CN103810037A - Job scheduling method and computing device - Google Patents
Job scheduling method and computing device Download PDFInfo
- Publication number
- CN103810037A CN103810037A CN201410012660.1A CN201410012660A CN103810037A CN 103810037 A CN103810037 A CN 103810037A CN 201410012660 A CN201410012660 A CN 201410012660A CN 103810037 A CN103810037 A CN 103810037A
- Authority
- CN
- China
- Prior art keywords
- head
- backfill
- queue
- time
- working time
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Images
Landscapes
- Debugging And Monitoring (AREA)
- Retry When Errors Occur (AREA)
Abstract
The invention provides a job scheduling method and a computing device. The job scheduling method includes the steps that when it is detected that computing resources are not enough to run a current head-of-line job, a corresponding backfilling window is computed; an estimated running time corresponding to non-head-of-line jobs is shortened, whether the non-head-of-line jobs can be backfilled or not is judged according to a shortened running time, and if yes, the corresponding non-head-of-line jobs are backfilled; when computation of the backfilling window is finished and remaining computing resources are not enough to run the current head-of-line job, a backfilled job being run currently is stopped, and check points are set for the stopped backfilled job; when the backfilled job is started again, computation of unfinished parts is carried out according to the check points in the backfilled job. The job scheduling method can improve the utilization ratio of system resources and shorten waiting time of jobs.
Description
Technical field
The present invention relates to calculation element technical field, relate in particular to a kind of job scheduling method and calculation element.
Background technology
Concurrent Job Dispatching system is the very important ingredient of supercomputer system always, and it distributes available computing node to concurrent program according to user's operation resource request.Concurrent job can take the computing node of distribution up to complete/failure of job run, or is killed by system process because the job run time exceedes specified operation estimated time of Run Script.Due to resource-constrained, concurrent job generally can not obtain applied for computational resource immediately entering after job queue, often will wait for a period of time.The main target of Concurrent Job Dispatching system is to reduce the average operation stand-by period, and the resource utilization of the system of maximization, All Jobs is kept to certain fairness simultaneously.
Through the development of decades, Concurrent Job Dispatching technology is quite ripe, rare major progress in recent years.Although academia has proposed some New Policies and new algorithm, for example dynamic partition technology and group scheduling technology.But these technology are because the restriction of reality use and other factors does not obtain large-scale application.Up to the present, the BACKFILL based on first serving first (First Come First Served, FCFS) is still the Concurrent Job Dispatching algorithm that most supercomputers and group system extensively adopt.First this algorithm is IBM SP1 concurrent computational system at that time design by Argonne National Laboratory of the U.S., and is achieved in EASY dispatching system.The basic thought of this algorithm is very simple: high-performance computer system is according to job priority scheduling queue in concurrent job operation, in system, can leave some and be difficult to the resource " hole " of utilizing, at this moment system can be found out the short operation of those applicable low priorities and be moved in advance under the prerequisite that does not affect high priority job run from waiting list.Thereby improve the utilization factor of system, shortened the average latency of operation.The Concurrent Job Dispatching system of main flow at present, if the default setting that LSF, Moab, Maui, PBS/Torque and LoadLeveler adopted is all the mutation based on FCFS BACKFILL.
Although BACKFILL can utilize the little operation of low priority to fill the resource " hole " that high priority operation stays, and has improved utilization factor.But the job run time Estimate that user submits to is in actual applications generally bigger than normal, causes a lot of operations can not backfill.Thereby cause the utilization factor of computer system still lower, a lot of resource free time, still have a lot of operations to wait in queue.
Visible, job scheduling scheme of the prior art can not effectively be utilized the computational resource of computing machine.
Summary of the invention
The invention provides a kind of job scheduling method and calculation element, improve the utilization factor of system resource, reduce the stand-by period of operation.
The invention provides a kind of job scheduling method, comprising:
In the time detecting that computational resource is not enough to move current head of the queue operation, calculate corresponding backfill window;
Shorten non-head of the queue operation and estimate corresponding working time, and whether can backfill according to estimating described non-head of the queue operation the working time after shortening, and in the time being judged as YES, by corresponding non-head of the queue operation backfill;
In the time that backfill window finishes and remains computational resource and be not enough to move current head of the queue operation, stop the backfill operation of current operation, and for Checkpointing in the backfill operation being terminated;
In the time again restarting described backfill operation, carry out the calculation task of part according to the checkpoint in described backfill operation.
Preferably, other the non-head of the queue operations outside the operation of described shortening head of the queue are estimated corresponding working time, specifically comprise:
To estimate to be multiplied by a default splitting factor p working time corresponding described non-head of the queue operation, wherein, 0<p<1.
Preferably, a default splitting factor p is multiplied by described estimation working time that will be corresponding described non-head of the queue operation, specifically comprises:
In the time judging that described non-head of the queue operation is estimated to be greater than preset value corresponding working time, estimate to be multiplied by a default splitting factor p working time that will be corresponding described non-head of the queue operation.
Preferably, described according to estimating described non-head of the queue operation and whether can backfill specifically comprise the working time after shortening:
Judging that computational resource that described non-head of the queue operation need to take is less than the computational resource of current residual, and while working time after shortening estimating to be less than described backfill window, judging that described non-head of the queue operation can backfill.
Preferably, described according to estimating described non-head of the queue operation and whether can backfill specifically comprise the working time after shortening:
Judging that computational resource that described non-head of the queue operation need to take is less than the computational resource of current residual, and the working time after shortening is while estimating to be greater than described backfill window, described in judging in the time that described backfill window finishes, whether non-head of the queue operation and described head of the queue operation can move simultaneously, if so, judge that described non-head of the queue operation can backfill.
The invention provides a kind of calculation element, comprising:
Backfill window calculation module, in the time detecting that computational resource is not enough to move current head of the queue operation, calculates corresponding backfill window;
Whether backfill module, estimates corresponding working time for shortening non-head of the queue operation, and can backfill according to estimating described non-head of the queue operation the working time after shortening, and in the time being judged as YES, by corresponding non-head of the queue operation backfill;
Computing module, when finishing at backfill window and remain computational resource and be not enough to move current head of the queue operation, stop the backfill operation of current operation, also add checkpoint for the backfill operation being terminated, and in the time again restarting described backfill operation, carry out the calculation task of part according to the checkpoint in described backfill operation.
Preferably, described backfill module, for estimating to be multiplied by a default splitting factor p working time corresponding described non-head of the queue operation, wherein, 0<p<1.
Preferably, described backfill module, specifically in the time judging that described non-head of the queue operation is estimated to be greater than preset value corresponding working time, estimates to be multiplied by a default splitting factor p working time that will be corresponding described non-head of the queue operation.
In the present invention, shorten the operation estimated time of non-head of the queue operation, thereby be beneficial to the backfill of non-head of the queue operation, improve the resource utilization of system.On the other hand, cause non-head of the queue operation still moving and then causing the computational resource of system can not move the situation of head of the queue operation in the time that backfill window finishes estimated time for fear of the operation because shortening non-head of the queue operation, in the present invention, also in the time that backfill window finishes and fruit remain computational resource and be not enough to stop while moving current head of the queue operation backfill operation, thereby guarantee the operation of head of the queue operation.In addition, in the present invention, can make calculation element in the time again restarting described backfill operation, only carry out the calculation task of part, and without repeatedly carry out identical calculation task for a backfill operation, so also can further improve the utilization factor of system resources in computation.
Accompanying drawing explanation
The schematic flow sheet of a kind of job scheduling method that Fig. 1 provides for the embodiment of the present invention one;
A kind of specific implementation of the job scheduling method that Fig. 2 provides for the embodiment of the present invention one;
A kind of implementation that shortens working time in the job scheduling method that Fig. 3 provides for the embodiment of the present invention and estimate;
Processing while ending backfill operation in the job scheduling method that Fig. 4 provides for the embodiment of the present invention;
The schematic diagram of a kind of application scenarios of the job scheduling method that Fig. 5 (a) provides for the embodiment of the present invention;
Fig. 5 (b) is for utilizing the design sketch that in prior art, classical BACKFILL carries out backfill;
The design sketch that Fig. 5 (c) carries out backfill for the job scheduling method that utilizes the embodiment of the present invention and provide;
The structural representation of a kind of calculation element that Fig. 6 provides for the embodiment of the present invention.
Embodiment
Below in conjunction with drawings and Examples, the specific embodiment of the present invention is further described.Following examples are only for technical scheme of the present invention is more clearly described, and can not limit the scope of the invention with this.
The embodiment of the present invention provides a kind of job scheduling method, and as shown in Figure 1, the method comprises:
Head of the queue operation in the present invention and non-head of the queue operation refer to respectively first operation in the queue after the operation for the treatment of operation is sorted according to priority and other operations except first operation.Backfill window refers to, what calculation element was estimated can provide the current needed time of the needed resource of head of the queue operation to calculation element.The backfill window here and the understanding of head of the queue operation are consistent with prior art.
In practical application, before step 101, in the time having concurrent job to move or submit to, according to the priority orders scheduling job queue of concurrent job; Arrange successively head of the queue job run, until job queue sky or head of the queue operation are because computational resource can not move not; If job queue not empty (being that computational resource is not enough to move current head of the queue operation) is the reserved resource of head of the queue operation.And according to just estimating corresponding working time at running job in system, calculate and set aside some time and backfill window size.Estimate the time that refers to that user need to move for this operation of an operation estimation the working time here.This process is consistent with prior art, no longer describes in detail at this.
The operation here should be rigidity operation, i.e. operation is all operating in during whole service on the processor/computing node of fixed qty, and each concurrent job can operate in any computing node that meets computation requirement, is not limited to specific computing node.This resource requirement is specified by user in job script.
Whether step 102, shortens non-head of the queue operation and estimates corresponding working time, and can backfill according to estimating described non-head of the queue operation the working time after shortening, and in the time being judged as YES, by corresponding non-head of the queue operation backfill.
After step 103, suspended backfill operation is put in job queue and waits for and being again performed, and a kind of possible situation is, in the time that current head of the queue operation completes, suspended backfill operation becomes head of the queue operation, now, after step 101, may directly perform step 104.Job scheduling method provided by the invention is an Event triggered process, in the time that corresponding condition is met, carries out corresponding step, in an Event triggered process, carries out the execution that must not depend on previous step in step 103 or step 104.
Due in step 101, calculation element calculate backfill window be calculation element estimate can provide the current needed time of the needed resource of head of the queue operation to calculation element, if backfill window finishes, but the residue computational resource deficiency of system, be likely because the operation of backfill does not finish to cause, now, the operation of backfill is stopped, to guarantee the preferential operation of head of the queue operation.
In the embodiment of the present invention, shorten the operation estimated time of non-head of the queue operation, thereby be beneficial to the backfill of non-head of the queue operation, improve the resource utilization of system.On the other hand, cause non-head of the queue operation still moving and then causing the computational resource of system can not move the situation of head of the queue operation in the time that backfill window finishes estimated time for fear of the operation because shortening non-head of the queue operation, in the present invention, if also residue computational resource is not enough to stop while moving current head of the queue operation backfill operation in the time that backfill window finishes, thereby guarantee the operation of head of the queue operation.On the other hand, by for backfill operation Checkpointing, and make calculation element in the time again restarting described backfill operation, only carry out the calculation task of part, can make calculation element without repeatedly carrying out identical calculation task for a backfill operation, so also can further improve the utilization factor of system resources in computation.
If Fig. 2 is the formalized description of the Concurrent Job Dispatching method based on checkpoint.Job scheduling method provided by the invention and classical operation BACKFILL are roughly the same, whether its key distinction is can backfill according to estimating to check the working time that estimate working time rather than user submits to after shortening in step 102, the expected concluding time of computational tasks.
Preferably, the non-head of the queue operation of shortening in above-mentioned step 102 is estimated corresponding working time, specifically comprises:
Step 1021, will estimate to be multiplied by a default splitting factor p working time corresponding described non-head of the queue operation, wherein, and 0<p<1.
Preferably, step 102 is specially: in the time judging that described non-head of the queue operation is estimated to be greater than preset value corresponding working time, estimate to be multiplied by a default splitting factor p working time that will be corresponding described non-head of the queue operation.Its specific implementation can be as shown in Figure 3.
Be made with like this consideration of several respects.First, estimate even if do not adjust the working time of operation, short operation is also relatively easily by backfill.Secondly, checkpoint and the expense of restarting are too large for short operation.The 3rd, the short operation program that user is used for test or debugs often.If the divided words of operational process, can reduce user's experience.P and preset value t
thrbe adjustable parameter, can control the radical degree of backfill.Find through test and statistics, such simple mechanisms work obtains fine, and system manager can be that parameter is selected a suitable value according to practical work experience.
Preferably, in above-mentioned arbitrary embodiment, " whether can the backfill according to estimating described non-head of the queue operation the working time after shortening " in described step 102, specifically comprises:
(a), judging that computational resource that described non-head of the queue operation need to take is less than the computational resource of current residual, and working time after shortening estimating to be less than described backfill window, judging that described non-head of the queue operation can backfill; Or,
(b), judging that computational resource that described non-head of the queue operation need to take is less than the computational resource of current residual, and the working time after shortening is while estimating to be greater than described backfill window, described in judging in the time that described backfill window finishes, whether non-head of the queue operation and described head of the queue operation can move simultaneously, if so, judge that described non-head of the queue operation can backfill.
For (a), can guarantee that the job run of backfill is in backfill window, thereby not affect the operation of head of the queue operation.For (b), even if estimate the working time after shortening to be greater than described backfill window, i.e. the operation of backfill operation may exceed backfill window, as long as the operation of backfill operation can not affect the operation of head of the queue operation, also can allow this operation backfill.
Fig. 4 has provided a kind of implementation of step 103, in the time setting aside some time (backfill window finishes) and there is no enough resources to head of the queue job run, and this algorithm operation.In this algorithm, backfill operation is done checkpoint and is stopped running, until head of the queue operation obtains enough computational resources.The operation being preempted is reentered into queue.In selecting to be preempted operation, we are from using the operation that computing node is many to start to select, to reduce the quantity that is preempted and does checkpoint operation.It is pointed out that the computing node number of each backfill operation use will be less than the high priority operation of head of the queue.Otherwise high priority operation will be scheduled at time point early.Therefore, the strategy of trying to be the first in the present invention is in fact a kind of optimal selection algorithm.
Below in conjunction with Fig. 5, job scheduling method provided by the invention is elaborated, as shown in Figure 5, each concurrent job represents by a rectangle, and horizontal and vertical dimension represents respectively the computation requirement of operation: processor/number of nodes of estimating working time and request.Gray area in each wait operation rectangle represents its actual run time.In system sometime, operation A and B move, and the C of operation simultaneously, D and E wait for operation in job queue, as shown in Fig. 5 (a).After A has moved, system does not have enough resources to move to head of the queue operation C.Dispatching system is according to being estimated as the reserved computational resource of C the working time of operation B.The moment of B end of run is exactly " setting aside some time " of C so.Finish to finish this time period to B from A and formed " backfill window ".
For classical BACKFILL (Fig. 5 (b)), because the estimation of D and E operation has all exceeded backfill window working time, two operations all can not be by backfill, although the actual run time of operation D is less than this window.D and E operation meeting obtain moving chance after C end of run, thereby have wasted the computational resource in backfill window.
For the BACKFILL based on checkpoint (Fig. 5 (c)), operation D and E can be by backfills.Operation D had completed operation before the arrival of setting aside some time.And operation E is a little preempted setting aside some time, its operational process is divided into two sections: E1 and E2.In the time that C finishes, E continues operation until finish.
This example has been shown the several advantages based on checkpoint dispatching algorithm: first, radical backfill strategy is benefited from the operation of over-evaluating, and has shortened the stand-by period and has improved system availability; Secondly, do the operation of checkpoint and also shortened the response time; Finally, the reserved resource of high priority operation is guaranteed.
Based on identical design, the embodiment of the present invention three also provides a kind of calculation element, by above-mentioned job scheduling method, concurrent job is dispatched, and as shown in Figure 6, this calculation element comprises:
The first computing module 601, in the time detecting that computational resource is not enough to move current head of the queue operation, calculates corresponding backfill window;
Whether backfill module 602, estimates corresponding working time for shortening non-head of the queue operation, and can backfill according to estimating described non-head of the queue operation the working time after shortening, and in the time being judged as YES, by corresponding non-head of the queue operation backfill;
The second computing module 603, when finishing at backfill window and remain computational resource and be not enough to move current head of the queue operation, stop the backfill operation of current operation, also add checkpoint for the backfill operation being terminated, and in the time again restarting described backfill operation, carry out the calculation task of part according to the checkpoint in described backfill operation.
Preferably, backfill module 602, for estimating to be multiplied by a default splitting factor p working time corresponding described non-head of the queue operation, wherein, 0<p<1.
Preferably, backfill module 602, specifically in the time judging that described non-head of the queue operation is estimated to be greater than preset value corresponding working time, estimates to be multiplied by a default splitting factor p working time that will be corresponding described non-head of the queue operation.
The above is only the preferred embodiment of the present invention; it should be pointed out that for those skilled in the art, do not departing under the prerequisite of the technology of the present invention principle; can also make some improvements and modifications, these improvements and modifications also should be considered as protection scope of the present invention.
Claims (8)
1. a job scheduling method, is characterized in that, comprising:
In the time detecting that computational resource is not enough to move current head of the queue operation, calculate corresponding backfill window;
Shorten non-head of the queue operation and estimate corresponding working time, and whether can backfill according to estimating described non-head of the queue operation the working time after shortening, and in the time being judged as YES, by corresponding non-head of the queue operation backfill;
In the time that backfill window finishes and remains computational resource and be not enough to move current head of the queue operation, stop the backfill operation of current operation, and for-the backfill operation Checkpointing that is terminated;
In the time again restarting described backfill operation, carry out the calculation task of part according to the checkpoint in described backfill operation.
2. the method for claim 1, is characterized in that, other the non-head of the queue operations outside the operation of described shortening head of the queue are estimated corresponding working time, specifically comprise:
To estimate to be multiplied by a default splitting factor p working time corresponding described non-head of the queue operation, wherein, 0<p<1.
3. method as claimed in claim 2, is characterized in that, a default splitting factor p is multiplied by described estimation working time that will be corresponding described non-head of the queue operation, specifically comprises:
In the time judging that described non-head of the queue operation is estimated to be greater than preset value corresponding working time, estimate to be multiplied by a default splitting factor p working time that will be corresponding described non-head of the queue operation.
4. whether the method for claim 1, is characterized in that, described according to estimating described non-head of the queue operation and can backfill specifically comprise the working time after shortening:
Judging that computational resource that described non-head of the queue operation need to take is less than the computational resource of current residual, and while working time after shortening estimating to be less than described backfill window, judging that described non-head of the queue operation can backfill.
5. whether the method for claim 1, is characterized in that, described according to estimating described non-head of the queue operation and can backfill specifically comprise the working time after shortening:
Judging that computational resource that described non-head of the queue operation need to take is less than the computational resource of current residual, and the working time after shortening is while estimating to be greater than described backfill window, described in judging in the time that described backfill window finishes, whether non-head of the queue operation and described head of the queue operation can move simultaneously, if so, judge that described non-head of the queue operation can backfill.
6. a calculation element, is characterized in that, comprising:
The first computing module, in the time detecting that computational resource is not enough to move current head of the queue operation, calculates corresponding backfill window;
Whether backfill module, estimates corresponding working time for shortening non-head of the queue operation, and can backfill according to estimating described non-head of the queue operation the working time after shortening, and in the time being judged as YES, by corresponding non-head of the queue operation backfill;
The second computing module, when finishing at backfill window and remain computational resource and be not enough to move current head of the queue operation, stop the backfill operation of current operation, also for the backfill operation Checkpointing for being terminated, and in the time again restarting described backfill operation, carry out the calculation task of part according to the checkpoint in described backfill operation.
7. calculation element as claimed in claim 6, is characterized in that,
Described backfill module, for estimating to be multiplied by a default splitting factor p working time corresponding described non-head of the queue operation, wherein, 0<p<1.
8. calculation element as claimed in claim 7, it is characterized in that, described backfill module, specifically in the time judging that described non-head of the queue operation is estimated to be greater than preset value corresponding working time, estimates to be multiplied by a default splitting factor p working time that will be corresponding described non-head of the queue operation.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410012660.1A CN103810037B (en) | 2014-01-10 | 2014-01-10 | Job scheduling method and computing device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410012660.1A CN103810037B (en) | 2014-01-10 | 2014-01-10 | Job scheduling method and computing device |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN103810037A true CN103810037A (en) | 2014-05-21 |
| CN103810037B CN103810037B (en) | 2017-02-15 |
Family
ID=50706846
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201410012660.1A Active CN103810037B (en) | 2014-01-10 | 2014-01-10 | Job scheduling method and computing device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN103810037B (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105718312A (en) * | 2016-01-20 | 2016-06-29 | 华南理工大学 | Multi-queue back-filling job scheduling method oriented to living organism gene sequencing calculation task |
| CN109710414A (en) * | 2018-12-29 | 2019-05-03 | 北京三快在线科技有限公司 | A kind of job scheduling method, device, equipment and storage medium |
| CN113448705A (en) * | 2021-06-25 | 2021-09-28 | 皖西学院 | Unbalanced job scheduling algorithm |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US7082606B2 (en) * | 2001-05-01 | 2006-07-25 | The Regents Of The University Of California | Dedicated heterogeneous node scheduling including backfill scheduling |
| CN102402461A (en) * | 2011-12-08 | 2012-04-04 | 曙光信息产业(北京)有限公司 | Balanced scheduling method based on operation scale |
-
2014
- 2014-01-10 CN CN201410012660.1A patent/CN103810037B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US7082606B2 (en) * | 2001-05-01 | 2006-07-25 | The Regents Of The University Of California | Dedicated heterogeneous node scheduling including backfill scheduling |
| CN102402461A (en) * | 2011-12-08 | 2012-04-04 | 曙光信息产业(北京)有限公司 | Balanced scheduling method based on operation scale |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105718312A (en) * | 2016-01-20 | 2016-06-29 | 华南理工大学 | Multi-queue back-filling job scheduling method oriented to living organism gene sequencing calculation task |
| CN105718312B (en) * | 2016-01-20 | 2018-10-30 | 华南理工大学 | More queues that calculating task is sequenced towards biological gene backfill job scheduling method |
| CN109710414A (en) * | 2018-12-29 | 2019-05-03 | 北京三快在线科技有限公司 | A kind of job scheduling method, device, equipment and storage medium |
| CN109710414B (en) * | 2018-12-29 | 2021-03-26 | 北京三快在线科技有限公司 | Job scheduling method, device, equipment and storage medium |
| CN113448705A (en) * | 2021-06-25 | 2021-09-28 | 皖西学院 | Unbalanced job scheduling algorithm |
Also Published As
| Publication number | Publication date |
|---|---|
| CN103810037B (en) | 2017-02-15 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106293919B (en) | A kind of the built-in tasks dispatching device and method of time trigger | |
| US10754706B1 (en) | Task scheduling for multiprocessor systems | |
| US9086922B2 (en) | Opportunistically scheduling and adjusting time slices | |
| CN109144710B (en) | Resource scheduling method, device and computer readable storage medium | |
| CN104536835B (en) | Self-adapting task scheduling method in a kind of Hot Spare real-time control system | |
| US9104491B2 (en) | Batch scheduler management of speculative and non-speculative tasks based on conditions of tasks and compute resources | |
| CN105550040B (en) | CPU resources of virtual machine preservation algorithm based on KVM platform | |
| US20150113252A1 (en) | Thread control and calling method of multi-thread virtual pipeline (mvp) processor, and processor thereof | |
| CN104111877A (en) | Thread dynamic deployment system and method based on thread deployment engine | |
| WO2015066979A1 (en) | Machine learning method for mapreduce task resource configuration parameters | |
| JPWO2005106623A1 (en) | CPU clock control device, CPU clock control method, CPU clock control program, recording medium, and transmission medium | |
| JP2009294712A (en) | Priority controller and priority control method | |
| WO2014101561A1 (en) | Method and device for implementing multi-application parallel processing on single processor | |
| CN104915253A (en) | Work scheduling method and work processor | |
| CN109660569B (en) | Multitask concurrent execution method, storage medium, device and system | |
| CN111142943A (en) | Automatic control concurrency method and device | |
| CN104820616A (en) | Task scheduling method and device | |
| CN103810037A (en) | Job scheduling method and computing device | |
| CN103049314A (en) | Real-time scheduling method for open system | |
| Lunacek et al. | The scaling of many-task computing approaches in python on cluster supercomputers | |
| JP2021504777A5 (en) | ||
| CN109739634A (en) | A kind of atomic task execution method and device | |
| US20130117757A1 (en) | Method and apparatus for scheduling application programs | |
| Gharajeh et al. | Heuristic-based task-to-thread mapping in multi-core processors | |
| CN112445614A (en) | Thread data storage management method, computer equipment and storage medium |
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 |