN201410309665A CN104102544B CN 104102544 B CN104102544 B CN 104102544B CN 201410309665 A CN201410309665 A CN 201410309665A CN 104102544 B CN104102544 B CN 104102544B
- Authority
- CN
- China
- Prior art keywords
- task
- resource
- public cloud
- cloud
- resource slot
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Expired - Fee Related
Links
Abstract
The present invention relates to a kind of Parallel Task Scheduling Cost Optimization Approach of multi-QoS constraint under mixing cloud environment, this method includes：Task scheduling, the task of private clound are rescheduled minimizes hiring cost with publicly-owned resource.In private clound method for scheduling task, resource is distributed to task according to the strategy of improved minimax, proposes a Fast Heuristic Algorithm-TSOPR.In task reschedules, completed according to RSD algorithms：Determine that task should be arranged into public cloud；Judge whether public cloud can meet deadline date constraint and budgetary restraints；If constraints can meet, system is that the task in operation is submitted to generate a dispatch list.The present invention can minimize the hiring cost of publicly-owned cloud resource under the premise of meeting budget control and constraining.
Description
Technical field
The present invention relates to the method for scheduling task under a kind of mixing cloud environment, more particularly to deadline date constraint and budget control
Restrict the Parallel Task Scheduling Cost Optimization Approach of multi-QoS constraint under the mixing cloud environment of beam.
Background technology
Cloud computing in recent years becomes a research topic to become more and more important.Many existing cloud service platforms, such as Google
Cloud computing platform, IBM indigo plants cloud computing platform, Amazon elastic calculations cloud and Microsoft cloud computing platforms have been proven that him
Success and provide a kind of mode of payable at sight i.e. to the public.Cloud environment can be simply divided into private clound and public cloud.Fortune
One privately owned cloud data center of row, mechanism of data center will buy, safeguard, manage and operate all software and hardware bases and set
It applies, but still is faced with the risk that supply falls short of demand.There is the continuous second largest Online Video in the 46 days dynamic monitoring U.S. of researcher point
Enjoy website Yahoo video workload hourly.Peak load is far longer than average value, but peak load is of short duration and can not
Prediction.If a private data center is intended to satisfy that institute's Constrained of workload, peak load will force owner in private
Have and invests more hardware resources in cloud.In this case, it can cause in most of the time hardware resource to be waste.I.e.
Paying public cloud i.e. can help us to handle these unpredictable workload peak values, and it is super to be only used in public cloud processing
It needs this period of load task to pay extra-pay, and there is no any extra resource in private clound.Therefore, if it is private
There is cloud to have existed, the method for building mixed cloud can be to avoid the waste of lower deployment cost and operation cost.And Parallel Task Scheduling
It is one of the significant challenge of mixed cloud.
Different scheduling strategies may change resource utilization, response time, reliability, operating cost and maintenance cost.
In addition, under the premise of system is abided by QoS and constrained, user usually prefers cost-effective mode and obtains resource.Mixing
Cloud selects different service levels to provide flexibility when submitting task for user.Since public cloud is that payable at sight is used, when privately owned
When cloud cannot meet user demand, finds and meet the publicly-owned cloud resource that QoS is constrained and cost-effectiveness is optimal as the one of mixed cloud
A important subject.Domestic and international researcher has done a large amount of research to Task matching and scheduling problem, and proposes
Many heuristic driving methods, however these methods cannot be applied both for isomerous environments such as grid or distributions mixed
It closes in cloud environment, and the mixed cloud task scheduling Cost Optimization Approach proposed at present is less.
Therefore, it is necessary to provide the Parallel Task Scheduling Cost Optimization Approach of multi-QoS constraint under mixing cloud environment to overcome
Above-mentioned the deficiencies in the prior art.
Invention content
The object of the invention, which is that, to be overcome above-mentioned the deficiencies in the prior art and provides more under a kind of present invention mixing cloud environment
Parallel Task Scheduling Cost Optimization Approach (the Multi-QoS Constraints Cost Optimal of QoS constraints
Algorithm for Parallel Task Scheduling in Hybrid Cloud, hereinafter referred to as MQCOHC), this method
The hiring cost of publicly-owned cloud resource can be minimized under the premise of meeting budget control and constraining.
The object of the invention is realized the technical solution adopted is that a kind of Parallel Task Scheduling mixing multi-QoS constraint under cloud environment
Cost Optimization Approach, this method include private clound task scheduling and public cloud task scheduling,
The private clound task scheduling includes：It is the operation each submitted according to the needs of operation in mixing cloud environment
Private clound resource slot is distributed, if private clound resource slot cannot meet the deadline date constraint of operation, according to the most later stage of operation
Limit constraint determines that the task of which operation should be dispatched to public cloud, private clound and two dispatch lists for generating the operation, one
It is a to dispatch one for public cloud scheduling for private clound；
The public cloud task scheduling includes that task reschedules and publicly-owned resource minimum hiring cost, wherein
The task reschedule including：Determine that task should be arranged into public cloud；Judge whether public cloud can be with
Meet deadline date constraint and budgetary restraints；If constraints can meet, system is that the task in operation is submitted to generate one
A dispatch list；
The publicly-owned resource minimize hiring cost resource slot according to resource quality, CPU total times, always use memory space
It charges with wastage in bulk or weight bandwidth, the hiring cost of publicly-owned cloud resource is minimized under the premise of meeting budget control and constraining.
In the above-mentioned technical solutions, the private clound task scheduling includes the following steps：
(1) J is definedi、PrRiWith RT { 1...m }；Wherein JiIndicate that the operation for needing to submit, the operation include n task,
PrRiIndicate distribution private clound resource slot, and the private privileges slot number amount be m, RT { 1...m } record a private privileges slot from
Currently become the available shortest time；
(2) according to data volume size DSi,jOperation JiAll tasks arranged according to descending；
(3) each task T is calculatedi,jEstimation execute time Eet [i, j, k]；
(4) task T is giveni,jResource slot is distributed, and records mapping；
(5) if all RT are both less than deadline date constraint, a private clound schedule of tasks is returned；Otherwise, sharp
With the selection of QoS comprehensive estimation methods and operation JiThe publicly-owned cloud resource that matches of safety and reliability.
In the above-mentioned technical solutions, determine that task should be arranged into public cloud and include the following steps：
(1) Z is definedi, PrRi, RT [1...m], JR；Wherein ZiIndicate operation JiIt is arranged in the scheduling run in private clound
Table, PrRiIndicate that private clound resource slot, resource slot quantity are m, RT [1...m] indicates one private privileges slot of record from current
Become the available shortest time, JR is the output result of QoS comprehensive estimation methods；
(2) task-set being assigned in public cloud will be needed to empty, meter TPPU=φ；
(3) task is executed into the size progress ascending sort of time according to estimation and obtains TPPR；
(4) it is more than the final coutoff time limit when the execution time of task on private privileges slot k, inquiry distribution is resource slot k's
The task-set is added to TPPU by task-set, i.e., preceding n task in private clound resource slot is moved on to TPPU from TPPR；
(5) when calculating the deadline of public cloud task less than the final coutoff limit, then a tasks of n ' are moved on to from TPPU
TPPR；
(6) two tuple ＜ Z are exportedi′,TPi＞, wherein Zi' it is operation JiDispatch list in public cloud, TPiIt is to be transferred to
Task-set in public cloud.
The hiring cost of publicly-owned cloud resource is minimized under the premise of meeting budget control and constraining in the above-mentioned technical solutions
Include the following steps：
(1) TP is definedi, PRT, tinit；Wherein TPiIt is operation JiThe task-set in public cloud is operated in, PRT is public cloud
The set of types of resource slot, tinitIt is the time initial value of public cloud resource slot；
(2) initializing variable, NRi=φ, Zi=φ, TotalCost=0, wherein NRiIndicate one group of public cloud resource slot,
ZiIndicate that the dispatch list of task in public cloud, TotalCost indicate the hiring cost of public cloud；
(3) suitable resource type is found for task in public cloud, which meets under conditions of deadline date constraint
Price is minimum；
(4) if best resource type is found, system can create an example types and be distributed in public cloud
Task distributes task T to the public cloud resource instancesi,jTo resource k and in ZiMiddle record mapping；
(5) T run on k is calculatedi,jCost and be added to TotalCost；
(6) it returns to the dispatch list of public cloud and correspondingly dispatches public cloud resource slot for the task in operation.
The present invention will be constrained with budget control about the deadline date using time estimation method and fast dispatch method is executed
Beam is introduced into method for scheduling task, and under the premise of meeting deadline date constraint and budget control constraint, it is public to minimize lease
There is the cost of cloud resource.
Description of the drawings
Fig. 1 is parallel task self-adaptable scheduling model under mixing cloud environment.
Fig. 2 is the flow chart of the Parallel Task Scheduling Cost Optimization Approach of multi-QoS constraint under present invention mixing cloud environment.
Specific implementation mode
The following further describes the present invention in detail with reference to the accompanying drawings and specific embodiments.
Mixed cloud can carry out control cost using its inner base facility and publicly-owned cloud resource, meet data security, peace
Specific requirements in terms of Quan Xing, performance and delay.The IT platforms of each enterprise have oneself network, server and storage hard
Part can thus form a private clound.The processing work load peak problem from the aspect of cost, privately owned cloud resource can
A mixing cloud environment is constituted to be dynamically added to a publicly-owned cloud resource.
Cloud alliance is intended to focus on cost-effectiveness and resource optimization, and all clouds cooperate unlimited to obtain under isomerous environment
Computing resource, therefore become new business opportunity.When the private clound of any one tissue reaches a specific workload threshold value,
Once demand, a publicly-owned cloud resource can be selected and become to mix cloud environment.
The present invention initially sets up parallel task self-adaptable scheduling model under mixing cloud environment as shown in Figure 1, and user passes through
One interface obtains resource, which allows user to set application program capacity, deadline date and the budget using public cloud.When
When one new job (task) is arrived, dispatcher components will be triggered, and then scheduler program obtains the information and resource of task,
Such as size of data, load.Different types of scheduler program may have different sub-components, the scheduling mechanism of proposition to contain
Four sub-components：Execute time Estimate algorithm, cost function, dynamic programming module and scheduling selection algorithm.Estimation executes the time
Pass through the information acquisition of task and resource with cost function.Wherein, algorithm for estimating is estimated to execute time, transmission time and complete respectively
At the time；Cost function generates its value at cost to different public cloud resource slots.Based on execution time Estimate algorithm and cost letter
It is several as a result, the scheduling mechanism of proposition calculates best resource using dynamic programming technique from cost value and in terms of completing the time limit
Configuration.Scheduling selection algorithm carrys out assigned tasks by using the result of dynamic programming component, ensures that the operation of all submissions can
It is completed with least cost and before the deadline date.These tasks are specifically dispatched to private clound or public cloud, this will depend on
In the result of method for scheduling and selecting.
The present invention mixes under cloud environment in the Parallel Task Scheduling Cost Optimization Approach of multi-QoS constraint, under QoS is constrained
Mission Scheduling is converted into a kind of multidimensional multiple-choice knapsack problem of variation, utilizes execution time estimation method and fast velocity modulation
Degree method, is preferably minimized task execution time, and the expense for renting publicly-owned cloud resource can be controlled and be a part of the budget.I.e.
MQCOHC maximizes the utilization rate of private clound, minimizes the hiring cost of public cloud.
Parallel task self-adaptable scheduling model parameter is such as given a definition under mixing cloud environment set forth above：
Under mixing cloud environment set forth above in parallel task self-adaptable scheduling model, user submits a job to mixing
Cloud, each operation include n task.Multiple tasks can be handled by a resource slot, and a resource slot is normally constructed to one
Virtual machine.Operation defined below and task：
Define 1 (operation JiWith deadline Di) operation i is defined as Ji, in a model, an operation i may be by n task
{Ti,1,Ti,2,...,Ti,j,...,Ti,nComposition, Ti,jIndicate j-th of task in operation i and 1≤j≤n.Each operation i
There are one deadline Di, this is user-defined, is the maximum execution time of operation i.
Define 2 (task Ti,j)Ti,jIt is operation JiA task, be user request base unit, a resource slot one
One task of secondary processing.Task is four-tuple Ti,j={ Di,SCi,j,SDi,j,Mi, wherein：
(1)DiIt is off the time limit, indicates user to operation JiThe deadline date of definition.Each operation JiThere are one the off periods
Limit Di, this is user-defined, is operation JiMaximum execution time.Operation JiEach task should be before the deadline date
It executes completely and returns result to user.If the operation deadline is more than the specified time limit, QoS constraints are just violated.
(2)SCi,jIt is workload, indicates task Ti,jWorkload size, be task Ti,jBy a standard resource slot
The time of execution.
(3)SDi,jIt is data volume size, indicates task Ti,jThe size of data.It influences the time of data transmission.Data volume
It is weighed with MBs.
(4)MiIt is executory cost, indicates to execute operation J in public cloudiExpense.
In the mixed cloud scheduling model of proposition, physical machine is that can have and the resource kernel of CPU slot number as much
Resource.Although a physical machine can be with the more slots of the CPU core of distribution ratio physical machine, more than supply
The efficiency of resource slot will significantly decline.Resource slot has different meters according to the computing capability of resource and the mode of resource-sharing
Calculation ability.According to existing cloud system, the publicly-owned cloud resource with different prices may have different performances.Resource defined below
Slot：
It defines 3 (resource slot) resource slot k and is expressed as Pk, PkIt is created by private clound or public cloud.Resource slot is with one seven
Tuple Pk={ pr μk,xk,yk,dtik,dtok,NB,LkIndicate.
(1)prμkIt is the computing capability of private clound resource slot k, indicates the computing capability of resource slot k million instructions per seconds.
(2)xkIt is the calculating cost of public cloud resource slot k, indicates the executory cost of every million instruction in public cloud.
(3)ykIt is the carrying cost of public cloud resource slot k, indicates to preserve cost of the data per MB in public cloud.
(4)dtikIt is to indicate the cost for inputting data into resource slot to the cost of resource slot k input datas.
(5)dtokIt is the cost of resource slot k output datas, indicates the cost from resource slot output data.
(6) NB is the network bandwidth between private clound and public cloud, it influences the transmission time of data.
(7)LkIt is the caching task in resource slot k, it includes multiple copies of task.When an operation is sent to private
There is cloud, the task of operation will be on automatic deployment to private clound resource slot.
In the mixed cloud Task Scheduling Model of proposition, private data center operations and maintenance cost are considered low-down
Therefore it can be ignored, be set as xk=0, yk=0, dtik=0 and dtok=0.However, the public cloud resource slot of lease has
Various prices.This is because there is different pricing strategies in different public cloud providers.If only considering operation and dimension
Cost is protected, the resource of usual public cloud is more than the resource-expensive of private clound.
Service quality (QoS) is an overall target, can provide different application programs different priority, use
Family, data flow or the data flow for ensureing certain performance.The mixed cloud scheduling model of proposition focuses on the execution time of QoS standards
(deadline date) and price (value at cost).Next the parameter of QoS needed for definition：
Define 4 (estimated time to completion Estk)EstkRepresent the resource slot k estimated times completed.It is by operating in resource slot k
On remaining workload size determine.Based on this estimation, when there is new task to be used for the resource, we can predict to estimate
Count the deadline.
5 (data transmission period Dtt) data transmissions are defined to be happened at when a resource slot k does not have task TijData when.
If necessary these data can be transferred to resource slot.Transmission time depends on network bandwidth NB.Data transmission period Dtt is fixed as follows
Justice：
It defines 6 (estimation executes time EEt) and estimates that executing times is equal to workload size SCi,jDivided by the calculating of resource slot k
Ability pkIn addition data transmission period Dtt.Estimation task is necessary in the execution time of different resource slot, can be in this way
The appropriate distribution resource of operation with specific QoS standard.Estimation executes time EEt and such as gives a definition：
The general publicly-owned cloud service providers of 7 (cost function CostF) are defined, there are three aspects for chargeable service：It calculates, storage
And data transmission.Therefore, cost function can pass through calculating task Ti,jWorkload size SCi,jCarry out computing cost, calculates and rent
With the price x of a resource slotk, data volume size SDi,jStorage overhead, rent the storage overhead y of storage servicek, data are big
Small SDi,jTransport overhead, data input expense dtikWith data output expense dtok.In private clound, any money is arranged in we
The expense of source slot is equal to 0.
CostF [i, j, k]=SCi,j×xk+SDi,j×yk+SDi,j×(dtik+dtok) (3)
It defines 8 (deadline date constraints) and gives an operation Ji={ Ti,1,Ti,2,...,Ti,n, k resource slot, most later stage
Limit is Di, operation JiDeadline be the last one resource slot complete task time be less than or equal to Di.Because of the deadline date
The resource slot for Activity Calculation is related only to, two selection variable a are arranged in wei,j,kAnd bk。ai,j,kExpression task Ti,jWhether
Resource slot k is distributed to, a is worked asi,j,k=1 indicates task Ti,jResource slot k is distributed to, a is worked asi,j,k=0, it is on the contrary.bkIndicating resource slot is
It is no to be used, work as bk=1 expression resource slot k is being used, and b is worked ask=0, it is on the contrary.Deadline date constraint must satisfy following public affairs
Formula：
It defines 9 (budget control constraints) and gives an operation Ji, budget MiIt is a user-defined variable, indicates operation
JiThe expense executed in public cloud.In other words, workload size is SCi,jIt is SD with data volume sizei,jTask Ti,jIt holds
Row is in public resource slot q ∈ { PuR1,PuR2,...,PuRmExpense is less than calculating cost xqWith carrying cost yqSum, i.e., in advance
Calculate Mi.Budget control constraint is such as given a definition：
Multidimensional multiple-choice knapsack problem DO-MMKP (the Dual-Objective Multi-dimension Multi-
choice Knapsack Problem)
Because this patent needs to distribute resource slot for each task, minimizes the executory cost in public cloud and minimize total
The execution time.Therefore, partner selection can be attributed to Bi-objective multidimensional multiple-choice knapsack problem, give an operation Ji
N task is contained, has m available resources slot (to be defined as resource slot Di＞ Estk), Bi-objective multidimensional more options problem can be as
Give a definition：
Define 10 Bi-objective multidimensional multiple-choice knapsack problem TDO-MMKP (The Dual-Objective Multi-
dimension Multi-choice Knapsack Problem)
Function)
The estimation execution time is set to be less than the deadline date
The present invention needs to distribute resource slot for each task, and a mixed cloud scheduling mechanism must construct dispatch list with full
The QoS of sufficient user is constrained and is maximally utilized private clound and minimize the cost using public cloud.Partner selection is mapped as
The multidimensional multiple-choice knapsack problem of one variation：1. in TDO-MMKP, each task of operation is mapped to a resource slot,
Each resource slot is mapped to an object in one group；2. in TDO-MMKP, the QoS constraint consistencies of each task to object institute
The resource needed.3. the first profit that cost function is mapped to object must optimize；4. the least estimated using resource slot is completed
Second of profit of time map to object must optimize；5. the deadline date constrains and budgetary restraints are looked at as in knapsack to use
The limitation of resource；6. in TDO-MMKP, whether it is selected come certain an object indicated in one group using selection variable；7. have compared with
The solution of the first good profit is preferable solution, but if two schemes have the first identical profit, is had
The scheme of better second of profit is more preferable.
Optimal solution is found for the scheduling of the n task with restrict on m resource slot, we enumerate
Go out in the case of being possible to, by selecting an object (resource slot) from each object group (task), it is corresponding then to calculate it
Profit.This solution is only feasible in n and m all smaller, and time complexity is O (nm).Unfortunate
It is that the resource slot quantity of publicly-owned cloud resource is very big.Therefore, it is mixed under cloud environment at one hardly possible to find most
Excellent solution.This must be solved the problems, such as using heuristic, we use improved Max-Min algorithms herein,
The time complexity of Max-Min algorithms is O (n2M), it is greatly lowered the time for finding optimal solution in this way, may be implemented double
The optimization of target.
The Parallel Task Scheduling Cost Optimization Approach that the present invention mixes multi-QoS constraint under cloud environment specifically includes following step
Suddenly：
S100, private clound task scheduling TSOPR (Task Scheduling On Private)
It is that the operation each submitted distributes private clound resource slot according to the needs of operation, if private in mixing cloud environment
There is cloud resource slot that cannot meet the deadline date constraint of operation, is constrained according to the deadline date of operation and determine appointing for which operation
Business should be dispatched to public cloud, private clound and two dispatch lists for generating the operation, dispatch one for private clound for one and be used for
Public cloud is dispatched.
When one group of private clound resource slot is distributed in operation and starts to execute, task dispatch will give each task to distribute
Resource slot.Based on above-mentioned definition 10, first object task scheduling optimizes to use resource as cost.Assuming that the private clound money used
Source slot is that expense is 0, so if private clound can handle the operation of submission, scheduler program can be focused directly at second mesh
Mark executes time-optimized.
The present invention uses the strategy of improved Max-Min, the maximal workload of the policy selection task and the minimum of task
Deadline.The present invention calculates a good solution without consuming by a Fast Heuristic Algorithm-TSOPR
The too many time.
The step of TSOPR methods
(1) J is definedi、PrRiWith RT { 1...m }；Wherein JiIndicate that the operation for needing to submit, the operation include n task,
PrRiIndicate distribution private clound resource slot, and the private privileges slot number amount be m, RT { 1...m } record a private privileges slot from
Currently become the available shortest time；
(2) according to data volume size DSi,jOperation JiAll tasks arranged according to descending；
(3) each task T is calculatedi,jEstimation execute time Eet [i, j, k]；
(4) task T is giveni,jResource slot is distributed, and records mapping；
(5) if all RT are both less than deadline date constraint, a private clound schedule of tasks is returned；Otherwise, sharp
With the selection of QoS comprehensive estimation methods and operation JiThe publicly-owned cloud resource that matches of safety and reliability, walked into RSD methods
Suddenly.
S200, public cloud task scheduling
Public cloud task scheduling include task reschedule with publicly-owned resource minimize hiring cost, specifically include following step
Suddenly：
S201, task reschedule RSD (Rescheduling Decision)
It is finished when private privileges slot is occupied, the operation of new incoming may need to be dispatched to public affairs to meet the constraint of its time limit
There is cloud.Problem is that we need for one dispatch list of Activity Calculation in mixed cloud.Schedule needs to meet the deadline date about
Beam rents public resource slot and minimizes cost and minimize the operation deadline.Here we reschedule technology using task
To solve the problems, such as this.When an operation fails to be arranged to private clound, which should be assigned to public cloud by some tasks,
Each task in this way in operation meets deadline date constraint.The scheduling mechanism of proposition completes this mesh using four steps
Mark：1. what task is system, which have to decide on, should be arranged into public cloud；2. system needs to judge whether public cloud can meet most
Time limit constraint and budgetary restraints afterwards；3. if constraints can meet, system should be that the task in operation is submitted to generate one
A dispatch list；4. minimizing the hiring cost of public cloud.
When an operation reaches private clound, scheduler program should be that this operation distributes private privileges and uses TSOPR
Algorithm is one dispatch list of Activity Calculation, can determine whether operation can run in private clound in this way.If it can, being based on
Scheduling is calculated, scheduler program directly transmits task and their data to corresponding private clound resource slot.Otherwise, calculating scheduling cannot
It fulfils assignment under deadline date constraints, scheduler program will call RSD algorithms, and it is publicly-owned to determine that task should be arranged into
Cloud.
RSD algorithms need three parameters from TSOPR arithmetic results.
The step of RSD methods
(1) Z is definedi, PrRi, RT [1...m], JR；Wherein ZiIndicate operation JiIt is arranged in the scheduling run in private clound
Table, PrRiIndicate that private clound resource slot, resource slot quantity are m, RT [1...m] indicates one private privileges slot of record from current
Become the available shortest time, JR is the output result of QoS comprehensive estimation methods；
(2) task-set being assigned in public cloud will be needed to empty, TPPU=φ；
(3) task is executed into the size progress ascending sort of time according to estimation and obtains TPPR；
(4) it is more than the final coutoff time limit when the execution time of task on private privileges slot k, inquiry distribution is resource slot k's
The task-set is added to TPPU by task-set, i.e., preceding n task in private clound resource slot is moved on to TPPU from TPPR；
(5) when calculating the deadline of public cloud task less than the final coutoff limit, then a tasks of n ' are moved on to from TPPU
TPPR；
(6) two tuple ＜ Z are exportedi′,TPi＞, wherein Zi' it is operation JiDispatch list in public cloud, TPiIt is to be transferred to
Task-set in public cloud.
Minimum hiring cost MRCPR (the Minimizing Renting Cost of Public of S202, publicly-owned resource
Resource)
After distribution task to public cloud, the scheduling mechanism of proposition is exactly to minimize hiring cost and form one in next step
Schedule of tasks in a public cloud.Problem is to be related to public cloud service model in the scheduler task of public cloud.User utilizes
The mode of current spot payment can rent any kind of resource slot.Resource slot is stored according to resource quality, CPU total times, total use
Space and wastage in bulk or weight bandwidth are charged.We use MRCPR algorithms, are minimized under the premise of meeting budget control constraint publicly-owned
The hiring cost of cloud resource.Input three parameters that algorithm needs, parameter TPiResult and other two parameters from RSD are come
Self scheduling mechanism data is safeguarded.
MRCPR methods include the following steps：
(1) this method needs to input three parameters, one of TPiResult of the parameter from RSD methods and other two
Maintenance of the parameter from scheduling mechanism metadata.Define TPi, PRT, tinit；Wherein TPiIt is operation JiIt operates in public cloud
Task-set, PRT are the set of types of public cloud resource slot, tinitIt is the time initial value of public cloud resource slot；
(2) initializing variable, NRi=φ, Zi=φ, TotalCost=0, wherein NRiIndicate one group of public cloud resource slot,
ZiIndicate that the dispatch list of task in public cloud, TotalCost indicate the hiring cost of public cloud；
(3) suitable resource type is found for task in public cloud, which meets under conditions of deadline date constraint
Price is minimum；
(4) if best resource type is found, system can create an example types and be distributed in public cloud
Task distributes task T to the public cloud resource instancesi,jTo resource k and in ZiMiddle record mapping；
(5) T run on k is calculatedi,jCost and be added to TotalCost；
(6) it returns to the dispatch list of public cloud and correspondingly dispatches public cloud resource slot for the task in operation.
Claims (1)
1. the Parallel Task Scheduling Cost Optimization Approach of multi-QoS constraint under a kind of mixing cloud environment, it is characterised in that including privately owned
Cloud task scheduling and public cloud task scheduling,
The private clound task scheduling includes：It is that the operation each submitted distributes according to the needs of operation in mixing cloud environment
Private clound resource slot, if private clound resource slot cannot meet the deadline date constraint of operation, about according to the deadline date of operation
Beam determines that the task of which operation should be dispatched to public cloud, private clound and two dispatch lists for generating the operation, a use
One is dispatched in private clound to dispatch for public cloud；The private clound task scheduling includes the following steps：
(1) J is definedi、PrRiWith RT { 1...m }, wherein JiIndicate that the operation i for needing to submit, operation i include n task, PrRi
Indicate distribution private clound resource slot, and the private clound resource slot quantity be m, RT { 1...m } record a private clound resource slot from
Currently become the available shortest time；
(2) according to data volume size DSi,jOperation JiAll tasks arranged according to descending；
(3) each task T is calculatedi,jEstimation execute time Eet [i, j, k]；
(4) task T is giveni,jResource slot is distributed, and records mapping；
(5) if all RT are both less than deadline date constraint, a private clound schedule of tasks is returned；Otherwise, QoS is utilized
Comprehensive estimation method selects and operation JiThe publicly-owned cloud resource that matches of safety and reliability；
The operation JiN task is contained, has m available resources slot, is defined as resource slot Di＞ Estk, Bi-objective multidimensional is more
Select permeability is such as given a definition：
Bi-objective multidimensional multiple-choice knapsack problem TDO-MMKP,
Minimizing cost function is：
Wherein, CostF [i, j, k]=SCi,j×xk+SDi,j×yk+SDi,j×(dtik+dtok),
SCi,jIt is workload, xkIt is the calculating cost of public cloud resource slot k, SDi,jIt is data volume size, ykIt is publicly-owned cloud resource
The carrying cost of slot k, dtikIt is to the cost of resource slot k input datas, dtokIt is the cost of resource slot k output datas；
It minimizes estimation and executes the time
Wherein,NB is between private clound and public cloud
Network bandwidth, pkFor the computing capability of resource slot k；ai,j,kExpression task Ti,jWhether resource slot k is distributed to；bkIndicate resource slot
Whether used；
The estimation execution time is set to be less than the deadline date：
EstkRepresent the resource slot k estimated times completed；
Cost function will be less than estimated cost simultaneouslyWherein MiIt is executory cost, table
Show and executes operation J in public cloudiExpense；
Partner selection is mapped as the multidimensional multiple-choice knapsack problem of a variation, including：1. in TDO-MMKP, operation
Each task is mapped to a resource slot, and each resource slot is mapped to an object in one group；2. in TDO-MMKP, often
Resource needed for the QoS constraint consistencies to object of a task；3. the first profit that cost function is mapped to object must optimize；
4. being mapped to second of profit of object using the least estimated deadline of resource slot must optimize；5. the deadline date constrain and
Budgetary restraints are looked at as the limitation of available resources in knapsack；6. in TDO-MMKP, indicated in one group using selection variable
Certain an object whether be selected；7. the solution of the first profit is the solution of selection, but if two schemes
There is the first identical profit, the scheme of better second of profit is the solution of selection；
The public cloud task scheduling include task reschedule with publicly-owned resource minimize hiring cost,
Wherein, the task reschedule including：Determine that task should be arranged into public cloud；Judge whether public cloud can be with
Meet deadline date constraint and budgetary restraints；If constraints can meet, system is that the task in operation is submitted to generate one
A dispatch list；Wherein, determine that task should be arranged into public cloud and include the following steps：
(1) Z is definedi, PrRi, RT [1...m], JR；Wherein ZiIndicate operation JiIt is arranged in the dispatch list run in private clound, PrRi
Indicate that private clound resource slot, resource slot quantity are m, RT [1...m] indicates that one private clound resource slot of record becomes from currently
Available shortest time, JR are the output results of QoS comprehensive estimation methods；
(2) task-set being assigned in public cloud will be needed to empty, meter TPPU=φ；
(3) task is executed into the size progress ascending sort of time according to estimation and obtains TPPR；
(4) it is more than the final coutoff time limit when the execution time of task on private clound resource slot k, inquiry distribution is appointed resource slot k's
Business collection, is added to TPPU by the task-set, i.e., preceding n task in private clound resource slot is moved on to TPPU from TPPR；
(5) when calculating the deadline of public cloud task less than the final coutoff time limit, then n ＇ tasks are moved on to TPPR from TPPU；
(6) two tuple ＜ Z ＇ are exportedi,TPi＞, wherein Z ＇iIt is operation JiDispatch list in public cloud, TPiBe be transferred to it is publicly-owned
Task-set on cloud；
The publicly-owned resource minimize hiring cost resource slot according to resource quality, CPU total times, always use memory space and total
Bandwidth is consumed to charge, the hiring cost of publicly-owned cloud resource is minimized under the premise of meeting budget control and constraining, wherein described
Include the following steps meeting the hiring cost for minimizing publicly-owned cloud resource under the premise of budget control constraint：
(1) TP is definedi, PRT, tinit；Wherein TPiIt is operation JiThe task-set in public cloud is operated in, PRT is public cloud resource slot
Set of types, tinitIt is the time initial value of public cloud resource slot；
(2) initializing variable, NRi=φ, Zi=φ, TotalCost=0, wherein NRiIndicate one group of public cloud resource slot, ZiTable
Show that the dispatch list of task in public cloud, TotalCost indicate the hiring cost of public cloud；
(3) suitable resource type is found for task in public cloud, which meets price under conditions of deadline date constraint
It is minimum；
(4) if best resource type is found, system can create an example types in public cloud and distribute task
To the public cloud resource instances, that is, distribute task Ti,jTo resource k and in ZiMiddle record mapping；
(5) T run on k is calculatedi,jCost and be added to TotalCost；
(6) it returns to the dispatch list of public cloud and correspondingly dispatches public cloud resource slot for the task in operation.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410309665.0A CN104102544B (en) | 2014-06-30 | 2014-06-30 | Mix the Parallel Task Scheduling Cost Optimization Approach of multi-QoS constraint under cloud environment |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410309665.0A CN104102544B (en) | 2014-06-30 | 2014-06-30 | Mix the Parallel Task Scheduling Cost Optimization Approach of multi-QoS constraint under cloud environment |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN104102544A CN104102544A (en) | 2014-10-15 |
| CN104102544B true CN104102544B (en) | 2018-08-03 |
Family
ID=51670716
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201410309665.0A Expired - Fee Related CN104102544B (en) | 2014-06-30 | 2014-06-30 | Mix the Parallel Task Scheduling Cost Optimization Approach of multi-QoS constraint under cloud environment |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN104102544B (en) |
Families Citing this family (23)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105656973B (en) * | 2014-11-25 | 2018-11-13 | 中国科学院声学研究所 | Method for scheduling task and system in a kind of distributed node group |
| CN104657220B (en) * | 2015-03-12 | 2018-04-17 | 广东石油化工学院 | Scheduling model and method based on deadline and expense restriction in mixed cloud |
| CN104793990B (en) * | 2015-04-21 | 2018-08-17 | 中国海洋大学 | A kind of multiple timings method for scheduling task and system |
| CN104778076B (en) * | 2015-04-27 | 2017-10-31 | 东南大学 | A kind of cloud service workflow schedule method |
| US10129101B2 (en) * | 2015-04-30 | 2018-11-13 | Futurewei Technologies, Inc. | Application driven and adaptive unified resource management for data centers with Multi-Resource Schedulable Unit (MRSU) |
| CN106973030A (en) * | 2016-01-14 | 2017-07-21 | 北京仿真中心 | A kind of cloud artificial resource dispatching method based on SLA |
| CN106056294A (en) * | 2016-06-06 | 2016-10-26 | 四川大学 | Hybrid cloud scientific workflow scheduling strategy based on task probability clustering and multi-constraint workflow division |
| CN106533839A (en) * | 2016-11-30 | 2017-03-22 | 郑州云海信息技术有限公司 | Application node monitoring method and apparatus of hybrid cloud |
| CN107040575B (en) * | 2016-11-30 | 2020-09-22 | 苏州浪潮智能科技有限公司 | Application distribution method and device on hybrid cloud |
| CN106603438B (en) * | 2016-12-21 | 2019-07-19 | 云南电网有限责任公司信息中心 | A kind of mixed cloud utilization of resources based on cost and distribution appraisal procedure |
| CN108733464B (en) * | 2017-04-18 | 2021-09-14 | 华为技术有限公司 | Method and device for determining scheduling scheme of computing task |
| CN107317836A (en) * | 2017-05-16 | 2017-11-03 | 北京交通大学 | One kind mixing cloud environment lower time appreciable request scheduling method |
| CN107797870A (en) * | 2017-11-10 | 2018-03-13 | 苏州铭冠软件科技有限公司 | A kind of cloud computing data resource dispatching method |
| CN107908458A (en) * | 2017-11-10 | 2018-04-13 | 苏州铭冠软件科技有限公司 | A kind of cloud computing data resource dispatching method for considering time and expense |
| CN108595255B (en) * | 2018-04-13 | 2022-01-21 | 武汉理工大学 | Workflow task scheduling method based on shortest path algorithm in geographically distributed cloud |
| CN110389816B (en) * | 2018-04-20 | 2023-05-23 | 伊姆西Ip控股有限责任公司 | Method, apparatus and computer readable medium for resource scheduling |
| CN109710392B (en) * | 2018-12-21 | 2023-08-01 | 万达信息股份有限公司 | Heterogeneous resource scheduling method based on hybrid cloud |
| CN109976890B (en) * | 2019-03-28 | 2023-05-30 | 东南大学 | Variable frequency method for minimizing heterogeneous private cloud computing resource energy consumption |
| WO2020219041A1 (en) * | 2019-04-24 | 2020-10-29 | Alibaba Group Holding Limited | Distributed resource allocation |
| SG11201910151UA (en) * | 2019-04-24 | 2020-11-27 | Advanced New Technologies Co Ltd | Distributed resource allocation |
| AU2019253789B1 (en) * | 2019-04-24 | 2020-10-08 | Advanced New Technologies Co., Ltd. | Distributed resource allocation |
| CN110674153B (en) * | 2019-09-25 | 2023-02-07 | 北京金山云网络技术有限公司 | Data consistency detection method and device and electronic equipment |
| CN117331706B (en) * | 2023-12-01 | 2024-02-13 | 国网江苏省电力有限公司南通供电分公司 | Calculation force optimization method and system in electric power data maintenance |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102780759A (en) * | 2012-06-13 | 2012-11-14 | 合肥工业大学 | Cloud computing resource scheduling method based on scheduling object space |
| CN103699448A (en) * | 2014-01-14 | 2014-04-02 | 浪潮(北京)电子信息产业有限公司 | Scheduling method based on time limit and budget in cloud computing environment |
-
2014
- 2014-06-30 CN CN201410309665.0A patent/CN104102544B/en not_active Expired - Fee Related
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102780759A (en) * | 2012-06-13 | 2012-11-14 | 合肥工业大学 | Cloud computing resource scheduling method based on scheduling object space |
| CN103699448A (en) * | 2014-01-14 | 2014-04-02 | 浪潮(北京)电子信息产业有限公司 | Scheduling method based on time limit and budget in cloud computing environment |
Non-Patent Citations (1)
| Title |
|---|
| 混合云环境下多QOS约束的任务调度算法研究;刘炎培等;《http://www.paper.edu.cn》;20131202;第3－5页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN104102544A (en) | 2014-10-15 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN104102544B (en) | Mix the Parallel Task Scheduling Cost Optimization Approach of multi-QoS constraint under cloud environment | |
| Madni et al. | Recent advancements in resource allocation techniques for cloud computing environment: a systematic review | |
| Sanaj et al. | Nature inspired chaotic squirrel search algorithm (CSSA) for multi objective task scheduling in an IAAS cloud computing atmosphere | |
| Singh et al. | QRSF: QoS-aware resource scheduling framework in cloud computing | |
| Gill et al. | BULLET: particle swarm optimization based scheduling technique for provisioned cloud resources | |
| Tan et al. | A trust service-oriented scheduling model for workflow applications in cloud computing | |
| Abrishami et al. | Cost-driven scheduling of grid workflows using partial critical paths | |
| Wang et al. | Adaptive scheduling for parallel tasks with QoS satisfaction for hybrid cloud environments | |
| US8583799B2 (en) | Dynamic cost model based resource scheduling in distributed compute farms | |
| US9020829B2 (en) | Quality of service aware scheduling for composite web service workflows | |
| CN108345501A (en) | A kind of distributed resource scheduling method and system | |
| Nesmachnow et al. | Efficient heuristics for profit optimization of virtual cloud brokers | |
| Kaur et al. | Deep‐Q learning‐based heterogeneous earliest finish time scheduling algorithm for scientific workflows in cloud | |
| De Coninck et al. | Dynamic auto-scaling and scheduling of deadline constrained service workloads on IaaS clouds | |
| Adabi et al. | Bi-level fuzzy based advanced reservation of Cloud workflow applications on distributed Grid resources | |
| Hirales-Carbajal et al. | A grid simulation framework to study advance scheduling strategies for complex workflow applications | |
| Durgadevi et al. | Resource allocation in cloud computing using SFLA and cuckoo search hybridization | |
| Chakravarthi et al. | TOPSIS inspired budget and deadline aware multi-workflow scheduling for cloud computing | |
| Gutierrez-Garcia et al. | Agent-based cloud bag-of-tasks execution | |
| CN1639688A (en) | Decentralized processing system, job decentralized processing method, and program | |
| Mastelic et al. | Predicting resource allocation and costs for business processes in the cloud | |
| Singh et al. | Deadline and cost based ant colony optimization algorithm for scheduling workflow applications in hybrid cloud | |
| Geeta et al. | Multi-objective cloud load-balancing with hybrid optimization | |
| Aruna et al. | FRAMEWORK FOR RANKING SERVICE PROVIDERS OF FEDERATED CLOUD ARCHITECTURE USING FUZZY SETS. | |
| Lee | Adaptive resource scheduling for workflows considering competence and preference |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20180803 Termination date: 20190630 |