CN103902374B - Cellular automation and empowerment directed hypergraph based cloud-computing task scheduling method - Google Patents
Cellular automation and empowerment directed hypergraph based cloud-computing task scheduling method Download PDFInfo
- Publication number
- CN103902374B CN103902374B CN201410137810.1A CN201410137810A CN103902374B CN 103902374 B CN103902374 B CN 103902374B CN 201410137810 A CN201410137810 A CN 201410137810A CN 103902374 B CN103902374 B CN 103902374B
- Authority
- CN
- China
- Prior art keywords
- tax
- cellular
- power
- node
- directed hypergraph
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Expired - Fee Related
Links
Landscapes
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
- Mobile Radio Communication Systems (AREA)
Abstract
The invention relates to a cellular automation and empowerment directed hypergraph based cloud-computing task scheduling method. The cloud-computing task scheduling method includes: describing resource demands and dependencies of tasks by adopting an empowerment directed hypergraph and generating a corresponding empowerment directed hypergraph document; starting a cellular automation based empowerment directed hypergraph partitioning program and partitioning the generated empowerment directed hypergraph; structuring task subsets according to a partitioning result of the empowerment directed hypergraph and subjecting the task subsets to mapping and scheduling through a MapReduce task scheduling model. By the cloud-computing task scheduling method, efficiency of task scheduling is effectively improved, performance of task partitioning is remarkably improved, and good practicability is achieved.
Description
Technical field
The present invention relates to the cloud computing task based on cellular automata and tax power Directed Hypergraph under a kind of cloud computing environment
Dispatching method.
Background technology
Cloud computing is as the conventional arts such as Distributed Calculation, parallel computation, grid computing and network programming model, distributed
The product of the new technique fusion development such as data storage technology, Intel Virtualization Technology, is the key of the Fashion of Future information industry innovation
Strategic technology and means, will have important strategic importance to China's developing new and high-tech industry.Cloud computing is passed through to calculate
Task is divided on large-scale low-cost server cluster so that people can be processed using the unused resource being distributed in various places
Complex application program, obtains high calculating quality with extremely low cost input.
On the premise of meeting cloud computing environment requirement, scattered application program task in a large number is divided into and multiple has one
Determine the task subset of restriction relation, be dispatched on different virtual machines, obtain than some other for grid computing or parallel count
The shorter time span of task scheduling algorithm calculated and more preferable running quality, are to realize cloud computing high performance key core skill
Art.There is the partitioning of several task, these partitionings are from dependence number minimum, division in the dividing system of prior art
The task number of task subset the different aspects such as is uniformly distributed to realize afterwards, mainly has partitioning based on migration, level embedding
The methods such as set partitioning, multilevel partitioning, cellular automata partitioning.
Partitioning based on migration.The method produces the random initial division of task first, and same task can not be simultaneously
Belong to two task subsets.In the migration optimizing phase, respectively choose a task from two task subsets and exchanged in pairs, this
Two tasks are belonging respectively to two different task subsets and Income Maximum, thus utilize exchange process to greatest extent every time
Improvement task divides quality.Once record cuts off the task division result reaching the minimum of a value moment, and have exchanged selected two
Individual task, in the remaining Optimal improvements of whole transition process, this two task lockings is made them no longer selected.Repeat
Said process until all possible task all pass through migration after, roll back to accumulated earnings maximum cut off minimum of a value when
Carve.The task division result that this partitioning obtains is unstable, and discreteness is very big, and therefore limiting this partitioning institute can solve problem
Scale.
Horizontal nest partitioning.The method selects a task first, and this task is put on number 0, then will own
Being connected with this task of task puts on number 1, does not also put on number for those afterwards, but and has put on appointing of number
The task that business is connected, the number being numbered connected task Jia 1.Until the task of half puts on number, label procedure is
Terminate.The set of tasks that those have put on number is set to a task subset, and other tasks are another task subset.This stroke
, only when the initiating task chosen is close to periphery, the task division result obtaining is relatively preferable, generally speaking this task for point-score
Division result is also unstable.
Multilevel partitioning.Karypis reaches millions of partition problems for node scale it is proposed that multilevel division
Concept, high-quality division can be obtained within the relatively short time.It is excellent that the method comprises roughening, initial division and migration
Change three phases.First, some tasks are combined together by it using random fit strategy, and the roughening obtaining next level course is appointed
Business figure, repeats this process till roughening task image is sufficiently small, that is, obtains a minimum task figure.Then, using partitioning
Minimum task figure is carried out, to dividing, obtain an initial division.Afterwards, minimum task figure is projected back in initiating task figure, every
During the refinement task of one level course divides, select the maximum task of financial value to carry out migration according to greedy principle and optimize, obtain
Task division result afterwards.Since the concept of multilevel division proposes, obtain widely paying attention to, and applied and draw in circuit
Point, many research fields such as cloud computing task scheduling.On May 27th, 2009 China Intellectual Property Office's Authorization Notice No.
Cn100492377c " is drawn based on the large scale integrated circuit of multilevel partitioning by what cold bright, Yu Songnian and Sun Lingyu declared
Divide method " Chinese invention patent, is moved with greedy principle because carrying out coupling using randomized policy in prior art
Move and optimize, lead to not the division fleeing from local optimum, there is provided a kind of improved extensive collection based on multilevel partitioning
Become circuit partitioning method, be effectively improved efficiency and the performance of large scale integrated circuit division.On April 9th, 2014, China was known
Know property right office Authorization Notice No. cn102693340b by Sun Lingyu, cold bright and hail positive declare " based on multilevel partitioning
Large scale integrated circuit division methods with empowerment hypergraph " Chinese invention patent, weigh non-directed graph as extensive for using tax
The Mathematical Modeling of IC partition problem, exists and assigns power non-directed graph optimal dividing and large scale integrated circuit optimal dividing
A kind of inconsistency, there is provided large scale integrated circuit division methods being weighed undirected hypergraph based on multilevel partitioning and tax, is entered
One step improves efficiency and the performance of large scale integrated circuit division.On October 16th, 2013 China Intellectual Property Office's Granted publication
Number cn102663216b by Sun Lingyu, the cold bright and positive " large scale integrated circuit based on node attribute function declared of hail
Core value calculating method " Chinese invention patent, using multilevel partitioning solve assign weigh undirected hypergraph partition problem roughening
In stage, there is provided the core value calculating method of the required large scale integrated circuit based on node attribute function.June 25 in 2014
Day China Intellectual Property Office application publication number cn103885839a by Sun Lingyu, cold bright and hail positive declare " based on many water
Flat partitioning and the cloud computing method for scheduling task assigning power Directed Hypergraph " Chinese invention patent, weigh Directed Hypergraph description using assigning
The resource requirement of task and dependence, and generate corresponding tax power Directed Hypergraph file;Then start and be based on multilevel division
The tax power Directed Hypergraph partition program of method, divides to the tax power Directed Hypergraph generating;Last foundation assigns power Directed Hypergraph
Division result constructs task subset, by mapreduce Task Scheduling Model, it is mapped and dispatches.June 18 in 2014
Day China Intellectual Property Office application publication number cn103870342a by Sun Lingyu, cold bright and positive " the cloud computing ring declared of hail
Task core value calculating method based on node attribute function in border " Chinese invention patent, assign being solved using multilevel partitioning
In the roughening stage of power Directed Hypergraph partition problem, there is provided the task based on node attribute function in required cloud computing environment
Core value calculating method.
Cellular automata partitioning.On December 25th, 2013 China Intellectual Property Office Authorization Notice No. cn102682176b's
" the large scale integrated circuit division methods based on cellular automata and empowerment hypergraph " declared by cold bright, Sun Lingyu and hail sun
Chinese invention patent, carries out mathematical modeling using cellular automata to assigning the undirected hypergraph partition problem of power.Wherein, cellular corresponds to
Assign the node weighed in undirected hypergraph, adjacent cellular corresponds to and adjoins the node that super side is comprised, and the state of cellular corresponds to and is located
Dividing subset, and then using quick cellular financial value and divide and cut off the computational methods of value, significantly reduce extensive
The space complexity of IC partition method and time complexity, and improve the performance of large scale integrated circuit division.
Content of the invention
In cloud computing environment according to the present invention, task scheduling includes the scheduling of Meta task and the scheduling of dependence task.Unit appoints
Separate between business, its scheduling does not consider the data correlation between task and priority constraint relationship, and therefore it simply partly solves
Resource of having determined isomerism and availability issue, lack general applicability.And there is priority dependence between dependence task it is desirable to
One task could start after must receiving its all predecessor task message to execute.
The present invention to construct the Mathematical Modeling of task partition problem using assigning power Directed Hypergraph, and task presentation is that tax power is oriented
The node of hypergraph, the priority dependence between task node is expressed as assigning the oriented super side in power Directed Hypergraph.Power of assigning is oriented super
The many-to-many relationship of figure provides the accurate means describing user task, and the process level user that its node corresponds to after decomposing appoints
Business, oriented super side correspond to task node between priority dependence, the task corresponding to tail end node on arbitrarily super side whole
The source terminal that predecessor task is included in this super side is concentrated.Compare weighted and directed diagraph and assign the undirected hypergraph of power, assign power Directed Hypergraph
Scheduling for dependence task provides more accurate model, can comprehensively represent the isomerism of cloud computing environment, distributivity, wide
The features such as domain property, thus improve accuracy and the execution efficiency of task scheduling.
For comparing weighted and directed diagraph, assign the dependence that power Directed Hypergraph is task node and provide more accurate mould
Type: each edge only connects two nodes, can only connect two tasks corresponding to the dependence between task node, and every super
Side can connect plural node, can connect plural corresponding to the priority dependence between task node
Business, that is, arbitrarily super while task corresponding to tail end node whole (two or more) predecessor task be included in this super while source
In subset.
Compare tax weigh undirected hypergraph for, assign power Directed Hypergraph be task node priority dependence provide more smart
True model: undirected super side connects plural node, but cannot represent the priority dependence between task node, and every
Oriented super side can connect plural node, super all direct precursor nodes in tail end node be included in this super while
Source terminal concentrate, corresponding to task node between priority dependence.
Present invention aims to the deficiency that prior art exists, one kind is provided to have based on cellular automata and tax power
To the cloud computing method for scheduling task of hypergraph, the time span that the task of being effectively shortened completes is it is achieved that cloud computing resources
Rationally utilize, provide efficient Task Scheduling Mechanism for cloud computing.For reaching above-mentioned purpose, the design of the present invention is as follows.
First, in cloud computing environment, scattered task in a large number is divided into multiple scales according to specific constraints less
Task subset, make each task subset after division calculate scale quite, but the dependence number between task subset reach
To minimum, thus farthest realizing the load balancing of cloud computing platform, and shorten the time span that whole task completes.Appoint
The optimization of business subset divides the key link as cloud computing Task Scheduling Mechanism, the operation to whole cloud computing environment for its result
Efficiency has important impact, can efficiently reduce resource free time, improves the utilization benefit of resource.
2nd, the optimization partition problem of the task subset of cloud computing task scheduling is converted to tax power Directed Hypergraph partition problem,
I.e. the partition problem that optimizes of task subset requires the task number that each task subset is comprised equal, oriented super corresponding to power of assigning
The equilibrium constraint of Graph partition problem, division result makes the dependence number between task subset reach minimum, corresponding
Always cut off in the minimum assigning power Directed Hypergraph partition problem.The division assigning power Directed Hypergraph requires each task subset to be comprised
Task number equal, corresponding to the equilibrium constraint under the conditions of multiple resource constraint, and division result makes these tasks
Dependence number between subset reaches minimum, and the minimum corresponding to multiple-objection optimization partition problem always cuts off.
3rd, describe resource requirement and the dependence of task using tax power Directed Hypergraph, and it is oriented to generate corresponding tax power
Hypergraph file;Then start the tax based on cellular automata and weigh Directed Hypergraph partition program, the tax power Directed Hypergraph generating is entered
Row divides;The last division result construction task subset according to tax power Directed Hypergraph, by mapreduce Task Scheduling Model pair
It is mapped and is dispatched.
4th, mathematical modeling is carried out to assigning power Directed Hypergraph partition problem using cellular automata, wherein cellular corresponds to and assigns
Node in power Directed Hypergraph, adjacent cellular corresponds to and adjoins the node that oriented super side is comprised, and the state of cellular corresponds to institute
Dividing subset, and then using quick cellular financial value and divide and cut off the computational methods of value, significantly reduce and be based on
Space complexity and the time complexity of Directed Hypergraph division methods is weighed in the tax of cellular automata.
According to above-mentioned inventive concept, the technical scheme is that and be achieved in that: a kind of based on cellular automata and
The cloud computing method for scheduling task assigning power Directed Hypergraph is it is characterised in that comprise the following steps that.
Step 1, class types degree is analyzed, the task that under input cloud computing environment, user submits to, and carries out type and class to it
The analysis of degree, determines parallelization degree and the feature of task.
Step 2, proceeding graininess decomposition, the parallelization degree according to user task and feature, and the resource of cloud computing is altogether
Enjoy the peculiar properties such as the method for salary distribution, user task is decomposed according to proceeding graininess rank.
Step 3, resource characteristicses are analyzed, the peculiar property such as resource-sharing method of salary distribution according to cloud computing, after decomposing
Task carries out resource characteristicses analysis.
Step 4, assigns power Directed Hypergraph file generated, according to the analysis result to task resource characteristic, sets up its money of description
The tax power Directed Hypergraph model of source demand and dependence, and it is oriented to save as tax power according to the file memory format improving compression
Hypergraph file.
Step 5, assigns power Directed Hypergraph and divides, and starts the tax based on cellular automata and weighs Directed Hypergraph partition program, reads
Assign power Directed Hypergraph file, stored to assigning power Directed Hypergraph using the memory compression storage format based on cellular automata,
The tax power Directed Hypergraph generating is divided, the division result finally giving is stored in tax power Directed Hypergraph and divides file
In.
Step 6, task subset construction, complete to draw the power Directed Hypergraph partition program of the tax based on cellular automata is detected
/ after, read corresponding division result from assigning power Directed Hypergraph division file, according to the division result assigning power Directed Hypergraph
Construction process level task subset.
Step 7, duty mapping is dispatched, and by mapreduce Task Scheduling Model, draws to based on tax power Directed Hypergraph optimization
The task subset dividing construction is mapped and is dispatched, and the task in cloud computing environment of realizing is submitted to and executed, and effectively equalizes
The load of cloud computing platform and shorten the time span that completes of whole task.
In above-mentioned step 4, the file memory format that the improvement compression of Directed Hypergraph is weighed in described tax is as follows.
Step 4.1, the 1st parameter of the 1st row of file format represents assigns the number m weighing oriented super side, the 2nd parameter generation
The number x assigning power node by table.
Step 4.2, the 2nd row of file format starts to represent one article of phase assigning the oriented super side of power to the often row of m+1 row
Pass information, the 1st numerical value is to assign the value information weighing oriented super side, and its remainder values is to assign the node information weighing oriented super side, its
In often row last numerical value represent assign weigh oriented super side tail end node information, and assign weigh oriented super side source node letter
Breath is in be assigned between the value information weighing oriented super side and tail end node information.
Step 4.3, the m+2 row of file format starts to represent the power assigning power node to the often row of m+x+1 row
Value information.
In above-mentioned step 5, the step that Directed Hypergraph partition program is weighed in the described tax based on cellular automata is as follows.
Step 5.1, reads and assigns power Directed Hypergraph file, using the memory compression storage format based on cellular automata to tax
Power Directed Hypergraph is stored.
Step 5.2, cellular initializes, and travels through each cellular and gives whole between cellular state in which 1 and n at random
Number, represents certain dividing subset in the middle of n dividing subset v1 ... vn residing for the corresponding node of cellular respectively, thus obtaining initial
Divide.
Step 5.3, initialization two dimension auxiliary array edg [n] [m], according to initial division, initialization two dimension auxiliary array
edg[n][m].
Step 5.4, calculates the value that cuts off of initial division, according to two dimension auxiliary array edg [n] [m], quickly calculates current drawing
That divides cuts off value.
Step 5.5, loop initialization, loop initialization counter count is 0.
Step 5.6, travels through whether each cellular terminates, if access do not terminated, that is, there is current cellular and is not accessed, then
Go to step 5.7;Otherwise access and terminate, go to step 5.13.
Step 5.7, the financial value of the current cellular of calculating, the state of the state according to current cellular and adjacent cellular, quickly
Calculate the financial value of current cellular.
Step 5.8, develop current cellular state, if the financial value of current cellular is more than zero, current cellular state is certain
Be turned to rollover states to from current state from, otherwise current cellular state with the upset probability that sets from current state from
It is turned to rollover states to.
Step 5.9, if current cellular state is turned to rollover states to from current state from, goes to step 5.10,
Otherwise go to step 5.6.
Step 5.10, updates two dimension auxiliary array edg [n] [m], all of its neighbor super side e of traversal cellular, executes edg
[from] [e] operation that subtracts 1, operation that edg [to] [e] Jia 1.
Step 5.11, update current division cuts off value, and according to two dimension auxiliary array edg [n] [m], quick calculating is current
Divide cuts off value.
Step 5.12, updates the optimal dividing having found, goes to step 5.6.
Step 5.13, cycle criterion, cycle counter count adds 1, if meeting count to reach the bar setting evolution number of times
When part 1 or all cellular all no longer change the condition 2 of oneself state, execution step 5.14, otherwise return to step 5.6.
Step 5.14, enters into equilibrium stage, runs the tax power Directed Hypergraph based on fm early-exit method and divides journey
Sequence: due to, in the tax power Directed Hypergraph partition process based on cellular automata, tax power Directed Hypergraph partition problem may be run counter to
Equilibrium constraint, therefore based on cellular automata tax power Directed Hypergraph divide solved on the basis of, operation is based on
The tax power Directed Hypergraph division methods of fm early-exit method, make division solution meet equilibrium constraint, thus obtaining assigning power
The division solution of Directed Hypergraph partition problem.
Step 5.15, the tax finally giving power Directed Hypergraph division result is stored in tax power Directed Hypergraph and divides file
In.
In above-mentioned step 5.1, the memory compression storage format based on cellular automata of Directed Hypergraph is weighed in described tax
As follows.
Step 5.1.1, corresponds to, using id storage of array cellular, the number information assigning node in power Directed Hypergraph, and id number
The size of group is the node number assigned in power Directed Hypergraph.
Step 5.1.2, using the status information of state storage of array cellular, and the size of state array is that tax power is oriented
Node number in hypergraph.
Step 5.1.3, corresponds to, using vwgts storage of array cellular, the value information assigning node in power Directed Hypergraph, and
The size of vwgts array is the node number assigned in power Directed Hypergraph.
Step 5.1.4, assigns the original position weighing oriented super side list using each node all of its neighbor of xadj storage of array
Information, that is, the final position of i-th node is that the original position of i+1 node subtracts 1, and the size of xadj array has for tax power
Add 1 to the node number in hypergraph, last element of xadj array is used for depositing the final position of last node.
Step 5.1.5, assigns, using each node all of its neighbor of adjncy storage of array, the list information weighing oriented super side, the
The adjacent tax of i node is weighed oriented super side list and is stored in adjncy array, from adjncy [xadj [i]] to adjncy
[xadj[i+1]-1].
Step 5.1.6, assigns the original position weighing the node list that oriented super side is comprised for every using eptr storage of array
Information, that is, j-th strip assign power oriented super while final position be jth+1 assign power oriented super while original position subtract 1, and eptr number
The size of group is that the tax oriented super edge strip number of power assigned in power Directed Hypergraph adds 1, and last element of eptr array is used for depositing
Assign the final position weighing oriented super side for one afterwards.
Step 5.1.7, assigns the list information weighing the comprised node in oriented super side for every using eind storage of array, wherein often
The tail end node that bar assigns the oriented super side of power only has 1, and every is assigned all direct precursor nodes weighing oriented super side tail end node
It is included in the source terminal concentration that oriented super side is weighed in this tax.The node list that j-th strip assigns the oriented super side of power is stored in eind array
In, from eind [eptr [j]] to eind [eptr [j+1] -1], the source node that wherein j-th strip assigns the oriented super side of power is eind
[eptr [j]] arrives eind [eptr [j+1] -2], and the tail end node that j-th strip assigns the oriented super side of power is eind [eptr [j+1] -1].
Step 5.1.8, assigns, using hewgts storage of array, the value information weighing oriented super side, and the size of hewgts array
Weigh oriented super edge strip number for assigning the tax in power Directed Hypergraph.
In above-mentioned step 5.3, the step of described initialization two dimension auxiliary array edg [n] [m] is as follows.
Step 5.3.1, two dimension auxiliary array edg [n] [m] resets.
Step 5.3.2, reads eptr array and the node letter that oriented super side is comprised is weighed in every tax of eind storage of array
Breath, calculates every based on initial division and assigns the node number weighing oriented super side in n dividing subset v1 ... vn, i.e. two-dimentional supplementary number
The n row of group edg [n] [m] is deposited m bar respectively and is assigned the node number weighing oriented super side in n dividing subset.
In above-mentioned step 5.4 and step 5.11, described quick calculating is current divide cut off value step as follows.
Step 5.4.1, divides and cuts off value clearing.
Step 5.4.2, traversal every is assigned and is weighed whether oriented super side terminates, if access do not terminated, that is, there is the power of tax oriented
Super side e is not accessed, then go to step 5.4.3;Otherwise access and terminate, return division and cut off value.
Step 5.4.3, if meet during the condition 2 of the condition 1 of edg [i] [e] >=1 and edg [j] [e] >=1 it is meant that
Assign and weigh the node number in dividing subset vi and vj for the oriented super side e both greater than equal to 1, you can judgement is assigned and weighed oriented super side e is two
Dwell side, and cut off dividing value and add up the upper current weights assigning the oriented super side of power;Otherwise judge that it is not amphibious for assigning the oriented super side e of power
Side, it is constant that division cuts off value.
Step 5.4.4, goes to step 5.4.2.
In above-mentioned step 5.7, the step of described quick calculating current cellular financial value is as follows.
Step 5.7.1, cellular financial value resets.
Step 5.7.2, reads current state from and the rollover states to of cellular.
Step 5.7.3, all of its neighbor of traversal cellular is assigned and is weighed oriented super side e, if two-dimensional array edg [from] [e] value is
1, then financial value is added and assign the weights weighing oriented super side e;If two-dimensional array edg [to] [e] value is 0, financial value is deducted
Assign the weights weighing oriented super side e.
Step 5.7.4, returns cellular financial value.
The present invention compared with prior art, has and obviously projects substantive distinguishing features and remarkable advantage as follows.
1st, improve the efficiency of circuit division.
The present invention based on cellular automata and assigns the cloud computing method for scheduling task weighing Directed Hypergraph, on the one hand passes through task
To the conversion assigning power Directed Hypergraph file, start the tax based on cellular automata and weigh Directed Hypergraph partition program, to the tax generating
Power Directed Hypergraph is divided, it is to avoid division methods are directly divided in task;On the other hand can be by setting unit
The upset probability parameter of cellular automaton is obtaining preferably after division result, then carries out mapping and the scheduling of task, thus effectively
Improve the efficiency of task scheduling, the time span that the task of shortening completes, it is achieved that the reasonable utilization of cloud computing resources, is
Cloud computing provides efficient Task Scheduling Mechanism.
2nd, improve the performance of task division.
The present invention based on cellular automata and assigns the task optimization division methods weighing Directed Hypergraph, by two dimension auxiliary array
Edg [n] [m] stores every super side in the node number of dividing subset it is achieved that dividing the computational methods cuing off value, effectively keeps away
Exempt to travel through the node in super side.Compare tax and weigh undirected hypergraph division methods and weighted and directed diagraph division methods, the method is effective
Find the task division result more excellent than prior art, reduce space complexity and time complexity, finally significantly carry
The performance that high task optimization divides.
Brief description
By the following reality to the present invention based on cellular automata and the cloud computing method for scheduling task assigning power Directed Hypergraph
Example combines the description of its accompanying drawing, it will be further appreciated that the purpose of the present invention, specific structural features and advantage.
Fig. 1 is the flow chart based on cellular automata and the cloud computing method for scheduling task assigning power Directed Hypergraph for the present invention.
Fig. 2 is the compression storage format based on cellular automata of the empowerment hypergraph of the present invention.
Fig. 3 is the flow chart of the empowerment hypergraph division methods based on cellular automata of the present invention.
Specific embodiment
In order to be more clearly understood that the cloud computing task tune based on cellular automata and tax power Directed Hypergraph for the present invention
The technology contents of degree method, describe in detail especially exemplified by following instance.
The flow chart based on cellular automata and the cloud computing method for scheduling task assigning power Directed Hypergraph of the present embodiment is such as
Shown in Fig. 1.Under cloud computing environment, the task 101 that input user submits to, user task is carried out with type and the analysis of class degree
102, determine parallelization degree and the feature of task;Parallelization degree according to user task and feature, and the money of cloud computing
The peculiar properties such as the method for salary distribution are shared in source, and user task is carried out decomposing 103 according to proceeding graininess rank;And then after to decomposing
Task carries out resource characteristicses analysis 104;According to the analysis result to task resource characteristic, set up and describe its resource requirement and dependence
The tax power Directed Hypergraph model 105 of relation;File memory format according to improving compression saves as tax power Directed Hypergraph file
109;Start the tax based on cellular automata and weigh Directed Hypergraph partition program 111, read and assign power Directed Hypergraph file 109, obtain
The tax power Directed Hypergraph 112 of the memory compression storage format based on cellular automata;Enter into the division stage, run and be based on cellular
The tax power Directed Hypergraph partition program 113 of automatic machine, divides to the tax power Directed Hypergraph of memory compression storage format;Enter
To equilibrium stage, run the tax based on fm early-exit method and weigh Directed Hypergraph partition program 114, make initially to assign power oriented
The division result of hypergraph meets equilibrium constraint, and the division result finally giving is stored as assigning power Directed Hypergraph division literary composition
Part 110;After the power Directed Hypergraph partition program 111 of the tax based on cellular automata is detected and completing to divide, from assigning, power is oriented
Hypergraph divides and reads corresponding division result in file 110, according to the division result construction process level task assigning power Directed Hypergraph
Subset 106;By mapreduce Task Scheduling Model, enter to based on the task subset assigning power Directed Hypergraph optimization division construction
Row mapping and scheduling 107;In cloud computing environment, to based on appointing in the task subset assigning power Directed Hypergraph optimization division construction
Business is submitted to and execution 108, and effectively equalize cloud computing platform loads and shorten the time span that whole task completes.
The tax of the present embodiment is weighed Directed Hypergraph and is improved the file memory format of compression referring to formerly technology [1] " g.
karypis and v. kumar. hmetis 1.5.3: a hypergraph partitioning package [r].
technical report, department of computer science, university of minnesota,
1998. " and formerly technology [2] " Sun Lingyu, cold bright, Guo Kaiqiang, Zhu Ping. a kind of vlsi is designed into the converting system of hypergraph
[j]. computer engineering and application, 2012, vol.29, issue.2, pages 7-16. ".With formerly technology [1] and
The identical point of first technology [2]: the 1st parameter of the 1st row of file format represents the number m on oriented tax Quan Chao side, the 2nd parameter
Represent the number x assigning power node;2nd row of file format starts to represent one article of oriented tax and weigh to the often row of m+1 row to surpass
The relevant information on side;The m+2 row of file format starts to represent a weights letter assigning power node to the often row of m+x+1 row
Breath.Distinctive points with formerly technology [1] and formerly technology [2]: the 2nd row of file format starts in m+1 row, except the 1st
Its remainder values outside numerical value are the node information on oriented tax Quan Chao side, and last numerical value of where each row represents oriented tax and weighs
The tail end node information on super side, and oriented tax Quan Chao while source node information be in oriented tax Quan Chao while value information and tail
Between leaf information.
The compression storage format based on cellular automata that Directed Hypergraph is weighed in the tax of the present embodiment is as shown in Figure 2.Legend 202
Shown cellular storage organization, wherein not only stores numbering, state and the weights of itself, also stored for adjacent tax and weighs oriented surpassing
The original position on side.Compression storage format based on cellular automata uses the corresponding tax power of id array 203 storage cellular oriented
The number information of node in hypergraph, and the size of id array 203 is the node number assigned in power Directed Hypergraph.Using state number
The status information of group 204 storage cellular, and the size of state array 204 is the node number assigned in power Directed Hypergraph.Use
The value information of vwgts array 205 storage node, and the size of vwgts array 205 is the node assigned in power Directed Hypergraph
Number.Using xadj array 206 store each node all of its neighbor assign weigh oriented super side list start position information, that is, i-th
The final position of node is that the original position of i+1 node subtracts 1, and the size of xadj array 206 is to assign in power Directed Hypergraph
Node number add 1, last element of xadj array 206 is used for depositing the final position of last node.Use
Adjncy array 207 stores each node all of its neighbor and assigns the list information weighing oriented super side, and the adjacent tax power of i-th node has
It is stored in adjncy array to the list of super side, from adjncy [xadj [i]] to adjncy [xadj [i+1] -1].Use
Hewgts array 208 stores assigns the value information weighing oriented super side, and the size of hewgts array 208 is to assign in power Directed Hypergraph
Tax weigh oriented super edge strip number.Store every using eptr array 209 and assign the initial of the node list that the oriented super side of power is comprised
Positional information, that is, j-th strip assign power oriented super while final position be jth+1 assign power oriented super while original position subtract 1, and
The size of eptr array 209 is that the tax oriented super edge strip number of power assigned in power Directed Hypergraph adds 1, last unit of eptr array 209
Element assigns, for depositing the last item, the final position weighing oriented super side.Store every using eind array 210 and assign the oriented super side of power
The list information of comprised node.Assume that group address is started from scratch, node numbering is started from scratch, then i-th node is adjacent
Assign the oriented super side list of power to be stored in adjncy array 207, from adjncy [xadj [i]] to adjncy [xadj [i+1] -1];
The adjacent node list that j-th strip assigns the oriented super side of power is stored in eind array 210, from eind [eptr [j]] to eind [eptr
[j+1] -1], the source node that wherein j-th strip assigns the oriented super side of power is that eind [eptr [j]] arrives eind [eptr [j+1] -2],
The tail end node that j-th strip assigns the oriented super side of power is eind [eptr [j+1] -1].The tax power Directed Hypergraph of legend 201 comprises altogether 7
Individual node and 8 assign and weigh oriented super sides, and the wherein weights of the 6th node are 7, have 2 adjacent oriented super sides f, h, wherein oriented
The corresponding weights of super side f are 4, and corresponding adjacent node is respectively node 7,3,6, and source node is node 7 and 3, and tail end is tied
Point is node 6;The corresponding weights of oriented super side h are 1, and corresponding adjacent node is respectively node 4,6, and source node is node
4, tail end node is node 6.
The tax power Directed Hypergraph partition program based on fm early-exit method of the present embodiment is referring to formerly technology [3]
“karypis g, aggarwal r, kumar v, shekhar s. multilevel hypergraph
partitioning: applications in vlsi domain[j]. ieee transactions on very large
scale integration systems, 1999, vol.7, issue.1, pages 69-79.”.
The tax based on cellular automata of the present embodiment weighs the flow chart of Directed Hypergraph division methods as shown in figure 3, step
As follows.
A01: read and assign power Directed Hypergraph file.
A02: cellular initializes.
A03: initialization two dimension auxiliary array edg [n] [m].
A04: calculate initial division cuts off value.
A05: loop initialization counter count is 0.
A06: travel through whether each cellular terminates, if access do not terminated, that is, there is current cellular and be not accessed, then turn step
Rapid a07;Otherwise access and terminate, go to step a13.
A07: calculate the financial value of current cellular.
A08: develop current cellular state.
A09: if current cellular state is turned to rollover states to from current state from, goes to step a10, otherwise turn
Step a 06.
A10: update two dimension auxiliary array edg [n] [m].
A11: update current division cuts off value.
A12: update the optimal dividing having found, go to step a06.
A13: cycle counter count adds 1, if meet count to reach the condition 1 setting evolution number of times or whole cellular
All no longer change oneself state condition 2 when, execution step a 14, otherwise return to step a 06.
A14: run the tax based on fm early-exit method and weigh Directed Hypergraph partition program.
A15: the tax finally giving power Directed Hypergraph division result is stored in tax power Directed Hypergraph and divides in file.
Claims (1)
1. a kind of cloud computing method for scheduling task based on cellular automata and tax power Directed Hypergraph is it is characterised in that concrete walk
Suddenly as follows:
Step 1, class types degree is analyzed, the task that under input cloud computing environment, user submits to, and it is carried out with type and class degree
Analysis, determines parallelization degree and the feature of task;
Step 2, proceeding graininess decomposition, the parallelization degree according to user task and feature, and the resource-sharing of cloud computing divide
The peculiar properties such as formula formula, decompose according to proceeding graininess rank to user task;
Step 3, resource characteristicses are analyzed, the peculiar property such as resource-sharing method of salary distribution according to cloud computing, to the task after decomposing
Carry out resource characteristicses analysis;
Step 4, assigns power Directed Hypergraph file generated, and according to the analysis result to task resource characteristic, setting up its resource of description needs
Ask and Directed Hypergraph model is weighed in the tax of dependence, and save as tax power Directed Hypergraph according to the file memory format improving compression
File;
Step 5, assigns power Directed Hypergraph and divides, and starts the tax based on cellular automata and weighs Directed Hypergraph partition program, reads the power of tax
Directed Hypergraph file, is stored to assigning power Directed Hypergraph using the memory compression storage format based on cellular automata, opposite
The tax power Directed Hypergraph becoming is divided, and the division result finally giving is stored in tax power Directed Hypergraph and divides in file;
Step 6, task subset construction, complete to divide it the power Directed Hypergraph partition program of the tax based on cellular automata is detected
Afterwards, divide from tax power Directed Hypergraph and file, read corresponding division result, weigh the division result construction of Directed Hypergraph according to tax
Process level task subset;
Step 7, duty mapping is dispatched, and by mapreduce Task Scheduling Model, divides structure to based on tax power Directed Hypergraph optimization
The task subset made is mapped and is dispatched, and the task in cloud computing environment of realizing is submitted to and execution, effectively equalizes cloud meter
Calculate the load of platform and shorten the time span that whole task completes;
In above-mentioned step 4, the file memory format that the improvement compression of Directed Hypergraph is weighed in described tax is as follows:
Step 4.1, the 1st parameter of the 1st row of file format represents assigns the number m weighing oriented super side, and the 2nd parameter represents
Assign the number x of power node;
Step 4.2, the 2nd row of file format starts to represent one bar of related letter assigning the oriented super side of power to the often row of m+1 row
Breath, the 1st numerical value is to assign the value information weighing oriented super side, and its remainder values is to assign the node information weighing oriented super side, wherein often
Last numerical value of row represents assigns the tail end node information weighing oriented super side, and assigns at the source node information weighing oriented super side
Between the value information assigning the oriented super side of power and tail end node information;
Step 4.3, the m+2 row of file format starts to represent a weights letter assigning power node to the often row of m+x+1 row
Breath;
In above-mentioned step 5, the step that Directed Hypergraph partition program is weighed in the described tax based on cellular automata is as follows:
Step 5.1, is read and assigns power Directed Hypergraph file, using the memory compression storage format based on cellular automata, tax power is had
Stored to hypergraph;
Step 5.2, cellular initializes, and travels through each cellular the integer giving between cellular state in which 1 and n at random, point
Do not represent certain dividing subset in the middle of n dividing subset v1 ... vn residing for the corresponding node of cellular, thus obtaining initial division;
Step 5.3, initialization two dimension auxiliary array edg [n] [m], according to initial division, initialization two dimension auxiliary array edg [n]
[m];
Step 5.4, calculates the value that cuts off of initial division, according to two-dimentional auxiliary array edg [n] [m], quickly calculates current division
Cut off value;
Step 5.5, loop initialization, loop initialization counter count is 0;
Step 5.6, travels through whether each cellular terminates, if access do not terminated, that is, there is current cellular and is not accessed, then turn step
Rapid 5.7;Otherwise access and terminate, go to step 5.13;
Step 5.7, calculates the financial value of current cellular, the state of the state according to current cellular and adjacent cellular, quick calculating
The financial value of current cellular;
Step 5.8, develop current cellular state, if the financial value of current cellular is more than zero, current cellular state is necessarily from working as
Front state from is turned to rollover states to, and otherwise current cellular state is overturn from current state from the upset probability setting
To rollover states to;
Step 5.9, if current cellular state is turned to rollover states to from current state from, goes to step 5.10, otherwise
Go to step 5.6;
Step 5.10, updates two dimension auxiliary array edg [n] [m], all of its neighbor super side e of traversal cellular, execution edg [from]
[e] operation that subtracts 1, operation that edg [to] [e] Jia 1;
Step 5.11, update current division cuts off value, and according to two dimension auxiliary array edg [n] [m], quick calculating is current to be divided
Cut off value;
Step 5.12, updates the optimal dividing having found, goes to step 5.6;
Step 5.13, cycle criterion, cycle counter count adds 1, if meet count reach set evolution number of times condition 1 or
When the whole cellular of person all no longer changes the condition 2 of oneself state, execution step 5.14, otherwise return to step 5.6;
Step 5.14, enters into equilibrium stage, the tax power Directed Hypergraph partition program based on fm early-exit method for the operation:
Due to, in the tax power Directed Hypergraph partition process based on cellular automata, may running counter to and assigning the flat of power Directed Hypergraph partition problem
Weighing apparatus constraints, therefore on the basis of the tax power Directed Hypergraph based on cellular automata divides and solved, runs and is based on fm
The tax power Directed Hypergraph division methods of early-exit method, make division solution meet equilibrium constraint, thus obtain tax power having
Division solution to hypergraph partition problem;
Step 5.15, the tax finally giving power Directed Hypergraph division result is stored in tax power Directed Hypergraph and divides in file;
In above-mentioned step 5.1, the memory compression storage format based on cellular automata that Directed Hypergraph is weighed in described tax is as follows:
Step 5.1.1, is corresponded to using id storage of array cellular and assigns the number information weighing node in Directed Hypergraph, and id array
Size is the node number assigned in power Directed Hypergraph;
Step 5.1.2, using the status information of state storage of array cellular, and the size of state array is to assign power Directed Hypergraph
In node number;
Step 5.1.3, corresponds to, using vwgts storage of array cellular, the value information assigning node in power Directed Hypergraph, and vwgts
The size of array is the node number assigned in power Directed Hypergraph;
Step 5.1.4, assigns, using each node all of its neighbor of xadj storage of array, the start position information weighing oriented super side list,
I.e. the final position of i-th node subtracts 1 for the original position of i+1 node, and the size of xadj array is that power of assigning is oriented super
The node number of in figure adds 1, and last element of xadj array is used for depositing the final position of last node;
Step 5.1.5, using each node all of its neighbor of adjncy storage of array assign weigh oriented super side list information, i-th
The adjacent tax of node is weighed oriented super side list and is stored in adjncy array, from adjncy [xadj [i]] to adjncy [xadj [i
+1]-1];
Step 5.1.6, assigns the start position information weighing the node list that oriented super side is comprised for every using eptr storage of array,
I.e. j-th strip assign power oriented super while final position be jth+1 assign power oriented super while original position subtract 1, and eptr array
Size is that the tax oriented super edge strip number of power assigned in power Directed Hypergraph adds 1, and last element of eptr array is used for depositing last
Bar assigns the final position weighing oriented super side;
Step 5.1.7, assigns the list information weighing the comprised node in oriented super side, wherein every tax for every using eind storage of array
The tail end node weighing oriented super side only has 1, and every all direct precursor nodes assigning power oriented super side tail end node all wrap
It is contained in the source terminal concentration that oriented super side is weighed in this tax;The node list that j-th strip assigns the oriented super side of power is stored in eind array, from
Eind [eptr [j]] arrives eind [eptr [j+1] -1], and the source node that wherein j-th strip assigns the oriented super side of power is eind [eptr
[j]] arrive eind [eptr [j+1] -2], the tail end node that j-th strip assigns the oriented super side of power is eind [eptr [j+1] -1];
Step 5.1.8, assigns, using hewgts storage of array, the value information weighing oriented super side, and the size of hewgts array is to assign
Oriented super edge strip number is weighed in tax in power Directed Hypergraph;
In above-mentioned step 5.3, the step of described initialization two dimension auxiliary array edg [n] [m] is as follows:
Step 5.3.1, two dimension auxiliary array edg [n] [m] resets;
Step 5.3.2, reads eptr array and the node information that oriented super side is comprised, base are weighed in every tax of eind storage of array
Calculate every in initial division and assign the node number weighing oriented super side in n dividing subset v1 ... vn, be i.e. two dimension auxiliary array edg
The n row of [n] [m] is deposited m bar respectively and is assigned the node number weighing oriented super side in n dividing subset;
In above-mentioned step 5.4 and step 5.11, described quick calculating is current divide cut off value step as follows:
Step 5.4.1, divides and cuts off value clearing;
Step 5.4.2, traversal every is assigned and is weighed whether oriented super side terminates, if access do not terminated, that is, exists to assign and weighs oriented super side e
It is not accessed, then go to step 5.4.3;Otherwise access and terminate, return division and cut off value;
Step 5.4.3, if met during the condition 2 of the condition 1 of edg [i] [e] >=1 and edg [j] [e] >=1 it is meant that assigning power
Oriented super side e is both greater than equal to 1 in the node number of dividing subset vi and vj, you can judge to assign power oriented super when e is amphibious,
And cut off, by dividing, the weights that oriented super side is weighed in the cumulative current tax of value;Otherwise judge assign power oriented super when e is not amphibious, draw
It is constant that value is cut in segmentation;
Step 5.4.4, goes to step 5.4.2;
In above-mentioned step 5.7, the step of described quick calculating current cellular financial value is as follows:
Step 5.7.1, cellular financial value resets;
Step 5.7.2, reads current state from and the rollover states to of cellular;
Step 5.7.3, all of its neighbor of traversal cellular is assigned and is weighed oriented super side e, if two-dimensional array edg [from] [e] value is 1,
Financial value is added and assigns the weights weighing oriented super side e;If two-dimensional array edg [to] [e] value is 0, financial value is deducted tax power
The weights of oriented super side e;
Step 5.7.4, returns cellular financial value.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201410137810.1A CN103902374B (en) | 2014-04-08 | 2014-04-08 | Cellular automation and empowerment directed hypergraph based cloud-computing task scheduling method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201410137810.1A CN103902374B (en) | 2014-04-08 | 2014-04-08 | Cellular automation and empowerment directed hypergraph based cloud-computing task scheduling method |
Publications (2)
Publication Number | Publication Date |
---|---|
CN103902374A CN103902374A (en) | 2014-07-02 |
CN103902374B true CN103902374B (en) | 2017-01-18 |
Family
ID=50993712
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201410137810.1A Expired - Fee Related CN103902374B (en) | 2014-04-08 | 2014-04-08 | Cellular automation and empowerment directed hypergraph based cloud-computing task scheduling method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN103902374B (en) |
Families Citing this family (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2016094687A1 (en) * | 2014-12-10 | 2016-06-16 | Kyndi, Inc. | Apparatus and method for combinatorial hypermap based data representations and operations |
CN105005503B (en) * | 2015-07-26 | 2017-12-01 | 孙凌宇 | Cloud computing load balancing method for scheduling task based on cellular automata |
CN105159762B (en) * | 2015-08-03 | 2018-09-07 | 冷子阳 | Heuristic cloud computing method for scheduling task based on Greedy strategy |
Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102663216A (en) * | 2012-05-16 | 2012-09-12 | 孙凌宇 | Core value calculating method of large-scale integrated circuit based on node attribute function |
CN102682176A (en) * | 2012-05-18 | 2012-09-19 | 冷明 | Method for dividing large-scale integrated circuit based on cellular automaton and empowerment hypergraph |
-
2014
- 2014-04-08 CN CN201410137810.1A patent/CN103902374B/en not_active Expired - Fee Related
Patent Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102663216A (en) * | 2012-05-16 | 2012-09-12 | 孙凌宇 | Core value calculating method of large-scale integrated circuit based on node attribute function |
CN102682176A (en) * | 2012-05-18 | 2012-09-19 | 冷明 | Method for dividing large-scale integrated circuit based on cellular automaton and empowerment hypergraph |
Non-Patent Citations (2)
Title |
---|
一种基于元胞自动机的无向图剖分优化算法;孙凌宇、冷明等;《计算机工程与应用》;20081231;第44卷(第24期);全文 * |
赋权有向图的最小生成树算法;孙凌宇、冷明等;《计算机工程与应用》;20100131;第36卷(第2期);全文 * |
Also Published As
Publication number | Publication date |
---|---|
CN103902374A (en) | 2014-07-02 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN103885839B (en) | Cloud computing task scheduling method based on multilevel division method and empowerment directed hypergraphs | |
Zhao et al. | User-based collaborative-filtering recommendation algorithms on hadoop | |
Naas et al. | A graph partitioning-based heuristic for runtime IoT data placement strategies in a fog infrastructure | |
Hsu et al. | Task allocation on a network of processors | |
Aridhi et al. | A MapReduce-based approach for shortest path problem in large-scale networks | |
He et al. | Parallel implementation of classification algorithms based on MapReduce | |
CN106919769A (en) | A kind of hierarchy type FPGA placement-and-routings method based on Hierarchy Method and empowerment hypergraph | |
Arfat et al. | Parallel shortest path graph computations of united states road network data on apache spark | |
CN102693340B (en) | Large scale integrated circuit partitioning method on basis of multilevel partitioning method and weighted hypergraph | |
Djidjev et al. | All-Pairs Shortest Path algorithms for planar graph for GPU-accelerated clusters | |
Selvitopi et al. | Locality-aware and load-balanced static task scheduling for MapReduce | |
CN106875320A (en) | The efficient visual analysis method of ship aeronautical data under cloud environment | |
Jangiti et al. | Scalable and direct vector bin-packing heuristic based on residual resource ratios for virtual machine placement in cloud data centers | |
Chakravarthi et al. | TOPSIS inspired cost-efficient concurrent workflow scheduling algorithm in cloud | |
CN104679966B (en) | Empowerment hypergraph optimization division methods based on Hierarchy Method and discrete particle cluster | |
CN103902374B (en) | Cellular automation and empowerment directed hypergraph based cloud-computing task scheduling method | |
Bakhthemmat et al. | Decreasing the execution time of reducers by revising clustering based on the futuristic greedy approach | |
Saprykin et al. | Large-scale multi-agent mobility simulations on a GPU: towards high performance and scalability | |
Maqsood et al. | Energy and communication aware task mapping for MPSoCs | |
CN105005503A (en) | Cellular automaton based cloud computing load balancing task scheduling method | |
CN103870342B (en) | Task core value calculating method based on node attribute function in cloud computing environment | |
Huang et al. | An efficient parallel method for batched OS-ELM training using MapReduce | |
Jiang et al. | Drscro: A metaheuristic algorithm for task scheduling on heterogeneous systems | |
Rahmaniani et al. | An asynchronous parallel benders decomposition method for stochastic network design problems | |
Bendechache et al. | Performance evaluation of a distributed clustering approach for spatial datasets |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
C14 | Grant of patent or utility model | ||
GR01 | Patent grant | ||
CF01 | Termination of patent right due to non-payment of annual fee | ||
CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20170118 Termination date: 20170408 |