CN106934537A - The sub- time limit based on the scheduling of reverse operation stream obtains optimization method - Google Patents
The sub- time limit based on the scheduling of reverse operation stream obtains optimization method Download PDFInfo
- Publication number
- CN106934537A CN106934537A CN201710121606.4A CN201710121606A CN106934537A CN 106934537 A CN106934537 A CN 106934537A CN 201710121606 A CN201710121606 A CN 201710121606A CN 106934537 A CN106934537 A CN 106934537A
- Authority
- CN
- China
- Prior art keywords
- task
- time
- resource
- sub
- time limit
- 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.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
- G06Q10/063—Operations research, analysis or management
- G06Q10/0631—Resource planning, allocation, distributing or scheduling for enterprises or organisations
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
- G06Q10/063—Operations research, analysis or management
- G06Q10/0633—Workflow analysis
Landscapes
- Business, Economics & Management (AREA)
- Human Resources & Organizations (AREA)
- Engineering & Computer Science (AREA)
- Strategic Management (AREA)
- Entrepreneurship & Innovation (AREA)
- Economics (AREA)
- Operations Research (AREA)
- Game Theory and Decision Science (AREA)
- Development Economics (AREA)
- Marketing (AREA)
- Educational Administration (AREA)
- Quality & Reliability (AREA)
- Tourism & Hospitality (AREA)
- Physics & Mathematics (AREA)
- General Business, Economics & Management (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The sub- time limit based on the scheduling of reverse operation stream obtains optimization method and belongs to grid or field of cloud calculation, it is an important parameter that the workflow of limited constraint dispatches former business sub- time limit on one group of fixed resource, the sub- time limit acquired in existing method can not ensure the completion of remaining task sometimes, and the present invention provides new method to obtain the reasonable available sub- time limit.For workflow schedule, the RHEFT methods of proposition are negated to weights and sorted to task first by the reverse thought of workflow, then earliest start time and deadline that task travels through resource acquisition inverse task one by one are taken out successively, are finally corresponded to and are obtained sub- time limit and late start time.This process not only considers remaining critical path depth, it is also contemplated that the influence of DAG degree of parallelisms so that each task is in the case where the sub- time limit constrains to still ensuring that remaining task has enough scheduling times after resource impact.Influence to the sub- time limit is changed by DAG degree of parallelisms, the sub- time limit for demonstrating present invention acquisition is more reasonable effective.
Description
Technical field
It is that a kind of workflow for being related to limited constraint exists specifically the invention belongs to grid computing, field of cloud calculation
The sub- time limit acquisition optimization method of former business is dispatched on one group of fixed resource.
Background technology
In the DAG workflows of limited constraint share one group of scheduling of static resource, the sub- time limit is every usually as judging
One of calculating parameter of individual task priority.In addition, relevant user dispatches the optimization of expense, in order to obtain most rational resource point
Match somebody with somebody, it is contemplated that each task allows the degree of Cost Optimization, the sub- time limit is typically an important indicator[1,2].Therefore, rationally obtain
The sub- time limit of each task has important practical significance.
With famous HEFT dispatching algorithms[3]Assuming that identical, it is assumed that a DAG multiple tasks node needs to be mapped to one group
Parallel Scheduling on heterogeneous distributed computing resource R.The node at each directed edge two ends corresponds between father, subtask, due to
Control and the presence of data dependence relation, subtask could perform after the completion of all of father's task.According to famous HEFT algorithms
It is assumed that each task niThe execution time in each computing resource spends known, and resource bandwidth is assumed to 1, cI, jExpression task
niIt is delivered to njData communication time spend, the intertask communication time in same resource of being mapped in is spent as 0.In DAG,
Not the having any father's task of the task is referred to as entrance task, is expressed as nentry, the task without any subtask is referred to as
Mouth task, is expressed as nexit.If given workflow contains more than one entrance task or export task in DAG, can
To generate the dummy entry task that an incidental expenses are taken.Cost is performed when such hypothesis does not interfere with workflow schedule[4]。
HEFT algorithms its key ideas is divided into two steps, and first is the data according to the execution time of task and with father's task
Transmission time is calculated this task to the ultimate range between export task, i.e., upward weights ranku(ni), such as formula
(1)：
Wherein：It is task niEach resource performs the average of time cost, succ (n in Ri) it is niDirect subtask
Set, njBelong to succ (ni), cI, jTask niWith its subtask njBetween data communication time spend,
Export task nexitUpward weights,It is export task nexitTime average is performed in each resource in R..
Task priority sequence is carried out according to this numerical value.
Second stage is the task ranking obtained according to first stage, and priority is most in choosing not scheduled task
Task high, travels through each processor, finds the processor that can earliest complete task, is inserted into available free time
In.
Existing document[1,2]Method on obtaining the DAG tasks sub- time limit is normally based on calculating current task and appoints to outlet
What the approximate critical path of business was defined, such as formula (2).
Wherein：succ(ni) it is task niAll direct subtask, njBelong to succ (ni),It is task niWith its institute
There is subtask njBetween data communication time spend average, LFT (nj) it is subtask njThe sub- time limit, min (wi) it is niNot
With the minimum execution time on processor.
This method for obtaining the sub- time limit in view of the how many pairs of influences in sub- time limit of parallel task, is not appointed weighing
Business urgency level aspect is not accurate enough, with larger improved space.
The content of the invention
The method that the present invention obtains the task sub- time limit for the DAG back schedulings proposed to limited constraint.The sub- time limit
Seek to find the permission deadline at the latest when all tasks are dispatched in one group of resource, that is, each task is as far as possible
Postpone and completing, until postponing again whole DAG can be caused to exceed the time limit.In order to maximum postponement for reaching each task completes, DAG is used
Reversely scheduling obtains earliest finish time and earliest start time one by one, is inverted by the time limit and can obtain the sub- time limit and transport at the latest
The row time started.The back scheduling strategy of DAG does reverse process to all of side in DAG first, i.e., direct subtask and directly
The role swap of father's task, is then scheduled using the HEFT algorithms for allowing each task to have earliest finish time again.It is actual
During scheduling, in order to more quickly obtain scheduling results of the reverse DAG using HEFT algorithms, and need not do actual reverse to DAG,
Only need to based on HEFT algorithms, realize the RHEFT methods of reverse HEFT scheduling.Comprise the following steps that：
Step (1)：Calculate each task niTo entrance task nentryMaximum average range parameter, that is, task ni
Reverse weights rankr(ni), such as formula (3), be reversely relative to HEFT algorithms in upward weights and define.
Wherein：prec(ni) represent task niAll direct father's task, npIt is task niOne of them direct father appoint
Business,It is task niTime average, c are performed in each resource in fixed resource group Rp,iFather's task npTo task niData lead to
The letter time spends, rankr(np) it is father's task npDownward weights.Because formula is to need iterative calculation, it is necessary first to calculated
Entrance task nentry, and nentryThere is no direct father's task, its reverse weights It is task nentry
In R time average is performed in each resource.
Step (2)：Task ranking.
According to the rank of each task in DAGr(ni) descending arrangement, obtain task to the dispatch map list of resource so that
Export task nexitPriority scheduling, entrance task nentryFinally dispatch.
Step (3)：Selection resource.
Take out first task in order from dispatch list, find that have earliest can the deadline from all resources
The resource timeslot of task.If first task is nexit, directly judging the minimum execution time of each resource can find resource；It is no
Then, due to first task n that be reverse, being taken outiAll subtasks have been mapped into resource, first to calculate niIt is all
Maximum delivered time of the subtask to each resource.Current task finds each resource on the basis of maximum delivered dependence is met
On available time slot, take the resource with the minimum end time and realize dispatch map.Such as one simple four tasks DAG's is anti-
To scheduling result as shown in figure 1, task niEarliest start time EST can be performed after being mapped to resource in the hope of itRHEFT(ni) and
Earliest finish time EFTRHEFT(ni)。
Step (4)：Calculating task sub- time limit and late start time.
Task niThe sub- time limit or latest finishing time subDL (ni) may be defined as：
subDL(ni)=D-ESTRHEFT(ni) (4)
Wherein：D is the DAG time limits
Task niLate Start LST (ni) may be defined as：
LST(ni)=D-EFTRHEFT(ni) (5)
Step (5)：Task n is removed in dispatch listi, repeat step (3) (4) (5),
Until all tasks of DAG complete the dispatch map to resource, each task computation result subDL (n are exportedi)、LST
(ni)。
Each task sub- time limit when being dispatched in one group of resource by limited DAG accessed by the present invention and open at the latest
Time beginning, due to being really to dispatch the result for obtaining, each task is remaining in DAG where completing before the sub- time limit ensure to appoint
Finishing on schedule for business, credibility is had more compared with the existing sub- time limit obtained with critical path thought.The acquisition in sub- time limit may be used also
It is further applicable in the algorithm of Cost Optimization, the determinating reference of resource is selected during as Cost Optimization.
Brief description of the drawings：
Fig. 1：Reverse DAG uses HEFT algorithmic dispatching result examples
Fig. 2：The low simple DAG sample datas of degree of parallelism
Fig. 3：Simple DAG sample datas with high degree of parallelism
Specific embodiment
The arthmetic statement that the sub- time limit of limited constraint workflow schedule obtains optimization method is as follows：
Input:Data between one group of computing resource R, the run time matrix W of each task of DAG in resource group, task
Transfer matrix C, DAG time limit D
1:The layer inverted order arrangement task according to where task, same layer is arranged by original order
2:Each task n is calculated according to formula (2)iReverse weights rankr(ni)
3:Non- mapping tasks list unMapList is obtained by the sequence of reverse weights to all tasks
4:WHILE(unMapList≠Φ)DO
5:Take out the task n of reverse maximum weighti
6:Resource has been mapped to according to all subtasks of DAG back scheduling thoughts (reverse father's task), n has been obtainediInstitute
There is subtask
7:To calculating n on each processoriIt is earliest can end time FT, (FT=subtasks perform the end time+with son appoint
The execution time on business data transfer time+processor), obtain duty mapping to the earliest finish time of each processor
EFTRHEFT(ni) and correspondence earliest start time ESTRHEFT(ni)。
8:From unMapList removal tasks ni
9:To GkEach task niSubDL (n are calculated according to formula (4)i) and formula (5) calculating Late Start LST
(ni)
10:END WHILE
11:Return to each task sub- time limit and late start time in DAG
DAG degree of parallelisms 0.5 (every layer of parallel task average and processor number relation) are set, simple DAG1 is generated at random
(A) such as Fig. 2, the circle in figure represents task number, and the edge direction between task is reversely, and side right weight is spent for data transfer time
Take, and give task execution time of each task on one group of processor and spend.
If the DAG time limits are 38.8, according to RHEFT algorithms, step by step calculation is carried out according to the implementation method of this patent：
Step (1)：By DAG sample datas given herein above, the reverse weights of each task are calculated based on formula (3)
rankr(ni)：
rankr(A1)：7.925 rankr(A2)：19.175 rankr(A3)：31.175
rankr(A4)：20.95 rankr(A5)：43.6
Step (2)：According to the rank of each task in DAGd(ni) to task ranking, obtain task scheduling order is descending：
A5, A3, A4, A2, A1, are put into dispatch map list successively.
Step (3)：Take out a task in order from dispatch list, finding to have from all resources can complete earliest
The resource timeslot of the task of time.
It is assumed that subtask is first dispatched after reversely, subtask passes to father's task, and scheduled sons all for each resource are appointed
Business transfers data to the maximum delivered time up to the resource.Current task finds each resource on the basis of transitive dependency is met
On available time slot, take the resource with earliest finish time EFT and realize dispatch map, while obtaining approximate earliest start time
EST.A5 since 0, in R2With minimum execution time 9.1, therefore selection resource R2.Second wheel can take out A3 when taking task, by
In back transfer, A5 broadcasts data to the A3 times and spends as 9.1, selects R1、R3、R4On permission start time be 9.1+3=
12.1, R2Upper passing time is 0, it is allowed to which start time is 9.1, therefore A3 is respectively in four processor earliest finish times
32.3rd, 27.9,31.7,31.9, therefore selection resource R2.Other tasks the like, one calculating of often wheel selection is finally obtained
EST, EFT of A1.Back scheduling result with DAG1 (A) is illustrated as shown in Fig. 2 can be in the hope of each task in scheduling process
The resource concrete outcome of reverse execution earliest start time and earliest finish time, EST, EFT and corresponding selection is shown in Table 1：
Table 1
Step (4)：Each task sub- time limit subDL and late start time LST is calculated according to formula (4), (5), and it is public
Formula (2) calculating task subDL is contrasted, and comparing result is shown in Table 2.
Table 2
It is the sub- time limit obtained from two class methods, smaller so that most number of tasks are less than in every layer of tasks in parallel degree of DAG
During processor number, formula (2) method calculates the sub- time limit that typically greater than RHEFT methods of each task sub- time limit are obtained.
There is bigger difference in the task sub- time limit on other RHEFT method same layers, and size order is consistent with reverse weights, and formula
(2) method calculates the sub- time limit same layer size Relatively centralized of each task, does not consider the relation with weights.When using same
When degree of parallelism generates the DAG of more multitask at random (20/50/100 task), still with such rule.If in addition, by public affairs
The sub- time limit of the task A1 that formula (2) is calculated is 3.9, the execution time minimum 7.7 for task A1 on each processor,
So being had no idea within the sub- time limit performs the task at all.In fact, work as being scheduled to DAG1 (A) using HEFT algorithms
When, the longest finishing time for obtaining is 36.4<38.8, also can be just to complete all tasks in the explanation time limit, thus also may be used
To find out that formula (2) calculates the unreasonable of sub- time limit method.
If the degree of parallelism that setting randomly generates DAG brings up to 2 (about 2 times of processor numbers generate each layer task at random),
Processor number is still much larger than using tasks in parallel number in four processors, i.e. some layers, correspondence has high degree of parallelism
DAG2 (B) structures and data parameters such as Fig. 3, the time limit is 27.Correspondence is still calculated the sub- time limit and is carried out using two methods
Contrast, comparing result is shown in Table 3.From this comparing result it will be seen that the task sub- time limit that RHEFT methods are obtained only has n6
The sub- time limit be greater than existing method, and this task is obviously more than other tasks in each processor average performance times,
Belong to a task in critical path.When processor number keeps constant, degree of parallelism is still 2, continues to randomly generate more
During the DAG of task (20/50/100 task), equally, the task sub- time limit that existing method only has in critical path is less than
The task sub- time limit that RHEFT methods are obtained.Therefore, when DAG tasks have high degree of parallelism and more than processor number, RHEFT
The task sub- time limit that the task sub- time limit that method is obtained, typically larger than existing method was obtained.
Table 3
Be can see from the contrast of the two, based on RHEFT algorithms obtain the task sub- time limit it is more dispersed, size order with it is anti-
Consistent to weights, constraint can be more reasonable.It is lower than its priority after the completion of each task closes on its sub- moment in time limit execution
Task still can perform completion, and be based on the sub- time limit of formula (2) acquisition due to the concurrency for not having consideration task, high preferential
Level task is closed on its sub- remaining time in time limit and is then possible to can not be completed all residue tasks.
Bibliography
[1]Abrishami S,Naghibzadeh M,Epema D H J.Deadline-constrained
workflow scheduling algorithms for Infrastructure as a Service Clouds[J]
.Future Generation Computer Systems,2013,29(1):158-169.
[2]Jia Y,Buyya R,Chen K T.Cost-Based Scheduling of Scientific
Workflow Application on Utility Grids[C]//International Conference on E-
Science and Grid Computing.DBLP,2006:8pp.-147.
[3]Topcuoglu H.,Hariri S.and Min-You W..Performance-effective and
Low-complexity Task Scheduling for Heterogeneous Computing.IEEE Transactions
on Parallel and Distributed Systems,2002(13):260-274.
[4]Sakellariou R.,Zhao H..A hybrid heuristic for DAG scheduling on
heterogeneous systems[C]//Parallel and Distributed Processing Symposium,
2004.Proceedings.International.IEEE,2004:111.
Claims (1)
1. optimization method is obtained based on the sub- time limit that reverse operation stream is dispatched, it is characterised in that step is as follows：
Step (1)：Calculate each task niTo entrance task nentryMaximum average range parameter, that is, task niIt is anti-
To weights rankr(ni), such as formula (1), be reversely relative to HEFT algorithms in upward weights and define；
Wherein：prec(ni) represent task niAll direct father's task, npIt is task niOne of them direct father's task,For
Task niTime average, c are performed in each resource in Rp,iFather's task npTo task niData communication time spend, rankr
(np) it is father's task npDownward weights；
Because formula is to need iterative calculation, it is necessary first to calculate entrance task nentry, and nentryThere is no direct father's task, its
Reverse weightsIt is task nentryTime average is performed in each resource in R；
Step (2)：Task ranking；
According to each task n in DAGiRankr(ni) descending arrangement, obtain task to the dispatch map list of resource so that go out
Mouth task nexitPriority scheduling, entrance task nentryFinally dispatch；
Step (3)：Selection resource；
Take out task n in order from dispatch listi, the time slot of the resource of the task with earliest finish time is found from R；
If niIt is first nentryTask, directly judging the minimum execution time of each resource in R can just find resource；Otherwise, due to anti-
To task niAll subtasks have been mapped into resource, first to calculate the n that goes out on missionsiMaximum of all subtasks to each resource
Passing time；Current task is meeting the available time slot found on the basis of maximum delivered is relied in each resource, takes with most
The resource of small end time realizes dispatch map；Task niIt is mapped to after resource and performs earliest finish time EFT in the hope of itRHEFT
(ni) and time started ESTRHEFT(ni)；
Step (4)：Calculating task sub- time limit and late start time；
Task niThe sub- time limit or latest finishing time subDL (ni) be defined as：
subDL(ni)=D-ESTRHEFT(ni) (2)
Wherein：D is the DAG time limits
Task niLate Start LST (ni) be defined as：
LST(ni)=D-EFTRHEFT(ni) (3)
Step (5)：Task n is removed in dispatch listi, repeat step (3) (4) (5), until all tasks of DAG are completed to resource
Dispatch map, exports each task computation result subDL (ni)、LST(ni)。
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710121606.4A CN106934537A (en) | 2017-03-02 | 2017-03-02 | The sub- time limit based on the scheduling of reverse operation stream obtains optimization method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710121606.4A CN106934537A (en) | 2017-03-02 | 2017-03-02 | The sub- time limit based on the scheduling of reverse operation stream obtains optimization method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN106934537A true CN106934537A (en) | 2017-07-07 |
Family
ID=59423669
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710121606.4A Pending CN106934537A (en) | 2017-03-02 | 2017-03-02 | The sub- time limit based on the scheduling of reverse operation stream obtains optimization method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106934537A (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109976890A (en) * | 2019-03-28 | 2019-07-05 | 东南大学 | A kind of conversion method minimizing the privately owned cloud computing resources energy consumption of isomery |
| CN110162393A (en) * | 2019-05-30 | 2019-08-23 | 奇瑞汽车股份有限公司 | Method for scheduling task, device and storage medium |
| CN110689262A (en) * | 2019-09-25 | 2020-01-14 | 中国人民解放军战略支援部队航天工程大学 | Space-based information system task scheduling method and device and electronic equipment |
| CN111612411A (en) * | 2020-04-02 | 2020-09-01 | 中能国际建筑投资集团有限公司 | Target task detection method, system, device and storage medium |
| CN114691342A (en) * | 2022-05-31 | 2022-07-01 | 蓝象智联(杭州)科技有限公司 | Method and device for realizing priority scheduling of federated learning algorithm component and storage medium |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103838627A (en) * | 2014-03-18 | 2014-06-04 | 北京工业大学 | Workflow dispatching method based on workflow throughput maximization |
| CN104834995A (en) * | 2015-04-20 | 2015-08-12 | 安徽师范大学 | Workflow bidirectional scheduling method based on cloud computing |
-
2017
- 2017-03-02 CN CN201710121606.4A patent/CN106934537A/en active Pending
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103838627A (en) * | 2014-03-18 | 2014-06-04 | 北京工业大学 | Workflow dispatching method based on workflow throughput maximization |
| CN104834995A (en) * | 2015-04-20 | 2015-08-12 | 安徽师范大学 | Workflow bidirectional scheduling method based on cloud computing |
Non-Patent Citations (4)
| Title |
|---|
| XIUJIE XU等: "Expansion slot backfill scheduling for concurrent workflows with deadline on heterogeneous resources", 《CLUSTER COMPUT》 * |
| XIU-JIE XU等: "Hybrid Scheduling Deadline-Constrained Multi-DAGs Based on Reverse HEFT", 《2016 INTERNATIONAL CONFERENCE ON INFORMATION SYSTEM AND ARTIFICIAL INTELLIGENCE》 * |
| 田国忠，肖创柏，谢军奇: "有期限约束的多DAG共享资源的调度及公平费用优化方法", 《计算机学报》 * |
| 田国忠: "多DAG共享资源调度的若干问题研究", 《中国博士学位论文全文数据库》 * |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109976890A (en) * | 2019-03-28 | 2019-07-05 | 东南大学 | A kind of conversion method minimizing the privately owned cloud computing resources energy consumption of isomery |
| CN109976890B (en) * | 2019-03-28 | 2023-05-30 | 东南大学 | Variable frequency method for minimizing heterogeneous private cloud computing resource energy consumption |
| CN110162393A (en) * | 2019-05-30 | 2019-08-23 | 奇瑞汽车股份有限公司 | Method for scheduling task, device and storage medium |
| CN110162393B (en) * | 2019-05-30 | 2023-06-27 | 奇瑞汽车股份有限公司 | Task scheduling method, device and storage medium |
| CN110689262A (en) * | 2019-09-25 | 2020-01-14 | 中国人民解放军战略支援部队航天工程大学 | Space-based information system task scheduling method and device and electronic equipment |
| CN111612411A (en) * | 2020-04-02 | 2020-09-01 | 中能国际建筑投资集团有限公司 | Target task detection method, system, device and storage medium |
| CN111612411B (en) * | 2020-04-02 | 2023-05-30 | 中能国际高新科技研究院有限公司 | Target task detection method, system, device and storage medium |
| CN114691342A (en) * | 2022-05-31 | 2022-07-01 | 蓝象智联(杭州)科技有限公司 | Method and device for realizing priority scheduling of federated learning algorithm component and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US9785468B2 (en) | Finding resource bottlenecks with low-frequency sampled data | |
| CN106934537A (en) | The sub- time limit based on the scheduling of reverse operation stream obtains optimization method | |
| CN111381950A (en) | Task scheduling method and system based on multiple copies for edge computing environment | |
| CN106447173A (en) | Cloud workflow scheduling method supporting any flow structure | |
| CN106991006B (en) | Support the cloud workflow task clustering method relied on and the time balances | |
| CN106951313A (en) | The sub- time limit acquisition methods of Multi-workflow shared resource cooperative scheduling | |
| Taheri et al. | Hopfield neural network for simultaneous job scheduling and data replication in grids | |
| Rafsanjani et al. | A new heuristic approach for scheduling independent tasks on heterogeneous computing systems | |
| Arabnejad et al. | Budget constrained scheduling strategies for on-line workflow applications | |
| Kumar et al. | Delay-based workflow scheduling for cost optimization in heterogeneous cloud system | |
| CN114579282A (en) | Task scheduling method and device | |
| Choi et al. | VM auto-scaling methods for high throughput computing on hybrid infrastructure | |
| Nasr et al. | Task scheduling algorithm for high performance heterogeneous distributed computing systems | |
| Smanchat et al. | Scheduling parameter sweep workflow in the Grid based on resource competition | |
| US8281313B1 (en) | Scheduling computer processing jobs that have stages and precedence constraints among the stages | |
| Caron et al. | Multi-criteria malleable task management for hybrid-cloud platforms | |
| CN115098240A (en) | Multiprocessor application scheduling method and system and storage medium | |
| Sun et al. | Dynamic task flow scheduling for heterogeneous distributed computing: algorithm and strategy | |
| Aida et al. | Performance enhancement of scheduling algorithm in heterogeneous distributed computing systems | |
| Wang et al. | A novel deadline and budget constrained scheduling heuristics for computational grids | |
| Jula et al. | A hybrid genetic algorithm-gravitational attraction search algorithm (HYGAGA) to solve grid task scheduling problem | |
| Mao et al. | Hierarchical model-based associate tasks scheduling with the deadline constraints in the cloud | |
| Rauchecker et al. | High-performance computing for scheduling decision support: A parallel depth-first search heuristic | |
| Singh | Hybrid metaheuristic based scheduling with job duplication for cloud data centers | |
| Althoubi et al. | Response-Time Tail Mitigation in Cloud Systems |
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 | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20170707 |
|
| WD01 | Invention patent application deemed withdrawn after publication |