CN110489229A - A kind of multiple target method for scheduling task and system - Google Patents
A kind of multiple target method for scheduling task and system Download PDFInfo
- Publication number
- CN110489229A CN110489229A CN201910645478.2A CN201910645478A CN110489229A CN 110489229 A CN110489229 A CN 110489229A CN 201910645478 A CN201910645478 A CN 201910645478A CN 110489229 A CN110489229 A CN 110489229A
- Authority
- CN
- China
- Prior art keywords
- task
- individual
- virtual machine
- value
- indicate
- 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.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 59
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 162
- 238000012545 processing Methods 0.000 claims abstract description 21
- 230000035772 mutation Effects 0.000 claims description 35
- 231100000219 mutagenic Toxicity 0.000 claims description 33
- 230000003505 mutagenic effect Effects 0.000 claims description 33
- 238000004364 calculation method Methods 0.000 claims description 14
- 238000009826 distribution Methods 0.000 claims description 8
- 238000004590 computer program Methods 0.000 claims description 4
- 239000011800 void material Substances 0.000 claims description 4
- 238000002474 experimental method Methods 0.000 description 20
- 238000005457 optimization Methods 0.000 description 20
- 230000008569 process Effects 0.000 description 10
- 238000003860 storage Methods 0.000 description 10
- 238000013459 approach Methods 0.000 description 6
- 238000010586 diagram Methods 0.000 description 6
- 230000006870 function Effects 0.000 description 6
- 230000008901 benefit Effects 0.000 description 5
- 230000006872 improvement Effects 0.000 description 5
- 238000012360 testing method Methods 0.000 description 5
- 230000003044 adaptive effect Effects 0.000 description 4
- 230000000052 comparative effect Effects 0.000 description 4
- 230000001965 increasing effect Effects 0.000 description 4
- 230000000694 effects Effects 0.000 description 3
- 201000004569 Blindness Diseases 0.000 description 2
- 230000015572 biosynthetic process Effects 0.000 description 2
- 230000008859 change Effects 0.000 description 2
- 230000007423 decrease Effects 0.000 description 2
- 230000007812 deficiency Effects 0.000 description 2
- 238000013461 design Methods 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 238000011156 evaluation Methods 0.000 description 2
- 230000005484 gravity Effects 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 230000000717 retained effect Effects 0.000 description 2
- 238000003786 synthesis reaction Methods 0.000 description 2
- 238000005303 weighing Methods 0.000 description 2
- 230000006978 adaptation Effects 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 238000003491 array Methods 0.000 description 1
- 230000001174 ascending effect Effects 0.000 description 1
- 238000009412 basement excavation Methods 0.000 description 1
- 230000006399 behavior Effects 0.000 description 1
- 230000033228 biological regulation Effects 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- 238000010835 comparative analysis Methods 0.000 description 1
- 238000005265 energy consumption Methods 0.000 description 1
- 230000002708 enhancing effect Effects 0.000 description 1
- 230000010429 evolutionary process Effects 0.000 description 1
- 230000002068 genetic effect Effects 0.000 description 1
- 230000000977 initiatory effect Effects 0.000 description 1
- 230000010354 integration Effects 0.000 description 1
- 230000009916 joint effect Effects 0.000 description 1
- 230000009191 jumping Effects 0.000 description 1
- 239000011159 matrix material Substances 0.000 description 1
- 238000013508 migration Methods 0.000 description 1
- 230000005012 migration Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000010355 oscillation Effects 0.000 description 1
- 230000002028 premature Effects 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 238000004904 shortening Methods 0.000 description 1
- 239000007787 solid Substances 0.000 description 1
- 238000007619 statistical method Methods 0.000 description 1
- 230000004083 survival effect Effects 0.000 description 1
- 238000012795 verification Methods 0.000 description 1
- 239000002699 waste material Substances 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F9/00—Arrangements for program control, e.g. control units
- G06F9/06—Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
- G06F9/46—Multiprogramming arrangements
- G06F9/50—Allocation of resources, e.g. of the central processing unit [CPU]
- G06F9/5005—Allocation of resources, e.g. of the central processing unit [CPU] to service a request
- G06F9/5027—Allocation of resources, e.g. of the central processing unit [CPU] to service a request the resource being a machine, e.g. CPUs, Servers, Terminals
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F9/00—Arrangements for program control, e.g. control units
- G06F9/06—Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
- G06F9/46—Multiprogramming arrangements
- G06F9/50—Allocation of resources, e.g. of the central processing unit [CPU]
- G06F9/5005—Allocation of resources, e.g. of the central processing unit [CPU] to service a request
- G06F9/5027—Allocation of resources, e.g. of the central processing unit [CPU] to service a request the resource being a machine, e.g. CPUs, Servers, Terminals
- G06F9/505—Allocation of resources, e.g. of the central processing unit [CPU] to service a request the resource being a machine, e.g. CPUs, Servers, Terminals considering the load
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F9/00—Arrangements for program control, e.g. control units
- G06F9/06—Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
- G06F9/46—Multiprogramming arrangements
- G06F9/50—Allocation of resources, e.g. of the central processing unit [CPU]
- G06F9/5061—Partitioning or combining of resources
- G06F9/5077—Logical partitioning of resources; Management or configuration of virtualized resources
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/12—Computing arrangements based on biological models using genetic models
- G06N3/126—Evolutionary algorithms, e.g. genetic algorithms or genetic programming
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2209/00—Indexing scheme relating to G06F9/00
- G06F2209/50—Indexing scheme relating to G06F9/50
- G06F2209/508—Monitor
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Software Systems (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Life Sciences & Earth Sciences (AREA)
- Health & Medical Sciences (AREA)
- Bioinformatics & Computational Biology (AREA)
- Biophysics (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Computational Linguistics (AREA)
- Genetics & Genomics (AREA)
- Physiology (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a kind of multiple target method for scheduling task and systems, by carrying out fuzzy clustering processing respectively to task and virtual machine using fuzzy clustering algorithm, and by the way that the Task clustering set of acquisition is matched with virtual machine cluster set, obtain matching set, using differential evolution algorithm, by the task schedule in matching set to the virtual machine in matching set, obtain initial schedule result, according to the load value of virtual machine in initial schedule result, virtual machine is divided into heavy duty set, equally loaded set and light collections of loads, selection needs the task-set of degree of readjustment from heavy duty set, it obtains readjustment degree task-set and uses Q value-based algorithm, by the task schedule in readjustment degree task-set to the virtual machine in light collections of loads, solve the technical problem bad using the scheduling strategy of task with traditional dispatching algorithm acquisition, it is logical It crosses and is redistributed using Q value method progress Local resource task, realize the load balancing of global resource, improve resource utilization.
Description
Technical field
The present invention relates to field of cloud computer technology, in particular to a kind of multiple target method for scheduling task and system.
Background technique
Cloud computing is a kind of market-oriented emerging technology based on network Development, becomes the hot spot of academic research in recent years
Topic.With the riseing year by year of its occupation rate of market, the expansion of resource extent, huge user volume and task amount also one after another and
Extremely, this, which just gives how rationally efficiently dispatch to task in cloud environment, increases difficulty.And design excellent scheduling strategy
The service satisfaction for improving the runnability of cloud platform, shortening task response-time and improving user is all played can not
Or scarce effect.
User submit to the task of cloud platform scale and quantity it is very huge, and multiple users submit task between
There is no the constraint relationship, task exists in parallel form, and how the task of these substantial amounts is reasonably assigned to magnanimity
Resource node in execute be a difficulty and complex process.Cloud task schedule belongs to multiple target combinatorial optimization problem, by a variety of
The joint effect of factor, task with traditional dispatching algorithm often only realize the optimization of simple target in scheduling process, and algorithm
The result scheme for the scheduling strategy that can have certain deficiency, therefore be obtained in the biggish task of processing quantity itself is inadequate
It is ideal.
Cloud computing resources have biggish otherness, they are not only there is the difference in geographical location, and more there is performances
On difference.Heterogeneous resource is become an entirety by virtualization integration by cloud computing system, and constituting one may be implemented
Shared resource pool, when there are can measure to distribute to its corresponding virtual resource according to the demand of user when mission requirements by user.In
In huge resource set by the larger task-set of quantity be allocated easily there are blindness and limitation.In addition, times that user submits
Business also has various preferences, such as: calculation type task is partial to the scientific algorithm of high miscellaneous degree；Storage-type task needs a large amount of memory spaces
To store large data；Bandwidth type task communication data volume is big to need high bandwidth to access；Data-intensive task needs a large amount of
It calculates and storage resource carries out big data statistical analysis excavation etc..Therefore, it should by the resource priority with different type advantage
Distribute to the task with corresponding preference.
Under cloud computing environment, the target of user and cloud service provider is different.User more pays close attention to task
Deadline, and cloud service provider often focuses more on the utilization rate in resource, task execution cost and energy consumption etc..Cause
This, how going to execute by task schedule to suitable resource and being one the needs of meeting each side under cloud environment simultaneously is worth grinding
The problem of studying carefully.
Summary of the invention
A kind of multiple target method for scheduling task provided by the invention and system, are solved and are obtained using task with traditional dispatching algorithm
The bad technical problem of the scheduling strategy obtained.
In order to solve the above technical problems, a kind of multiple target method for scheduling task proposed by the present invention includes:
Using fuzzy clustering algorithm fuzzy clustering processing is carried out to task and virtual machine respectively, obtains Task clustering set
It clusters and gathers with virtual machine, and Task clustering set is matched with virtual machine cluster set, obtain matching set；
The task schedule in matching set to the virtual machine in matching set is obtained initial using differential evolution algorithm
Scheduling result；
According to the load value of virtual machine in initial schedule result, virtual machine is divided into heavy duty set, equally loaded set
And light collections of loads；
Selection needs the task-set of degree of readjustment from heavy duty set, obtains readjustment degree task-set；
Using Q value-based algorithm, by the task schedule in readjustment degree task-set to the virtual machine in light collections of loads.
Further, Task clustering set is matched with virtual machine cluster set, obtaining matching set includes:
According to computing capability, bandwidth and the memory capacity of virtual machine in virtual machine cluster set, weighted calculation obtains empty
The resource performance of quasi- machine；
According to the average value of the resource performance of all virtual machines in virtual machine cluster set, virtual machine cluster set is obtained
Comprehensive resources performance；
According to the length of task, size and output data size in Task clustering set, weighted calculation obtains task
Resource requirement；
According to the average value of the resource requirement of tasks all in Task clustering set, the comprehensive money of Task clustering set is obtained
Source demand；
According to virtual machine cluster set comprehensive resources performance and Task clustering set comprehensive resources demand between away from
From, Task clustering set is gathered with virtual machine cluster and is matched, acquisition matching set.
Further, using differential evolution algorithm, by the task schedule in matching set to matching gather in virtual machine,
Obtaining initial schedule result includes:
According to task quantity, the virtual machine quantity, mission number in matching set, differential evolution is calculated using one-dimension array
The individual of population is encoded in method；
Mutagenic factor is calculated according to current evolutionary generation and intersects the factor；
According to mutagenic factor, in conjunction with a variety of Mutation Strategies, mutation operation is carried out to population at individual, obtains variation individual；
According to the component for intersecting selecting predictors parent individuality or variation individual, obtains and intersect individual；
According to the component for intersecting factor Negative selection parent individuality or variation individual, anti-cross individual is obtained；
According to task completion time, task execution cost and virtual machine load balancing degrees, Proper treatment is determined；
Target individual, variation individual are obtained according to Proper treatment, intersect the individual appropriateness value of individual, anti-cross, and according to
Appropriateness value obtains next-generation individual, to obtain initial schedule result to next-generation individual decoding.
Further, mutagenic factor is calculated according to current evolutionary generation and intersects the specific formula of the factor are as follows:
Wherein, F (t) indicates that mutagenic factor, CR (t) indicate to intersect the factor, and w (t) indicates weight factor, and t indicates generation of evolving
Number, T indicate population maximum evolutionary generation, FmaxIndicate mutagenic factor maximum value, FminIndicate mutagenic factor minimum value, CRminIt indicates
Intersect factor minimum value, CRmaxIt indicates to intersect factor maximum value.
Further, according to mutagenic factor, in conjunction with a variety of Mutation Strategies, mutation operation is carried out to population at individual, is become
The specific formula of different individual are as follows:
Wherein, λ indicates weight factor, and u indicates impact factor, vi(t) the corresponding variation of population present age evolutionary generation is indicated
Individual, t indicate that evolutionary generation, T indicate population maximum evolutionary generation, xiIndicate current individual, xr1(t)、xr2(t)、xr3(t)、xr4
(t) the population present age evolutionary generation chosen at random corresponding four are represented and is different from xiIndividual, xbest(t) contemporary population is indicated
In optimal individual, F indicates mutagenic factor.
Further, target individual, variation individual, the appropriateness for intersecting individual, anti-cross individual are obtained according to Proper treatment
Value, and it is worth the calculation formula for obtaining next-generation individual according to appropriateness specifically:
Wherein, f indicates Proper treatment, and f=min [a ln (totalTime)+b ln (total Cost)+c ln
(load)], xi t+1Indicate next-generation target individual, ui t+1It indicates next-generation and intersects individual, f (ui t+1) indicate next-generation intersection
The appropriateness value of body, vi t+1Indicate next-generation variation individual, f (vi t+1) indicate that the appropriateness of next-generation variation individual is worth, hi t+1Under expression
Generation anti-cross individual, f (hi t+1) indicate that the appropriateness of next-generation anti-cross individual is worth, f (xi t) indicate the suitable of contemporary target individual
Angle value, xbestIt indicates moderately to be worth optimal individual, x in contemporary populationavgIndicate the intermediate value of all individuals in contemporary population,
TotalTime expression task completion time, totalCost expression task run cost overhead, load expression load balancing degrees, a,
B, c is weight factor, meets condition a+b+c=1.
Further, in Q value-based algorithm Q value calculation formula specifically:
Wherein, QiIt indicates to the Q value after j-th of virtual machine distribution task, vmpjIndicate in light collections of loads j-th it is virtual
The processing capacity of machine, vmTimejIndicate the execution time of task on j-th of virtual machine in light collections of loads, taskTimeijIt indicates
By task execution time of i-th of the task immigration in readjustment degree task-set into light collections of loads on j-th of virtual machine.
Further, mutagenic factor maximum value Fmax=1, mutagenic factor minimum value Fmin=0.4, intersecting factor minimum value is
CRmin=0.6, intersection factor maximum value is CRmax=0.9.
A kind of multiple target task scheduling system proposed by the present invention includes:
Memory, processor and storage on a memory and the computer program that can run on a processor, processor
The step of realizing multiple target method for scheduling task of the invention when executing computer program.
Compared with the prior art, the advantages of the present invention are as follows:
Multiple target method for scheduling task provided by the invention and system, by utilizing fuzzy clustering algorithm to task and void
Quasi- machine carries out fuzzy clustering processing respectively, obtains Task clustering set and virtual machine cluster set, and by Task clustering set with
Virtual machine cluster set is matched, and matching set is obtained, and using differential evolution algorithm, the task schedule in matching set is arrived
Virtual machine in matching set obtains initial schedule as a result, according to the load value of virtual machine in initial schedule result, by virtual machine
It is divided into heavy duty set, equally loaded set and light collections of loads, selection needs the task of degree of readjustment from heavy duty set
Collection obtains readjustment degree task-set and uses Q value-based algorithm, by the task schedule in readjustment degree task-set into light collections of loads
Virtual machine solves the bad technical problem of scheduling strategy obtained using task with traditional dispatching algorithm, by using Q value method into
Row Local resource task is redistributed, and the load balancing of global resource is realized, and improves resource utilization.
Detailed description of the invention
Fig. 1 is the flow chart of the multiple target method for scheduling task of the embodiment of the present invention one；
Fig. 2 is the flow chart of the multiple target method for scheduling task of the embodiment of the present invention two；
Fig. 3 is the embodiment of the present invention for the cloud task and virtual machine assumed, after carrying out fuzzy clustering to task and resource
Cluster result；
Fig. 4 is the embodiment of the present invention for the cloud task and virtual machine assumed, the task schedule schematic diagram of acquisition；
Fig. 5 is the embodiment of the present invention for the cloud task and virtual machine assumed, decodes the task scheduling approach of acquisition；
Fig. 6 is the crossover operation procedure chart of the embodiment of the present invention two；
Fig. 7 is the reversed crossover operation procedure chart of the embodiment of the present invention two；
Fig. 8 is the course of work of the t times iteration of population in the IADE algorithm of the embodiment of the present invention two；
Fig. 9 is the flow chart of the multiple target method for scheduling task of the embodiment of the present invention three；
Figure 10 is that the embodiment of the present invention three compares without DE algorithm execution time result figure after cluster DE algorithm and cluster；
Figure 11 is the task number different task deadline comparison diagram of the embodiment of the present invention three；
Figure 12 is the virtual machine number different task deadline comparison diagram of the embodiment of the present invention three；
Figure 13 is the task execution expense comparison diagram of the embodiment of the present invention three；
Figure 14 is the load balancing value comparison diagram of the embodiment of the present invention three；
Figure 15 is the multiple target task scheduling system block diagram of the embodiment of the present invention.
Appended drawing reference:
10, memory；20, processor.
Specific embodiment
To facilitate the understanding of the present invention, the present invention is made below in conjunction with Figure of description and preferred embodiment more complete
Face meticulously describes, but the protection scope of the present invention is not limited to the following specific embodiments.
The embodiment of the present invention is described in detail below in conjunction with attached drawing, but the present invention can be defined by the claims
Implement with the multitude of different ways of covering.
Embodiment one
Referring to Fig.1, the multiple target method for scheduling task that the embodiment of the present invention one provides, comprising:
Step S101 carries out fuzzy clustering processing to task and virtual machine using fuzzy clustering algorithm respectively, is appointed
Business cluster set and virtual machine cluster set, and Task clustering set is matched with virtual machine cluster set, it is matched
Set；
Step S102, using differential evolution algorithm, by the task schedule in matching set to matching gather in virtual machine,
Obtain initial schedule result；
Virtual machine is divided into heavy duty set, equilibrium according to the load value of virtual machine in initial schedule result by step S103
Collections of loads and light collections of loads；
Step S104, selection needs the task-set of degree of readjustment from heavy duty set, obtains readjustment degree task-set；
Step S105, it is using Q value-based algorithm, the task schedule in readjustment degree task-set is virtual into light collections of loads
Machine.
Multiple target method for scheduling task provided in an embodiment of the present invention, by utilizing fuzzy clustering algorithm to task and void
Quasi- machine carries out fuzzy clustering processing respectively, obtains Task clustering set and virtual machine cluster set, and by Task clustering set with
Virtual machine cluster set is matched, and matching set is obtained, and using differential evolution algorithm, the task schedule in matching set is arrived
Virtual machine in matching set obtains initial schedule as a result, according to the load value of virtual machine in initial schedule result, by virtual machine
It is divided into heavy duty set, equally loaded set and light collections of loads, selection needs the task of degree of readjustment from heavy duty set
Collection obtains readjustment degree task-set and uses Q value-based algorithm, by the task schedule in readjustment degree task-set into light collections of loads
Virtual machine solves the bad technical problem of scheduling strategy obtained using task with traditional dispatching algorithm, by using Q value method into
Row Local resource task is redistributed, and the load balancing of global resource is realized, and improves resource utilization.
Specifically, by user in this present embodiment submit task have multiple types and what the embodiment of the present invention was studied
Scheduling problem towards be self task, in order to establish an efficient scheduling model, therefore make following hypothesis:
(1) different user upload task between it is mutually indepedent and do not have the constraint relationship；
(2) total task number that user submits is much larger than the number of resources in cloud environment；
(3) virtual machine executes task by prerequisite variable principle non-preemptive；
(4) independent task is undecomposable, can only be assigned to a virtual machine.
According to hypothesis it is found that data dependence relation is not present between each subtask in independent task set, it is assumed that use
The task that family is submitted to cloud platform has n subtask, and subtask is numbered in order, and independent task collection shares T expression,
Task model is defined as:
T={ t1,t2,t3,...,tn} (1)
ti={ tId,tLength,tFileSize,tOutputSize} (2)
Wherein, tIdThe number of expression task, to each task of unique identification；tLengthThe length of expression task, unit are used
MI (Million Instructions) is indicated；And tFileSizeWhat is represented is the size of task, and unit is indicated with MB；
tOutputSizeRepresentative needs output data size.
The Heterogeneous Computing resource of cloud computation data center is abstracted composition cluster, it is assumed that in the resource pool of cloud data center
There are the m virtual machines with different performance, is indicated with set VM, vmjIndicate that j-th of virtual machine, resource model may be expressed as:
VM={ vm1,vm2,vm3,...,vmm} (3)
vmj={ vmId,vmMips,vmPENumber,vmStor,vmBW} (4)
Wherein, virtual machine performance information collection includes: vmIdThe id for indicating virtual machine, is the unique identification of every virtual machine；
vmMipsIndicate the instruction execution number (million instruction per second) of virtual machine；vmPENumberIndicate CPU quantity, vmStorIndicate virtual machine
Memory capacity；vmBWIndicate virtual machine amount of bandwidth.
vmjComputing capability can be found out by following formula:
vmjcompute=vmPENumber*vmMips (5)
Each task holding on assigned virtual machine can be calculated after virtual machine collection is assigned in task-set
The row time is indicated with matrix ETC are as follows:
Wherein m, n are respectively virtual machine and task quantity, etcj,iFor task tiIn virtual machine vmjOn the execution time.
In multi-objective optimization question, between each influence factor be restrict each other cause also deposit between optimization aim
It may result under other optimization target values that it is restricted in the raising of restricting relation, one of optimization target values
Drop, all targets obtain optimal often impossible simultaneously.Therefore, the solution of multi-objective optimization question needs to comprehensively consider institute
There is target, so that they are optimal as far as possible in the condition restricted each other.
In cloud computing environment, the deadline of task is user and the emphasis mesh that cloud service provider is paid close attention to jointly
Mark.As consumer, task charge costs are the focal points of user, and the deadline for reducing task can reduce user's indirectly
Charge costs.Meanwhile cloud computing belongs to nature of business, for provider, realizes that money can be improved in the load balancing of system
Source utilization rate reduces the waiting time of user, promotes the satisfaction of user.Therefore, comprehensively consider user and and service provider
The interests of both sides, the present embodiment choose task schedule optimization aim include the user task deadline, task run expense with
And the load balancing of virtual machine, seek so that they while the scheduling scheme being optimal, both ensure that user for clothes in this way
Business quality requirement, and it is able to ascend resource utilization, avoid the waste of cloud resource.
(1) task completion time
The execution time of single virtual machine j is that the task execution of all distribution thereon completes the time it takes, can be with table
It is shown as:
Virtual machine in cloud environment executes in a parallel fashion, and therefore, total time spent by task execution is institute
Have and spend the time used in time longest virtual machine in the virtual machine for carrying out execution task, then task completion time can indicate
Are as follows:
TotalTime=max (vmTimej) (8)
(2) task run cost overhead
Task run cost overhead is defined as the sum of task execution expense and storage expense in the present embodiment.
The expense of execution task be all virtual machine tasks execute time and unit strike price product and, can be with table
It is shown as:
In formula, vmTimejIndicate the execution time for the task that virtual machine j is distributed, vmjCPUpriceThat indicate is virtual machine j
Unit time strike price.
Task tiIn virtual machine vmjOn storage expense can indicate are as follows:
StorCosti,j=tiFileSize*vmjStorprice (10)
In formula, tiFileSizeThe size of memory shared by expression task i, vmjStorpriceThat indicate is virtual machine vmjUnit it is big
Small storage price.
Then the total storage expense of task is the sum of all virtual machine task storage expenses:
After aggregative formula (9) and formula (11), cost overhead totalCost can be denoted as when task run:
Total Cost=CPU Cost+Stor Cost (12)
(3) virtual machine loads
The load of virtual machine j is indicated in the present embodiment with having executed the total time of distributed task, virtual machine vmjIt is a certain
The load at moment can be formulated as:
Then the total load of whole virtual machines and average load may be expressed as: in system
Load balancing degrees are indicated with standard deviation in system, may be defined as:
Loading condition of the load balancing degrees load to quantify cluster virtual machine in cloud system, a certain moment all calculating sections
Load is 0 when point load is identical, and at this moment cluster virtual machine is in ideal load balancing state, and load is smaller to illustrate that load is more equal
Weighing apparatus.
The target of task scheduling strategy of the present embodiment based on multiple-objection optimization are as follows:
Embodiment two
Referring to Fig. 2, multiple target method for scheduling task provided by Embodiment 2 of the present invention, comprising:
Step S201 carries out fuzzy clustering processing to task and virtual machine using fuzzy clustering algorithm respectively, is appointed
Business cluster set and virtual machine cluster set, and Task clustering set is matched with virtual machine cluster set, it is matched
Set.
With the rapid development of cloud computing, the scale of resource cluster constantly expands.Task schedule under current cloud environment is past
Toward being directed to all resources, relationship between task and resource is not considered, more than the resource and when task amount is big, by task point
The time overhead for being fitted on suitable resource node will increase, this can directly be finally completed time generation large effect to task,
Also result in the decline of task execution efficiency.In view of the blindness of task choosing resource has to a certain degree task execution situation
Influence, therefore, the present embodiment when carrying out task schedule, setting pretreatment stage simultaneously to task and resource carry out cluster draw
Divide and solves.
When to user task and cloud resource cluster, node as resource performance Attribute class is classified as one kind, will be appointed
Being engaged in, resource requirement is similar to be classified as one kind, and after division, each small set of tasks corresponds to a small set of resource nodes, simultaneously
Reduce the scale of task and resource.In task schedule, task can accurately select corresponding calculate node collection, search money
The time in source will shorten, and then can reduce task waiting time, so that assigning process becomes efficient, it is also possible that high-performance
Node processing complex task demand in resource set, the simple task of node processing in low performance resource set.
Resource under cloud computing environment has isomerism while ambiguity is it is also obvious that usually not stringent between them
It divides, therefore is divided the uncertainty that can more accurately reflect resource using fuzzy clustering method.The present embodiment uses
Fuzzy clustering method completes the cluster to task and resource.
Specifically, the present embodiment carries out fuzzy clustering processing to task and virtual machine using fuzzy clustering algorithm respectively,
Task clustering set and virtual machine cluster set are obtained, and Task clustering set is matched with virtual machine cluster set, is obtained
Set, which must be matched, includes:
Step S2011, according to computing capability, bandwidth and the memory capacity of virtual machine in virtual machine cluster set, weighting
Calculate the resource performance for obtaining virtual machine.
Specifically, the virtual machine vm in the present embodiment virtual machine cluster setjResource performance pass through vmjEvery ability
Synthesis show that formula is as follows:
In formula, parameter a, b, c are to respectively indicate to virtual machine vmjCalculated performance, bandwidth and the empirical coefficient of storage.
Step S2012 obtains virtual machine according to the average value of the resource performance of all virtual machines in virtual machine cluster set
Cluster the comprehensive resources performance of set.
Specifically, the comprehensive resources performance of the present embodiment virtual machine cluster set is the resource of all virtual machines in the set
The average value of performance, formula are as follows:
In formula, parameter n indicates the virtual machine quantity in the resource set.
Step S2013, according to the length of task, size and output data size, weighted calculation in Task clustering set
The resource requirement of acquisition task.
Specifically, in the present embodiment Task clustering set task i resource requirement calculation formula:
Parameter a, b, c in formula respectively indicate task tiFor the demand size of calculating, bandwidth and storage capacity.
Step S2014 obtains Task clustering collection according to the average value of the resource requirement of tasks all in Task clustering set
The comprehensive resources demand of conjunction, specific formula are as follows:
Step S2015, according to the comprehensive resources need of the comprehensive resources performance of virtual machine cluster set and Task clustering set
The distance between ask, Task clustering set is matched with virtual machine cluster set, obtains matching set.
Specifically, the present embodiment is according to the comprehensive resources demand t of Task clustering setREWith the synthesis of virtual machine cluster set
Resource performance rGPThe distance between d be that task-set selects suitable resource set, more can apart from the smaller resource for illustrating the set
Meet the mission requirements in set of tasks.
D=| rGP-tRE| (22)
By classifying to task and resource in pretreatment stage, as shown in figure 3, reducing to be matched appoint simultaneously
The scale of business and resources of virtual machine, can be greatly reduced the match time in scheduling process, improve task schedule efficiency
And user satisfaction, it is adjusted in matching set later using the multiple target task scheduling strategy of improved differential evolution algorithm
Degree.
Step S202, according to task quantity, the virtual machine quantity, mission number in matching set, using one-dimension array pair
The individual of population is encoded in differential evolution algorithm.
When carrying out task schedule using differential evolution algorithm, need to encode the individual of population.The present embodiment
Coding mode are as follows: task quantity is M, and number is [0, M-1]；Virtual machine quantity is N, and number is [0, N-1].Population at individual set
It is indicated with several one-dimension arrays, the length of one-dimension array is task quantity, and the index of array is the number of task, array indexing
Corresponding value is the number of virtual machine, in the range of [0, N-1], indicates task schedule that number is array index to array index
Virtual machine in corresponding value goes to execute.Each array represents a kind of task scheduling approach, and population scale NP then forms NP kind tune
Degree scheme.For algorithm in random initializtion population, the virtual machine number result of generation is that decimal leads to the optimal scheduling side obtained
Case is also not in the form of integer coding, it is therefore desirable to further to coded treatment.Assuming that have user have submitted cloud task be T1,
T2, T3, T4, T5, T6 }, it is { VM1, VM2, VM3 } that system, which has virtual machine,.When the optimal solution individual UVR exposure that algorithm is run is X
(2.5122,0.3196,1.8350,0.8705,2.2845,0.2445), it is every to individual to obtain the individual of integer coding form
One-dimensional component calculatesIndividual is as shown in Figure 4 after integer coding.Decoding obtains final task scheduling approach, such as Fig. 5 institute
Show.
Step S203 calculates mutagenic factor according to current evolutionary generation and intersects the factor.
Mutation operation is mutated to population at individual to increase the diversity of population.By being disturbed to target individual
It is dynamic, variation individual is generated, can prevent population from falling into local optimum.Mutation operation step is in the t times iteration, from population
Two mutually different individuals of random selection, which subtract each other, to be generated stochastic difference vector and is subject to after weight and randomly selected kind of third
Group's individual generates variation individual by sum operation, for vector individual for some in parent population, obtains through variation
Individual viCalculation are as follows:
vi(t)=xr1(t)+F*(xr2(t)-xr3(t)) (23)
In formula, t represents the algebra of current Evolution of Population, and serial number r1, r2, r3 are from population at individual set randomly selected three
A individual with the different index of current individual i serial number, while guaranteeing individual xr1、xr2、xr3It is respectively different, with random
Individual xr1As base vector, pass through xr2And xr3Random vector difference weighting after carry out disturbance generate variation vector.F be variation because
Son, value are the key that guarantee that individual mutates for controlling the influence degree of stochastic difference vector between [0,2].When
Two individual vector differences more hour illustrates that individual similarity degree is bigger, and entire population gradually approaches optimal solution.Mutagenic factor F value
Size the global optimizing ability of algorithm can be had an impact.When F value is smaller, algorithm is more biased towards in local search, may solution
The regional scope of search is smaller；When F value is larger, in Local Extremum search stagnation behavior, but biggish F will not occur for algorithm
It is slack-off that value will lead to convergence rate.
At this stage, DE algorithm has the Mutation Strategy of many different modes, is generally indicated with the form of " DE/a/b ",
A indicates the individual for participating in mutation operation how is chosen from population at individual, and general value is rand, best and current.Its
In, rand indicates that the individual for participating in mutation operation is random individual, and best indicates that mutation operation is mainly based upon current optimal solution
Corresponding individual, current indicate that mutation operation is to be operated using the current target individual of population as base vector；B indicates to become
The item number for the Difference Terms used in ETTHER-OR operation, generally integer.Mode in standard DE algorithm is indicated with " DE/rand/1 ", is removed
There are also five kinds of Mutation Strategies shown in table 1 for DE algorithm except this.
1 DE algorithm Mutation Strategy of table
X involved in Mutation Strategyr1、xr2、xr3、xr4、xr5Indicate different and different from current individual in population
Random individual vector, xbestIndicate the corresponding individual vector of adaptive optimal control angle value in population, F is mutagenic factor.
For different Mutation Strategies, the performance difference of differential evolution algorithm is bigger, DE/rand/1 and DE/rand/2
Mutation Strategy base vector is random individual, and search range is wide, and optimal value search speed is slower；DE/best/1 and DE/best/2 with
Optimum individual is base vector in contemporary population, and search range is confined near optimum individual, Premature convergence easily occurs；DE/
The convergence of current-to-best/2 and DE/rand-to-best/2 Mutation Strategy and search capability are more balanced.
The mutagenic factor F of differential evolution algorithm plays a part of adjusting to the search scale and population diversity of algorithm, calculates
Method be to the setting of mutagenic factor F it is very sensitive, usual F is value in the real number of [0,2], for controlling the amplitude of Difference Terms.
If the inappropriate new variation individual that can not generate that will lead in the variation stage is arranged in F, individual similarity increases and gradually becomes
Together, while F also influences convergence speed of the algorithm: when F setting is bigger than normal, the difference degree between individual increases, and algorithm at this time is searched
Rope range also will increase, close to the mode of global random searching, the matter of the slow solution for causing efficiency of algorithm too low and acquiring of convergence
It measures not high；When F setting is less than normal, algorithm search scale at this time reduces, local search enhancing, although convergence rate is accelerated
Algorithm can be made to fall into locally optimal solution.Therefore, mutagenic factor F is rationally set for the quality and convergence rate of algorithm optimal solution
It influences great.In standard difference evolution algorithm, F is arranged to fixed constant, all keeps solid during entire iterative solution
Fixed constant, this will lead to algorithm and is unable to satisfy each period during evolution for the size requirements of parameter, cause algorithm easily
Global optimum can not be obtained by falling into local optimum, while convergence rate is also relatively slow.In order to solve DE algorithm appearance this
A little problems, the present embodiment sets F to adaptively, with the size of the automatic setting value of evolutionary process.Most start iteration in algorithm
When, F is set as the larger value, this allows algorithm to have biggish search space to guarantee that the diversification of population can be avoided global optimum
The escape of value.The value of F is allowed constantly to reduce with being continuously increased for evolutionary generation in the later period, algorithm can be there are feasible at this time
The a small range of solution carries out the careful search in part, avoids optimal solution from being destroyed, promotes the precision of search.The improvement formula of F are as follows:
F (t)=Fmax-(Fmax-Fmin)*w(t) (25)
The size of F value is configured according to current evolutionary generation, with the increase of evolutionary generation t, weight factor w (t)
Value increase, the size for the F value newly assigned successively decreases, it is therefore an objective to which later stage of evolution is reinforced locally on the basis of the more excellent solution that algorithm obtains
Search, enables the algorithm to rapidly converge to optimal solution.The valid interval of F is [0.4,1], therefore mutagenic factor in the present embodiment
Maximum value Fmax=1, mutagenic factor minimum value Fmin=0.4.
Crossover operation is further to regulate and control to it, in order to increase the multiplicity of population at individual after mutation operation
Property.Crossover operation passes through preset strategy decision variation individual ViThe middle correspondence position which dimension component to replace target individual with
The dimension component set, CR decide that each dimension component value source of intersection individual and update ratio, this step largely influence
After in algorithm population diversity.
In formula, for crossover probability CR value in section [0,1], the size of CR determines the probability of the component intersected, CR
Bigger, the information content of exchange is bigger, intersects each dimension component of individual more from variation individual；CR is smaller, intersects individual
Component can more from parent individuality, this be maintained the diversity of population to each dimension.D represents the dimension of individual, rand
(D) random number in the dimensional extent of individual is generated, j is used to guarantee in intersecting step, intersects individual and is at least once handed over
Fork operation, it is ensured that intersect individual and not exactly the same evolve to avoid population at individual of parent individuality is stagnated, can not approach most
The figure of merit.Specific crossover operation process is as shown in Figure 6.In Fig. 6, each component for intersecting individual is determined according to intersection factor CR
Value, judgement are taken from parent target individual vector or variation individual vector.
The crossover operation of DE algorithm can increase the diversity of population at individual, and the parameter intersection factor CR in Crossover Strategy determines
The size degree determined filial generation and parent, intersect the dimension component exchanged between individual.CR is bigger to illustrate to make a variation variable contribution rate more
Height, disturb it is bigger, intersect vector each dimension component in belong to variation vector each dimension component shared by specific gravity it is bigger, this can make
The diversity of population reduces, and is conducive to algorithm and carries out local search to accelerate convergent rate；CR value is smaller, is more conducive to
Keep population diversity and ability of searching optimum.CR is unable to satisfy for preset parameter and was evolving in standard difference evolution algorithm
The preference that CR size is arranged in each stage in journey.Therefore, need during evolution according to operating condition, constantly change intersect because
The value of sub- CR, innovatory algorithm specifically:
CR (t)=CRmin+(CRmax-CRmin)*w(t) (27)
Intersect size that the adaptive impovement strategy of factor CR is newly worth with mutagenic factor F, CR also according to current evolution generation
Number t is configured, and with the increase of evolutionary generation t, the value of weight factor w (t) increases, and the size for the CR value newly assigned also is in increase
Trend can accelerate algorithm in the convergence rate of later stage of evolution being moderately worth in this way, while guarantee convergent precision.The present embodiment
Middle intersection factor minimum value is CRmin=0.6, intersection factor maximum value is CRmax=0.9.
Step S204, in conjunction with a variety of Mutation Strategies, carries out mutation operation to population at individual, is become according to mutagenic factor
Different individual.
In DE algorithm, single variation mode is often difficult to meet the requirement of Evolution of Population, therefore can use a variety of plans
The mode slightly combined integrates respective feature.DE/rand/1 mode can guarantee population diversity, and ability of searching optimum compared with
By force, it is not easy to fall into local optimum, but convergence rate is slower；And DE/best/1 mode convergence rate is very fast.The present embodiment is improved
Mutation Strategy has closed both classical variation modes, can combine search capability and convergence, the present embodiment is improved
Mode are as follows:
vi(t)=u*xr1(t)+(1-u)*xbest(t)+F*(xr2(t)-xr3(t)) (29)
Wherein, t indicates evolutionary generation, T population maximum evolutionary generation, xiIndicate current individual, xr1、xr2、xr3It represents random
Three chosen are different from xiIndividual, xbestIndicate individual optimal in contemporary population.
The size of variable u changes with the number of iterations t, and u is parabola in section [0,1] image, and early period, nominal growth was slow,
So that u value also slowly becomes smaller.Therefore in the Evolution of Population of early period, u → 1, variation mode can tend to DE/ to the full extent
Rand/1, algorithm possesses stronger global search performance population at individual diversification is guaranteed at this time.In later period u → 0,
Mutation Strategy tends to DE/best/1, and the performance of algorithm can reach on search capability, stability, speed of searching optimization at this time
It is preferable balanced, it is slow to avoid algorithm preconvergence speed.Different phase of algorithm during Evolution of Population after improving
It can be balanced between search capability and convergence rate with adaptively selected different Mutation Strategy.
Simultaneously in order to preferably control the Evolutionary direction of later period population, the present embodiment be added on the basis of original one it is optimal
Difference value, while weight factor is added to stochastic difference and optimal difference and controls respective contribution rate.Improved Mutation Strategy
Formula are as follows:
Wherein, λ indicates weight factor, and u indicates impact factor, vi(t) the corresponding variation of population present age evolutionary generation is indicated
Individual, t indicate that evolutionary generation, T indicate population maximum evolutionary generation, xiIndicate current individual, xr1(t)、xr2(t)、xr3(t)、xr4
(t) the population present age evolutionary generation chosen at random corresponding four are represented and is different from xiIndividual, xbest(t) contemporary population is indicated
In optimal individual, F indicates mutagenic factor.
Specifically, the λ in the present embodiment is weight factor, for controlling stochastic difference and optimal difference respectively for variation
Individual disturbance size, λ value range is section [0,1], and size is slowly incremented by with Evolution of Population algebra, so that stochastic difference exists
Weight early period of evolving is bigger, and percentage contribution is maximum in variation vector, and algorithm can be reinforced to global search dynamics.After evolution
Phase population at individual can restrain, so that the potential range of optimal solution reduces, in order to accelerate to search for the rate of optimal solution, therefore need at this time
The diversity of population at individual is reduced, increases the specific gravity of optimal difference in corresponding formula, enables the algorithm to attached in optimum individual
Nearly oscillation search enhances later period population local optimal searching ability.U indicates impact factor, and size changes with the number of iterations t, and u is in area
Between [0,1] image be parabola, u is that parameter is used to control the Mutation Strategy of differential evolution algorithm, because formula has merged two kinds
Mutation Strategy, as evolutionary generation can control tendency.U → 1, variation mode can tend to DE/rand/1 to the full extent,
Algorithm possesses stronger global search performance population at individual diversification is guaranteed at this time.In later period u → 0, make a variation plan
Slightly tend to DE/best/1, the performance of algorithm can reach preferable on search capability, stability, speed of searching optimization at this time
It is slow to avoid algorithm preconvergence speed for equilibrium.
Step S205 is obtained according to the component for intersecting selecting predictors parent individuality or variation individual and is intersected individual, according to friendship
The component of factor Negative selection parent individuality or variation individual is pitched, anti-cross individual is obtained；
Step S206 determines appropriateness letter according to task completion time, task execution cost and virtual machine load balancing degrees
Number.
The present embodiment is based on the characteristics of cloud platform business prototype, realizes task completion time, executory cost and virtual
The collaboration of three targets of machine load balancing optimizes, and multi-objective optimization question is converted to monocular by using the mode of weighted sum
Optimization problem is marked to handle.Because of these three mesh target values of task completion time, task execution cost and virtual machine load balancing
Number of levels is it is possible that very big gap, in order to their balanced influences for fitness function value, so needing elder generation
The standardization processing of target value is carried out, the mode of the standardization processing of the target value of the present embodiment is taken respectively to these three values
Right logarithm, the mode for reusing weighted sum later design Proper treatment.Formula is as follows:
F=min [a ln (totalTime)+b ln (total Cost)+c ln (load)] (31)
Wherein f is the objective function of the present embodiment task schedule, in scheduling process target be realize improved difference into
Change the minimum of algorithm IADE Proper treatment f value, a, b, c in formula are weight factor, meet condition a+b+c=1, weight factor
To control the influence in task scheduling process for task completion time, cost, resource load to comprehensive appropriateness value size
Degree.
Step S207 obtains target individual, variation individual, the appropriateness for intersecting individual, anti-cross individual according to Proper treatment
Value, and be worth according to appropriateness and obtain next-generation individual, to obtain initial schedule result to next-generation individual decoding.
What the individual choice step of DE algorithm utilized is nature " survival of the fittest " thought.First according to the appropriateness of definition
Function obtains the appropriateness value of individual, the carry out greediness selection according to the size of value to individual later, if minimization problem is then fitted
Angle value is smaller to illustrate that individual is more outstanding, therefore when the fitness value of new individual is smaller than original individual, knows from experience for original and washed in a pan
It eliminates, new individual can be retained to the next generation, so, the worst individual situation in population is to maintain and original identical fitness
Value, and individual will not be deteriorated.Therefore, population at individual can be continuously available optimization, and final convergence is stablized.
In formula, f (x) for the problem of being solved objective function, present embodiment assumes that algorithm solve be minimization problem.
In the selection operation step of standard DE algorithm, participating in compare two individuals is target individual and crossover operation
The intersection individual of generation, crossover operation further adjusts each dimension component for the individual vector that mutation operation obtains, so that a
Body is more diversified so that algorithm can search optimal solution in a big way.But it may also can be generated in mutation operation step
Globally optimal solution, and since directly crossover operation can be carried out to variation individual after mutation operation, variation is substituted in crossover operation
The local component of vector, this is likely to result in outstanding solution and is destroyed.
In addition, parent individuality or variation can be selected according to crossover probability CR in the crossover operation step of standard DE algorithm
The part component composition of body intersects individual vector, and carries out Negative selection according to crossover probability CR and then may be constructed another not
With intersection individual vector, as shown in Fig. 7 thickened portion, and due in standard DE algorithm only just with the therein of generation
The individual vector of one intersection, this may also will lead to the outstanding loss for solving individual component.
Therefore, in order to avoid destroying the potential excellent individual of population, algorithm is enable to obtain current optimal solution as far as possible.
Standard DE algorithm is improved, in the selection operation step of improved DE algorithm, from variation vector vi, intersect vector ui, it is anti-
To intersection vector hiAnd object vector xiModerately the optimal individual of value enters the next generation for middle selection, and improved selection strategy is such as
Shown in formula (33):
In the selection operation step of standard DE algorithm, when the appropriateness value for intersecting individual is bigger than the appropriateness value of parent individuality
When, parent individuality is retained, and the present embodiment improves this.When variation vector vi, intersect vector ui, reversed intersect vector hi
Appropriateness value than parent object vector xiAppropriateness value it is big when, it is a as the next generation using the mean value of optimum individual and intermediate value individual
Body accelerates convergence speed of the algorithm and population diversity with this.Selection strategy further progress in formula (33) is improved, most
The formula of selection strategy is as follows after improving eventually:
Wherein, xbestIt indicates moderately to be worth optimal individual, x in contemporary populationavgIndicate all individuals in contemporary population
Intermediate value, xavgIndividual component xavg,jCalculation formula is as follows:
According to it is above-mentioned to mutagenic factor, intersect the factor, variation and selection strategy improved procedure, the calculation of the present embodiment IADE
Method pseudocode is as shown in table 2:
Table 2
The course of work of the t times iteration of population in IADE algorithm, as shown in Figure 8.The present embodiment passes through to standard DE algorithm
It is existing to be easy to be limited to local optimum, the problems such as late convergence is slow, and parameter setting is difficult, propose a kind of improved difference
Evolution algorithm IADE, the algorithm improvement mutagenic factor F of standard DE algorithm, the strategy for intersecting factor CR, variation and selection.It will
The DE algorithm of IADE algorithm and other variation modes is by 8 classics benchmark functions tests, by Experimental comparison results,
It is concluded that IADE algorithm will be better than the DE algorithm of other variation modes in convergence rate with the ability for jumping out local optimum.
Differential evolution algorithm (Differential Evolution, DE) is the global optimization approach novel as one kind,
Because its is high-efficient, strong robustness, realize simple due to be widely used.The present embodiment is for DE algorithm iteration later period convergence speed
It spends the deficiencies of slack-off, parameter relies on strong setting difficulty to be improved, proposes improved DE algorithm (Improved
Adaptation Differential Evolution, IADE), the algorithm improvement zoom factor F of standard DE algorithm, intersect
Probability CR and variation and selection strategy adjust parameter adaptively with the iterative evolution of population, each to meet
Period requires the difference of parameter value, and the performance of algorithm after improving is demonstrated by benchmark function.
Since problem of load balancing has similar place with Q value method (Q-value) Modelling for Seats Distribution Problem: Modelling for Seats Distribution Problem
Purpose is that the number of covers that can make each side assigned as far as possible is directly proportional to its own population, and load balancing makes each resource as far as possible
Being distributed for task is directly proportional to resource offer ability.Therefore the present embodiment improves standard Q value method according to Mission Scheduling,
It is applied in the solution of task schedule problem of load balancing.
With population piRepresent the processing capacity VM of virtual machine VMPj, VM processing capacity VM in the present embodimentPjIt is big
It is small, it is measured with the storage of virtual machine, calculating, bandwidth ability, because having in multiple-objection optimization to task execution expense
It considers, so the price of the unit time of VM to be also added herein.In summary, VMjProcessing capacity formula indicate are as follows:
The number of covers n currently distributed in standard Q value method formulaiWith the execution time vmTime of task on current VMjTable
Show, adds one to be changed to plus the task execution time moved on the virtual machine the number of covers in standard Q value method formula
taskTimeij:
Q value method, improved Q value method formula are improved according to the different of processing capacity between different VM are as follows:
Start to execute dispatching algorithm after completing cluster in pretreatment stage, carries out task tune parallel in multiple tasks set
Degree, the corresponding resource set of a task-set form a small-sized cloud computing environment.The task of the same category is in resource requirement
Aspect has similitude, and cluster also makes task distribution be provided with certain purpose.But work as the task mistake of a certain classification of user
When more, its corresponding resource set overload can be made, and other resource set underloads are even idle, equally will cause system
Whole load imbalance problem, leads to the growth of task completion time, so that resource utilization is not high, it is full to thereby reduce user
Meaning degree and the operation cost for improving cloud supplier.
Currently based in the task schedule research of cluster, the laod unbalance problem as caused by clustering is not solved,
The present embodiment improves this, and after cluster and task are assigned, carries out local task according to the load condition of resource
Secondary readjustment degree.In the first stage, the present embodiment is scheduled using improved differential evolution algorithm, by balancing resource load
Degree, task completion time and executory cost are as regulation goal, and after the completion of dispatching at this time, it is equal that load is substantially in resource set
Weighing apparatus state.But can may also have load imbalance between different resource collection, in order to realize the load balancing of system entirety resource
Carrying out two times scheduling, cpu busy percentage is excessively high but memory usage is low or cpu busy percentage is low but memory is sharp to avoid occurring
With the phenomenon that rate is excessively high.So needing to carry out global load balancing, namely when certain class is also in the void of the state of high load
When quasi- machine, the task schedule of the virtual machine is handled to other class low-load state resources, so that the overall load of system
More equilibrium state can be reached.
Specifically, the present embodiment realizes that load balancing part readjustment degree can be mainly divided into three steps: the first step, according to
VM current load condition is grouped VM in system；Second step is the task that selection needs to migrate from heavily loaded VM set,
It places it in task queue to be migrated；In the third step, by the VM of the task rescheduling in task queue to low-load with
Realize global whole load balancing.
Virtual machine is divided into heavy duty set, equilibrium according to the load value of virtual machine in initial schedule result by step S208
Collections of loads and light collections of loads.
The current loading condition that each virtual machine is calculated after maiden mission distribution, is divided into three according to the load condition
Queue is respectively as follows: overload virtual machine queue, balanced virtual machine queue and underloading virtual machine queue.That removes from overload VM appoints
Business needs to find the VM of suitable underload, because only that such VM can just distribute to new task.Sentenced with adaptive threshold
The mode for determining virtual machine load condition is grouped the virtual machine in system:
In formula, loadmaxFor VM maximum load value, loadminFor VM minimum load value, loadmidBy minimum and maximum negative
Load value is calculated, loadupperTo overload boundary value, loadlowerTo be lightly loaded boundary value.
Virtual machine is grouped according to threshold value:
(1) it is lightly loaded virtual machine VM-_Over queue: load section [loadmin, loadlower) indicate task node load
It is relatively light, more multitask can be undertaken.
(2) balanced virtual machine VM-_Normal queue: load section [loadlower, loadupper] indicate task node
Load is best, and utilization of resources situation is best.
(3) virtual machine VM_Light queue is overloaded: load section (loadupper, loadmax] indicate task node load
Overweight, assigned task amount is excessive at this time.
Step S209, selection needs the task-set of degree of readjustment from heavy duty set, obtains readjustment degree task-set.
After resource is grouped by the present embodiment according to load condition, need to pending on the virtual machine in heavy condition
Task in task queue is selected, and selected task is rescheduled on the virtual machine of light condition executes later,
So that the virtual machine of system is in the state of load balancing after readjustment degree.
In order to carry out load balancing operation, it is necessary to first determine overload VM task list to be migrated, preferentially judge heavy burden
It is more than or equal to load with the presence or absence of task execution time on the VM of load statej-loadupperTask, and if it exists, directly by this
Business is moved out, and virtual machine can reach load balancing state, if it does not exist the task of the type, will be overloaded appointing in VM task queue
According to scale, ascending order is arranged from small to large for business, task list to be transferred is added in the smallest task of scale every time, while judging to appoint
Business load, is added task list to be transferred for next task if still overloading, successively recurrence, bears until the load of VM is less than
Carry threshold value loadupperStop when the upper limit.
Step S210, it is using Q value-based algorithm, the task schedule in readjustment degree task-set is virtual into light collections of loads
Machine.
Specifically, when selecting virtual machine in light collections of loads, improved Q value method is dispatched using task based access control to determine and close
Suitable virtual machine, the destination node shifted with tasks clear.Preferentially task immigration is executed into the big virtual machine of Q value.
The load condition of heavy duty and low-load resource collection VM is updated after task immigration.After migration, if resource is negative
It carries and is in moderate state, then deleted from heavy duty resource collection.The virtual machine of remaining high load condition is iterated negative
Balanced weight scheduling operation is carried, until the resources of virtual machine collection of low-load state is combined into sky.
Embodiment three
As shown in figure 9, the multiple target method for scheduling task of the present embodiment three includes:
Step1: carrying out fuzzy clustering processing to independent task and virtual machine using fuzzy clustering algorithm respectively, according to appoint
Business characteristic and virtual machine feature are divided into each class；
Step2: according to the comprehensive performance of set of tasks resource requirement size and virtual machine set gathered between
Match；
Step3: successively by the task in set according to the multiple target task scheduling strategy tune of improved differential evolution algorithm
It spends in the resources of virtual machine in corresponding virtual machine set, the scheduling result of acquisition initiating task to virtual machine；
Step4: calculating the load condition of virtual machine, is grouped according to load value to virtual machine, is divided into three groups:
Heavy duty, balanced and light load；
Step5: determining task-set to be reassigned, by the task schedule of the overload on heavily loaded virtual machine to underloading virtual machine
Upper operation, the task based access control that determines the use of of target virtual machine dispatches improved Q value method in scheduling process again.Task has been reassigned
At the scheduling scheme of the final task of rear acquisition to virtual machine；
Step6: task execution and completion.
The present embodiment task scheduling strategy MODE is described in detail below shown in table 3:
Table 3
Actual conditions of the present embodiment in view of the common interest and task schedule of user and cloud service provider in cloud platform
The optimization aim for determining the present embodiment is gathered using fuzzy clustering algorithm according to the characteristic of task and resource simultaneously in pretreatment stage
Class reduces the exploration space of task, accelerates task dispensing rate.Using improved Q value method, solve between resource set due to
Load imbalance problem caused by cluster improves the utilization of resources so that entire cloud system reaches the state of load relative equilibrium
Rate.
In order to test the multiple target task schedule based on cluster and improved differential evolution algorithm of the present embodiment proposition
Tactful MODE is used to solve the validity of cloud computing Mission Scheduling, and the present embodiment tests MODE by emulation experiment
Card, and MODE and other dispatching algorithms are subjected to experimental result comparative analysis.
It is proposed that the multiple target task scheduling strategy based on cluster and improved differential evolution algorithm exists to test in the present embodiment
Performance advantage when cloud Mission Scheduling is solved, the present embodiment is by emulation testing, the task schedule that the present embodiment is proposed
Tactful MODE and Min-Min algorithm, standard DE algorithm, heredity (Genetic Algorithm, GA) algorithm carry out performance comparison.
The experimental situation of the present embodiment task scheduling strategy is as shown in table 4:
Table 4
The data of each virtual machine and task that the present embodiment is used in emulation experiment are all randomly generated, table 5,
6 show specific data value range:
Table 5
Table 6
Table 7 is the parameter setting of each algorithm:
Table 7
The present embodiment uses task completion time (Makespan), load balancing value to the task scheduling strategy of proposition
Three indexs of (Load balancing level), task execution cost overhead (Cost) carry out Experimental comparison and evaluation.
(1) cluster operation comparative experiments
When being scheduled to task, when in face of user task amount is big, system resource number is more, task choosing resource when
Between expense can increase, cause cloud platform task schedule efficiency lower, therefore, in scheduling process reduce resource selection range just
With necessity, and cluster operation is a kind of effective means for reducing range of choice.The present embodiment compares without cluster DE algorithm
With DE algorithm execution time after cluster, comparison result is as shown in Figure 10, and wherein horizontal axis indicates task number in scheduling every time, the longitudinal axis
Indicate the execution time of algorithm.From experimental data, cluster than no cluster DE algorithm when being executed between on have and obviously mention
It is high.
(2) task completion time comparative experiments
Task completion time (Makespan) is a kind of index of evaluation handling capacity of isomery cloud computing system, it is to weigh
Measure from first task initially enter cloud platform to a last task execution at the end of time spend.
Experiment one: the independent task of different number is scheduled on 10 virtual machines with dispatching algorithm, in order to avoid
The contingency of experiment, repeats 10 experiment statistics comparisons respectively, and experimental result is as shown in figure 11.
As can be seen from Figure 11 in the case where virtual machine quantity remains unchanged, the task completion time of each algorithmic dispatching
All as the mode in direct ratio that increases of task amount increases.When task amount is smaller, each algorithm task completion time difference is not
Greatly；Become larger when with task amount, starts to show significant difference.Min-Min finds local optimum by greedy mode, and DE and
GA algorithm belongs to evolution algorithm, with the obvious advantage in processing large-scale data Combinatorial Optimization.DE algorithm is for GA algorithm
Optimal solution Approximation effect is more significant, and GA algorithm is the choosing that the individual after individual intersection and variation is controlled according to fitness value
Select, when handling minimization problem, the small probability that is selected of individual of fitness value be will increase, and DE algorithm variation vector be by
The random vector difference weighted array of parent individuality generates, and intersects generation with parent individuality and intersect individual vector and directly and his father
It is selected for individual, is just occurred when only more excellent than parent individuality value.Therefore, for optimization problem, DE algorithm ratio GA algorithm is received
Hold back that speed is faster and more accurate, the task completion time of DE algorithm is better than GA algorithm.Standard DE algorithm since parameter is fixed,
The reduction for relatively easily leading to search efficiency and solving precision, to DE algorithm improvement parameter in MODE strategy, so that adaptive
The needs for meeting Evolution of Population each stage should be changed, and mixovariation strategy has been used to carry out mutation operation, make algorithmic statement
Speed is accelerated solving precision and is got higher, while reducing time overhead of the task on matching virtual machine by cluster, with task
The increase of amount has more advantage.
Experiment two: 150 independent tasks are scheduled on the virtual machine of different number with dispatching algorithm, in order to avoid
The contingency of experiment, repeats 10 experiment statistics comparisons respectively, and experimental result is as shown in figure 12.
In figure 12 it can be seen that in the case where task quantity remains unchanged, the task completion time of each algorithm with
The inversely proportional mode that increases of virtual machine quantity reduce, it is seen that the quantity of resource can largely affect the completion of task
Time.When resource abundance, task completion time reduces trend and slows down.Standard DE algorithm compared with GA algorithm be not much different but
Standard DE algorithm is slightly excellent, and the task completion time of MODE is better than always other algorithms of comparison.
(3) task completes Cost comparisons' experiment
The profit that cloud service provider can be measured by task execution cost overhead (Cost), in the case where certain task amount
The profit of the smaller cloud service provider of Cost is higher.
The cost of the task execution of each algorithm is compared, in order to avoid the contingency of experiment, is repeated respectively
10 experiment statistics comparisons, experimental result are as shown in figure 13.
Task expense increases with task quantity and is increased as can be seen from Figure 13.When task quantity is smaller, standard DE, GA
And MODE algorithm performance gap is little, when task quantity increases, the task expense of MODE is significantly less than other algorithms.Min-
The task execution cost of Min algorithm is apparently higher than other algorithms, because Min-Min greediness characteristic always assigns the task to execution
The stronger virtual machine of ability is to seek time optimal, but the cost overhead of such virtual machine can be big.
(4) load balancing degrees comparative experiments
Load balancing value (Load balancing level), to measure resource utilization ratio.Task scheduling algorithm
The load balancing between node must be fully taken into account in carry out task distribution, keeps each node load balanced as far as possible, avoid appointing
Business waits operation all in the node of better performances.
It is compared for VM load, experiment measures system resource with the standard deviation of the task execution time of each VM
Overall load equilibrium situation.In order to avoid the contingency of experiment, repeat 10 experiment statistics comparisons respectively, experimental result is such as
Shown in Figure 14.
Figure 14 show the size of the load balancing value when the task of virtual machine load different number is scheduled, Cong Tuzhong
It can be seen that increasing with task amount, the load of virtual machine can be increased with it, since Min-Min algorithm is only by the execution of task
Time as optimization aim, has ignored considering for balancing resource load, therefore load characteristic is worst.Standard DE algorithm is in scheduling
It is slightly more excellent than GA algorithm, and standard DE algorithm leads to solving precision and speed due to the limitation of preset parameter and single Mutation Strategy
Lower than IADE, so that MODE scheduling strategy is relatively slow compared with other algorithmic load growth rate and load value is more excellent.
By carrying out performance verification and with the comparative analyses of other algorithms it is found that for independently appointing in cloud environment to algorithm
It is engaged in scheduling problem, the present embodiment strategy MODE not only can be improved the execution efficiency of task, shorten the deadline but also can be with
The cost overhead of task execution is reduced, so that system keeps higher resource utilization.
The present embodiment is tested by the task scheduling strategy MODE that emulation experiment proposes the present embodiment, and will
It carries out emulation experiment comparison with standard DE algorithm, GA algorithm, Min-Min algorithm, shows this implementation by contrast and experiment
The MODE policing algorithm that example proposes is all with good performance in deadline, load balancing, executory cost.
Referring to Fig.1 5, the multiple target task scheduling system that the embodiment of the present invention proposes, comprising:
Memory 10, processor 20 and it is stored in the computer journey that can be run on memory 10 and on processor 20
Sequence, wherein processor 20 realizes the step of multiple target method for scheduling task that the present embodiment proposes when executing computer program.
The specific work process and working principle of the multiple target task scheduling system of the present embodiment can refer in the present embodiment
Multiple target method for scheduling task the course of work and working principle.
These are only the preferred embodiment of the present invention, is not intended to restrict the invention, for those skilled in the art
For member, the invention may be variously modified and varied.All within the spirits and principles of the present invention, it is made it is any modification,
Equivalent replacement, improvement etc., should all be included in the protection scope of the present invention.
Claims (9)
1. a kind of multiple target method for scheduling task, which is characterized in that the described method includes:
Using fuzzy clustering algorithm fuzzy clustering processing is carried out to task and virtual machine respectively, obtains Task clustering set and void
Quasi- machine cluster set, and the Task clustering set is matched with virtual machine cluster set, obtain matching set；
It is obtained using differential evolution algorithm by the task schedule in the matching set to the virtual machine in the matching set
Initial schedule result；
According to the load value of virtual machine in the initial schedule result, the virtual machine is divided into heavy duty set, equally loaded
Set and light collections of loads；
Selection needs the task-set of degree of readjustment from the heavy duty set, obtains readjustment degree task-set；
Using Q value-based algorithm, by the task schedule in the readjustment degree task-set to the virtual machine in the light collections of loads.
2. multiple target method for scheduling task according to claim 1, which is characterized in that by the Task clustering set and institute
It states virtual machine cluster set to be matched, obtaining matching set includes:
According to computing capability, bandwidth and the memory capacity of virtual machine in virtual machine cluster set, weighted calculation obtains empty
The resource performance of quasi- machine；
According to the average value of the resource performance of all virtual machines in virtual machine cluster set, the virtual machine cluster set is obtained
The comprehensive resources performance of conjunction；
According to the length of task, size and output data size in the Task clustering set, weighted calculation obtains task
Resource requirement；
According to the average value of the resource requirement of all tasks in the Task clustering set, the comprehensive of the Task clustering set is obtained
Close resource requirement；
According between the comprehensive resources performance of virtual machine cluster set and the comprehensive resources demand of the Task clustering set
Distance, the Task clustering set and the virtual machine cluster set are matched, matching set is obtained.
3. multiple target method for scheduling task according to claim 2, which is characterized in that differential evolution algorithm is used, by institute
The task schedule in matching set is stated to the virtual machine in the matching set, obtaining initial schedule result includes:
According to task quantity, the virtual machine quantity, mission number in the matching set, differential evolution is calculated using one-dimension array
The individual of population is encoded in method；
Mutagenic factor is calculated according to current evolutionary generation and intersects the factor；
According to the mutagenic factor, in conjunction with a variety of Mutation Strategies, mutation operation is carried out to population at individual, obtains variation individual；
According to the component of the intersection selecting predictors parent individuality or variation individual, obtains and intersect individual；
According to the component of the intersection factor Negative selection parent individuality or variation individual, anti-cross individual is obtained；
According to task completion time, task execution cost and virtual machine load balancing degrees, Proper treatment is determined；
Target individual, the variation individual, the appropriateness value for intersecting individual, anti-cross individual are obtained according to the Proper treatment, and
Next-generation individual is obtained according to the appropriateness value, to obtain initial schedule result to the next-generation individual decoding.
4. multiple target method for scheduling task according to claim 3, which is characterized in that calculated according to current evolutionary generation
Mutagenic factor and the specific formula for intersecting the factor are as follows:
Wherein, F (t) indicates that mutagenic factor, CR (t) indicate to intersect the factor, and w (t) indicates weight factor, and t indicates evolutionary generation, T
Indicate population maximum evolutionary generation, FmaxIndicate mutagenic factor maximum value, FminIndicate mutagenic factor minimum value, CRminIt indicates to intersect
Factor minimum value, CRmaxIt indicates to intersect factor maximum value.
5. multiple target method for scheduling task according to claim 4, which is characterized in that according to the mutagenic factor, in conjunction with
A variety of Mutation Strategies carry out mutation operation to population at individual, obtain the specific formula of variation individual are as follows:
Wherein, λ indicates weight factor, and u indicates impact factor, vi(t) the corresponding variation individual of population present age evolutionary generation, t are indicated
Indicate that evolutionary generation, T indicate population maximum evolutionary generation, xiIndicate current individual, xr1(t)、xr2(t)、xr3(t)、xr4(t) generation
The population present age evolutionary generation that table is chosen at random corresponding four are different from xiIndividual, xbest(t) indicate optimal in contemporary population
Individual, F indicate mutagenic factor.
6. multiple target method for scheduling task according to claim 5, which is characterized in that obtain mesh according to the Proper treatment
Mark individual, the variation individual, the appropriateness value for intersecting individual, anti-cross individual, and next-generation is obtained according to the appropriateness value
The calculation formula of body specifically:
Wherein, f indicates Proper treatment, and f=min [aln (totalTime)+bln (totalCost)+cln (load)], xi t+1
Indicate next-generation target individual, ui t+1It indicates next-generation and intersects individual, f (ui t+1) indicate the next-generation appropriateness value for intersecting individual,
vi t+1Indicate next-generation variation individual, f (vi t+1) indicate that the appropriateness of next-generation variation individual is worth, hi t+1Indicate next-generation anti-cross
Individual, f (hi t+1) indicate that the appropriateness of next-generation anti-cross individual is worth, f (xi t) indicate that the appropriateness of contemporary target individual is worth, xbestTable
Show and is moderately worth optimal individual, x in contemporary populationavgIndicate the intermediate value of all individuals in contemporary population, totalTime is indicated
Task completion time, totalCost indicate task run cost overhead, load indicate load balancing degrees, a, b, c be weight because
Son meets condition a+b+c=1.
7. -6 any multiple target method for scheduling task according to claim 1, which is characterized in that Q value in the Q value-based algorithm
Calculation formula specifically:
Wherein, QiIt indicates to the Q value after j-th of virtual machine distribution task, vmpjIndicate in the light collections of loads j-th it is virtual
The processing capacity of machine, vmTimejIndicate the execution time of task on j-th of virtual machine in the light collections of loads, taskTimeij
Indicate the task by i-th of task immigration in the readjustment degree task-set into the light collections of loads on j-th of virtual machine
Execute the time.
8. multiple target method for scheduling task according to claim 7, it is characterised in that:
The mutagenic factor maximum value Fmax=1, the mutagenic factor minimum value Fmin=0.4, the intersection factor minimum value is
CRmin=0.6, the intersection factor maximum value is CRmax=0.9.
9. a kind of multiple target task scheduling system, the system comprises:
Memory (10), processor (20) and it is stored in the computer that can be run on memory (10) and on processor (20)
Program, which is characterized in that the processor (20) realizes any institute of the claims 1 to 8 when executing the computer program
The step of stating method.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910645478.2A CN110489229B (en) | 2019-07-17 | 2019-07-17 | Multi-target task scheduling method and system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910645478.2A CN110489229B (en) | 2019-07-17 | 2019-07-17 | Multi-target task scheduling method and system |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110489229A true CN110489229A (en) | 2019-11-22 |
| CN110489229B CN110489229B (en) | 2020-06-30 |
Family
ID=68547336
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910645478.2A Active CN110489229B (en) | 2019-07-17 | 2019-07-17 | Multi-target task scheduling method and system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110489229B (en) |
Cited By (14)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111176792A (en) * | 2019-12-31 | 2020-05-19 | 华为技术有限公司 | Resource scheduling method, device and related equipment |
| CN111208796A (en) * | 2020-04-21 | 2020-05-29 | 天津开发区精诺瀚海数据科技有限公司 | Workshop production operation scheduling method based on clustering niche genetic algorithm |
| CN111382941A (en) * | 2020-03-09 | 2020-07-07 | 河海大学常州校区 | Parallel task scheduling method with multiple constraint conditions |
| CN111818119A (en) * | 2020-05-04 | 2020-10-23 | 国网浙江省电力有限公司 | Data service resource optimization scheduling method based on improved spider cluster |
| CN112187891A (en) * | 2020-09-16 | 2021-01-05 | 国网浙江省电力有限公司信息通信分公司 | Load optimization method and device of edge computing node set based on multiple services |
| CN112492651A (en) * | 2020-11-23 | 2021-03-12 | 中国联合网络通信集团有限公司 | Resource scheduling scheme optimization method and device |
| CN112579280A (en) * | 2020-12-30 | 2021-03-30 | 奇瑞汽车股份有限公司 | Scheduling method and device of cloud resources and computer storage medium |
| CN112667379A (en) * | 2020-12-29 | 2021-04-16 | 深圳Tcl新技术有限公司 | Task scheduling method and server |
| CN112882805A (en) * | 2021-01-26 | 2021-06-01 | 上海应用技术大学 | Profit optimization scheduling method based on task resource constraint |
| CN113361150A (en) * | 2021-08-11 | 2021-09-07 | 华东交通大学 | Multi-objective optimization method and system for urban train operation |
| CN113938813A (en) * | 2020-06-29 | 2022-01-14 | ***通信有限公司研究院 | Task processing method and device and storage medium |
| CN114546623A (en) * | 2022-03-01 | 2022-05-27 | 淮安市第二人民医院 | Task scheduling method and system based on big data system |
| CN114900525A (en) * | 2022-05-20 | 2022-08-12 | 中国地质大学（北京） | Method and system for deflecting data stream |
| CN116401071A (en) * | 2023-06-07 | 2023-07-07 | 江西财经大学 | Resource allocation method and system for edge calculation |
Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP1536335A2 (en) * | 2003-10-30 | 2005-06-01 | Alcatel | Intelligent scheduler for multi-level exhaustive scheduling of tasks |
| CN101788920A (en) * | 2010-02-05 | 2010-07-28 | 北京航空航天大学 | CPU virtualization method based on processor partitioning technology |
| CN102968344A (en) * | 2012-11-26 | 2013-03-13 | 北京航空航天大学 | Method for migration scheduling of multiple virtual machines |
| CN104331321A (en) * | 2014-10-09 | 2015-02-04 | 冷明 | Cloud computing task scheduling method based on tabu search and load balancing |
| CN105094989A (en) * | 2015-08-18 | 2015-11-25 | 南京邮电大学 | Self-adaption QoS granularity partition method based on fuzzy clustering |
| CN105677470A (en) * | 2016-01-07 | 2016-06-15 | 中国联合网络通信集团有限公司 | Virtual machine task scheduling method and virtual machine system |
| CN108089918A (en) * | 2017-12-06 | 2018-05-29 | 华中科技大学 | A kind of figure computational load equalization methods towards heterogeneous server structure |
| CN109522104A (en) * | 2018-10-15 | 2019-03-26 | 华南理工大学 | Utilize the method for differential evolution algorithm optimization two goal task of Iaas scheduling |
| CN109976909A (en) * | 2019-03-18 | 2019-07-05 | 中南大学 | Low delay method for scheduling task in edge calculations network based on study |
-
2019
- 2019-07-17 CN CN201910645478.2A patent/CN110489229B/en active Active
Patent Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP1536335A2 (en) * | 2003-10-30 | 2005-06-01 | Alcatel | Intelligent scheduler for multi-level exhaustive scheduling of tasks |
| CN101788920A (en) * | 2010-02-05 | 2010-07-28 | 北京航空航天大学 | CPU virtualization method based on processor partitioning technology |
| CN102968344A (en) * | 2012-11-26 | 2013-03-13 | 北京航空航天大学 | Method for migration scheduling of multiple virtual machines |
| CN104331321A (en) * | 2014-10-09 | 2015-02-04 | 冷明 | Cloud computing task scheduling method based on tabu search and load balancing |
| CN105094989A (en) * | 2015-08-18 | 2015-11-25 | 南京邮电大学 | Self-adaption QoS granularity partition method based on fuzzy clustering |
| CN105677470A (en) * | 2016-01-07 | 2016-06-15 | 中国联合网络通信集团有限公司 | Virtual machine task scheduling method and virtual machine system |
| CN108089918A (en) * | 2017-12-06 | 2018-05-29 | 华中科技大学 | A kind of figure computational load equalization methods towards heterogeneous server structure |
| CN109522104A (en) * | 2018-10-15 | 2019-03-26 | 华南理工大学 | Utilize the method for differential evolution algorithm optimization two goal task of Iaas scheduling |
| CN109976909A (en) * | 2019-03-18 | 2019-07-05 | 中南大学 | Low delay method for scheduling task in edge calculations network based on study |
Non-Patent Citations (1)
| Title |
|---|
| 董丽丽 等: "云计算中基于差分进化算法的任务调度研究", 《计算机工程与应用》 * |
Cited By (25)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2021136137A1 (en) * | 2019-12-31 | 2021-07-08 | 华为技术有限公司 | Resource scheduling method and apparatus, and related device |
| CN111176792B (en) * | 2019-12-31 | 2023-11-17 | 华为技术有限公司 | Resource scheduling method and device and related equipment |
| CN111176792A (en) * | 2019-12-31 | 2020-05-19 | 华为技术有限公司 | Resource scheduling method, device and related equipment |
| CN111382941B (en) * | 2020-03-09 | 2022-08-02 | 河海大学常州校区 | Parallel task scheduling method with multiple constraint conditions |
| CN111382941A (en) * | 2020-03-09 | 2020-07-07 | 河海大学常州校区 | Parallel task scheduling method with multiple constraint conditions |
| CN111208796A (en) * | 2020-04-21 | 2020-05-29 | 天津开发区精诺瀚海数据科技有限公司 | Workshop production operation scheduling method based on clustering niche genetic algorithm |
| CN111818119A (en) * | 2020-05-04 | 2020-10-23 | 国网浙江省电力有限公司 | Data service resource optimization scheduling method based on improved spider cluster |
| CN111818119B (en) * | 2020-05-04 | 2023-10-24 | 国网浙江省电力有限公司 | Improved spider cluster-based data service resource optimization scheduling method |
| CN113938813A (en) * | 2020-06-29 | 2022-01-14 | ***通信有限公司研究院 | Task processing method and device and storage medium |
| CN112187891A (en) * | 2020-09-16 | 2021-01-05 | 国网浙江省电力有限公司信息通信分公司 | Load optimization method and device of edge computing node set based on multiple services |
| CN112187891B (en) * | 2020-09-16 | 2022-06-10 | 国网浙江省电力有限公司信息通信分公司 | Load optimization method and device of edge computing node set based on multiple services |
| CN112492651A (en) * | 2020-11-23 | 2021-03-12 | 中国联合网络通信集团有限公司 | Resource scheduling scheme optimization method and device |
| CN112492651B (en) * | 2020-11-23 | 2023-07-21 | 中国联合网络通信集团有限公司 | Resource scheduling scheme optimization method and device |
| CN112667379B (en) * | 2020-12-29 | 2024-06-07 | 深圳Tcl新技术有限公司 | Task scheduling method and server |
| CN112667379A (en) * | 2020-12-29 | 2021-04-16 | 深圳Tcl新技术有限公司 | Task scheduling method and server |
| CN112579280A (en) * | 2020-12-30 | 2021-03-30 | 奇瑞汽车股份有限公司 | Scheduling method and device of cloud resources and computer storage medium |
| CN112579280B (en) * | 2020-12-30 | 2022-11-29 | 奇瑞汽车股份有限公司 | Cloud resource scheduling method and device and computer storage medium |
| CN112882805A (en) * | 2021-01-26 | 2021-06-01 | 上海应用技术大学 | Profit optimization scheduling method based on task resource constraint |
| CN113361150B (en) * | 2021-08-11 | 2021-12-24 | 华东交通大学 | Multi-objective optimization method and system for urban train operation |
| CN113361150A (en) * | 2021-08-11 | 2021-09-07 | 华东交通大学 | Multi-objective optimization method and system for urban train operation |
| CN114546623B (en) * | 2022-03-01 | 2022-12-27 | 淮安市第二人民医院 | Task scheduling method and system based on big data system |
| CN114546623A (en) * | 2022-03-01 | 2022-05-27 | 淮安市第二人民医院 | Task scheduling method and system based on big data system |
| CN114900525A (en) * | 2022-05-20 | 2022-08-12 | 中国地质大学（北京） | Method and system for deflecting data stream |
| CN116401071A (en) * | 2023-06-07 | 2023-07-07 | 江西财经大学 | Resource allocation method and system for edge calculation |
| CN116401071B (en) * | 2023-06-07 | 2023-08-29 | 江西财经大学 | Resource allocation method and system for edge calculation |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110489229B (en) | 2020-06-30 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110489229A (en) | A kind of multiple target method for scheduling task and system | |
| Mansouri et al. | Hybrid task scheduling strategy for cloud computing by modified particle swarm optimization and fuzzy theory | |
| Alkayal et al. | Efficient task scheduling multi-objective particle swarm optimization in cloud computing | |
| Houssein et al. | Task scheduling in cloud computing based on meta-heuristics: review, taxonomy, open challenges, and future trends | |
| CN110737529B (en) | Short-time multi-variable-size data job cluster scheduling adaptive configuration method | |
| Ben Alla et al. | A novel task scheduling approach based on dynamic queues and hybrid meta-heuristic algorithms for cloud computing environment | |
| CN107273209B (en) | Hadoop task scheduling method based on minimum spanning tree clustering improved genetic algorithm | |
| CN112286677B (en) | Resource-constrained edge cloud-oriented Internet of things application optimization deployment method | |
| CN104239141B (en) | Optimizing and scheduling task method based on workflow critical path in data center | |
| Hüsken et al. | Structure optimization of neural networks for evolutionary design optimization | |
| CN110096349A (en) | A kind of job scheduling method based on the prediction of clustered node load condition | |
| CN111124689A (en) | Dynamic allocation method for container resources in cluster | |
| CN112685138B (en) | Multi-workflow scheduling method based on multi-population hybrid intelligent optimization in cloud environment | |
| CN105744006A (en) | Particle swarm optimization user request dispatching method facing multi-type service | |
| CN108009003A (en) | A kind of multiple target virtual machine adaptive location system of selection, distributed cloud system | |
| Chakravarthi et al. | TOPSIS inspired budget and deadline aware multi-workflow scheduling for cloud computing | |
| CN111813500B (en) | Multi-target cloud workflow scheduling method and device | |
| US20140278695A1 (en) | Systems and methods for prioritizing funding of projects | |
| CN109871488A (en) | A kind of Web service construction method and Web service for merging availability and user preference | |
| Kodli et al. | Hybrid Max-Min Genetic Algorithm for Load Balancing and Task Scheduling in Cloud Environment. | |
| He | Optimization of edge delay sensitive task scheduling based on genetic algorithm | |
| Keivani et al. | A review of recent methods of task scheduling in cloud computing | |
| Liu et al. | A pareto-based particle swarm optimization algorithm for multi-objective location routing problem. | |
| CN108650191B (en) | Decision method for mapping strategy in virtual network | |
| CN113157381B (en) | Multi-workflow scheduling method based on non-dominated sorting and owl search |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| TR01 | Transfer of patent right |
Effective date of registration: 20240625 Address after: Room 1001, Comprehensive Building, Futian Xingye Building, No. 377 Jinma Road, Furong North Road Street, Kaifu District, Changsha City, Hunan Province, 410000 Patentee after: Hunan Zhongkan Beidou Research Institute Co.,Ltd. Country or region after: China Address before: 410022 No. 98 Hongshan Road, Kaifu District, Changsha City, Hunan Province Patentee before: CHANGSHA University Country or region before: China |