CN107656799A - The workflow schedule method of communication and calculation cost is considered under a kind of more cloud environments - Google Patents
The workflow schedule method of communication and calculation cost is considered under a kind of more cloud environments Download PDFInfo
- Publication number
- CN107656799A CN107656799A CN201711077227.6A CN201711077227A CN107656799A CN 107656799 A CN107656799 A CN 107656799A CN 201711077227 A CN201711077227 A CN 201711077227A CN 107656799 A CN107656799 A CN 107656799A
- Authority
- CN
- China
- Prior art keywords
- mrow
- msub
- particle
- msubsup
- mtd
- 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
Classifications
-
- 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
- G06F9/44—Arrangements for executing specific programs
- G06F9/455—Emulation; Interpretation; Software simulation, e.g. virtualisation or emulation of application or operating system execution engines
- G06F9/45533—Hypervisors; Virtual machine monitors
- G06F9/45558—Hypervisor-specific management and integration aspects
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L67/00—Network arrangements or protocols for supporting network services or applications
- H04L67/01—Protocols
- H04L67/10—Protocols in which an application is distributed across nodes in the network
-
- 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
- G06F9/44—Arrangements for executing specific programs
- G06F9/455—Emulation; Interpretation; Software simulation, e.g. virtualisation or emulation of application or operating system execution engines
- G06F9/45533—Hypervisors; Virtual machine monitors
- G06F9/45558—Hypervisor-specific management and integration aspects
- G06F2009/4557—Distribution of virtual machine instances; Migration and load balancing
Abstract
The present invention relates to a kind of workflow schedule method that communication and calculation cost are considered under more cloud environments.This method is based on workflow self structure and performs characteristic, and the correlative factor of current cloud resource Environment communication and Executing Cost, the operation of random two-point crossover and random single-point mutation operation thought based on genetic algorithm, improve the diversity during Evolution of Population, virtual resources are integrated, consider data communication cost and task computation cost, optimize resource utilization, under the premise of the workflow deadline is met, its Executing Cost is reduced.The inventive method meets to the workflow deadline in the case of fluctuating factor presence and Executing Cost control aspect has superperformance performance, under the premise of the workflow deadline is met as far as possible, reduce its Executing Cost.
Description
Technical field
The invention belongs to parallel and distributed high-performance computing field workflow schedule method, and in particular to a kind of cloudy
The workflow schedule method of communication and calculation cost is considered under environment.
Background technology
With the continuous development of cloud computing technology, there are ' the public affairs more that multiple cloud service providers coexist in current cloud in the market
Have cloud ' situation.The property of cloud computing elastic supply virtual resource and pay-for-use, it is (following to handle extensive scientific workflow
Referred to as ' workflow ') facility is provided.However, the task scheduling under cloud isomerous environment is a NP-hard problem, workflow itself
Subtask between complicated Time Dependent and data dependence relation be present, and many differences between multiple cloud service providers be present
Different (such as charge mechanism, example types, communication bandwidth etc.), it is therefore desirable to which a kind of suitable dispatching method is meeting work as far as possible
Under the premise of streaming quality (Quality of Service, QoS), its Executing Cost is reduced.Workflow under current cloud environment
Dispatching method is to do some improvement on the basis of the workflow schedule algorithm of traditional distributed computing environment (such as grid) mostly, not
Consider cloud environment self-characteristic.Or some dispatching methods are only considered in static single cloud environment, simple pursue performs the time
Target is minimized, research is not deployed to the cost optimization scheduling problem of the workflow (such as deadline) of belt restraining.
In recent years, the workflow schedule under traditional distributed environment has been widely studied.Workflow under grid environment
Scheduling scheme, typically by heuristic or meta-heuristic dispatching algorithm, reach Optimization Work stream and perform the time, meet QoS need
Ask.The purpose of improving the resource utilization under grid environment.However, cloud computing environment and grid environment are in resource provisioning and resource
Greatest differences be present in charge mechanism.Service due to the two caused motivation, the scheme of deployment and offer etc. is not
It is identical, so the workflow schedule method under grid environment can not be simply applied to cloud computing environment.Part research work is set
The cost driving workflow schedule algorithm of cloud environment lower band time and deadline constraint is counted, but is only considered mostly a kind of
Type of virtual machine, real cloud environment is not met.In addition, there is correlative study work to propose that the single workflow based on PSO algorithms is adjusted
Degree scheme, the cost optimization problem for deadline constraint and the execution time optimal problem of budgetary restraints deploy respectively for they
Research.But the type of virtual machine and quantity of its work are fixed, cloud environment elastic supply property is not met.At present under cloud environment
Workflow schedule work consider single cloud service provider, not to more cloud environments deploy study.
Scheduling problem is driven on cost of more cloud environments with deadline constraint workflow, does not have phase also both at home and abroad at present
Close research work.Wherein, the research work of most correlation be on single cloud environment exist fluctuating factor based on the deadline
The workflow schedule research of constraint, the work handle the workflow subtask scheduling side of the overall situation using traditional PS O dispatching methods
Case.But in current research work, also it is not directed to be directed to the band deadline that communication and calculation cost are considered under how publicly-owned cloud environment
Phase constrains research-on-research stream scheduling method.
The content of the invention
It is an object of the invention to provide under a kind of more cloud environments consider communication and calculation cost workflow schedule method,
This method can meet in the case of fluctuating factor presence and Executing Cost control aspect have well to the workflow deadline
Performance, under the premise of the workflow deadline is met as far as possible, reduce its Executing Cost.
To achieve the above object, the technical scheme is that:Communication and calculation cost are considered under a kind of more cloud environments
Workflow schedule method, based on workflow self structure and characteristic is performed, and current cloud resource Environment communication and Executing Cost
Correlative factor, based on genetic algorithm random two-point crossover operation and random single-point mutation operation thought, improve Evolution of Population
During diversity, integrate virtual resources, consider data communication cost and task computation cost, optimize resource utilization,
Under the premise of the workflow deadline is met, its Executing Cost is reduced.
In an embodiment of the present invention, the workflow schedule method is implemented as follows,
It is S=(Re, Map, T by the definition of workflow schedule methodtotal,Ctotal), wherein Re represents that one group of needs enables
Resources of virtual machine Re={ vm1,vm2,...,vmr, Map={ (ti,vmj)|ti∈V,vmj∈ Re } represent that workflow neutron is appointed
The corresponding resources of virtual machine Re of business mapping relations, TtotalThe execution deadline of expression workflow, and CtotalThen represent workflow
Total Executing Cost;Workflow is represented with directed acyclic graph G (V, E), and wherein V represents to include the vertex set of n task
{t1,t2,...,tn, and E then data dependence relation { e between expression task12,e13,...,eij};Per data dependence edge eij=
(ti,tj) represent subtask tiWith subtask tjBetween data dependence relation, wherein subtask t be presentiIt is subtask tjIt is direct
Pioneer's node, and subtask tjIt is then subtask tiImmediate successor node;
Every virtual machine has corresponding type of virtual machine spi, and corresponding start-up time Tls (vmi) and close moment
Tle(vmi);After the completion of scheduled when subtask, have and time started AST (t is actually performed corresponding to one groupi) and actual execution
Deadline AET (ti), and go out on missions and will not produce again and communication data;Therefore, workflow execution deadline TtotalWith it is right
The total Executing Cost C answeredtotalRespectively as shown by the following formula:
The first half of formula (2) represents the Executing Cost of virtual machine, and latter half represents data communication cost;λpIt is cloud
The service initialization that service provider p provides for it is specifically asked a price the unit time, subtask tkIt is subtask tjFollow-up section
Point, p (tj) and p (tk) represent to perform t respectivelyjAnd tkService provider;Work as tjAnd tkPerformed by same cloud service provider
When, sjkFor 0, i.e., data communicate between not producing cloud, otherwise sjkFor 1;
Based on above related definition, the workflow schedule problem of more cloud environment lower band deadline constraints, table can be formalized
Formula (3) is shown as, its core concept is to pursue Executing Cost CtotalWhile minimum, make execution time TtotalLess than or equal to work
Make stream deadline D (w);
Then, following algorithm is performed, to realize Executing Cost CtotalWhile minimum, make execution time TtotalLess than etc.
In workflow deadline D (w):
S1:It is as follows to initialize relevant parameter:Population Size 100, maximum iteration 1000, inertia weight factor w and recognize
Know factor parameter c1_ start=0.9, c1_ end=0.2, c2_ start=0.4, c2_ end=0.9, generate initial population;
S2:According to the fitness function of particle mapping policy, and particle, i.e. formula (4), (5) and (6) calculates each grain
Fitness value under sub- different situations, the minimum particle of fitness value is therefrom selected as population global optimum particle, by first
Each particle is arranged to its own history optimal particle in generation;
Wherein, it is feasible solution that formula (4), which represents a particle, and another particle is the particle in the case of infeasible solution
Fitness function, formula (5) represent the fitness function of the particle in the case that two particles are all feasible solution, formula (6) generation
Two particles of table be all infeasible solution in the case of particle fitness function;
S3:According to particle more new formula (7) to (10) more new particle;
Wherein, formula (7) represents particle i in the update mode of t, Mu() represents mutation operation, Cg() and Cp() table
Show crossover operation,And gBesttIt is particle i itself history optimal location and whole population after t iteration respectively
History optimum position;Formula (8) represents the update mode of inertia portion, r1It is the random number between 0 to 1;Formula (9), (10)
The update mode of individual cognition and social recognition part, r are represented respectively2And r3It is the random number between 0 to 1;
S4:The fitness value of each particle is recalculated, if the fitness value of current particle is optimal less than its own history
Value, then be updated to its own history optimal particle by new particle;
S5:If the fitness value of current particle is less than the fitness value of population global optimum particle, the particle is updated
For population global optimum particle;
S6:Check whether and meet algorithm end condition, if it is satisfied, then algorithm terminates;Conversely, go to S3.
Compared to prior art, the invention has the advantages that:This method can fluctuating factor presence in the case of,
The workflow deadline is met and Executing Cost control aspect has superperformance performance, is meeting workflow cut-off as far as possible
Under the premise of date, its Executing Cost is reduced.
Brief description of the drawings
Fig. 1 workflow schedule frame diagrams.
Fig. 2 particle code patterns.
The mutation operator operation diagram of Fig. 3 inertia portions.
The crossover operator operation diagram of Fig. 4 individuals (society) cognition part.
The algorithm flow chart of Fig. 5 dispatching methods.
Embodiment
Below in conjunction with the accompanying drawings, technical scheme is specifically described.
The workflow schedule method of communication and calculation cost is considered under a kind of more cloud environments of the present invention, based on workflow certainly
Body structure and perform characteristic, and the correlative factor of current cloud resource Environment communication and Executing Cost, based on genetic algorithm with
Machine two-point crossover operates and random single-point mutation operation thought, improves the diversity during Evolution of Population, integrates virtualization money
Source, consider data communication cost and task computation cost, optimize resource utilization, under the premise of the workflow deadline is met,
Reduce its Executing Cost;It is implemented as follows,
It is S=(Re, Map, T by the definition of workflow schedule methodtotal,Ctotal), wherein Re represents that one group of needs enables
Resources of virtual machine Re={ vm1,vm2,...,vmr, Map={ (ti,vmj)|ti∈V,vmj∈ Re } represent that workflow neutron is appointed
The corresponding resources of virtual machine Re of business mapping relations, TtotalThe execution deadline of expression workflow, and CtotalThen represent workflow
Total Executing Cost;Workflow is represented with directed acyclic graph G (V, E), and wherein V represents to include the vertex set of n task
{t1,t2,...,tn, and E then data dependence relation { e between expression task12,e13,...,eij};Per data dependence edge eij=
(ti,tj) represent subtask tiWith subtask tjBetween data dependence relation, wherein subtask t be presentiIt is subtask tjIt is direct
Pioneer's node, and subtask tjIt is then subtask tiImmediate successor node;
Every virtual machine has corresponding type of virtual machine spi, and corresponding start-up time Tls (vmi) and close moment
Tle(vmi);After the completion of scheduled when subtask, have and time started AST (t is actually performed corresponding to one groupi) and actual execution
Deadline AET (ti), and go out on missions and will not produce again and communication data;Therefore, workflow execution deadline TtotalWith it is right
The total Executing Cost C answeredtotalRespectively as shown by the following formula:
The first half of formula (2) represents the Executing Cost of virtual machine, and latter half represents data communication cost;λpIt is cloud
The service initialization that service provider p provides for it is specifically asked a price the unit time, subtask tkIt is subtask tjFollow-up section
Point, p (tj) and p (tk) represent to perform t respectivelyjAnd tkService provider;Work as tjAnd tkPerformed by same cloud service provider
When, sjkFor 0, i.e., data communicate between not producing cloud, otherwise sjkFor 1;
Based on above related definition, the workflow schedule problem of more cloud environment lower band deadline constraints, table can be formalized
Formula (3) is shown as, its core concept is to pursue Executing Cost CtotalWhile minimum, make execution time TtotalLess than or equal to work
Make stream deadline D (w);
Then, following algorithm is performed, to realize Executing Cost CtotalWhile minimum, make execution time TtotalLess than etc.
In workflow deadline D (w):
S1:It is as follows to initialize relevant parameter:Population Size 100, maximum iteration 1000, inertia weight factor w and recognize
Know factor parameter c1_ start=0.9, c1_ end=0.2, c2_ start=0.4, c2_ end=0.9, generate initial population;
S2:According to the fitness function of particle mapping policy, and particle, i.e. formula (4), (5) and (6) calculates each grain
Fitness value under sub- different situations, the minimum particle of fitness value is therefrom selected as population global optimum particle, by first
Each particle is arranged to its own history optimal particle in generation;
Wherein, it is feasible solution that formula (4), which represents a particle, and another particle is the particle in the case of infeasible solution
Fitness function, formula (5) represent the fitness function of the particle in the case that two particles are all feasible solution, formula (6) generation
Two particles of table be all infeasible solution in the case of particle fitness function;
S3:According to particle more new formula (7) to (10) more new particle;
Wherein, formula (7) represents particle i in the update mode of t, Mu() represents mutation operation, Cg() and Cp() table
Show crossover operation,And gBesttIt is particle i itself history optimal location and whole population after t iteration respectively
History optimum position;Formula (8) represents the update mode of inertia portion, r1It is the random number between 0 to 1;Formula (9),
(10) update mode of individual cognition and social recognition part, r are represented respectively2And r3It is the random number between 0 to 1;
S4:The fitness value of each particle is recalculated, if the fitness value of current particle is optimal less than its own history
Value, then be updated to its own history optimal particle by new particle;
S5:If the fitness value of current particle is less than the fitness value of population global optimum particle, the particle is updated
For population global optimum particle;
S6:Check whether and meet algorithm end condition, if it is satisfied, then algorithm terminates;Conversely, go to S3.
Fig. 1 is the workflow schedule frame diagram that the present invention defines.It mainly includes workflow, more cloud environments, and cost
Drive scheduler.
Workflow w is represented with directed acyclic graph G (V, E), and wherein V represents to include the vertex set { t of n task1,
t2,...,tn, and E then data dependence relation { e between expression task12,e13,...,eij}.Per data dependence edge eij=(ti,
tj) represent task tiWith task tjBetween data dependence relation, wherein task t be presentiIt is task tjImmediate predecessor (father) node,
And task tjIt is then task tiImmediate successor (son) node.During workflow schedule, a task must be in its all elder generation
After drive node is executed, the task could start to perform.In the directed acyclic graph of some given representative workflow
In, the task of no pioneer's node is referred to as ' entering task ', similarly, the task of no descendant node call ' gone out on missions '.Often
Individual workflow w has a corresponding deadline D (w), when some dispatching method can perform before the corresponding deadline
Into the workflow, then a kind of feasible solution is called.
Multiple cloud service provider P={ p, q ..., r } in more cloud environments be present, service provider p provides a variety of virtual
Machine example types Sp={ sp1,sp2,...,spk}.Each type of virtual machine instance has its specific computing capability and storage
Ability, present invention assumes that virtual machine has enough memory spaces to store transmission data, therefore this during subtasking
Text is primarily upon virtual machine computing capability (i.e. CPUs quantity).Subtask tiIn virtual machine vmpijOn estimation perform the time be
Exe_T(ti,vmpij), execution cost performance of the Given task on different type virtual machine is different.
Virtual machine is in initial start-up, it is necessary to which certain initialization starts time Tboot(vmpij) match somebody with somebody to carry out initialization
Put.During workflow schedule, this virtual machine initialization time should be paid attention to, because it is to workflow schedule scheme
Formation can produce significant impact.Likewise, after the completion of being performed when all subtasks on virtual machine, corresponding virtual machine is not
It is to immediately close off, but waits until that all subtasks are the output data full communication of itself to its younger generation's task pair on virtual machine
Untill on the virtual machine answered.Under more cloud environments, service initialization that cloud service provider p provides for it specifically ask a price unit when
Between λpEvery kind of type of virtual machine spiThere is corresponding unit interval price cpi。
The infrastructure of same cloud service provider is generally all concentrated in smaller area, and different cloud service providers
Infrastructure it is then apart from each other, it is therefore assumed that bandwidth will be faster than bandwidth between the cloud between different clouds in the cloud of single cloud.Cloud service
In provider p, data are from subtask tiIt is transferred to subtask tjCloud in call duration time be Tintra(eij, p), and data take in cloud
The call duration time being engaged between provider p and q is Tinter(eij, p, q), as shown by the following formula.
Subtask tiIt is transferred to subtask tjData volume size be Data (eij), Bintra(p) be cloud p cloud in bandwidth,
And Binter(p, q) is then bandwidth between cloud between cloud p and cloud q.Assuming that the bandwidth in single virtual machine is infinitely great, therefore when two
Subtask is assigned to when being performed on same virtual machine, Tintra(eij, p) value be 0.
Data communication cost between different clouds will influence whether final scheduling decision, cp,qRepresent from cloud p communications 1GB
Data volume to cloud q needed for unit price.The present invention does not consider generation caused by the business such as monitoring resource, data storage and load balancing
Valency, because these can be ignored compared with calculation cost or data communication cost at a low price.
The purpose of scheduler is on the premise of deadline constraint is met, minimizes workflow execution cost, the execution
Cost includes the data communication cost between the calculation cost of virtual machine and subtask.The definition of whole scheduling scheme for S=(Re,
Map,Ttotal,Ctotal), wherein Re represents the resources of virtual machine Re={ vm that one group of needs enables1,vm2,...,vmr, Map=
{(ti,vmj)|ti∈V,vmj∈ Re } represent that subtask corresponds to resources of virtual machine Re mapping relations, T in workflowtotalRepresent
The execution deadline of workflow, and CtotalThen represent the total Executing Cost of workflow.Every virtual machine has corresponding virtual machine
Type spi, and corresponding start-up time Tls (vmi) and close moment Tle (vmi).After the completion of scheduled when subtask, have
Time started AST (t is actually performed corresponding to one groupi) and actual execution deadline AET (ti), and go out on missions and will not produce again
And communication data.Therefore, workflow execution deadline TtotalWith corresponding total Executing Cost CtotalRespectively such as below equation institute
Show.
The first half of formula (4) represents the Executing Cost of virtual machine, and latter half represents data communication cost.λpIt is cloud
The service initialization that service provider p provides for it is specifically asked a price the unit time, subtask tkIt is subtask tjFollow-up section
Point, p (tj) and p (tk) represent to perform t respectivelyjAnd tkService provider.Work as tjAnd tkPerformed by same cloud service provider
When, sjkFor 0 (data communicate between not producing cloud), otherwise sjkFor 1.
Based on above related definition, the workflow schedule problem of more cloud environment lower band deadline constraints, table can be formalized
Formula (5) is shown as, its core concept is to pursue Executing Cost CtotalWhile minimum, make execution time TtotalLess than or equal to work
Make stream deadline D (w).
PSO algorithms are a kind of zoogeny computing techniques based on flock of birds social action, it be nineteen ninety-five by
Eberhart and Kennedy are proposed jointly.Particle is particularly important in PSO algorithms, the candidate of each particle representing optimized problem
Solution, they can be moved in the range of whole problem space.Each particle with certain speed mobile update moving direction of oneself,
The speed is by this tripartite's face of the history optimum position of particle own situation, the optimal historical position of particle itself and whole population
Ring.In order to judge the superiority-inferiority solved produced by diverse location of each particle in problem space, fitness function is introduced to comment
Estimate the solution quality of each particle.Each particle is to determine that they are according to ambient particles and itself by the position and speed of its own
Experience in problem search space the position of continuous iteration renewal adjustment oneself and speed.Its medium velocity is according to formula (6)
It is updated, position is updated according to formula (7).
Wherein, t represents current iterations,WithRepresent respectively speed of i-th of particle in the t times iteration and
Position, it usually needs define a maximal rate VmaxTo limit particle rapidity, make search result in problem solution space.
And gBesttIt is particle i itself history optimal location and the history optimum position of whole population after t iteration respectively.w
It is inertia weight, influence size of the iteration speed to present speed before it is determined is most important to convergence.c1
And c2It is perception factor, they embody cognitive learning of the current particle to itself history optimal value and population global history optimal value
Ability.r1And r2It is two stochastic variables of the scope between 0 to 1, for strengthening the randomness in iterative search procedures.
Method provided by the invention is mainly used in Fig. 1 scheduler.The inventive method mainly includes representation, initial
Change the parts such as resource pool, fitness function, particle more new strategy, the mapping of particle to scheduling result and parameter setting, by with
Lower content is specifically discussed.
Representation
Algorithm search efficiency and performance are improved, it is necessary to a kind of good coded system.The evaluation criterion of coding strategy is mainly examined
Consider three its viability, completeness and nonredundancy basic principles.The present invention uses cloud provider-example types-instantiation
Nested mode coding work stream scheduling problem.One particle represents a scheduling scheme of workflow under more cloud environments, particle i
In the position of tAs shown in formula (8).
WhereinDispensing position of k-th of subtask in t is represented, as shown in formula (9).(p,spj,
vmpjr) to represent that the subtask is assigned to example types in cloud p be spjR-th of instantiation on.Each section on particle
Point position is nested being divided into 3 small minute positions, represents cloud service provider, example types and instantiation respectively, therefore encodes empty
Between size be 3 times of subtask quantity.When initializing population, each random initializtion is 0 right to its for small point of position of node of particle
Answer the integer value between maximum.Fig. 2 is particle code pattern, and displaying scheduling includes the particle coding plan of 8 subtask workflows
Slightly, wherein assuming that more cloud environments include 3 cloud service providers, and each cloud service provider is provided which 8 kinds of example types.Cause
This, p coordinate values are the s from 0 to 2pjCoordinate value is from 0 to 7.As can be seen from Figure 2, subtask t1It is s to be assigned to type in cloud 000's
Virtual machine vm000。
Initialize resource pool
The elastic supply pattern of cloudy environmental resource, causes algorithm can not obtain initial resource set.For PSO algorithms
Speech, the size for initializing resource pool will determine the scope of search space, algorithm complex and workflow execution performance played
Key effect.When initialization resource pool is too small, it may appear that the workflow that originally can be completed before the deadline, due to resource
Lack and can not complete in time.When initialize resource pool it is too big when, PSO coding potential solution it is excessively huge, make algorithm can not and
When restrain.A kind of initialization Resource Allocation Formula of simple possible, it is to distribute all categories in more cloud environments for each subtask
Each one of virtual machine, can so ensure the diversity and integrality of search space.However, the initialization resource of this scheme
Pond RintialSize be n*Numtype(vm), search space is bigger, increases algorithm complex.
Wherein n be workflow w in subtask quantity, Numtype(vm)For the example types quantity of all cloud service providers
Summation, it defines such as formula (10).Numvm(p)The example types quantity provided by cloud service provider p.
For further compression search space, while keep the diversity of original potential solution, present invention design initialization resource
Pond RintialSize be | Spar(w)|*Numtype(vm), wherein Spar(w)It is that maximum in workflow w can parallel subtasks set.By
In except Spar(w)In subtask, other subtasks all can and Spar(w)In subtask dependence directly or indirectly be present and close
System, so the initialization resource policy, which can ensure that each subtask has, selects a kind of chance of type instance, so as to ensure
The diversity of potential solution, while reduce search space.
Fitness function
The fitness function of particle is the superiority-inferiority for evaluating two particles that compare, usual less fitness function
Particle is more excellent corresponding to value.Because the particle coding strategy of early stage is unsatisfactory for viability principle, that is, the execution of workflow occurs
Time exceedes the corresponding deadline, so needing to distinguish to feasible solution and more than the fitness function of the infeasible solution of deadline
Definition.The present invention judges two particles good and bad fitness function point, three kinds of different situations definition.
1 one particles of situation are feasible solutions, and another particle is infeasible solution.Feasible solution is selected free from controvery, and it is suitable
The definition of response function is as shown in formula (11).
2 two particles of situation are all feasible solutions.The relatively low particle of Executing Cost is selected, its fitness function is defined as follows:
3 two particles of situation are all infeasible solutions.Selection performs time less particle, because more having after the particle evolution
Feasible solution may be changed into.The definition of its fitness function is as shown in formula (13).
Particle more new strategy
As shown in formula (6), PSO includes three cores:Inertia portion, individual cognition part and social recognition portion
Point.To overcome Premature Convergence defect existing for traditional PS O algorithms, ADPSOGA algorithms introduce the variation of genetic algorithm and intersect behaviour
Make, operation is updated to appropriate section in formula (6).Particle i is shown in the update mode such as formula (14) of t, wherein
Mu() represents mutation operation, Cg() and Cp() represents crossover operation.
Mutation operation thought in inertia portion combination genetic algorithm in formula (6), the update mode such as public affairs of inertia portion
Shown in formula (15), wherein r1It is the random number between 0 to 1.Mu() randomly selects a point of position in particle, irregular to change it
Tantile, and new value must be all in corresponding threshold value.Fig. 3 shows the mutation operation to Fig. 2 encoded particles, randomly chooses particle
One point of position mp1, the value on mp1 positions are updated to (1,2,0) by (0,1,2), and the variation meets dispatching criterion.
Individual cognition and social recognition part in formula (6) combine crossover operation of genetic algorithms thought, and it updates result
Respectively as shown in formula (16) and (17).r2And r3It is the random number between 0 to 1, Cp() (or Cg()) random selection particle
Two points of positions, the numerical value between the numerical value between point position and corresponding pBest (or gBest) point position is intersected.Fig. 4 displayings
The crossover operation of people (or society) cognition part, randomly generates two crossover locations (i.e. cp1 and cp2), by particle cp1 and cp2
Value between position replaces with values of the pBest (or gBest) in the section.
Mapping of the particle to scheduling result
Design is mapped to the false code of workflow schedule process from encoded particles.The algorithm input include workflow w, initially
Change resource pool RintialWith encoded particles X.First, to scheduling scheme S=(Re, Map, Ttotal,Ctotal) four elements initialization.
After initialization, calculate each subtask correspond to different type example estimation perform time matrix Exe_T [| w | × | Rintial|],
Element Exe_T [i] [j] in matrix represents subtask tiIn virtual machine mvjOn estimation perform the time.Between calculating subtask
Data volume single cloud and it is cloudy between estimation call duration time, Tintra[i] [j] represents subtask t in single cloudiCaused data
Subtask t is arrived in amount communicationjRequired estimation time, Tinter[i] [j] [p] [q] represents subtask tiCaused data volume is from cloud p
Communicate to cloud q subtask tjThe required estimation time.
Operated more than, obtained obtaining the full detail of candidate solution from encoded particles at present.Progressively scan particle X's
Each point of position, Re corresponding to generation and Map set.Based on ' representation ', the coding of particle divides position to correspond to subtask, divides position
It is worth corresponding instance resource, it is thus determined that subtask tiIt is assigned to example rX(i)On.Need to calculate subtask tiEstimation when starting
Between STti, here in two kinds of situation:
A) subtask tiIt is truly to enter task, i.e., it does not have immediate predecessor subtask.As virtual machine rX(i)When available, son is appointed
Be engaged in tiExecution is got started, it estimates time started STtiFor virtual machine rX(i)Lease time LETrX(i).In addition, it is necessary to sentence
Disconnected virtual machine rX(i)Whether have turned on, if do not opened, need to start virtual machine, the lease time LETr of virtual machineX(i)I.e.
For the initialization time T of virtual machineboot(rX(i))。
B) subtask tiIt is not into task, i.e., it has one or more father's tasks.Subtask tiNot only need to wait resource space
Idle just be can perform, and its all father's tasks carrying must be waited to complete, and will be produced data communication and be arrived virtual machine rX(i)It is upper
Perform.Call and calculate subtask tiStand-by period and the false code of data communication cost, while consider virtual machine rX(i)Whether by
Start.
Subtask t is calculatediEstimation time started STti, it is necessary to according to its estimation on a virtual machine perform time and
Data communication time, to calculate subtask tiEstimation end time ETti.Calculating for data communication time, it is necessary to according to
Thereafter lifetime task tcWhether with subtask tiDistribution determines there are three kinds of situations here in same cloud:
a)tcWith tiPerformed on same virtual machine, then call duration time transfer is 0.
b)tcWith tiPerformed in same cloud but on different virtual machine, then call duration time transfer is Tintra[i]
[c]。
c)tcWith ti(performed respectively in different clouds on such as cloud p and cloud q), then call duration time transfer is Tinter[ip]
[cq]。
Subtask tiIt is dispatched to virtual machine rX(i), its time started STtiWith end time ETtiMapping Deng four elements is closed
System is added in Map set.Subsequently determine whether virtual machine rX(i)Whether it is added in lease resource Re, if be not added,
Then it is added.Virtual machine rX(i)Newest lease time, equal to subtask tiEstimated time to completion.Finally, according to formula
(3) and formula (4) distinguishes total execution time of calculation workflow and total Executing Cost.Scheduling scheme corresponding to exports coding particle
S。
Design calculates subtask tiStand-by period and communication cost false code.First, stand-by period T is initializedwaitWith
Data communication cost Ctranfer.Subtask tiStand-by period is equal to data communication time maximum in its all father's task.Group is appointed
Be engaged in tiWith his father's task tpWhen being assigned in different clouds, data communication cost is just considered.
Parameter setting
The inertia weight factor w of formula (6) can determine PSO convergences and search capability.When w is smaller, calculate
Method has stronger local search ability;Otherwise, algorithm has stronger ability of searching optimum.At algorithm performs initial stages, more focus on
The diversity and particle ability of searching optimum of problem space search, with search deeply, stage more focuses on Local Search side
The ability in face.Therefore, inertia weight factor w weights should increasing and gradually decrease with algorithm iteration number.Formula (18)
It is classical inertia weight factor adjustable strategies.Wherein, wmaxAnd wminIt is the maximum and minimum set during w initialization respectively
Value, iterscurAnd itersmaxThe maximum iteration of current algorithm iterations and initializing set is then represented respectively.
Classical inertia weight factor adjustable strategies above, w change are only relevant with iterations, it is impossible to meet very well real
The characteristic non-linear, complicated and changeable of border problem.Inertia weight factor w weights size should with population particle evolution without
Inertia weight factor adjustable strategies that are disconnected to develop, therefore building a kind of the good and bad of current population particle of basis and adaptively adjust.
As shown in formula (19), the strategy adjusts inertia power based on the difference degree between current particle and global history optimal particle
Repeated factor size.Wherein div (Xt-1,gBestt-1) represent particle Xt-1With global history optimal particle gBestt-1Between difference
Divide the digit size of position, T is the number size of subtask in workflow.
As div (Xt-1) value it is smaller when, represent particle Xt-1And gBestt-1Between difference degree it is smaller, so w should be reduced
Weights, to ensure that particle can more preferably be searched in a small range, find optimization solution;Otherwise, it should increase w weights, make particle
Search space become big, to quickly find optimization solution space.Therefore, inertia weight factor w weight computing formula updates such as
Under:
In addition, two perception factor c of algorithm1And c2It is configured using linear increase and decrease strategy.Its update mode such as formula
(21) and shown in formula (22), wherein c1_ start and c2_ start represents parameter c respectively1And c2The initial value of iteration, c1_end
And c2_ end represents parameter c respectively1And c2The end value of iteration.
Fig. 5 is the algorithm flow chart of dispatching method of the present invention, and its concrete operation step is as follows:
Step 1:The relevant parameter initialized in dispatching method is as follows:Population Size 100, maximum iteration 1000, it is used to
The property parameter such as weight factor and perception factor c1_ start=0.9, c1_ end=0.2, c2_ start=0.4, c2_ end=0.9,
Generate initial population.
Step 2:Calculated according to particle mapping policy, and formula (11), (12) and (13) under each particle different situations
Fitness value, the minimum particle of fitness value is therefrom selected as population global optimum particle, by each particle in the first generation
It is arranged to its own history optimal particle.
Step 3:According to particle more new formula (14) to (17) more new particle.
Step 4:The fitness value of each particle is recalculated, if the fitness value of current particle is less than its own history most
The figure of merit, then new particle is updated to its own history optimal particle.
Step 5:If the fitness value of current particle is less than the fitness value of population global optimum particle, by the particle more
New is population global optimum particle.
Step 6:Check whether and meet algorithm end condition, if it is satisfied, then algorithm terminates;Conversely, go to step 3.
Above is presently preferred embodiments of the present invention, all changes made according to technical solution of the present invention, caused function are made
During with scope without departing from technical solution of the present invention, protection scope of the present invention is belonged to.
Claims (2)
1. the workflow schedule method of communication and calculation cost is considered under a kind of more cloud environments, it is characterised in that:Based on workflow
Self structure and execution characteristic, and the correlative factor of current cloud resource Environment communication and Executing Cost, based on genetic algorithm
Random two-point crossover operation and random single-point mutation operation thought, improve the diversity during Evolution of Population, integrate virtualization
Resource, consider data communication cost and task computation cost, optimize resource utilization, meeting workflow deadline premise
Under, reduce its Executing Cost.
2. the workflow schedule method of communication and calculation cost is considered under a kind of more cloud environments according to claim 1, its
It is characterised by:The workflow schedule method is implemented as follows,
It is S=(Re, Map, T by the definition of workflow schedule methodtotal,Ctotal), wherein Re represents the void that one group of needs enables
Plan machine resource Re={ vm1,vm2,...,vmr, Map={ (ti,vmj)|ti∈V,vmj∈ Re } represent subtask pair in workflow
Answer resources of virtual machine Re mapping relations, TtotalThe execution deadline of expression workflow, and CtotalThen represent that workflow is always held
Row cost;Workflow is represented with directed acyclic graph G (V, E), and wherein V represents to include the vertex set { t of n task1,
t2,...,tn, and E then data dependence relation { e between expression task12,e13,...,eij};Per data dependence edge eij=(ti,
tj) represent subtask tiWith subtask tjBetween data dependence relation, wherein subtask t be presentiIt is subtask tjImmediate predecessor
Node, and subtask tjIt is then subtask tiImmediate successor node;
Every virtual machine has corresponding type of virtual machine spi, and corresponding start-up time Tls (vmi) and close moment Tle
(vmi);After the completion of scheduled when subtask, have and time started AST (t is actually performed corresponding to one groupi) and actually perform completion
Time AET (ti), and go out on missions and will not produce again and communication data;Therefore, workflow execution deadline TtotalWith it is corresponding
Total Executing Cost CtotalRespectively as shown by the following formula:
<mrow>
<mi>T</mi>
<mi>t</mi>
<mi>o</mi>
<mi>t</mi>
<mi>a</mi>
<mi>l</mi>
<mo>=</mo>
<munder>
<mrow>
<mi>m</mi>
<mi>a</mi>
<mi>x</mi>
</mrow>
<mrow>
<msub>
<mi>t</mi>
<mi>i</mi>
</msub>
<mo>&Element;</mo>
<mi>W</mi>
</mrow>
</munder>
<mo>{</mo>
<mi>A</mi>
<mi>E</mi>
<mi>T</mi>
<mrow>
<mo>(</mo>
<msub>
<mi>t</mi>
<mi>i</mi>
</msub>
<mo>)</mo>
</mrow>
<mo>}</mo>
<mo>-</mo>
<mo>-</mo>
<mo>-</mo>
<mrow>
<mo>(</mo>
<mn>1</mn>
<mo>)</mo>
</mrow>
</mrow>
The first half of formula (2) represents the Executing Cost of virtual machine, and latter half represents data communication cost;λpIt is cloud service
The service initialization that provider p provides for it is specifically asked a price the unit time, subtask tkIt is subtask tjDescendant node, p
(tj) and p (tk) represent to perform t respectivelyjAnd tkService provider;Work as tjAnd tkWhen being performed by same cloud service provider, sjk
For 0, i.e., data communicate between not producing cloud, otherwise sjkFor 1;
Based on above related definition, the workflow schedule problem of more cloud environment lower band deadlines constraint, can formalization representation be
Formula (3), its core concept are to pursue Executing Cost CtotalWhile minimum, make execution time TtotalLess than or equal to workflow
Deadline D (w);
<mrow>
<mtable>
<mtr>
<mtd>
<mrow>
<mi>M</mi>
<mi>i</mi>
<mi>n</mi>
<mi>i</mi>
<mi>m</mi>
<mi>i</mi>
<mi>z</mi>
<mi>e</mi>
<mi> </mi>
<msub>
<mi>C</mi>
<mrow>
<mi>t</mi>
<mi>o</mi>
<mi>t</mi>
<mi>a</mi>
<mi>l</mi>
</mrow>
</msub>
</mrow>
</mtd>
</mtr>
<mtr>
<mtd>
<mrow>
<mi>s</mi>
<mi>u</mi>
<mi>b</mi>
<mi>j</mi>
<mi>e</mi>
<mi>c</mi>
<mi>t</mi>
<mi> </mi>
<mi>t</mi>
<mi>o</mi>
<mi> </mi>
<msub>
<mi>T</mi>
<mrow>
<mi>t</mi>
<mi>o</mi>
<mi>t</mi>
<mi>a</mi>
<mi>l</mi>
</mrow>
</msub>
<mo>&le;</mo>
<mi>D</mi>
<mrow>
<mo>(</mo>
<mi>w</mi>
<mo>)</mo>
</mrow>
</mrow>
</mtd>
</mtr>
</mtable>
<mo>-</mo>
<mo>-</mo>
<mo>-</mo>
<mrow>
<mo>(</mo>
<mn>3</mn>
<mo>)</mo>
</mrow>
</mrow>
Then, following algorithm is performed, to realize Executing Cost CtotalWhile minimum, make execution time TtotalLess than or equal to work
Flow deadline D (w):
S1:It is as follows to initialize relevant parameter:Population Size 100, maximum iteration 1000, inertia weight factor w and cognition because
Subparameter c1_ start=0.9, c1_ end=0.2, c2_ start=0.4, c2_ end=0.9, generate initial population;
S2:According to the fitness function of particle mapping policy, and particle, i.e. formula (4), (5) and (6) calculates each particle not
Fitness value with the case of, the minimum particle of fitness value is therefrom selected as population global optimum particle, by the first generation
Each particle is arranged to its own history optimal particle;
<mrow>
<mi>f</mi>
<mi>i</mi>
<mi>t</mi>
<mi>n</mi>
<mi>e</mi>
<mi>s</mi>
<mi>s</mi>
<mo>=</mo>
<msub>
<mi>C</mi>
<mrow>
<mi>t</mi>
<mi>o</mi>
<mi>t</mi>
<mi>a</mi>
<mi>l</mi>
<mrow>
<mo>(</mo>
<msub>
<mi>X</mi>
<mi>i</mi>
</msub>
<mo>)</mo>
</mrow>
</mrow>
</msub>
<mo>-</mo>
<mo>-</mo>
<mo>-</mo>
<mrow>
<mo>(</mo>
<mn>5</mn>
<mo>)</mo>
</mrow>
</mrow>
<mrow>
<mi>f</mi>
<mi>i</mi>
<mi>t</mi>
<mi>n</mi>
<mi>e</mi>
<mi>s</mi>
<mi>s</mi>
<mo>=</mo>
<msub>
<mi>T</mi>
<mrow>
<mi>t</mi>
<mi>o</mi>
<mi>t</mi>
<mi>a</mi>
<mi>l</mi>
<mrow>
<mo>(</mo>
<msub>
<mi>X</mi>
<mi>i</mi>
</msub>
<mo>)</mo>
</mrow>
</mrow>
</msub>
<mo>-</mo>
<mo>-</mo>
<mo>-</mo>
<mrow>
<mo>(</mo>
<mn>6</mn>
<mo>)</mo>
</mrow>
</mrow>
Wherein, it is feasible solution that formula (4), which represents a particle, another particle be infeasible solution in the case of particle adaptation
Function is spent, formula (5) represents the fitness function of the particle in the case that two particles are all feasible solution, and formula (6) represents two
Individual particle be all infeasible solution in the case of particle fitness function;
S3:According to particle more new formula (7) to (10) more new particle;
<mrow>
<msubsup>
<mi>X</mi>
<mi>i</mi>
<mi>t</mi>
</msubsup>
<mo>=</mo>
<msub>
<mi>c</mi>
<mn>2</mn>
</msub>
<mo>&CirclePlus;</mo>
<msub>
<mi>C</mi>
<mi>g</mi>
</msub>
<mrow>
<mo>(</mo>
<msub>
<mi>c</mi>
<mn>1</mn>
</msub>
<mo>&CirclePlus;</mo>
<msub>
<mi>C</mi>
<mi>p</mi>
</msub>
<mo>(</mo>
<mrow>
<mi>w</mi>
<mo>&CirclePlus;</mo>
<msub>
<mi>M</mi>
<mi>u</mi>
</msub>
<mrow>
<mo>(</mo>
<msubsup>
<mi>X</mi>
<mi>i</mi>
<mrow>
<mi>t</mi>
<mo>-</mo>
<mn>1</mn>
</mrow>
</msubsup>
<mo>)</mo>
</mrow>
<mo>,</mo>
<msubsup>
<mi>pBest</mi>
<mi>i</mi>
<mrow>
<mi>t</mi>
<mo>-</mo>
<mn>1</mn>
</mrow>
</msubsup>
</mrow>
<mo>)</mo>
<mo>,</mo>
<msup>
<mi>gBest</mi>
<mrow>
<mi>t</mi>
<mo>-</mo>
<mn>1</mn>
</mrow>
</msup>
<mo>)</mo>
</mrow>
<mo>-</mo>
<mo>-</mo>
<mo>-</mo>
<mrow>
<mo>(</mo>
<mn>7</mn>
<mo>)</mo>
</mrow>
</mrow>
<mrow>
<msubsup>
<mi>A</mi>
<mi>i</mi>
<mi>t</mi>
</msubsup>
<mo>=</mo>
<mi>w</mi>
<mo>&CirclePlus;</mo>
<msub>
<mi>M</mi>
<mi>u</mi>
</msub>
<mrow>
<mo>(</mo>
<msubsup>
<mi>X</mi>
<mi>i</mi>
<mrow>
<mi>t</mi>
<mo>-</mo>
<mn>1</mn>
</mrow>
</msubsup>
<mo>)</mo>
</mrow>
<mo>=</mo>
<mfenced open = "{" close = "">
<mtable>
<mtr>
<mtd>
<mrow>
<msub>
<mi>M</mi>
<mi>u</mi>
</msub>
<mrow>
<mo>(</mo>
<msubsup>
<mi>X</mi>
<mi>i</mi>
<mrow>
<mi>t</mi>
<mo>-</mo>
<mn>1</mn>
</mrow>
</msubsup>
<mo>)</mo>
</mrow>
</mrow>
</mtd>
<mtd>
<mrow>
<msub>
<mi>r</mi>
<mn>1</mn>
</msub>
<mo><</mo>
<mi>w</mi>
</mrow>
</mtd>
</mtr>
<mtr>
<mtd>
<msubsup>
<mi>X</mi>
<mi>i</mi>
<mrow>
<mi>t</mi>
<mo>-</mo>
<mn>1</mn>
</mrow>
</msubsup>
</mtd>
<mtd>
<mrow>
<mi>e</mi>
<mi>l</mi>
<mi>s</mi>
<mi>e</mi>
</mrow>
</mtd>
</mtr>
</mtable>
</mfenced>
<mo>-</mo>
<mo>-</mo>
<mo>-</mo>
<mrow>
<mo>(</mo>
<mn>8</mn>
<mo>)</mo>
</mrow>
</mrow>
<mrow>
<msubsup>
<mi>B</mi>
<mi>i</mi>
<mi>t</mi>
</msubsup>
<mo>=</mo>
<msub>
<mi>c</mi>
<mn>1</mn>
</msub>
<mo>&CirclePlus;</mo>
<msub>
<mi>C</mi>
<mi>p</mi>
</msub>
<mrow>
<mo>(</mo>
<msubsup>
<mi>A</mi>
<mi>i</mi>
<mi>t</mi>
</msubsup>
<mo>,</mo>
<msup>
<mi>pBest</mi>
<mrow>
<mi>t</mi>
<mo>-</mo>
<mn>1</mn>
</mrow>
</msup>
<mo>)</mo>
</mrow>
<mo>=</mo>
<mfenced open = "{" close = "">
<mtable>
<mtr>
<mtd>
<mrow>
<msub>
<mi>C</mi>
<mi>p</mi>
</msub>
<mrow>
<mo>(</mo>
<msubsup>
<mi>A</mi>
<mi>i</mi>
<mi>t</mi>
</msubsup>
<mo>,</mo>
<msup>
<mi>pBest</mi>
<mrow>
<mi>t</mi>
<mo>-</mo>
<mn>1</mn>
</mrow>
</msup>
<mo>)</mo>
</mrow>
</mrow>
</mtd>
<mtd>
<mrow>
<msub>
<mi>r</mi>
<mn>2</mn>
</msub>
<mo><</mo>
<msub>
<mi>c</mi>
<mn>1</mn>
</msub>
</mrow>
</mtd>
</mtr>
<mtr>
<mtd>
<msubsup>
<mi>A</mi>
<mi>i</mi>
<mi>t</mi>
</msubsup>
</mtd>
<mtd>
<mrow>
<mi>e</mi>
<mi>l</mi>
<mi>s</mi>
<mi>e</mi>
</mrow>
</mtd>
</mtr>
</mtable>
</mfenced>
<mo>-</mo>
<mo>-</mo>
<mo>-</mo>
<mrow>
<mo>(</mo>
<mn>9</mn>
<mo>)</mo>
</mrow>
</mrow>
<mrow>
<msubsup>
<mi>C</mi>
<mi>i</mi>
<mi>t</mi>
</msubsup>
<mo>=</mo>
<msub>
<mi>c</mi>
<mn>2</mn>
</msub>
<mo>&CirclePlus;</mo>
<msub>
<mi>C</mi>
<mi>g</mi>
</msub>
<mrow>
<mo>(</mo>
<msubsup>
<mi>B</mi>
<mi>i</mi>
<mi>t</mi>
</msubsup>
<mo>,</mo>
<msup>
<mi>gBest</mi>
<mrow>
<mi>t</mi>
<mo>-</mo>
<mn>1</mn>
</mrow>
</msup>
<mo>)</mo>
</mrow>
<mo>=</mo>
<mfenced open = "{" close = "">
<mtable>
<mtr>
<mtd>
<mrow>
<msub>
<mi>C</mi>
<mi>g</mi>
</msub>
<mrow>
<mo>(</mo>
<msubsup>
<mi>B</mi>
<mi>i</mi>
<mi>t</mi>
</msubsup>
<mo>,</mo>
<msup>
<mi>gBest</mi>
<mrow>
<mi>t</mi>
<mo>-</mo>
<mn>1</mn>
</mrow>
</msup>
<mo>)</mo>
</mrow>
</mrow>
</mtd>
<mtd>
<mrow>
<msub>
<mi>r</mi>
<mn>3</mn>
</msub>
<mo><</mo>
<msub>
<mi>c</mi>
<mn>2</mn>
</msub>
</mrow>
</mtd>
</mtr>
<mtr>
<mtd>
<msubsup>
<mi>B</mi>
<mi>i</mi>
<mi>t</mi>
</msubsup>
</mtd>
<mtd>
<mrow>
<mi>e</mi>
<mi>l</mi>
<mi>s</mi>
<mi>e</mi>
</mrow>
</mtd>
</mtr>
</mtable>
</mfenced>
<mo>-</mo>
<mo>-</mo>
<mo>-</mo>
<mrow>
<mo>(</mo>
<mn>10</mn>
<mo>)</mo>
</mrow>
</mrow>
Wherein, formula (7) represents particle i in the update mode of t, Mu() represents mutation operation, Cg() and Cp() represents to hand over
Fork operation,And gBesttIt is particle i itself history optimal location and the history of whole population after t iteration respectively
Optimum position;Formula (8) represents the update mode of inertia portion, r1It is the random number between 0 to 1;Formula (9), (10) are respectively
Represent the update mode of individual cognition and social recognition part, r2And r3It is the random number between 0 to 1;
S4:The fitness value of each particle is recalculated, if the fitness value of current particle is less than its own history optimal value,
New particle is updated to its own history optimal particle;
S5:If the fitness value of current particle is less than the fitness value of population global optimum particle, the particle is updated to plant
Group global optimum particle;
S6:Check whether and meet algorithm end condition, if it is satisfied, then algorithm terminates;Conversely, go to S3.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201711077227.6A CN107656799B (en) | 2017-11-06 | 2017-11-06 | Workflow scheduling method considering communication and computing cost under multi-cloud environment |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201711077227.6A CN107656799B (en) | 2017-11-06 | 2017-11-06 | Workflow scheduling method considering communication and computing cost under multi-cloud environment |
Publications (2)
Publication Number | Publication Date |
---|---|
CN107656799A true CN107656799A (en) | 2018-02-02 |
CN107656799B CN107656799B (en) | 2021-06-25 |
Family
ID=61119821
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201711077227.6A Active CN107656799B (en) | 2017-11-06 | 2017-11-06 | Workflow scheduling method considering communication and computing cost under multi-cloud environment |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN107656799B (en) |
Cited By (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108647771A (en) * | 2018-05-07 | 2018-10-12 | 福建师范大学 | The layout method of research-on-research flow data under a kind of mixing cloud environment |
CN110033076A (en) * | 2019-04-19 | 2019-07-19 | 福州大学 | Mix the Work stream data layout method below cloud environment to cost optimization |
CN110264097A (en) * | 2019-06-26 | 2019-09-20 | 哈尔滨理工大学 | More scientific workflows based on cloud environment concurrently execute dispatching method |
CN110519317A (en) * | 2018-05-21 | 2019-11-29 | 华为技术有限公司 | A kind of data transmission method and equipment |
CN112492032A (en) * | 2020-11-30 | 2021-03-12 | 杭州电子科技大学 | Workflow cooperative scheduling method under mobile edge environment |
CN114925935A (en) * | 2022-06-21 | 2022-08-19 | 福州大学 | Multi-workflow scheduling method for time delay constraint in cloud edge environment |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104484230A (en) * | 2014-12-31 | 2015-04-01 | 中国科学院遥感与数字地球研究所 | Multiple satellite data centre workflow scheduling algorithm on basis of near data calculation principle |
CN105068863A (en) * | 2015-07-16 | 2015-11-18 | 福州大学 | Cost-driven scheduling method for workflow with deadline constraints in cloudy environment |
CN106648831A (en) * | 2016-12-08 | 2017-05-10 | 杭州电子科技大学 | Cloud workflow scheduling method based on firefly algorithm and dynamic priority algorithm |
-
2017
- 2017-11-06 CN CN201711077227.6A patent/CN107656799B/en active Active
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104484230A (en) * | 2014-12-31 | 2015-04-01 | 中国科学院遥感与数字地球研究所 | Multiple satellite data centre workflow scheduling algorithm on basis of near data calculation principle |
CN105068863A (en) * | 2015-07-16 | 2015-11-18 | 福州大学 | Cost-driven scheduling method for workflow with deadline constraints in cloudy environment |
CN106648831A (en) * | 2016-12-08 | 2017-05-10 | 杭州电子科技大学 | Cloud workflow scheduling method based on firefly algorithm and dynamic priority algorithm |
Non-Patent Citations (4)
Title |
---|
MACIEJ MALAWSKI等: "Cost- and Deadline-Constrained Provisioning for", 《SC12: PROCEEDINGS OF THE INTERNATIONAL CONFERENCE ON HIGH PERFORMANCE COMPUTING, NETWORKING, STORAGE AND ANALYSIS》 * |
SOBHAN OMRANIAN-KHORASANI等: "Deadline Constrained Load Balancing Level Based Workflow Scheduling for Cost", 《2017 2ND IEEE INTERNATIONAL CONFERENCE ON COMPUTATIONAL INTELLIGENCE AND APPLICATIONS》 * |
林兵等: "多云环境下基于代价驱动的科学工作流调度策略", 《模式识别与人工智能》 * |
郭文彩: "基于遗传算法的网格服务工作流调度的研究", 《计算机应用》 * |
Cited By (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108647771A (en) * | 2018-05-07 | 2018-10-12 | 福建师范大学 | The layout method of research-on-research flow data under a kind of mixing cloud environment |
CN110519317A (en) * | 2018-05-21 | 2019-11-29 | 华为技术有限公司 | A kind of data transmission method and equipment |
CN110519317B (en) * | 2018-05-21 | 2021-02-12 | 华为技术有限公司 | Data transmission method and equipment |
CN110033076A (en) * | 2019-04-19 | 2019-07-19 | 福州大学 | Mix the Work stream data layout method below cloud environment to cost optimization |
CN110033076B (en) * | 2019-04-19 | 2022-08-05 | 福州大学 | Workflow data layout method for cost optimization in mixed cloud environment |
CN110264097A (en) * | 2019-06-26 | 2019-09-20 | 哈尔滨理工大学 | More scientific workflows based on cloud environment concurrently execute dispatching method |
CN112492032A (en) * | 2020-11-30 | 2021-03-12 | 杭州电子科技大学 | Workflow cooperative scheduling method under mobile edge environment |
CN114925935A (en) * | 2022-06-21 | 2022-08-19 | 福州大学 | Multi-workflow scheduling method for time delay constraint in cloud edge environment |
Also Published As
Publication number | Publication date |
---|---|
CN107656799B (en) | 2021-06-25 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN107656799A (en) | The workflow schedule method of communication and calculation cost is considered under a kind of more cloud environments | |
Mansouri et al. | Hybrid task scheduling strategy for cloud computing by modified particle swarm optimization and fuzzy theory | |
Wang et al. | A heuristic genetic algorithm for subcontractor selection in a global manufacturing environment | |
Trivedi et al. | Hybridizing genetic algorithm with differential evolution for solving the unit commitment scheduling problem | |
Zhou et al. | A review of methods and algorithms for optimizing construction scheduling | |
CN108880663A (en) | Incorporate network resource allocation method based on improved adaptive GA-IAGA | |
CN101237469B (en) | Method for optimizing multi-QoS grid workflow based on ant group algorithm | |
CN110737529A (en) | cluster scheduling adaptive configuration method for short-time multiple variable-size data jobs | |
Peng et al. | Deep reinforcement learning approach for capacitated supply chain optimization under demand uncertainty | |
Kanakasabapathy et al. | Evolutionary tristate PSO for strategic bidding of pumped-storage hydroelectric plant | |
CN104065745A (en) | Cloud computing dynamic resource scheduling system and method | |
CN106055395A (en) | Method for constraining workflow scheduling in cloud environment based on ant colony optimization algorithm through deadline | |
CN105550746A (en) | Training method and training device of machine learning model | |
CN104516785B (en) | A kind of cloud computing resources dispatch system and method | |
Zhao et al. | QoS-aware web service selection with negative selection algorithm | |
CN110517002A (en) | Production control method based on intensified learning | |
CN106357739A (en) | Two-stage composition and scheduling method specific to lot-sizing cloud service request | |
Wang et al. | QoS-aware service selection using an incentive mechanism | |
Heidari et al. | A QoS-aware technique for computation offloading in IoT-edge platforms using a convolutional neural network and Markov decision process | |
CN109523178A (en) | A kind of O&M method and device towards power communication scene | |
CN109684088A (en) | The remote sensing big data of cloud platform resource constraint quickly handles method for scheduling task | |
CN109710372A (en) | A kind of computation-intensive cloud workflow schedule method based on cat owl searching algorithm | |
Yu et al. | Efficient scheduling in crowdsourcing based on workers' mood | |
Balla et al. | Reliability-aware: task scheduling in cloud computing using multi-agent reinforcement learning algorithm and neural fitted Q. | |
CN110008023A (en) | Cloud computing system budgetary restraints Random Task dispatching method based on genetic algorithm |
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 |