CN105511866B - Resource constraint dispatching optimization method based on parallel organization cognition technology - Google Patents
Resource constraint dispatching optimization method based on parallel organization cognition technology Download PDFInfo
- Publication number
- CN105511866B CN105511866B CN201510863693.1A CN201510863693A CN105511866B CN 105511866 B CN105511866 B CN 105511866B CN 201510863693 A CN201510863693 A CN 201510863693A CN 105511866 B CN105511866 B CN 105511866B
- Authority
- CN
- China
- Prior art keywords
- search
- scheduling
- parallel
- resource constraint
- task
- 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.)
- Active
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F8/00—Arrangements for software engineering
- G06F8/20—Software design
- G06F8/22—Procedural
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F8/00—Arrangements for software engineering
- G06F8/30—Creation or generation of source code
- G06F8/31—Programming languages or programming paradigms
Abstract
The invention discloses a kind of resource constraint dispatching optimization methods of parallel organization cognition technology, including step：The operation order for setting initial ranging task in parallel search task, sets the high level operation enumeration order of remaining search mission；The structural information for obtaining data flow diagram in parallel search delimit the level wherein respectively operated based on resource constraint scheduling；Wherein, whether meet at least one constraints according to operation, abandon the search space for the operation parallel search task to adjust for meeting constraints；Collaborative framework is established in parallel search task, constraints is shared between search mission；After any search mission obtains optimal scheduling sequence or the search of all search missions, output optimal scheduling sequence.The present invention can efficiently obtain the optimal solution of the workflow resource allocation schedule under resource constraint, improve the success rate for detecting optimal solution, further improve the efficiency of search optimal scheduling.
Description
Technical field
The invention belongs under computer realm more particularly to a kind of resource constraint based on parallel organization cognition technology
Dispatch optimization method.
Background technology
Traditional scheduling optimization method is designed to mostly for finding an approximate optimal solution, using following technology：
High Level Synthesis (High-level Synthesis, abridge HLS)：Also known as higher synthesis, C synthesize (C
Synthesis), (Electronic System Level synthesis, abridge electronic system layering synthesis ESL
Synthesis), it is to convert the algorithm level of circuit design specification or behavioral scaling description to circuit knot under certain constraints
The method and process of structure description.High Level Synthesis is also known as Behavioral Synthesis, algorithm level synthesis etc..It is enabled a designer to more
High-level carry out Electronic Design, more rapidly effectively in higher level design verification and emulation, and the work of lower level is by work
Tool is automatically performed, to allow the digital circuitry design engineer can to have more energy and more fully condition goes to carry out
Best design scheme is sought in the search of design space.The process of HLS usually substantially include pretreatment, compiling, conversion, scheduling,
Several parts such as distribution, controller, synthesis, RTL, generation and decompiling.Compiling, conversion portion determine the compatibility of software
And ease for use, it dispatches (schedule) and distribution (binding) essentially dictates performance, resource size of RTL of generation etc..
Branch-and-bound (branch and bound) algorithm is a kind of solution for searching for problem on the solution space tree of problem
Method.Branch-bound algorithm searches for solution space tree using the preferential method of breadth First or minimum consuming, also, in branch-and-bound
In algorithm, each slip-knot point only has an opportunity as extension node.Using branch-bound algorithm to the solution space tree of problem
It scans for, its search strategy is：All child's nodes of current extensions node are generated first；Then it is tied in the child of generation
In point, the node of feasible solution (or optimal solution) can not possibly be generated by abandoning those；Again by remaining child's node join work ode table；
Finally select next slip-knot point as new extension node from ode table living.So cycle, until finding the feasible of problem
It is sky to solve (optimal solution) or ode table living.
Data flow diagram (Data Flow Diagram)：Abbreviation DFD, it is from data transfer and machining angle, to graphically
Come the logic function of expression system, data in the logic flow direction and logical conversion process of internal system, is structured system analysis
The main representation aids of method and a kind of graphic technique for indicating software model.
Parallel computation (parallel computing)：It generally refers to many instructions and is able to the calculating pattern being carried out at the same time.
Under the premise of carrying out at the same time, can be solved with concurrent fashion later by the procedure decomposition of calculating at fraction.
Under resource constraint, traditional scheduling optimization method has shortcoming below：It is big for the optimization of scheduling
The monokaryon of processor is also only utilized in majority, and the research in terms of High Level Synthesis parallel computation is without the letter using structuring
Breath, even without the influence for considering that different operation dispatching sequence brings.Therefore, in order to overcome drawbacks described above in the prior art,
The present invention proposes a kind of resource constraint dispatching optimization method based on parallel organization cognition technology, and this method more has
The optimization method of effect can greatly improve the efficiency for finding optimal scheduling, save the expense of time and energy consumption.
Invention content
The present invention proposes a kind of method of the resource constraint dispatching perceived based on parallel organization, including step：
Step 1：The operation order for setting initial ranging task in parallel search task, sets the height of remaining search mission
Layer operation enumeration order；
Step 2：The structural information for obtaining data flow diagram in parallel search delimit wherein each behaviour based on resource constraint scheduling
The level of work；Wherein, whether meet at least one constraints according to the operation, discarding meets the operation of constraints to adjust
The search space of whole institute's parallel search task；
Step 3：Collaborative framework is established in the parallel search task, constraints is shared between search mission；
Step 4：It is defeated after any described search task obtains optimal scheduling sequence or the search of all search missions
Go out optimal scheduling sequence.
In the method for the resource constraint dispatching based on parallel organization perception proposed by the present invention, according to weighting
Critical path depth sets the high level operation enumeration order of remaining search mission.
In the method for the resource constraint dispatching based on parallel organization perception proposed by the present invention, the weighting
Critical path depth is related with the time loss of operation itself and corresponding dispatching sequence.
It is at least one in the method for the resource constraint dispatching based on parallel organization perception proposed by the present invention
Constraints is：The expected upper limit of incomplete scheduling is equal to optimal scheduling length.
It is at least one in the method for the resource constraint dispatching based on parallel organization perception proposed by the present invention
Constraints is：The scheduling time of at least one operation is more than the scheduling time of current optimal scheduling.
It is at least one in the method for the resource constraint dispatching based on parallel organization perception proposed by the present invention
Constraints is：The scheduling time of all operations is not less than the scheduling time of current optimal scheduling.
In the method for the resource constraint dispatching based on parallel organization perception proposed by the present invention, the constraint
Condition includes the Border condition in the length of current global optimum's scheduling, the interior optimal scheduling length of each grouping and each grouping.
The beneficial effects of the present invention are：The workflow resource allocation schedule under resource constraint can efficiently be obtained
Optimal solution the success rate for detecting optimal solution is improved by the synergistic application of parallel computation.The present invention is during beta pruning
The structural information factor for also having fully considered the corresponding data flow diagram of workflow further improves the effect of search optimal scheduling
Rate.
Description of the drawings
Fig. 1 is the flow chart of the resource constraint dispatching optimization method of parallel organization cognition technology of the present invention.
Fig. 2 is the result figure being split to initial ranging task in embodiment.
Fig. 3 (a) is the structural information that data flow diagram is excavated in embodiment.
Fig. 3 (b) is a kind of optimal scheduling of data flow diagram G under current schedules.
Fig. 3 (c) is a unfinished scheduling of data flow diagram G.
Fig. 4 is the pseudocode of search mission.
Specific implementation mode
In conjunction with following specific examples and attached drawing, the present invention is described in further detail.The process of the implementation present invention,
Condition, experimental method etc. are among the general principles and common general knowledge in the art, this hair in addition to the following content specially referred to
It is bright that content is not particularly limited.
The resource constraint dispatching optimization method of parallel organization cognition technology of the present invention carries out initial ranging task
Divide, and improve the success rate of search mission by way of changing high level operation enumeration order, while utilizing data flow diagram
Structural information removes meaningless branch, then is updated by the cooperation between search mission as soon as possible, can greatly reduce optimizing
The total time-consuming and total energy consumption of journey.As shown in Figure 1, this method comprises the following steps：
Step 1：The operation order for setting initial ranging task in parallel search task, sets the height of remaining search mission
Layer operation enumeration order.The step is made by changing the method for the enumeration order of operating procedure with completing multidirectional parallel search
The generation of parallel search and constraints is obtained positioned at the different location of search space, optimal tune is found to substantially increase
The chance of degree sequence.
Step 2：The structural information for obtaining data flow diagram in parallel search delimit wherein each behaviour based on resource constraint scheduling
The level of work；Wherein, whether meet at least one constraints according to the operation, discarding meets the operation of constraints to adjust
The search space of whole institute's parallel search task.The step is based on traditional branch-and-bound (branch and bound) algorithm base
On plinth, by excavating the structural information of data flow diagram, for resource constraint problem (RCS, Resource Constrained
Scheduling a kind of more efficient level key-machine) is proposed；Different from conditional branch key-machine, when one it is imperfect
Scheduling its expected upper limit when being equal to optimal scheduling length, can be by comparing the scheduling time of part operation and its structure letter
It is timely abandoned (and conditional branch key-machine can not accomplish this point) by breath.Therefore, which may be used one kind more
Positive mode trims search space, to more efficiently find optimal solution.
Step 3：Collaborative framework is established in the parallel search task, constraints is shared between search mission.This
The each parallel search task of sample can have multiple constraintss to be effectively prevented from depth recursive search, so that need
Overall time greatly reduces.
Step 4：It is defeated after any described search task obtains optimal scheduling sequence or the search of all search missions
Go out optimal scheduling sequence.
Wherein, initial ranging task is a workflow task for having disposed concrete configuration, different initial ranging tasks
With the different execution time, execute the information such as resource and Executing Cost.
Wherein, optimal scheduling sequence be by being modeled to workflow instance, in the case where meeting certain resource constraint,
Through searching for obtained completion required by task time shortest dispatching sequence.
As shown in Fig. 2, designing an operation in the segmentation of initial ranging task, carried out by way of changing enumeration order
Parallel search.Assuming that there is n parallel search task (i.e. t0 ..., tn-1).It is more preferable than the performance of sequential search in order to ensure, if
Operation order and traditional heuristic branch-and-bound (B&B) method sequence for determining initial ranging task t0 are identical.For remaining
Parallel search task, in order to ensure in top generation effective sub- constraints as much as possible and delete as soon as possible useless
Branch upsets the sequence of high level operation as far as possible under the premise of ensureing to operate dependence.It is passed in the present invention using change
System BULB sort it is obtained weighting critical path depth mode realize.Assuming that B (i) is the binary representation of integer i,
B(i)[j](i,j>0) indicate that the jth bit digital of B (i) is so beaten i-th of parallel search task using following strategy
Random operation order：
If 1) B (i) [j]=1, then the corresponding jth layer operation sequence of data flow diagram should just be upset at random.For
Each of the j layers weighting critical path depth CPw (G (op)) for operating it should become [rand () mod N]/(N+1)-CPw
(G) × j, rand () are a random integers, and N is the total number operated in data flow diagram.
If 2) B (i) [j]=0, then the weighting critical path depth (CPw (G)) of the corresponding jth layer of data flow diagram is answered
This subtracts CPw (G) × j.Therefore, as in Fig. 2, task 0 is initial schedule sequence, and the first layer dispatching sequence of task 1 is disturbed,
The second layer of task 2 is disturbed, and the first layer of task 3 is disturbed with the second layer.Its dispatching sequence after upsetting is according to changing
Weighting critical path depth after change carries out.
As shown in Fig. 3 (a), in the parallel quick execution of search mission, the present invention is based on traditional branch-bound algorithm
On the basis of, by excavating the structural information of data flow diagram, a kind of more efficient level demarcation is proposed for resource constraint problem
Algorithm.Fig. 3 (a) illustrates the basic thought of the middle-level key-machine of the present invention：Layering.One layering refers to a collection on side
It closes, figure can be divided into two parts by it：A part is operated comprising all inputs, and another part includes that all output operates.
It re-defines K layers of all operations and refers to all set for passing through the K layers of input node to K+1 layers of side.Therefore divide in Fig. 3 (a)
For two layers, correspond to：All operations of first layer：{V1,V2,V3}；All operations of the second layer：{V1,V4}.Fig. 3 (b) gives
A kind of optimal scheduling (Sbsf, up to the present) of data flow diagram G, Fig. 3 (c) give a unfinished scheduling S of G.Figure
In mark the nodename and corresponding scheduling time in the bracket that each operates.Assuming that all nodes of first layer are all in S
All scheduling is completed, and other nodes not yet determine.In order to avoid meaningless search, traditional B&B methods (such as BULB)
Only the length of Sbsf is compared with the upper limit of S (it is 3 helpless that the two is equal at present).But in Fig. 3 (b) and Fig. 3
(c) in, the scheduling time of all first layer operations of Sbsf is all poor unlike S, this scheduling structure information (i.e. part operation and phase
Clock arrangement caused by relationship between the pre action answered) it can be used for trimming search space.
It is a unfinished scheduling for given data flow diagram a G, S, Sbsf is optimal scheduling so far.
Assuming that OPk is the set of K layer operations, if all operations have all been scheduled in OPk, just when following condition is set up
Level can be used to delimit method to be trimmed：
1) OPk has been scheduled and has sent；
2) scheduling time of all K layer operations corresponding nodes is small unlike in Sbsf in S；
3) scheduling time of K layer operations corresponding node, at least one was bigger than in Sbsf in S.
For a given data flow diagram, it is assumed that i-th layer comprising K operate, i.e. OPk=opi1, opi2 ...,
Opik }, and K layer operations are finished by all scheduling.If Sbsf indicates that optimal scheduling up to the present, S indicate current
Scheduling.Trimming condition is delimited when i layer operations meet level, then there is an operation opi j (1≤j≤k) so that S
(opi j)>Sbsf(opi j).If OP is the operational set comprising OPk and its pre action, and their clock step
As the control rooms S, then local optimum dispatches Sl, OP would not be more preferable than Sbsf.Since Sbsf is as B&B algorithms are passed
Return and constantly newer optimal scheduling, then current scheduling S can be cut off.
As shown in Fig. 4 pseudocodes, between search mission in collaboration update, in order to more effectively share constraints, constraint
Condition includes the Border condition in the length of current global optimum's scheduling, the interior optimal scheduling length of each grouping and each grouping.
The present invention has developed a special collaborative framework, while multiple level being supported to delimit trimming (multi-level bound
Pruning) and the minimum upper limit (minimum ω) is shared between parallel search task.In this frame, searching parallel
Rope task carries out piecemeal according to parallel search task ID (task IDs), for example several tasks with adjacent ID can be divided into
One group, this is because there is parallel search task the low level of same sequence to operate, but high level operation sequence is different.Cause
This, the constraints of parallel search task sharing has different high-level scheduling times, so as to act synergistically, with effective
Avoid the recursive search of deep layer.Each task block shares a local constraint pond to share the constraint of parallel search task
Condition.In the event of new constraints, the same task search mission in the block can update the local restriction of oneself with it
Pond.In order to make to cooperate between all task blocks, the present invention constructs a global data structures and is called optimum, can be with
(i.e. current global optimum's scheduling length, is equal to the length for the optimal scheduling that inquiry and update search mission are found so far
Minimum ω) and its ID (corresponding to schedule sequences comprising it).Any one parallel task is being found than current optimum
Scheduling can update this data structure when more preferably dispatching, it can also periodically inquire optimum scheduling, and more preferable at it
In the case of update local ω numerical value, to reduce the search space of parallel task.Therefore, frame can very big journey on the whole
Reduce the search time of parallel task in degree ground.When any one parallel search task finds optimal scheduling, entire frame can
To terminate, and export optimal scheduling
Resource constraint dispatching optimization method proposed by the present invention based on parallel-structure cognition technology can compared with
The optimal scheduling under resource constraint is obtained in the short time, greatly saves the resources such as time, energy consumption, promotes user's body
It tests.
The protection content of the present invention is not limited to above example.Without departing from the spirit and scope of the invention, originally
Field technology personnel it is conceivable that variation and advantage be all included in the present invention, and with appended claims be protect
Protect range.
Claims (6)
1. a kind of resource constraint dispatching optimization method of parallel organization cognition technology, which is characterized in that including step：
Step 1：The operation order for setting initial ranging task in parallel search task, according to weighting critical path depth setting
The high level operation enumeration order of remaining search mission；
Step 2：The structural information for obtaining data flow diagram in parallel search is wherein respectively operated based on resource constraint scheduling delimitation
Level；Search space is trimmed using the level key-machine proposed on the basis of based on traditional branch-bound algorithm, wherein according to
Whether meet at least one constraints according to the operation, abandon the operation for meeting constraints to adjust parallel search task
Search space；
Step 3：Collaborative framework is established in the parallel search task, constraints is shared between search mission；
Step 4：After any described search task obtains optimal scheduling sequence or the search of all search missions, output is most
Excellent dispatching sequence.
2. the resource constraint dispatching optimization method of parallel organization cognition technology as described in claim 1, feature exist
In the weighting critical path depth is related to the time loss of operation itself and dispatching sequence.
3. the resource constraint dispatching optimization method of parallel organization cognition technology as described in claim 1, feature exist
In at least one constraints is：The expected upper limit of incomplete scheduling is equal to optimal scheduling length.
4. the resource constraint dispatching optimization method of parallel organization cognition technology as described in claim 1, feature exist
In at least one constraints is：The scheduling time of at least one operation is more than the scheduling time of current optimal scheduling.
5. the resource constraint dispatching optimization method of parallel organization cognition technology as described in claim 1, feature exist
In at least one constraints is：The scheduling time of all operations is not less than the scheduling time of current optimal scheduling.
6. the resource constraint dispatching optimization method of parallel organization cognition technology as described in claim 1, feature exist
In the constraints includes being total in the length of current global optimum's scheduling, the interior optimal scheduling length of each grouping and each grouping
Enjoy boundary condition.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510863693.1A CN105511866B (en) | 2015-12-01 | 2015-12-01 | Resource constraint dispatching optimization method based on parallel organization cognition technology |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510863693.1A CN105511866B (en) | 2015-12-01 | 2015-12-01 | Resource constraint dispatching optimization method based on parallel organization cognition technology |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN105511866A CN105511866A (en) | 2016-04-20 |
| CN105511866B true CN105511866B (en) | 2018-11-09 |
Family
ID=55719876
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201510863693.1A Active CN105511866B (en) | 2015-12-01 | 2015-12-01 | Resource constraint dispatching optimization method based on parallel organization cognition technology |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN105511866B (en) |
Families Citing this family (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2019136758A1 (en) * | 2018-01-15 | 2019-07-18 | 深圳鲲云信息科技有限公司 | Hardware optimization method and system of artificial intelligence processing apparatus, storage medium and terminal |
| CN115809620B (en) * | 2022-11-25 | 2023-11-14 | 芯华章科技(北京)有限公司 | Method for simulating logic system design, electronic device and storage medium |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101923472A (en) * | 2010-09-16 | 2010-12-22 | 复旦大学 | Structure optimization method and system based on higher-level intermediate representation (HIR) |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US7366706B2 (en) * | 2004-11-23 | 2008-04-29 | International Business Machines Corporation | Method and apparatus for solution-template based deployment and management of an integration solution |
-
2015
- 2015-12-01 CN CN201510863693.1A patent/CN105511866B/en active Active
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101923472A (en) * | 2010-09-16 | 2010-12-22 | 复旦大学 | Structure optimization method and system based on higher-level intermediate representation (HIR) |
Non-Patent Citations (1)
| Title |
|---|
| 分支定界算法的分布并行化研究;李一明等;《计算机应用》;20060331;第26卷(第3期);全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN105511866A (en) | 2016-04-20 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Liu et al. | A branching-process-based method to check soundness of workflow systems | |
| CN110321999A (en) | Neural computing figure optimization method | |
| CN103488537B (en) | Method and device for executing data ETL (Extraction, Transformation and Loading) | |
| Li et al. | A hybrid Pareto-based local search algorithm for multi-objective flexible job shop scheduling problems | |
| CN102866912A (en) | Single-instruction-set heterogeneous multi-core system static task scheduling method | |
| Ma et al. | Multivoltage floorplan design | |
| CN103235743A (en) | Method for scheduling multi-target testing task based on decomposition and optimal solution following strategies | |
| CN105450684B (en) | Cloud computing resource scheduling method and system | |
| CN103488531B (en) | A kind of software and hardware mixed real-time application dispatching method based on polycaryon processor and FPGA | |
| CN101901161A (en) | Energy consumption related software/hardware partition-oriented hierarchical control and data flow graph modeling method | |
| CN105511866B (en) | Resource constraint dispatching optimization method based on parallel organization cognition technology | |
| Laalaoui et al. | Pre-run-time scheduling in real-time systems: Current researches and artificial intelligence perspectives | |
| Chen | Efficiency improvement in explicit enumeration for integer programming problems | |
| CN100531070C (en) | Network resource scheduling simulation system | |
| CN106055862A (en) | Novel efficient heuristic-type two-stage parallel branch-and-bound method | |
| CN111090613B (en) | Low-complexity hardware and software partitioning and scheduling method based on graph partitioning | |
| CN103810041A (en) | Parallel computing method capable of supporting dynamic compand | |
| CN102253837A (en) | Object tree-based software framework designing technology | |
| CN111190711B (en) | BDD combined heuristic A search multi-robot task allocation method | |
| Abraham et al. | Efficient design space exploration in PICO | |
| Deniziak et al. | Hardware/software co-synthesis of distributed embedded systems using genetic programming | |
| CN102571962B (en) | Service composition path construction method for cloud manufacturing | |
| US20070083411A1 (en) | Schedule coordinates method for project scheduling | |
| Mollajafari et al. | A repair-less genetic algorithm for scheduling tasks onto dynamically reconfigurable hardware | |
| Xiao et al. | Parallel custom instruction identification for extensible processors |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |