CN106126878A - The coarse granule parallel method of electromagnetic functional material optimization design and system - Google Patents
The coarse granule parallel method of electromagnetic functional material optimization design and system Download PDFInfo
- Publication number
- CN106126878A CN106126878A CN201610413026.8A CN201610413026A CN106126878A CN 106126878 A CN106126878 A CN 106126878A CN 201610413026 A CN201610413026 A CN 201610413026A CN 106126878 A CN106126878 A CN 106126878A
- Authority
- CN
- China
- Prior art keywords
- optimization
- coarse granule
- functional material
- calculating
- electromagnetic
- 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
Classifications
-
- G—PHYSICS
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16Z—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS, NOT OTHERWISE PROVIDED FOR
- G16Z99/00—Subject matter not provided for in other main groups of this subclass
Landscapes
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The present invention relates to functional material design and high-performance computing sector, particularly relate to coarse granule parallel method and the system of electromagnetic functional material optimization design.The method includes: obtaining with coarse granule is the serial version independently executing module, and is basic performance element with coarse granule, it is thus achieved that parallel version.Further, using randomly ordered allocation strategy to process multiple calculating tasks, this randomly ordered allocation strategy can thoroughly upset the allocation order of all calculating tasks, obtains the new sequence of calculating task.Then, each calculation procedure uses the allocation strategy application of first to file elder generation to distribute calculating task to be allocated under new sequence, and calculates, until all calculating tasks complete.Finally, statistical computation result also carries out targeted process, result is compared with expection optimization aim, it may be judged whether reach optimization aim.This coarse granule parallel method effectively solves the problem that in present stage electromagnetic functional material optimization design, parallel efficiency calculation is low.
Description
Technical field
The present invention relates to functional material design and high-performance computing sector, particularly relate to electromagnetic functional material optimization design
Coarse granule parallel method and system.
Background technology
Electromagnetic functional material, owing to possessing certain specific electromagnetic response feature, is widely used in various Functional Unit device,
Pivotal role is played, development and the shape of NPD projects to new and high technology in major fields such as information technology, electromagnetic protection, national defence
One-tenth has great importance.For obtaining electromagnetic functional material more wide-band, more excellent electromagnetic response feature, need according in advance
The electromagnetic response technical specification set, aspect carries out strict optimization design from material medium selection to stereochemical structure etc..Electromagnetism
Functional material optimization design is required for optimized algorithm and each of provides combination of media and the stereochemical structure of formation thereof, sets up electricity
Magnetic functional material is individual, calculates its electromagnetic response under a series of frequencies.For realizing the electricity of certain electromagnetic response technical specification
Magnetic functional material optimization designs, and is required for a large amount of Different Individual and carries out extensive Numerical Calculation of Electromagnetic Fields.This kind of extensive number
Value calculates owing to different calculated examples have different structure, thus generates the subdivision grid of varying number, causes different calculating in fact
The computation complexity of example is the most reciprocity, calculates for the most reciprocity this kind of magnanimity, needs high efficiency parallel calculating method to design, fully
Consider the most reciprocity of different instances computation complexity, improve parallel efficiency calculation as far as possible.
Conventional parallel calculates the most parallel for single calculated examples, say, that all processes carry out body one by one simultaneously
Calculating under some optimizes frequency, the calculating section in a large amount of circulations realizes parallel, and parallel granule is the thinnest, so causes
There is the exchange of substantial amounts of data between different processes, reduce parallel efficiency；Secondly as different processes calculate progress difference, no
A large amount of wait occurs when can avoid in needs data sharing and synchronize, thus causes overall parallel efficiency the lowest；Furthermore, due to list
In individual example calculation, substantial portion of calculating process has sequencing, data to have dependency, therefore for single calculated examples also
During row, have substantial portion of calculating cannot parallelization, this is also serious reduces overall parallel efficiency.
Summary of the invention
It is an object of the invention to provide coarse granule parallel method and the system of a kind of electromagnetic functional material optimization design, with
Solve the problem that in present stage electromagnetic functional material optimization design, parallel efficiency calculation is low.
The invention provides the coarse granule parallel method of a kind of electromagnetic functional material optimization design, comprising:
Step 1: according to the optimization aim of electromagnetic functional material, and electromagnetic functional material frequency-response characteristic curve is continuous
Property principle, choose more than one and optimize frequency, and selected optimized algorithm；
Step 2: generate more than one initial optimization according to described optimized algorithm individual；Described initial optimization individuality is electromagnetism
The structure of functional material；
Step 3: build and perform described initial optimization individuality electromagnetic response under all optimization frequencies and individual goal letter
Count the coarse granule calculated, and acquisition includes described coarse grained serial version；
Step 4: on the basis of serial version, is basic performance element with coarse granule, it is thus achieved that multiple processes calculate simultaneously
The parallel version of multiple calculating tasks；
Step 5: utilize described optimized algorithm to generate more than one and optimize individuality, optimizes individual under each optimization frequency
Electromagnetic response and optimization individual goal function calculate and form the calculating task that coarse granule performs, and use randomly ordered allocation strategy
The calculating task performing coarse granule processes, and obtains the new sequence of calculating task；
Step 6: each calculation procedure uses calculating to be allocated under the first to file elder generation allocation strategy application new sequence of distribution to appoint
Business, and call corresponding coarse granule and calculate；
Step 7: judge whether there is calculating task to be allocated in new sequence, if existing, returning step 6, otherwise, carrying out
Step 8；
All result of calculations are carried out targeted process by step 8: collect all result of calculation, and by result with pre-
Phase optimization aim compares, if being not reaching to optimization aim, carrying out step 5, otherwise, terminating.
In certain embodiments, being preferably, after step 3, before step 4, described coarse granule parallel method also includes:
In debugging described serial version, can coarse granule properly functioning；If can be properly functioning, carry out step 4；If being not normally functioning, enter
Row step 3.
In described step 6, described first to file elder generation allocation strategy is: the first calculation procedure is in application distribution the first calculating task
Time, generate the first status file and the first calculating task is marked, if described first status file exists, then show first
Calculating task has been allocated, and the first calculation procedure applies for distributing the second calculating task automatically；The generation side of described status file
Formula is: file locking and unlocking technology.
Described file locking with unlocking technology is:
If status file is to write locking state, then before described status file is unlocked, all to described status file
The thread locked all can be blocked；
If status file is reading locking state, all threads locked described status file in a read mode can
Obtain access right, will be blocked with the thread that described status file is locked by WriteMode；
If status file is in the lock status of reading mode, and other thread is had with WriteMode, described status file to be locked,
Described status file can block the request that other reading modes lock.
Described step 2 includes: generates more than one according to described optimized algorithm and optimizes individuality；Optimization number of individuals is defined as
Calculate number of tasks.The most described step 7 judging, whether there is calculating task to be allocated in new sequence includes: statistics file locks
The calculating task status quantity of documents generated with unlocking technology；If calculating task status quantity of documents and described calculating number of tasks phase
With, the most all calculating tasks are allocated, and otherwise, there is calculating task to be allocated.
Described electromagnetic functional material is to have the electromagnetic response biomaterials meeting particular requirement in particular frequency range, bag
Include: absorbing material, electromagnetic shielding material, frequency selective material or AF panel material.Described optimized algorithm includes differential evolution
Algorithm, ant group algorithm or genetic Optimization Algorithm.
The present invention proposes the coarse granule parallel system of a kind of electromagnetic functional material optimization design, comprising:
Chosen module, for the optimization aim according to electromagnetic functional material, and electromagnetic functional material frequency-response characteristic
The successional principle of curve, chooses more than one and optimizes frequency, and selected optimized algorithm；
Generation module, individual for generating more than one initial optimization according to described optimized algorithm；Described initial optimization
Body is the structure of electromagnetic functional material；
Build module, perform described initial optimization individuality electromagnetic response under all optimization frequencies and individuality for building
Object function calculate coarse granule, and obtain include described coarse grained serial version；Further, on the basis of serial version,
It it is basic performance element with coarse granule, it is thus achieved that parallel version；
Optimize and order module, be used for utilizing described optimized algorithm to generate more than one and optimize individuality, optimize individual often
Electromagnetic response and optimization individual goal function under individual optimization frequency calculate and form the calculating task that coarse granule performs, and use at random
Calculating task is processed by the allocation strategy of sequence, obtains the new sequence of calculating task；
Distribution module, is used for dispatching each calculation procedure and uses under the first to file elder generation allocation strategy application new sequence of distribution to be allocated
Calculating task, and call corresponding coarse granule and calculate；
Judge module, for judging whether there is calculating task to be allocated in new sequence；
Comparison module, is used for collecting all result of calculation, all result of calculations carries out targeted process, and process is tied
Fruit compares with expection optimization aim.
The coarse granule parallel method of the electromagnetic functional material optimization design that the embodiment of the present invention provides and system, with existing skill
Art is compared, and first acquisition comprises the serial version of the standalone module that coarse granule is formed and is that basic performance element is carried out with coarse granule
Parallel, rather than the calculating section of a large amount of circulations of prior art realizes parallel, and then improves parallel efficiency calculation, wherein coarse granule
Perform one and optimize individual electromagnetic response under all optimization frequencies and the calculating of individual goal function.On this basis, use
The calculating task that coarse granule is performed by randomly ordered allocation strategy processes.This randomly ordered allocation strategy can thoroughly be beaten
Disorderly the allocation order of all calculating tasks, can avoid relative complex calculating task to concentrate on some node and cause the plenty of time
Disk read-write, thus realize peak memory summation that the task that each calculating node calculates simultaneously takies by process number and all moulds
Type takies the meansigma methods of peak memory rather than peak determines.Then, each calculation procedure uses first to file elder generation allocation strategy Shen
Please distribute calculating task, and calculate, until all calculating tasks complete.This first to file elder generation allocation strategy calculates according to each
The performance of process carries out dynamic calculation task distribution, thus avoids different node, the different calculation procedure of calculating due to meter
The speed of calculation speed causes and waits as long for.
From the foregoing, it will be observed that the coarse granule parallel method that the electromagnetic functional material optimization that the present invention passes through designs, effectively solve
The problem that in present stage electromagnetic functional material optimization design, parallel efficiency calculation is low.
Accompanying drawing explanation
Fig. 1 is the coarse granule parallel method step signal of electromagnetic functional material optimization design in one embodiment of the invention
Figure；
Fig. 2 is random assortment strategic process schematic diagram in one embodiment of the invention；
Fig. 3 is first to file elder generation allocation strategy schematic flow sheet in one embodiment of the invention；
Fig. 4 is file labelling technique labelling schematic flow sheet in one embodiment of the invention.
Detailed description of the invention
For making the purpose of the embodiment of the present invention, technical scheme and advantage clearer, below in conjunction with the embodiment of the present invention
In accompanying drawing, the technical scheme in the embodiment of the present invention is clearly and completely described, it is clear that described embodiment is
A part of embodiment of the present invention rather than whole embodiments.Based on the embodiment in the present invention, ordinary skill people
The every other embodiment that member is obtained on the premise of not making creative work, broadly falls into the scope of protection of the invention.
For the problem that parallel efficiency calculation in the electromagnetic functional material optimization design of present stage is low, the present invention proposes one
Plant the coarse granule parallel method of electromagnetic functional material optimization design.As it is shown in figure 1, it specifically includes:
Step 1: according to the optimization aim of electromagnetic functional material, and electromagnetic functional material frequency-response characteristic curve is continuous
Property principle, choose more than one and optimize frequency, and selected optimized algorithm；
Wherein, electromagnetic functional material is to have the electromagnetic response biomaterials meeting particular requirement in particular frequency range,
Including: absorbing material, electromagnetic shielding material, frequency selective material or AF panel material, optimized algorithm includes that differential evolution is calculated
Method, ant group algorithm or genetic Optimization Algorithm；
Step 2: generate more than one initial optimization according to optimized algorithm individual, optimizes the knot that individuality is electromagnetic functional material
Structure, initial optimization individuality is equivalent to the initial value needed for program optimization；
Step 3: build and perform optimization individuality electromagnetic response under all optimization frequencies and individual goal function meter
Calculate coarse granule, and obtain include described coarse grained serial version；
Wherein, coarse granule is carried out individual electromagnetic response and an individual goal function meter under all selected frequencies of optimization
The program code calculated, serial version is not only to include coarse granule, also includes optimizing individual parameters, i.e. serial version can
Carry out the reading of customer parameter, judge whether to reach optimization aim, form new optimization individuality, and output according to optimized algorithm
Result of calculation etc.；
Step 4: on the basis of serial version, is basic performance element with coarse granule, it is thus achieved that multiple processes calculate simultaneously
The parallel version of multiple calculating tasks, parallel multiple processes use same coarse granule to go to perform calculating task；
Step 5: utilize optimized algorithm to generate the more than one individuality that optimizes, the individual electricity under each optimization frequency of optimization
Magnetic response and individual goal function calculate and form the calculating task that coarse granule performs, and use randomly ordered allocation strategy to thick
The calculating task that grain performs processes, and obtains the new sequence of calculating task；
Randomly ordered allocation strategy is: the order of the task of calculating is upset at random, produces new random sequence；Calculate and appoint
Business is one and optimizes individual electromagnetic response on all optimization frequencies and the calculating of individual goal function；
Step 6: each calculation procedure uses calculating to be allocated under the first to file elder generation allocation strategy application new sequence of distribution to appoint
Business, and call corresponding coarse granule and calculate；
First to file elder generation allocation strategy is: the first calculation procedure, when application distribution the first calculating task, generates the first state
File is marked first, if the first status file exists, then shows that the first calculating task has been allocated, and first calculates
Process applies for distributing the second calculating task automatically；
Step 7: judge whether there is calculating task to be allocated in new sequence, if existing, returning step 6, otherwise, carrying out
Step 8；
All result of calculations are carried out targeted process by step 8: collect all result of calculation, and by result with pre-
Phase optimization aim compares, if being not reaching to optimization aim, carrying out step 5, otherwise, terminating.
In step 2 above, optimizing the structure that individuality is electromagnetic functional material, wherein, the structure of electromagnetic functional material is not
Only refer to the combined material structure of electromagnetic functional material, also include that the space structure of each combined material, different medium are compound
Structure, medium composition etc..Assume to optimize the composite construction absorbing material of 10 layers, then 10 layer material, every layer thickness, every layer of selection etc.
Etc. can be formed countless versions may, every kind may be all the structure of a kind of composite wave-suction material.
After step 3, before step 4, described coarse granule parallel method also includes: debug in described serial version thick
Can grain properly functioning；If can be properly functioning, carry out step 4；If being not normally functioning, carry out step 3.In judging serial version
Whether all coarse granules can be parallel, are to determine that obtained serial version can carry out parallel optimization, i.e. can walk
Rapid 3 below step.Otherwise, can only again revise coarse granule block code, to ensure that the code performing calculating task realizes.
In steps of 5, first to file elder generation allocation strategy is realized by file mark technology, and the generation side of status file
Formula is: file locking and unlocking technology.
As shown in Figure 4, file locking ensures once can only a technology process to read by file read-write lock with unlocking technology
Or write same calculating task；And have the feature that
If status file is to write locking state, then before status file is unlocked, all lines that status file is locked
Journey all can be blocked；
If status file is reading locking state, all threads locked status file in a read mode can obtain
Access right, will be blocked with the thread that status file is locked by WriteMode；
If status file is in the lock status of reading mode, and other thread is had with WriteMode, status file to be locked, state
File can block the request that other reading modes lock.
Further, step 2 includes: generates more than one according to optimized algorithm and optimizes individuality, and determine calculating number of tasks, calculates
Number of tasks is for optimizing number of individuals.
Judge whether the calculating task that all coarse granules perform completes to include the most in step 7: by file locking and solution
The calculating task status quantity of documents that lock technology generates is identical with calculating number of tasks, and the most all calculating tasks complete, and otherwise, do not have
Complete.Further, in definable parallel computation, a process in multiple processes is host process, and host process is responsible for appointing all calculating
Business is assigned to each process, is responsible for collecting the result of calculation (target function value) of each process and completing optimized algorithm according to object function
Each step.According to the result of calculation collected, host process also will judge whether calculating task completes.
In above-mentioned steps, electromagnetic response calculate refer to obtain one optimize individual response under frequency (for
Different Optimization purpose, response here can be the suction ripple usefulness of material, shield effectiveness, AF panel usefulness etc.) and carry out
Calculate, including the amount being modeled, recycling numerical computations acquisition description response characteristic according to individual architectural feature.In program
In compiling procedure, by electromagnetic response optimizing individual (specific composite construction electromagnetic functional material) each under each frequency and individual
Body object function calculates and writes as an independent program module, it is simple to the serial meter in each process in follow-up parallel procedure
Calculate.Further, using independent program module (optimizes individual calculation procedure module under all selected frequencies) as parallel
Elementary cell be known as that coarse granule is parallel rather than circulation thinner inside stand-alone program module is as parallel elementary cell.
Meanwhile, during optimizing individuality calculating, in order to form description electromagnetic functional material response spy in optimizing frequency range
The curve levied and the calculating carrying out discrete point in frequency, in order to form the smooth curve of reflection response characteristic, need to calculate than typical case
Optimize the frequency that frequency is much more, such as the suction ripple usefulness in the range of calculating 1GHz～18GHz, use the sampling frequency of step-length 0.1GHz
Point, needs to calculate 181 points, and after having calculated, the suction ripple usefulness of 181 points being drawn, (abscissa is that frequency, vertical coordinate are for counting
The suction ripple usefulness calculated), formed and inhale ripple efficacy curve.
Randomly ordered allocation strategy and first to file elder generation allocation strategy are collectively referred to as stochastic and dynamic distribution and calculate task strategy, its
In " at random " refer to random assortment strategy, " dynamically " distribution refers to first to file elder generation allocation strategy.Randomly ordered allocation strategy is
For calculating task, as illustrated in fig. 2, it is assumed that there are 100 calculating tasks, numbered 1,2 ..., 100, random assortment refers to
The most by number 1,2,3 ... it is sequentially allocated, but distributes by random order.The purpose of do so is to avoid complicated calculations to concentrate
Calculating on node at some, such as 100 calculate task, front 50 computation model complexity, need 20GB internal memory, rear 50 calculating
Model is simple, it is only necessary to 10GB internal memory.These 100 calculating tasks are completed, if pressed by 2 calculating nodes having 32GB internal memory
Order-assigned, each node distributes two and calculates task, then having the substantial portion of time is that two complex tasks calculate simultaneously, peak
Saving as 40GB in value, cause the disk read-write of plenty of time, randomly ordered allocation strategy then can reduce significantly or keep away
Exempt from this problem.First to file elder generation allocation strategy is then for calculating for node, and first distributing corresponding with first to file is to divide in advance
Join, generally according to the task of calculating number and calculation procedure number be allocated to each calculation procedure in advance, after being assigned,
Can not change during parallel computation below.As it is shown on figure 3, first to file elder generation allocation strategy is not disposably to distribute, but
Calculating progress according to different processes realizes dynamically distribution, has calculated a model in certain process, dynamically distribution next one meter
Before calculating model, needing to check that calculating task has been allocated, the task that only calculates is not previously allocated just from unassigned
Task in distribute next task give this process.Such as 100 calculate task, and 10 processes, if allocated in advance, are then every
Individual course allocation 10 calculates task (either order-assigned or random assortment), no matter occurs what situation during calculating
The calculating task of each process is first distributed all without change, first to file, is that the performance according to each process calculating task is carried out
Dynamic distribution, so can avoid different calculating node, different process waits as long for owing to the speed calculating speed causes,
This wait, probably due to the computing capability that causes due to hardware configuration difference of different node is different, ultimately results in the identical number of calculating
The CPU time of the required by task of amount is different, it is also possible to because the complexity difference of different computation model causes similarly configuring joint
The CPU time that point calculates different model is different.
Such as, differential evolution algorithm is utilized to optimize 3～10 layers of composite construction absorbing material of a kind of band frequency-selective surfaces,
Computation model shows, same type calculates owing to optimizing individual computation model structure different with complexity, causes mesh generation to produce
Raw element number has larger difference, thus Different Optimization individuality calculates the internal memory needed and also has larger difference.Statistical result table
Bright, if using second-order finite element to calculate, under same mesh subdivision parameter facilities, 3 layers of composite junction that optimized algorithm produces
Structure, with 1 layer of frequency-selective surfaces, the model of thickness 5mm, calculates required memory peak value minimum, about 8GB；Optimized algorithm produces
10 layers of raw composite construction, with 6 layers of frequency-selective surfaces, the model of thickness 5mm, calculate required memory peak value maximum, exceed
20GB.If an each node memory of cluster is 48GB, this cluster is utilized to use second-order finite element parallel computation, simplest
Model can open 6 processes simultaneously, and the most complicated model then can only open 2 processes simultaneously, and otherwise system can be hard by part
Disk space as virtual memory for program.At present conventional hard disk drive HDD R/W speed of hard disc at about 80MB/s,
And the read or write speed of physical memory has the raising of more than Radix Achyranthis Bidentatae, such as, for the server memory of DDR3 1333MHz, its data
Transfer rate reaches 10.6GB/s.This comparative result illustrates, if the process that parallel computation is opened is too much, causes calculating process
Middle part hard-disc storage space is taken as virtual memory to read, and program operation speed will be made to reduce more than Radix Achyranthis Bidentatae.For avoiding meter
During calculation, part hard-disc storage space is taken as the phenomenon that virtual memory reads, when the process of unlatching, need to consider each enter
The maximum memory that may need during Cheng Yunhang, determines the maximum process number that each node can be opened on this basis.If used
Commonsense method coarse granule parallel computation, then every node multipotency opens 2 processes.Test result indicate that, use the present invention with
The distribution of motor-driven state calculates task strategy, and every node opens 4 processes, and memory usage is for a long time more than 80%, substantially accomplishes
Dislocation in Peak Memory Usage equalization and internal memory peak value time of occurrence.
The file locking using C++ to realize is as follows with the typical code of unlocking technology:
1) creating mark file also to lock, other processes can not read/write this document:
out.open(filename,_SH_DENYRW)；
2) reading mark file also to lock, other processes can not write this document:
ifstream in(filename,ios::in,_SH_DENYWR)；
3) writing mark file locking, other processes can not read/write this document:
out.open(filename,ios::app,_SH_DENYRW)；
Reader writer's problem is there is, say, that the access to some resource can exist two kinds of possibilities in some programs
Situation, a kind of is that to access must be exclusive row, it is simply that the exclusive meaning, this is referred to as write operation；Another kind of situation is visited exactly
The mode of asking can be shared, that is multiple thread can be had to go to access certain resource simultaneously, this is referred to as read operation.This
Individual problem model is to extend out from the read-write operation to file.
Read-Write Locks and mutex (mutex) are all the mechanism that thread synchronization is conventional, and Read-Write Locks has higher compared with mutex
The suitability, has higher concurrency, multiple thread can be had simultaneously to take the Read-Write Locks of reading mode, but can only have a line
The Read-Write Locks of Cheng Zhanyong WriteMode, three kinds of states of Read-Write Locks:
1. when Read-Write Locks is to write locking state, before this lock is unlocked, all lines attempting this lock is locked
Journey all can be blocked
2. when Read-Write Locks is when reading locking state, and all threads attempting to lock it in a read mode can obtain
Access right, but will be blocked to the thread that it locks with WriteMode
3., when the Read-Write Locks lock status at reading mode, lock if there being other thread to attempt with WriteMode, Read-Write Locks
The request that reading mode subsequently is locked would generally be blocked, reading mode lock so can be avoided to take for a long time, and the WriteMode lock waited
Ask then long-term blockage.
For the coarse granule parallel method of above-mentioned electromagnetic functional material optimization design, the invention also discloses a kind of electromagnetism merit
The coarse granule parallel system of energy optimization of material design, comprising:
Chosen module, for the optimization aim according to electromagnetic functional material, and electromagnetic functional material frequency-response characteristic
The successional principle of curve, chooses more than one and optimizes frequency, and selected optimized algorithm；
Generation module, individual for generating more than one initial optimization according to optimized algorithm；Optimizing individuality is function solenoid
The structure of material；
Build module, perform initial optimization individuality electromagnetic response under all optimization frequencies and individual goal for building
Function calculate coarse granule, and obtain include described coarse grained serial version；Further, on the basis of serial version, with slightly
Granule is basic performance element, it is thus achieved that parallel version；
Optimize and order module, be used for utilizing described optimized algorithm to generate more than one and optimize individuality, optimize individual often
Electromagnetic response and optimization individual goal function under individual optimization frequency calculate and form the calculating task that coarse granule performs, and use at random
Calculating task is processed by the allocation strategy of sequence, obtains the new sequence of calculating task；
Distribution module, is used for dispatching each calculation procedure and uses under the first to file elder generation allocation strategy application new sequence of distribution to be allocated
Calculating task, and call corresponding coarse granule and calculate；
Judge module, for judging whether there is calculating task to be allocated in new sequence；
Comparison module, is used for collecting all result of calculation, all result of calculations carries out targeted process, and process is tied
Fruit compares with expection optimization aim.
These are only the preferred embodiments of the present invention, be not limited to the present invention, for those skilled in the art
For Yuan, the present invention can have various modifications and variations.All within the spirit and principles in the present invention, any amendment of being made,
Equivalent, improvement etc., should be included within the scope of the present invention.
Claims (9)
1. the coarse granule parallel method of an electromagnetic functional material optimization design, it is characterised in that including:
Step 1: according to the optimization aim of electromagnetic functional material, and electromagnetic functional material frequency-response characteristic curve is successional
Principle, chooses more than one and optimizes frequency, and selected optimized algorithm；
Step 2: generate more than one initial optimization according to described optimized algorithm individual；Described initial optimization individuality is function solenoid
The structure of material；
Step 3: build and perform described initial optimization individuality electromagnetic response under all optimization frequencies and individual goal function meter
Calculate coarse granule, and obtain include described coarse grained serial version；
Step 4: on the basis of serial version, is basic performance element with coarse granule, it is thus achieved that parallel version；
Step 5: utilize described optimized algorithm to generate more than one and optimize individuality, optimize individual electromagnetism under each optimization frequency
Response and optimization individual goal function calculate and form the calculating task that coarse granule performs, and use randomly ordered allocation strategy to meter
Calculation task processes, and obtains the new sequence of calculating task；
Step 6: each calculation procedure uses the allocation strategy application of first to file elder generation to distribute calculating task to be allocated under new sequence, and
Call coarse granule to calculate；
Step 7: judge whether there is calculating task to be allocated in new sequence, if existing, returning step 6, otherwise, carrying out step
8；
All result of calculations are carried out targeted process by step 8: collect all result of calculation, and by excellent with expection for result
Change target to compare, if being not reaching to optimization aim, carrying out step 5, otherwise, terminating.
2. the coarse granule parallel method of electromagnetic functional material optimization design as claimed in claim 1, it is characterised in that in step
After 3, before step 4, described coarse granule parallel method also includes:
In debugging described serial version, can coarse granule properly functioning；
If can be properly functioning, carry out step 4；
If being not normally functioning, carry out step 3.
3. the coarse granule parallel method of electromagnetic functional material optimization design as claimed in claim 1, it is characterised in that described step
In rapid 6, described first to file elder generation allocation strategy is: the first calculation procedure, when application distribution the first calculating task, generates the first shape
First calculating task is marked by state file, if described first status file exists, has then shown the first calculating task
Allocated, the first calculation procedure applies for distributing the second calculating task automatically；
The generating mode of described status file is: file locking and unlocking technology.
4. the coarse granule parallel method of electromagnetic functional material optimization design as claimed in claim 3, it is characterised in that described literary composition
Part locking with unlocking technology is:
If status file is to write locking state, then before described status file is unlocked, all described status file is locked
Thread all can be blocked；
If status file is reading locking state, all threads locked described status file in a read mode can obtain
Access right, will be blocked with the thread that described status file is locked by WriteMode；
If status file is in the lock status of reading mode, and other thread is had with WriteMode, described status file to be locked, described
Status file can block the request that other reading modes lock.
5. the coarse granule parallel method of electromagnetic functional material optimization design as claimed in claim 3, it is characterised in that described step
Rapid 2 include:
Generate more than one according to described optimized algorithm and optimize individuality；
It is defined as optimization number of individuals calculating number of tasks.
6. the coarse granule parallel method of electromagnetic functional material optimization design as claimed in claim 5, it is characterised in that described step
Judge in rapid 7 that whether there is calculating task to be allocated in new sequence includes:
The calculating task status quantity of documents that statistics file locking generates with unlocking technology；
If it is identical with described calculating number of tasks to calculate task status quantity of documents, the most all calculating tasks are allocated, and otherwise, exist
Calculating task to be allocated.
7. the coarse granule parallel method of the electromagnetic functional material optimization design as described in any one of claim 1-6, its feature exists
In, described electromagnetic functional material is particular frequency range, there is the material meeting particular requirement electromagnetic response feature, including: inhale
Wave material, electromagnetic shielding material, frequency selective material or AF panel material.
8. the coarse granule parallel method of the electromagnetic functional material optimization design as described in any one of claim 1-6, its feature exists
In, described optimized algorithm includes differential evolution algorithm, ant group algorithm or genetic Optimization Algorithm.
9. the coarse granule parallel system of an electromagnetic functional material optimization design, it is characterised in that including:
Chosen module, for the optimization aim according to electromagnetic functional material, and electromagnetic functional material frequency-response characteristic curve
Successional principle, chooses more than one and optimizes frequency, and selected optimized algorithm；
Generation module, individual for generating more than one initial optimization according to described optimized algorithm；Described initial optimization individuality is
The structure of electromagnetic functional material；
Build module, perform described initial optimization individuality electromagnetic response under all optimization frequencies and individual goal for building
Function calculate coarse granule, and obtain include described coarse grained serial version；Further, on the basis of serial version, with slightly
Granule is basic performance element, it is thus achieved that parallel version；
Optimize and order module, be used for utilizing described optimized algorithm to generate more than one and optimize individuality, optimize individual each excellent
Change the electromagnetic response under frequency and calculate the calculating task forming coarse granule execution, use randomly ordered allocation strategy to appoint calculating
Business processes, and obtains the new sequence of calculating task；
Distribution module, for dispatch each calculation procedure use under the first to file elder generation allocation strategy application new sequence of distribution to be allocated based on
Calculation task, parallel version is called corresponding coarse granule and is calculated；
Judge module, for judging whether there is calculating task to be allocated in new sequence；
Comparison module, is used for collecting all result of calculation, and all result of calculations carry out targeted process, and by result with
Expection optimization aim compares.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610413026.8A CN106126878B (en) | 2016-06-13 | 2016-06-13 | The coarse granule parallel method and system of electromagnetic functional material optimization design |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610413026.8A CN106126878B (en) | 2016-06-13 | 2016-06-13 | The coarse granule parallel method and system of electromagnetic functional material optimization design |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN106126878A true CN106126878A (en) | 2016-11-16 |
| CN106126878B CN106126878B (en) | 2018-04-06 |
Family
ID=57270549
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610413026.8A Active CN106126878B (en) | 2016-06-13 | 2016-06-13 | The coarse granule parallel method and system of electromagnetic functional material optimization design |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106126878B (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111881643A (en) * | 2020-06-08 | 2020-11-03 | 北京智芯仿真科技有限公司 | Integrated circuit layout optimization method based on frequency domain electromagnetic response and capable of automatically starting and stopping process |
| CN111898332A (en) * | 2020-06-08 | 2020-11-06 | 北京智芯仿真科技有限公司 | Frequency domain simulation adaptive frequency point extraction and calculation method for very large scale integrated circuit |
| CN113128142A (en) * | 2021-04-27 | 2021-07-16 | 北京智芯仿真科技有限公司 | Integrated circuit electromagnetic simulation coarse grain parallel method and device based on file lock |
-
2016
- 2016-06-13 CN CN201610413026.8A patent/CN106126878B/en active Active
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111881643A (en) * | 2020-06-08 | 2020-11-03 | 北京智芯仿真科技有限公司 | Integrated circuit layout optimization method based on frequency domain electromagnetic response and capable of automatically starting and stopping process |
| CN111898332A (en) * | 2020-06-08 | 2020-11-06 | 北京智芯仿真科技有限公司 | Frequency domain simulation adaptive frequency point extraction and calculation method for very large scale integrated circuit |
| CN111898332B (en) * | 2020-06-08 | 2022-05-10 | 北京智芯仿真科技有限公司 | Frequency domain simulation adaptive frequency point extraction and calculation method for very large scale integrated circuit |
| CN113128142A (en) * | 2021-04-27 | 2021-07-16 | 北京智芯仿真科技有限公司 | Integrated circuit electromagnetic simulation coarse grain parallel method and device based on file lock |
Also Published As
| Publication number | Publication date |
|---|---|
| CN106126878B (en) | 2018-04-06 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN102902512B (en) | A kind of multi-threading parallel process method based on multi-thread programming and message queue | |
| Zhu et al. | Synchronization state buffer: supporting efficient fine-grain synchronization on many-core architectures | |
| CN106055401B (en) | Magnanimity calculates the parallel automatic start-stop and calculating task dynamic allocation method of coarse granule | |
| Camp et al. | Streamline integration using MPI-hybrid parallelism on a large multicore architecture | |
| Welton et al. | Mr. scan: Extreme scale density-based clustering using a tree-based network of gpgpu nodes | |
| US20120159507A1 (en) | Compiling apparatus and method of a multicore device | |
| CN106095574B (en) | Magnanimity calculates coarse granule Parallel Implementation and calculating task stochastic and dynamic distribution method | |
| US20100211753A1 (en) | Parallel garbage collection and serialization without per-object synchronization | |
| US20090327377A1 (en) | Copying entire subgraphs of objects without traversing individual objects | |
| Hendler et al. | A scalable lock-free stack algorithm | |
| CN102103525A (en) | Controlling time stamp counter (TSC) offsets for mulitple cores and threads | |
| CN106126878B (en) | The coarse granule parallel method and system of electromagnetic functional material optimization design | |
| CN102214086A (en) | General-purpose parallel acceleration algorithm based on multi-core processor | |
| Sassone et al. | Matrix scheduler reloaded | |
| CN102929711B (en) | A kind of implementation method of real-time transactional memory of software | |
| Su et al. | Bringing modular concurrency control to the next level | |
| Sharma et al. | How to databasify a blockchain: the case of hyperledger fabric | |
| Aydonat et al. | Serializability of transactions in software transactional memory | |
| WO2017131784A1 (en) | Hybrid concurrency control | |
| CN107844436A (en) | The organization and management method of dirty data, system and storage system in a kind of caching | |
| WO2011058657A1 (en) | Parallel computation device, parallel computation method, and parallel computation program | |
| Park et al. | Quantifying the performance and energy-efficiency impact of hardware transactional memory on scientific applications on large-scale NUMA systems | |
| CN106445735A (en) | Data recovery method and system for VMware virtual machine | |
| CN109656868A (en) | A kind of internal storage data transfer method between CPU and GPU | |
| CN106484532B (en) | GPGPU parallel calculating method towards SPH fluid simulation |
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 | ||
| TR01 | Transfer of patent right | ||
| TR01 | Transfer of patent right |
Effective date of registration: 20200324 Address after: Room 02b-350, building B (2f), No.28, information road a, Haidian District, Beijing 100000 Patentee after: Beijing Smart core Simulation Technology Co., Ltd Address before: 100034, room 1, No. 2028, Hu Ying Road, sheepskin City, Beijing, Xicheng District Patentee before: BEIJING WEIZHI JIACHEN TECHNOLOGY DEVELOPMENT Co.,Ltd. |