CN107015856A - Task scheduling approach generation method and device under cloud environment in scientific workflow - Google Patents
Task scheduling approach generation method and device under cloud environment in scientific workflow Download PDFInfo
- Publication number
- CN107015856A CN107015856A CN201710202301.6A CN201710202301A CN107015856A CN 107015856 A CN107015856 A CN 107015856A CN 201710202301 A CN201710202301 A CN 201710202301A CN 107015856 A CN107015856 A CN 107015856A
- Authority
- CN
- China
- Prior art keywords
- task
- node
- path
- scheduled
- scheduling approach
- 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.)
- Withdrawn
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
- G06F9/46—Multiprogramming arrangements
- G06F9/48—Program initiating; Program switching, e.g. by interrupt
- G06F9/4806—Task transfer initiation or dispatching
- G06F9/4843—Task transfer initiation or dispatching by program, e.g. task dispatcher, supervisor, operating system
- G06F9/4881—Scheduling strategies for dispatcher, e.g. round robin, multi-level priority queues
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F9/00—Arrangements for program control, e.g. control units
- G06F9/06—Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
- G06F9/46—Multiprogramming arrangements
- G06F9/50—Allocation of resources, e.g. of the central processing unit [CPU]
- G06F9/5005—Allocation of resources, e.g. of the central processing unit [CPU] to service a request
- G06F9/5027—Allocation of resources, e.g. of the central processing unit [CPU] to service a request the resource being a machine, e.g. CPUs, Servers, Terminals
- G06F9/5038—Allocation of resources, e.g. of the central processing unit [CPU] to service a request the resource being a machine, e.g. CPUs, Servers, Terminals considering the execution order of a plurality of tasks, e.g. taking priority or time dependency constraints into consideration
Landscapes
- Engineering & Computer Science (AREA)
- Software Systems (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses the task scheduling approach generation method in scientific workflow under a kind of cloud environment and device, this method comprises the following steps, step 1, and acquisition needs scheduled task；Step 2, describe to treat scheduled task by the directed acyclic graph of Weighted Coefficients；Step 3, in the directed acyclic graph, traveled through from start node to end node, obtain all paths；Step 4, calculate the time loss in each path, according to the descending order of the time loss by all path descending sorts, determine the priority in path；Step 5, task scheduling approach is generated according to the priority orders in path；The device includes task acquisition module, task description module, node traverses module, paths ordering module and schemes generation module.The present invention can make full use of the free time of processor, improve the concurrency of scientific workflow tasks carrying, improve processor utillization, shorten the deadline of scientific workflow task.
Description
Technical field
The present invention relates to scientific workflow task scheduling technique field, specifically for, the present invention is section under cloud environment
Learn the task scheduling approach generation method and device in workflow.
Background technology
Scientific workflow (Scientific Workflow, SWF) is a kind of general type of new application developed in recent years,
Data, service and the Software tool of integrated scientific research personnel, construction and collaboration distributed delays are supported, and there is provided a scientific algorithm
The management platform of Complicated Flow and automatic running.With the Business Stream of traditional Control-oriented stream (Business Workflow,
BWF) Comparatively speaking, Data Flow Oriented is one of important feature of scientific workflow.
As science and technology is growing, SWF has evolved into data and the general type of computation-intensive application, its data
Producing speed and scientific algorithm becomes increasingly complicated, and conventional computing environment is difficult to meet SWF to calculate demand.Therefore, cloud computing
A kind of brand-new deployment and execution are provided by research-on-research stream application of the storage resource of high performance computing resource and magnanimity
Mode.
Although cloud environment can provide infinite calculating and storage capacity in theory, user can rent it as desired
Computing resource.But, computing resource provide scheme change may relate to example establishment, distribution and data move, it is necessary to
The cost of correlation is paid, and the execution efficiency of scientific workflow and the expense paid may be influenceed.
So, it is all very important to user and administrative staff to formulate a rational original execution plan generation.Will
Workflow task be reasonably mapped in computing resource be workflow execution basis, the also execution efficiency with workflow and execution
Cost is closely related, and the process is referred to as：The executive plan generation of workflow, its key is task scheduling to suitable calculating
In resource, in the prior art, the scientific workflow dispatching algorithm of main flow is based on heuristic mutation operations, and it mainly includes clustering algorithms
With the dispatching algorithm based on genetic algorithm etc..It is that a cluster is mapped on virtual reality machine based on clustering algorithms, but exists
Between cluster the problem of task duplication；Although Gene hepatitis B vaccine can pass through selection, the tight scheduling for intersecting, making a variation in dispatching algorithm
Obtain preferably convergence time, but it in the selection process randomness it is very high, it is impossible to determine next schedule job in advance, and
Intersect and generally require largely to calculate time overhead and computing cost during making a variation.
Therefore, in view of prior art exist the wasting of resources, the problems such as time overhead is big, how to generate suitable task and adjust
Degree scheme becomes those skilled in the art's technical problem urgently to be resolved hurrily and the emphasis studied all the time.
The content of the invention
For overcome prior art exist the wasting of resources, time overhead is big the problems such as, the invention provides section under cloud environment
The task scheduling approach generation method and device in workflow are learned, resource utilization is not only increased, and shorten science work
Make the run time flowed.
To realize above-mentioned technical purpose, the invention discloses the life of the task scheduling approach in scientific workflow under cloud environment
Into method, the task scheduling approach generation method comprises the following steps,
Step 1, obtain and need scheduled task；
Step 2, describe to treat scheduled task by the directed acyclic graph of Weighted Coefficients；In the directed acyclic graph, circle
For task node, represent being scheduled for task, the leading to while for communication, between the task that representative is scheduled of two circles of connection
Letter, the direction on the side represents the order of tasks carrying；
Step 3, in the directed acyclic graph, traveled through from start node to end node, obtain all paths；
Step 4, the time loss in each path is calculated, all paths are dropped according to the descending order of the time loss
Sequence sequence, the priority for determining path；
Step 5, task scheduling approach is generated according to the priority orders in path.
By the above-mentioned path priority sequencing schemes comprising task node, by task node, according to priority order is adjusted
Degree, the present invention is advanced by the earliest finish time of scientific workflow, the number of its required processor is reduced, it is achieved thereby that carrying
High resource utilization, the purpose for shortening scientific workflow run time.
Further, in step 5, the task scheduling approach meets following schedulable condition：Treat that scheduled task does not have
It was scheduled, and treats that the corresponding task node of scheduled task has not had predecessor node or the corresponding task of predecessor node
It is scheduled.
By above-mentioned schedulable condition, repetitive schedule is avoided the occurrence of in the task scheduling approach that generation can be made, from
And it is advanced by the earliest finish time of scientific workflow task.
Further, after current task is chosen, step 5 also includes judging the corresponding current task node of current task
Predecessor node number the step of：
If predecessor node number=0, in task scheduling approach, current task is distributed to first in cloud environment
Processor；
If the predecessor node number=1 and corresponding task of the predecessor node has been scheduled, in task scheduling approach
In, current task is individually distributed to the second processing machine in cloud environment, and by where the corresponding task of the predecessor node
All Task Duplications on processor are to the second processing machine where current task；
If the predecessor node number >=2 and corresponding task of all forerunner's nodes has been scheduled, in task dispatching party
In case, all predecessor nodes of current task node are reached current according to from start node along each path by each predecessor node
The total time of task node carries out descending sort, forms father's task scheduling queue, by current task and father's task merging of head of the queue,
It is described to merge into：By head of the queue father's task all Task Duplications on treaters to processor where current task.
Can determine that by way of above-mentioned determination predecessor node number needs the number of processor in task scheduling approach, because
This, the present invention can determine that needs how many processor before task scheduling.Meanwhile, it can be reduced by modes such as duplication, merging
The quantity of processor, improves the utilization rate of processor.
Further, in step 5, if predecessor node number >=2, the father's task for meeting insertion condition is adjusted
Spend on the processor where father's task node outside the head of the queue in queue is inserted into current task；The insertion condition is：Currently
The free time of processor where task is more than father's task outside father's task node the calculating time overhead, and head of the queue outside head of the queue
The earliest finish time of node the earliest start time earlier than current task node.
It is negligible due to being communicated between the business on same processor, therefore appointed some fathers by above-mentioned interleaved plan
The free time that business is inserted on the processor where current task and made some father's tasks to manage machine in this place performs, after execution
Inform that current task is performed, notifying time greatly shortens, almost nil, so as to reduce intertask communication expense, shorten whole
The execution time of individual workflow task, and reduce the quantity of processor.
Further, in step 5, if current path be not present meet default schedulable condition treat scheduled appoint
Business, then select next paths according to the priority orders in path, until the scheduling scheme for needing scheduled task is whole
Generation.
Further, in step 5, until after all path processing terminate, generating task scheduling approach；If the task of generation
There is the scheduling scheme without practical significance in scheduling scheme, then delete the scheduling scheme of no practical significance, retain remaining task tune
Degree scheme.
Further, in step 2, in the directed acyclic graph, predecessor task node correspondence task the execution time earlier than
The execution time of current task node correspondence task.
Further, in step 2, in the directed acyclic graph, including between task computation time overhead mark and task
Call duration time expense is marked.
Further, in step 4, the time loss in each path is on the task computation time overhead on path and path
Call duration time expense sum between task.
Another goal of the invention of the present invention is the provision of the task scheduling side under a kind of cloud environment in scientific workflow
Case generating means, the task scheduling approach generating means include：
Task acquisition module, scheduled task is needed for obtaining；
Task description module, describes to treat scheduled task by the directed acyclic graph of Weighted Coefficients；The directed acyclic graph
In, circle is task node, represents scheduled task, two circles of connection while for communication, represent being scheduled for task
Between communication, the direction on the side represents the order of tasks carrying；
Node traverses module, in the directed acyclic graph, for being traveled through from start node to end node, is owned
Path；
Paths ordering module, the time loss for calculating each path, according to the order that the time loss is descending
By all path descending sorts, determine the priority in path；
Schemes generation module, task scheduling approach is generated according to the priority orders in path.
By the above-mentioned path priority sequencing schemes comprising task node, by task node, according to priority order is adjusted
Degree, the present invention is advanced by the earliest finish time of scientific workflow, the number of its required processor is reduced, it is achieved thereby that carrying
High resource utilization, the purpose for shortening scientific workflow run time.
Beneficial effects of the present invention are：The present invention can make full use of the free time of processor, improve scientific workflow
The concurrency of tasks carrying, improves processor utillization, shortens the deadline of scientific workflow task, does sth. in advance scientific workflow
Earliest finish time.
Brief description of the drawings
Fig. 1 is the task scheduling approach generation method schematic flow sheet in scientific workflow under cloud environment of the present invention.
Fig. 2 is the task scheduling approach generating means structural representation in scientific workflow under cloud environment of the present invention.
Fig. 3 is directed acyclic graph composition schematic diagram of the present invention using Weighted Coefficients.
Embodiment
Detailed explanation and illustration is carried out to the generation method and device of the present invention with reference to Figure of description.
As shown in Figure 1, 2, 3, the invention discloses the generation of the task scheduling approach in scientific workflow under a kind of cloud environment
Method, the present invention is innovatively by list scheduling, Task Duplication, task Intercalation, and can be applicable to that processor number do not limit is same
Task scheduling strategy is performed under structure environment.For convenience of the explanation of postorder, the present invention is defined as follows.
The executive plan generation of scientific workflow is reasonably to be mapped to workflow task in computing resource, and the present invention is related to
And computing resource be processor；Mapping, T={ t are represented with Fi, tiIt is the task in workflow, i=1,2,3 ..., n, n is appoints
Business number }；V={ vj, vjIt is the processor in cloud environment, j=1,2,3 ..., m, m is processor number }；
F (T, V)=<ti,vj>, ti∈ T, vj∈V}；
Specifically, task scheduling approach generation method of the invention comprises the following steps.
Step 1, obtain and need scheduled task, which of scientific workflow task need to be read in advance to be needed to be adjusted
Degree.
Step 2, describe to treat scheduled task by the directed acyclic graph of Weighted Coefficients, including each task composition and task it
Between correspondence.In directed acyclic graph, circle is task node, represents scheduled task, connects the side of two circles
For the communication between the scheduled task of communication side, representative, the direction on side represents the order of tasks carrying, and directed acyclic graph is also wrapped
The call duration time expense included between task computation time overhead mark and task is marked, specific as follows.
There is the relation mutually restricted, such as, predecessor task node pair in the communication between the task of each in scientific workflow
The execution time of task should be corresponded to earlier than current task node by answering the execution time of task, specifically, as shown in figure 3, of the invention
Described using the directed acyclic graph (DAG) of Weighted Coefficients, as shown in figure 3, i.e. G (T, E, M, W) come represent the dependence between task close
System.Wherein, T represents the set of task node in workflow, ti∈ T represent a task in workflow；E represents task node
Between communicate side set, eij∈ E represent ti、tkBetween side, tiIt is tkForerunner's node；M represents the collection of intertask communication expense
Close, mik∈ M represent task tiWith task tkBetween call duration time expense, call duration time expense is marked at by the side in DAG；W tables
Show the set of task computation expense, ci∈ W represent tiCalculating time overhead, task computation time overhead is marked at the circle in DAG
Circle is other.On the basis of above-mentioned directed acyclic graph, and it is defined as follows.
Utilize est (ti) and eft (ti) task t is represented respectivelyiEarliest start time and earliest finish time, start knot
Earliest start time est (the t of pointi)=0, task tiEft (t on earliest finish timei)=est (ti)+ci, whole workflow
Earliest finish time makespan=eft (tend), wherein, tendTo end task.
For task ti, parameter pre (ti) represent task tiPredecessor node.In processor (vm) on, by a upper task
The free time that end of run runs beginning to next task is defined as Free, is expressed as Free (tpre(ti),i, vm)=est
(ti, vm)-eft(pre(ti, vm))
Parameter pxThe path in DAG figures is represented, span represents the deadline in path, Span (px)=Σti∈Px(ci+
mij), reach (tpre(ti),i) represent task tiPredecessor task reach task tiTime, reach (tpre(ti),i) there are two kinds of feelings
Condition.The first situation：Task tiAnd its predecessor task is on same virtual machine, now reach (tpre(ti),i)=Σti∈px
(cpre(ti)+ci)+mpre(ti),i；Second of situation：Task tiAnd its predecessor task is not on a virtual machine, now, reach
(tpre(ti),i) and deadline Span (px) identical.
Step 3, in directed acyclic graph, traveled through from start node to end node, all possible road in traversal DAG figures
Footpath, and then obtain all paths.
Step 4, the time loss Span values in each path are calculated, according to the descending order of time loss by all paths
Descending sort, the priority for determining path, such as, priority is represented by way of generating heat source order list, by when
Between consume maximum path as critical path, preferentially judge whether the task in critical path is scheduled；Each path when
Between consumption be the task computation time overhead on path and the call duration time expense sum between task, formula is expressed as：Span
(px)=Σti∈Px(ci+mij)。
Step 5, task scheduling approach is generated according to the priority orders in path, circulating path, until at all path
After reason terminates, task scheduling approach is generated, then when traffic control starts, the scheme generated according to the present invention is scheduled, and is pressed
Task is assigned on processor by order；Specifically, during schemes generation, if current path is not present meets default
Schedulable condition treats scheduled task, then according to the list of heat source order, it is next according to the priority orders selection in path
Paths, until the scheduling scheme for needing scheduled task is all generated.
To realize scientific and rational scheduling, task scheduling approach of the invention meets following schedulable condition：Wait to be scheduled
Task be not scheduled, and treat that the corresponding task node of scheduled task does not have predecessor node or predecessor node corresponding
Task has been scheduled.Meet the task t of the two conditionsiIt can just be scheduled, then select other in current path meet
The task of condition, this is met if since next paths, continuing selection if the task of condition is not met in current path
The task of part.
To determine that the task in scientific workflow is needed in the quantity of processor, this step 5, in task tiAfter being chosen,
Present invention additionally comprises the step of the predecessor node number for judging the corresponding current task node of current task：
If predecessor node number=0, in task scheduling approach, current task is distributed to first in cloud environment
Processor.
If the predecessor node number=1 and corresponding task of the predecessor node has been scheduled, in task scheduling approach
In, current task individually distributed to the second processing machine in cloud environment, and by the processing where the corresponding task of predecessor node
All Task Duplications on machine are to the second processing machine where current task.
If the predecessor node number >=2 and corresponding task of all forerunner's nodes has been scheduled, in task dispatching party
In case, all predecessor nodes of current task node are reached current according to from start node along each path by each predecessor node
The total time of task node carries out descending sort, forms father's task scheduling queue, by current task and father's task merging of head of the queue,
The merging is interpreted as：By head of the queue father's task all Task Duplications on treaters to processor where current task,
And update the deadline of current task.The present invention combines the thought of list scheduling and Task Duplication, selects rational father's task
Replicated, and then expense, the quantity of reduction processor between reduction task.Moreover, based between task in same processor
The negligible principle of expense, during the free time that some tasks are sent to the processor of existing task and machine are managed in this place
Between perform, if predecessor node number >=2, the father's task section outside head of the queue that will be met in father's task scheduling queue of insertion condition
Point is inserted on the processor where current task, updates the deadline for the father's task for meeting insertion condition；Insertion condition is：
The free time of processor where current task is more than the father outside father's task node the calculating time overhead, and head of the queue outside head of the queue
The earliest finish time of task node the earliest start time earlier than current task node.The insertion condition is formulated
For：Free(ti, vk)≥cpre(ti)And eft (pre (ti),vk)<est(ti, vk).Until after all path processing terminate, generation
Task scheduling approach；There is the scheduling scheme without practical significance in task scheduling approach if generated, delete no practical significance
Scheduling scheme, retain remaining task scheduling approach.
The free time piece of suitable father's task insertion process machine is judged that all fathers appoint outside all heads of the queue by the present invention successively
Business node, untill current processor free time deficiency is unsatisfactory for father's task computation time overhead or no father's task.This
The mode of kind can improve the utilization rate of processor, reduce the quantity of processor.Above-mentioned scheduling scheme can be regarded as the pre- of task
Distribution, can be by duty mapping to suitable processor, and can determine that the quantity of processor.
It should be noted that " the first processor ", " second processing machine " that is related in the present invention is not to number or excellent
The restriction of first level, simply plays the purpose for distinguishing nonidentical processor, and according to mission requirements, " second processing machine " can be many
It is individual.
The generation method of the above-mentioned scheduling scheme of correspondence, as shown in Figures 2 and 3, the invention also discloses under a kind of cloud environment
Task scheduling approach generating means in scientific workflow, task scheduling approach generating means include：
Task acquisition module, needs scheduled task for obtaining, which of scientific workflow need to be read in advance
Task needs scheduled.
Task description module, describes to treat scheduled task by the directed acyclic graph of Weighted Coefficients；In directed acyclic graph, circle
Enclose as task node, represent scheduled task, two circles of connection while for communication, represent between being scheduled for task
Communication, the direction on side represents the order of tasks carrying；Communication between the task of each in scientific workflow has what is mutually restricted
Relation, such as, the execution time that predecessor task node corresponds to task should correspond to the execution time of task earlier than current task node.
Node traverses module, in directed acyclic graph, for being traveled through from start node to end node, obtains all roads
Footpath.
Paths ordering module, the time loss for calculating each path, according to the descending order of time loss by institute
There is path descending sort, determine the priority in path；Such as, represent preferential by way of generating heat source order list
Level, using the maximum path of time loss as critical path, preferentially judges whether the task in critical path is scheduled；Each road
The time loss in footpath is the task computation time overhead on path and the call duration time expense sum between task.
Schemes generation module, task scheduling approach is generated according to the priority orders in path, circulating path, until all
After path processing terminates, task scheduling approach is generated, then when traffic control starts, the scheme generated according to the present invention is adjusted
Task, is assigned on processor by degree in order；Specifically, during schemes generation, met if current path is not present
Default schedulable condition treats scheduled task, then is selected according to the list of heat source order, according to the priority orders in path
Next paths are selected, until the scheduling scheme for needing scheduled task is all generated.
With reference to concrete example to the task scheduling approach generation method in scientific workflow under cloud environment of the present invention and
Device carries out detailed explanation and illustration.With reference to Figure of description 3 and following path dispatch lists, the priority in path is suitable
After sequence is determined, pre-scheduling is carried out to task.
| Priority | Task path px | Deadline (Span) | Scheduled task |
| 1 | t1, t2, t7, t9 | 26 | t1、t2、t7 |
| 2 | t1, t4, t8, t9 | 24 | t4 |
| 3 | t1, t3, t8, t9 | 23 | t3、t8 |
| 4 | t1, t2, t6, t9 | 22 | t6、t9 |
| 5 | t1, t7, t9 | 16 | —— |
| 6 | t1, t5 | 13 | t5 |
For task path p1, the task t in circulating path 11, t first1Without father's task and without forerunner's section in path 1
Point, then t1Schedulable condition is met, secondly t1There is no predecessor node in DAG figures, so by t1It is separately positioned at a processor
On, then v11={ t1}；Continuation task t2, the t in path 12Father's task be scheduled and t2It was not scheduled, and met scheduling
Condition, the t in DAG figures2Only one of which predecessor node, by all tasks on the processor where the corresponding task of predecessor node
The second processing machine where current task is copied to, then v12={ t1、t2}；Continuation task t7, the t in path 17It is not scheduled
Cross and t7Father's task be scheduled, meet schedulable condition, the t in DAG figures7There are two predecessor node t1、t2And before the two
Drive node to be scheduled, t is calculated respectively1、t2Reach t7Time, reach (t1,7)=2+3=5, wherein " 2 " are t1Calculate
Time overhead, " 3 " are t1To t7Call duration time expense；reach(t2,7)=2+3+6=11, wherein " 2 " are t1The calculating time opens
Pin, " 3 " are t2Calculate time overhead, t2And t1It is placed on same processor, so, call duration time expense therebetween
Almost nil, " 6 " are t2To t7Call duration time expense；Then by t7With t2Merge and update the deadline, then v13={ t1、t2、t7}；
Continue the next task t in path9, the t in path 19Father's task be scheduled, but the t in DAG figures9Predecessor node
t6、t8It is not scheduled, so jumping out path 1.
For task path p2, continue cycling through path 2.t1It has been be scheduled that, skip t1.Continuation task t4, t4Meet scheduling bar
Part, and t4The predecessor node number of corresponding node is 1 and the corresponding task of predecessor node has been scheduled, by predecessor node correspondence
Task where processor on all Task Duplications to the second processing machine where current task, then v14={ t1、t4}.After
Continuous task t8, task t8Although not being scheduled, the task t in DAG figures8Predecessor node t3It is not scheduled, then jumps
Cross task t8；Similarly, task t is skipped9。
For task path p3, continue cycling through path 3, t1It has been scheduled, has skipped t1；As the task t being recycled in path 33
When, task t3It was not scheduled and the corresponding task of predecessor node has been scheduled, and met schedulable condition, task t3Forerunner saves
Point number is 1, by predecessor node t1All Task Duplications on processor where corresponding task are to where current task
Second processing machine, then v15={ t1、t3}；As the task t being recycled in path 38, now t8Predecessor task t3、t4Adjusted
Degree, calculates t respectively3、t4Reach t8Time.reach(t3,8)=2+3+5=10, reach (t4,8)=2+4+3=9.T3、t4
Sorted { t by arrival time3、t4, by t8With t3Merge and update deadline, now, processor v16={ t1、t3、t8, eft
(t3)=5, est (t8)=9, Free (t are obtained according to formula 23,8, vk)=9-5=4 >=c4=4, and eft (t4)=6<est(t8)
=9 meet insertion condition, by t4Insert reason machine task queue v16={ t1、t3、t8In, then the v after inserting16={ t1、t3、t4、
t8, continue next task t9, schedulable condition is not met, then jumps out path 3, continues path 4.
For task path p4, continue cycling through path 4, t1、t2It has been be scheduled that, then skipped, task t6Only one of which forerunner appoints
Be engaged in t2, then by t2Institute is on treaters from t2Task Duplication before is arrived and t6On single processor together, v17={ t1、
t2、t6}.Continue, t9The same t of method8It is identical, t6Reach t9Time be 15, t7Reach t9Time be 14, t8Reach t9When
Between be 16, then by t8And t9Merge, then v18={ t1、t3、t4、t8、t9, and t6、t7Insertion condition is not met.
For task path p5, path 5 is continued cycling through, without the node for meeting schedulable condition, path p is skipped5。
For task path p6, path 6 is continued cycling through, the t in path 65Father's task be scheduled and t5It is not scheduled
Cross, meet schedulable condition, the t in DAG figures5Only one of which predecessor node, so by the place where the corresponding task of predecessor node
All Task Duplications on reason machine are to the second processing machine where current task, then v19={ t1、t5}。
It is by scheduling result obtained above, v11={ t1, v12={ t1、t2, v13={ t1、t2、t7, v14={ t1、
t4, v15={ t1、t3, v16={ t1、t3、t4、t8, v17={ t1、t2、t6, v18={ t1、t3、t4、t8、t9, v19={ t1、
t5}；There is scheduling scheme without practical significance or result in above-mentioned scheduling result, only task is repeated, such as
v11={ t1, v12={ t1、t2Etc., so screening out scheduling knot of its whole task in the carrying into execution a plan of other processors
Really, then the end product for actual task scheduling scheme is：v1={ t1、t3、t4、t8、t9, v2={ t1、t2、t7, v3=
{t1、t2、t6, v4={ t1、t5, and makespan=eft (t9)=16.
It should be noted that for convenience of present invention explanation and describe, in DAG figures, tiIt can be regarded as the task corresponding
Node.Moreover, it relates to various " times " be relative time, can be by task t1At the beginning of between be interpreted as 0.
In the description of this specification, reference term " one embodiment ", " some embodiments ", " the present embodiment ", " specific
The description of example " or " some examples " etc. mean to combine the specific features that the embodiment or example describe, structure, material or
Feature is contained at least one embodiment of the present invention or example.In this manual, to the schematic representation of above-mentioned term
Necessarily it is directed to identical embodiment or example.Moreover, specific features, structure, material or the feature of description can be
Combined in an appropriate manner in any one or more embodiments or example.In addition, in the case of not conflicting, this area
Technical staff can be carried out the feature of the not be the same as Example described in this specification or example and non-be the same as Example or example
With reference to and combination.
The foregoing is merely illustrative of the preferred embodiments of the present invention, is not intended to limit the invention, all in essence of the invention
Any modification, equivalent substitution and simple modifications for being made in content etc., should be included in the scope of the protection.
Claims (10)
1. the task scheduling approach generation method under cloud environment in scientific workflow, it is characterised in that：The task scheduling approach
Generation method comprises the following steps,
Step 1, obtain and need scheduled task；
Step 2, describe to treat scheduled task by the directed acyclic graph of Weighted Coefficients；In the directed acyclic graph, circle is to appoint
The scheduled task of business node, representative, communication while for communication, between the task that representative is scheduled of two circles of connection,
The direction on the side represents the order of tasks carrying；
Step 3, in the directed acyclic graph, traveled through from start node to end node, obtain all paths；
Step 4, the time loss in each path is calculated, all path descendings are arranged according to the descending order of the time loss
Sequence, the priority for determining path；
Step 5, task scheduling approach is generated according to the priority orders in path.
2. the task scheduling approach generation method under cloud environment according to claim 1 in scientific workflow, its feature exists
In：In step 5, the task scheduling approach meets following schedulable condition：Treat that scheduled task was not scheduled, and treat
The corresponding task node of scheduled task does not have predecessor node or the corresponding task of predecessor node to be scheduled.
3. the task scheduling approach generation method under cloud environment according to claim 2 in scientific workflow, its feature exists
In：After current task is chosen, step 5 also includes the predecessor node number for judging the corresponding current task node of current task
The step of：
If predecessor node number=0, in task scheduling approach, current task is distributed to the first processing in cloud environment
Machine；
, will in task scheduling approach if the predecessor node number=1 and corresponding task of the predecessor node has been scheduled
Current task individually distributes to the second processing machine in cloud environment, and by the processor where the corresponding task of the predecessor node
On all Task Duplications to the second processing machine where current task；
If the predecessor node number >=2 and corresponding task of all forerunner's nodes has been scheduled, in task scheduling approach,
All predecessor nodes of current task node are reached into current task according to from start node along each path by each predecessor node
Total time of node carries out descending sort, forms father's task scheduling queue, by current task and father's task merging of head of the queue, described
Merge into：By head of the queue father's task all Task Duplications on treaters to processor where current task.
4. the task scheduling approach generation method under cloud environment according to claim 3 in scientific workflow, its feature exists
In：In step 5, if predecessor node number >=2, the head of the queue in father's task scheduling queue of insertion condition will be met
Outer father's task node is inserted on the processor where current task；The insertion condition is：Processing where current task
The free time of machine is more than the earliest completion that father's task node outside head of the queue calculates father's task node outside time overhead, and head of the queue
Earliest start time of the time earlier than current task node.
5. the task scheduling approach under the cloud environment according to any claim in claim 2 to 4 in scientific workflow
Generation method, it is characterised in that：In step 5, if current path be not present meet default schedulable condition wait be scheduled
Task, then next paths are selected according to the priority orders in path, until needing the scheduling scheme of scheduled task
All generations.
6. according to the task scheduling approach generation method in scientific workflow under claim 1 or 4 or described cloud environment, it is special
Levy and be：In step 5, until after all path processing terminate, generating task scheduling approach；If generated in task scheduling approach
In the presence of the scheduling scheme without practical significance, then delete the scheduling scheme of no practical significance, retain remaining task scheduling approach.
7. the task scheduling approach generation method under cloud environment according to claim 1 in scientific workflow, its feature exists
In：In step 2, in the directed acyclic graph, the execution time of predecessor task node correspondence task is earlier than current task node pair
Answer the execution time of task.
8. the task scheduling approach generation method under cloud environment according to claim 1 in scientific workflow, its feature exists
In：In step 2, in the directed acyclic graph, including the call duration time expense mark between task computation time overhead mark and task
Note.
9. the task scheduling approach generation method under cloud environment according to claim 8 in scientific workflow, its feature exists
In：In step 4, when the time loss in each path is the communication between the task on task computation time overhead and path on path
Between expense sum.
10. the task scheduling approach generating means under cloud environment in scientific workflow, it is characterised in that：The task scheduling approach
Generating means include：
Task acquisition module, scheduled task is needed for obtaining；
Task description module, describes to treat scheduled task by the directed acyclic graph of Weighted Coefficients；In the directed acyclic graph, circle
Enclose as task node, represent scheduled task, two circles of connection while for communication, represent between being scheduled for task
Communication, the direction on the side represents the order of tasks carrying；
Node traverses module, in the directed acyclic graph, for being traveled through from start node to end node, obtains all roads
Footpath；
Paths ordering module, the time loss for calculating each path, according to the descending order of the time loss by institute
There is path descending sort, determine the priority in path；
Schemes generation module, task scheduling approach is generated according to the priority orders in path.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710202301.6A CN107015856A (en) | 2017-03-30 | 2017-03-30 | Task scheduling approach generation method and device under cloud environment in scientific workflow |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710202301.6A CN107015856A (en) | 2017-03-30 | 2017-03-30 | Task scheduling approach generation method and device under cloud environment in scientific workflow |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN107015856A true CN107015856A (en) | 2017-08-04 |
Family
ID=59445562
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710202301.6A Withdrawn CN107015856A (en) | 2017-03-30 | 2017-03-30 | Task scheduling approach generation method and device under cloud environment in scientific workflow |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107015856A (en) |
Cited By (21)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107943561A (en) * | 2017-12-14 | 2018-04-20 | 长春工程学院 | A kind of scientific workflow method for scheduling task towards cloud computing platform |
| CN108628672A (en) * | 2018-05-04 | 2018-10-09 | 武汉轻工大学 | Method for scheduling task, system, terminal device and storage medium |
| CN108647084A (en) * | 2018-05-08 | 2018-10-12 | 武汉轻工大学 | Efficiency cloud method for scheduling task |
| CN108733473A (en) * | 2018-05-11 | 2018-11-02 | 北京航天发射技术研究所 | A kind of positioning based on VxWorks aims at the control method of integration apparatus task |
| CN108762899A (en) * | 2018-05-16 | 2018-11-06 | 武汉轻工大学 | A kind of cloud task rescheduling method and apparatus |
| CN108897625A (en) * | 2018-07-06 | 2018-11-27 | 陈霖 | Method of Scheduling Parallel based on DAG model |
| CN109783213A (en) * | 2018-12-28 | 2019-05-21 | 杭州电子科技大学 | The workflow fault-tolerant scheduling method of reliability is directed under a kind of edge calculations environment |
| CN109918182A (en) * | 2019-01-23 | 2019-06-21 | 中国人民解放军战略支援部队信息工程大学 | More GPU task dispatching methods under virtualization technology |
| CN110019986A (en) * | 2018-09-27 | 2019-07-16 | 福建福链科技有限公司 | The method for sequencing and terminal of message in a kind of directed acyclic graph |
| CN110084507A (en) * | 2019-04-24 | 2019-08-02 | 信雅达系统工程股份有限公司 | The scientific workflow method for optimizing scheduling of perception is classified under cloud computing environment |
| CN110147270A (en) * | 2019-04-15 | 2019-08-20 | 平安普惠企业管理有限公司 | Database dispatching method, device, computer equipment and storage medium |
| CN110489214A (en) * | 2019-06-19 | 2019-11-22 | 南京邮电大学 | The dynamic task allocation of data-intensive workflow in cloud environment |
| CN111026534A (en) * | 2019-12-10 | 2020-04-17 | 浙江工业大学 | Workflow execution optimization method based on multi-population genetic algorithm in cloud computing environment |
| CN111798063A (en) * | 2020-07-09 | 2020-10-20 | 北京理工大学 | Research and development resource allocation method based on operation double-layer redundancy time reduction |
| CN111861412A (en) * | 2020-07-27 | 2020-10-30 | 上海交通大学 | Completion time optimization-oriented scientific workflow scheduling method and system |
| CN112162727A (en) * | 2020-09-16 | 2021-01-01 | 深圳晶泰科技有限公司 | Cloud high-performance scientific computing workflow design control system and user graphical interface |
| CN112363819A (en) * | 2020-12-02 | 2021-02-12 | 深圳市房多多网络科技有限公司 | Big data task dynamic scheduling method and device and computing equipment |
| CN112379998A (en) * | 2020-10-20 | 2021-02-19 | 广东工业大学 | Switching time delay optimization method based on task fusion and task scheduling |
| CN112633753A (en) * | 2020-12-30 | 2021-04-09 | 广东赛诺科技股份有限公司 | Dynamic work order sharing system |
| CN113742089A (en) * | 2021-11-04 | 2021-12-03 | 苏州浪潮智能科技有限公司 | Method, device and equipment for distributing neural network computing tasks in heterogeneous resources |
| CN117610320A (en) * | 2024-01-23 | 2024-02-27 | 中国人民解放军国防科技大学 | Directed acyclic graph workflow engine cyclic scheduling method, device and equipment |
-
2017
- 2017-03-30 CN CN201710202301.6A patent/CN107015856A/en not_active Withdrawn
Cited By (31)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107943561B (en) * | 2017-12-14 | 2019-06-11 | 长春工程学院 | A kind of scientific workflow method for scheduling task towards cloud computing platform |
| CN107943561A (en) * | 2017-12-14 | 2018-04-20 | 长春工程学院 | A kind of scientific workflow method for scheduling task towards cloud computing platform |
| CN108628672A (en) * | 2018-05-04 | 2018-10-09 | 武汉轻工大学 | Method for scheduling task, system, terminal device and storage medium |
| CN108647084A (en) * | 2018-05-08 | 2018-10-12 | 武汉轻工大学 | Efficiency cloud method for scheduling task |
| CN108733473A (en) * | 2018-05-11 | 2018-11-02 | 北京航天发射技术研究所 | A kind of positioning based on VxWorks aims at the control method of integration apparatus task |
| CN108762899B (en) * | 2018-05-16 | 2020-05-15 | 武汉轻工大学 | Cloud task rescheduling method and device |
| CN108762899A (en) * | 2018-05-16 | 2018-11-06 | 武汉轻工大学 | A kind of cloud task rescheduling method and apparatus |
| CN108897625A (en) * | 2018-07-06 | 2018-11-27 | 陈霖 | Method of Scheduling Parallel based on DAG model |
| CN110019986A (en) * | 2018-09-27 | 2019-07-16 | 福建福链科技有限公司 | The method for sequencing and terminal of message in a kind of directed acyclic graph |
| CN110019986B (en) * | 2018-09-27 | 2020-12-29 | 福建福链科技有限公司 | Method and terminal for sequencing messages in directed acyclic graph |
| CN109783213A (en) * | 2018-12-28 | 2019-05-21 | 杭州电子科技大学 | The workflow fault-tolerant scheduling method of reliability is directed under a kind of edge calculations environment |
| CN109783213B (en) * | 2018-12-28 | 2021-02-19 | 杭州电子科技大学 | Workflow fault tolerance scheduling method for reliability in edge computing environment |
| CN109918182B (en) * | 2019-01-23 | 2020-12-15 | 中国人民解放军战略支援部队信息工程大学 | Multi-GPU task scheduling method under virtualization technology |
| CN109918182A (en) * | 2019-01-23 | 2019-06-21 | 中国人民解放军战略支援部队信息工程大学 | More GPU task dispatching methods under virtualization technology |
| CN110147270A (en) * | 2019-04-15 | 2019-08-20 | 平安普惠企业管理有限公司 | Database dispatching method, device, computer equipment and storage medium |
| CN110084507A (en) * | 2019-04-24 | 2019-08-02 | 信雅达系统工程股份有限公司 | The scientific workflow method for optimizing scheduling of perception is classified under cloud computing environment |
| CN110489214A (en) * | 2019-06-19 | 2019-11-22 | 南京邮电大学 | The dynamic task allocation of data-intensive workflow in cloud environment |
| CN110489214B (en) * | 2019-06-19 | 2022-09-20 | 南京邮电大学 | Dynamic task allocation for data intensive workflows in a cloud environment |
| CN111026534A (en) * | 2019-12-10 | 2020-04-17 | 浙江工业大学 | Workflow execution optimization method based on multi-population genetic algorithm in cloud computing environment |
| CN111026534B (en) * | 2019-12-10 | 2023-12-22 | 浙江工业大学 | Workflow execution optimization method based on multiple group genetic algorithms in cloud computing environment |
| CN111798063A (en) * | 2020-07-09 | 2020-10-20 | 北京理工大学 | Research and development resource allocation method based on operation double-layer redundancy time reduction |
| CN111861412A (en) * | 2020-07-27 | 2020-10-30 | 上海交通大学 | Completion time optimization-oriented scientific workflow scheduling method and system |
| CN111861412B (en) * | 2020-07-27 | 2024-03-15 | 上海交通大学 | Completion time optimization-oriented scientific workflow scheduling method and system |
| CN112162727A (en) * | 2020-09-16 | 2021-01-01 | 深圳晶泰科技有限公司 | Cloud high-performance scientific computing workflow design control system and user graphical interface |
| CN112379998B (en) * | 2020-10-20 | 2022-11-15 | 广东工业大学 | Switching time delay optimization method based on task fusion and task scheduling |
| CN112379998A (en) * | 2020-10-20 | 2021-02-19 | 广东工业大学 | Switching time delay optimization method based on task fusion and task scheduling |
| CN112363819A (en) * | 2020-12-02 | 2021-02-12 | 深圳市房多多网络科技有限公司 | Big data task dynamic scheduling method and device and computing equipment |
| CN112633753A (en) * | 2020-12-30 | 2021-04-09 | 广东赛诺科技股份有限公司 | Dynamic work order sharing system |
| CN113742089A (en) * | 2021-11-04 | 2021-12-03 | 苏州浪潮智能科技有限公司 | Method, device and equipment for distributing neural network computing tasks in heterogeneous resources |
| CN117610320A (en) * | 2024-01-23 | 2024-02-27 | 中国人民解放军国防科技大学 | Directed acyclic graph workflow engine cyclic scheduling method, device and equipment |
| CN117610320B (en) * | 2024-01-23 | 2024-04-02 | 中国人民解放军国防科技大学 | Directed acyclic graph workflow engine cyclic scheduling method, device and equipment |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107015856A (en) | Task scheduling approach generation method and device under cloud environment in scientific workflow | |
| Nie et al. | A GEP-based reactive scheduling policies constructing approach for dynamic flexible job shop scheduling problem with job release dates | |
| Gamache et al. | A method for optimally solving the rostering problem | |
| El-Rewini et al. | Scheduling parallel program tasks onto arbitrary target machines | |
| Roshani et al. | A simulated annealing algorithm for multi-manned assembly line balancing problem | |
| Fan et al. | An improved genetic algorithm for flexible job shop scheduling problem considering reconfigurable machine tools with limited auxiliary modules | |
| CN101237469B (en) | Method for optimizing multi-QoS grid workflow based on ant group algorithm | |
| CN109582448A (en) | A kind of edge calculations method for scheduling task towards criticality and timeliness | |
| Lequy et al. | Assigning multiple activities to work shifts | |
| CN103279818A (en) | Method for cloud workflow scheduling based on heuristic genetic algorithm | |
| CN106934539B (en) | Workflow scheduling method with deadline and expense constraints | |
| Lu et al. | A resource investment problem based on project splitting with time windows for aircraft moving assembly line | |
| CN107943561B (en) | A kind of scientific workflow method for scheduling task towards cloud computing platform | |
| Colak et al. | Multi-mode resource-constrained project-scheduling problem with renewable resources: new solution approaches | |
| Li et al. | Single machine scheduling with two-agent for total weighted completion time objectives | |
| WO2022198754A1 (en) | Method for optimizing large-scale cloud service process | |
| CN110084507A (en) | The scientific workflow method for optimizing scheduling of perception is classified under cloud computing environment | |
| Đurasević et al. | Heuristic and metaheuristic methods for the unrelated machines scheduling problem: A survey | |
| Gong et al. | Energy-efficient flexible job shop scheduling problem considering discrete operation sequence flexibility | |
| Furmans | Models of heijunka-levelled kanban-systems | |
| CN109783189A (en) | A kind of quiescent operation stream scheduling method and device | |
| CN109635328A (en) | Integrated circuit layout method and distributed design approach | |
| CN109034551A (en) | Equilibrium of stock method for construction plan establishment | |
| Li | Improved NSGA-II for the job-shop multi-objective scheduling problem | |
| Moin | Hybrid genetic algorithms for vehicle routing problems with time windows |
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 | ||
| WW01 | Invention patent application withdrawn after publication | ||
| WW01 | Invention patent application withdrawn after publication |
Application publication date: 20170804 |