CN109635473B - Heuristic high-flux material simulation calculation optimization method - Google Patents
Heuristic high-flux material simulation calculation optimization method Download PDFInfo
- Publication number
- CN109635473B CN109635473B CN201811556665.5A CN201811556665A CN109635473B CN 109635473 B CN109635473 B CN 109635473B CN 201811556665 A CN201811556665 A CN 201811556665A CN 109635473 B CN109635473 B CN 109635473B
- Authority
- CN
- China
- Prior art keywords
- job
- execution
- unexecuted
- execution mode
- executed
- 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
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Computer Hardware Design (AREA)
- Evolutionary Computation (AREA)
- Geometry (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention provides a heuristic high-flux material simulation calculation optimization method, and belongs to the field of material science. The method comprises the steps of firstly, determining a shared execution mode of all operations in high-throughput material calculation simulation; then selecting one unexecuted operation from the model with the maximum number of unexecuted operations in the high-throughput material calculation simulation and executing the operation in an independent execution mode; and acquiring the next execution operation by utilizing the heuristic information until all operations in the high-flux material calculation simulation are executed. The invention uses the adjacent relation between different elements in the periodic table of elements and the atomic radius of different elements as heuristic information, which can greatly improve the efficiency of high-flux operation execution and greatly shorten the time of new material design.
Description
Technical Field
The invention belongs to the field of material science, and particularly relates to a heuristic high-flux material simulation calculation optimization method.
Background
At present, the acquisition of new materials is shifted from a traditional mode of finding new materials through a large number of experiments to a high-throughput calculation mode of designing new materials through a large number of simulation calculations, so that the acquisition efficiency of new materials can be greatly improved.
The high-throughput material calculation is a design that a large batch of calculation tasks can be rapidly completed at one time by means of calculation resources with strong capacity, and candidate materials meeting requirements are screened out through analysis of calculation results. Such computing tasks, in their particular form, exist as jobs on high performance computing systems, and are referred to as high throughput material computing jobs. Because the calculation amount is generally very large, how to optimize and improve the performance becomes a challenging problem. Currently, mainstream performance optimization methods are performed for a single high-throughput material calculation operation. The method has the same optimization effect on other similar jobs as long as the performance of one job can be improved. However, such methods have the disadvantages that they are optimized only from the local information of a single job, the physical internal association between different jobs is not considered, and the more significant optimization potential possibly brought by the analysis from the macroscopic and overall perspectives is neglected, so the optimization effect is easily limited by various local factors, and the optimization performance of the whole job cannot be improved more comprehensively and greatly.
High-throughput material calculation represented by material genetic engineering is mainly carried out at present on the automatic connection of all stages of calculation, typical projects are international AFlow and MP projects, but no good solution exists on how to greatly reduce the execution time of operation, particularly the execution time of optimization calculation-intensive operation.
Disclosure of Invention
The invention aims to overcome the defect that the prior art cannot fully optimize the execution time of calculation-intensive operation, and provides a heuristic high-flux material simulation calculation optimization method. The invention combines the characteristics of the high-flux material calculation simulation task, opens up a brand-new path to realize performance optimization, is easy to implement and can obtain very obvious effect. The efficiency of high flux operation execution can be greatly improved, and the time for designing a new material is greatly shortened.
The invention provides a heuristic high-flux material simulation calculation optimization method which is characterized by comprising the following steps of:
1) determining a shared execution mode of all the jobs in the high-flux material calculation simulation; the method comprises the following specific steps:
1-1) selecting two operations in any model of high-flux material calculation simulation, wherein doping elements of the two operations are adjacent elements, and the two operations are respectively marked as X and Y;
1-2) executing X in an independent execution mode;
1-3) Y shares X execution by adopting different sharing execution modes, and selects the sharing execution mode with the shortest execution time as the sharing execution mode adopted by all the jobs;
2) selecting one unexecuted operation from a model containing the largest number of unexecuted operations in the high-throughput material calculation simulation and executing the operation in an independent execution mode; the specific selection method is as follows:
in the model containing the maximum number of the unexecuted jobs, counting the number of the shared jobs which can be adjacent to each unexecuted job, selecting the unexecuted job with the maximum number of the shared jobs which can be adjacent to each unexecuted job, and executing the unexecuted job in an independent execution mode; if a plurality of unexecuted jobs with the same number of adjacent sharable jobs exist, one of the unexecuted jobs is randomly selected and then executed in an independent execution mode; wherein, the shared operation which can be adjacent to each operation is the operation of the doping element adjacent to the operation doped element;
3) acquiring the next execution operation by utilizing heuristic information until all operations in the high-flux material calculation simulation are executed; the method comprises the following specific steps:
3-1) determining all sharable execution job pairs, wherein the sharable execution job pairs meet the following requirements: two jobs in each job pair belong to the same model, wherein one job is executed and the other job is not executed;
3-2) determining sharable execution job pairs:
if the shareable execution job pair exists, calculating the difference between the atomic radii of the two doping elements of each job pair, selecting one job pair with the smallest absolute value of the difference, executing the unexecuted jobs in the job pair in the shared execution mode determined in the step 1), and returning to the step 2); if no shareable execution job pair exists and no executed job exists in the high-throughput material calculation simulation, returning to the step 2) again; and ending the method until all the operations in the high-flux material calculation simulation are finished.
The invention has the characteristics and beneficial effects that:
(1) the method provided by the invention is optimized by analyzing the internal correlation among a plurality of simulation operations in the high-throughput material calculation, and is an optimization method for focusing on the whole operation. The method is completely different from the prior single-job performance optimization method, but the prior local optimization method aiming at the single job can be independently combined into the method, and the two methods are in a complementary and superposed relationship.
(2) The optimization method provided by the invention has very obvious overall effect, and although the possible effect of a few operations is not obvious, the performance of most operations is obviously improved. Can greatly improve the efficiency of high-flux operation execution and greatly shorten the time of new material design
(3) The method only needs very little well-known information, namely the adjacent relation between different elements in the periodic table of the elements and the atomic radius of the different elements, can analyze the similarity degree between different operations and make optimization decision on the basis, and the decision-making method is simple and efficient.
Detailed Description
The invention provides a heuristic high-flux material simulation calculation optimization method, which is further detailed below by combining specific embodiments.
The invention provides a heuristic High-Throughput material Simulation calculation optimization method, which assumes that a High-Throughput material calculation Simulation (HTCS) is composed of M models, and the expression is as follows:
where M represents the total number of high-throughput material calculation models, Model i Representing the ith model.
Wherein the ith model may be represented by N i A job is composed, so that the Model can be further modeled i Expressed as:
wherein S is i,j A j-th job representing an i-th model;
for different operations corresponding to the same model, the doping elements are different from each other, and other aspects (referring to the initial settings of all input files) are the same.
How the job is calculated and optimized is specifically described below based on widely used VASP software. Two implementations of high-throughput operation are first defined here: an independent execution mode and a shared execution mode.
The independent execution mode is a mode in which the job is executed on the parallel computing system based on input files of the INCAR, POSCAR, POTCAR, and KPOINT files independently generated for the job at the time of design, and has no relation to other jobs.
The shared execution mode refers to two jobs A and B, wherein A is a completed job, B is a job to be executed, and B optimizes the running efficiency of A by sharing the result of the execution of A. Specifically, three different shared execution modes are shared CONTCAR execution, shared CHGCAR execution, and shared CONTCAR and CHGCAR execution at the same time. The execution of the B-shared-A CONTCAR is that B replaces POSCAR generated in the original design stage of B with the CONTCAR obtained after A is executed. B-share a's CHGCAR implementation means that B uses the CHGCAR of a's output as an additional input file to get its charge distribution, which requires modifying its original INCAR file at the same time, setting its icharge flag to 1. B-shared a CONTCAR and CHGCAR execution means that B executes with a's CONTCAR as its POSCAR, but also with a's CHGCAR as an additional input to get its charge distribution, while setting its icharge flag in its INCAR file to 1. The effect of the three shared execution modes will be different due to different operations.
In the following we define the concept of adjacent elements, for two elements EA and EB in the periodic table, if they are in the same row of the periodic table and are directly adjacent, they are called row-adjacent elements each other. If they are in the same column of the periodic table of elements and are immediately adjacent, they are mutually referred to as column-adjacent elements. Two elements, whether row-adjacent or column-adjacent, are referred to as adjacent elements.
The concept of shareable jobs is further defined below. For operations based on the same model, since they are different only in doping element, operations adjacent to each other in doping element are defined as mutually adjacently sharable operations.
The invention provides a heuristic high-flux material simulation calculation optimization method, which comprises the following steps:
1) determining a shared execution mode of all the jobs in the high-flux material calculation simulation; the method comprises the following specific steps:
1-1) selecting two operations in any model of high-flux material calculation simulation, wherein doping elements of the two operations are adjacent elements, and the two operations are respectively marked as X and Y;
1-2) executing X in an independent execution mode, and storing all results obtained after execution in an output file of the operation, wherein the output result can describe some basic characteristics of a given material design.
1-3) Y shares X execution by using three different shared execution modes, and selects the shared execution mode with the shortest execution time as the shared execution mode adopted by all the jobs.
If the workload is very large, a plurality of groups of different X and Y jobs can be selected from different models to carry out advanced test, and the sharing execution mode is selected according to the average effect.
2) Selecting one unexecuted operation from a model containing the largest number of unexecuted operations in the high-throughput material calculation simulation and executing the operation in an independent execution mode; the specific selection method is as follows:
in the model containing the maximum number of the unexecuted jobs, counting the number of the shared jobs which can be adjacent to each unexecuted job, selecting the unexecuted job with the maximum number of the shared jobs which can be adjacent to each unexecuted job, and executing the unexecuted job in an independent execution mode; if there are a plurality of unexecuted jobs, the number of the shared jobs which can be adjacent is the same. Randomly selecting one of the unexecuted jobs, and then executing the job in an independent execution mode;
3) acquiring the next execution operation by utilizing heuristic information until all operations in the high-flux material calculation simulation are executed; the method comprises the following specific steps:
3-1) determining all sharable execution job pairs, wherein the sharable execution job pairs meet the following requirements: two jobs in each job pair belong to the same model, wherein one job is executed and the other job is not executed; all job pairs < a1, B1>, …, < Am, Bm > … that meet the above conditions will be recorded.
3-2) determining sharable execution job pairs:
if the shareable execution job pair exists, calculating the difference between the atomic radii of the two doping elements of each job pair, selecting one job pair with the smallest absolute value of the difference, executing the unexecuted jobs in the job pair in the shared execution mode determined in the step 1), and returning to the step 2); if no shareable execution job pair exists and no executed job exists in the high-throughput material calculation simulation, returning to the step 2) again; and ending the method until all the operations in the high-flux material calculation simulation are finished.
Claims (1)
1. A heuristic high-flux material simulation calculation optimization method is characterized by comprising the following steps:
1) determining a shared execution mode of all the jobs in the high-flux material calculation simulation; the method comprises the following specific steps:
1-1) selecting two operations in any model of high-flux material calculation simulation, wherein doping elements of the two operations are adjacent elements, and the two operations are respectively marked as X and Y;
1-2) executing X in an independent execution mode;
1-3) Y shares X execution by adopting different sharing execution modes, and selects the sharing execution mode with the shortest execution time as the sharing execution mode adopted by all the jobs;
the shared execution mode includes: sharing a CONTCAR execution mode, sharing a CHGCAR execution mode, and sharing both a CONTCAR and CHGCAR execution mode simultaneously; the shared CONTCAR execution mode refers to that the original POSCAR of the unexecuted operation is replaced by the CONTCAR obtained by the executed operation for execution; the shared CHGCAR execution mode is that an unexecuted job obtains corresponding charge distribution by using CHGCAR output by an executed job as an additional input file, and simultaneously modifies an original INCAR file of the unexecuted job and sets an ICHARG mark in the INCAR file to be 1; the shared CONTCAR and CHGCAR execution mode means that the non-executed job is executed by using a CONTCAR obtained by the executed job to replace an original POSCAR of the non-executed job, the non-executed job uses the CHGCAR output by the executed job as an additional input file to obtain a corresponding charge distribution, and simultaneously modifies an original INCAR file of the non-executed job, and an ICHARG mark in the non-executed job is set to be 1;
2) selecting one unexecuted operation from a model containing the largest number of unexecuted operations in the high-throughput material calculation simulation and executing the operation in an independent execution mode; the specific selection method is as follows:
in the model containing the maximum number of the unexecuted jobs, counting the number of the shared jobs which can be adjacent to each unexecuted job, selecting the unexecuted job with the maximum number of the shared jobs which can be adjacent to each unexecuted job, and executing the unexecuted job in an independent execution mode; if a plurality of unexecuted jobs with the same number of adjacent sharable jobs exist, one of the unexecuted jobs is randomly selected and then executed in an independent execution mode; wherein the adjacently sharable operation of each operation is an operation in which the doping element is adjacent to the operation doping element;
3) acquiring the next execution operation by utilizing heuristic information until all operations in the high-flux material calculation simulation are executed; the method comprises the following specific steps:
3-1) determining all sharable execution job pairs, wherein the sharable execution job pairs meet the following requirements: two jobs in each job pair belong to the same model, wherein one job is executed and the other job is not executed;
3-2) determining sharable execution job pairs:
if the shareable execution job pair exists, calculating the difference between the atomic radii of the two doping elements of each job pair, selecting one job pair with the smallest absolute value of the difference, executing the unexecuted jobs in the job pair in the shared execution mode determined in the step 1), and returning to the step 2); if no shareable execution job pair exists and no executed job exists in the high-throughput material calculation simulation, returning to the step 2) again; and ending the method until all the operations in the high-flux material calculation simulation are finished.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811556665.5A CN109635473B (en) | 2018-12-19 | 2018-12-19 | Heuristic high-flux material simulation calculation optimization method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811556665.5A CN109635473B (en) | 2018-12-19 | 2018-12-19 | Heuristic high-flux material simulation calculation optimization method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109635473A CN109635473A (en) | 2019-04-16 |
| CN109635473B true CN109635473B (en) | 2022-08-02 |
Family
ID=66075503
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811556665.5A Active CN109635473B (en) | 2018-12-19 | 2018-12-19 | Heuristic high-flux material simulation calculation optimization method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109635473B (en) |
Families Citing this family (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110334055B (en) * | 2019-06-06 | 2023-04-11 | 苏州迈高材料基因技术有限公司 | Method for acquiring material calculation data |
| CN113127973B (en) * | 2021-04-16 | 2022-05-10 | 湖南大学 | CAE simulation technology-based multi-material intelligent material selection method and system and electronic equipment |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1319761A (en) * | 2000-03-15 | 2001-10-31 | 清华大学 | Apparatus and method for high-flux electric rotation detection |
| CN103631988A (en) * | 2013-10-22 | 2014-03-12 | 芜湖大学科技园发展有限公司 | Multi-user simulation data management platform for electric power system simulation |
| CN104992258A (en) * | 2012-07-05 | 2015-10-21 | 爱利门图供应链管理(开曼)有限公司 | Method and system for controlling supply chains |
| CN108123109A (en) * | 2016-11-28 | 2018-06-05 | 华为技术有限公司 | Lithium cobaltate cathode material and preparation method thereof and lithium rechargeable battery |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US9342355B2 (en) * | 2013-06-20 | 2016-05-17 | International Business Machines Corporation | Joint optimization of multiple phases in large data processing |
-
2018
- 2018-12-19 CN CN201811556665.5A patent/CN109635473B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1319761A (en) * | 2000-03-15 | 2001-10-31 | 清华大学 | Apparatus and method for high-flux electric rotation detection |
| CN104992258A (en) * | 2012-07-05 | 2015-10-21 | 爱利门图供应链管理(开曼)有限公司 | Method and system for controlling supply chains |
| CN103631988A (en) * | 2013-10-22 | 2014-03-12 | 芜湖大学科技园发展有限公司 | Multi-user simulation data management platform for electric power system simulation |
| CN108123109A (en) * | 2016-11-28 | 2018-06-05 | 华为技术有限公司 | Lithium cobaltate cathode material and preparation method thereof and lithium rechargeable battery |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109635473A (en) | 2019-04-16 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US20210373888A1 (en) | Multi-objective optimization method and system for master production plan of casting parallel workshops | |
| CN105446979B (en) | Data digging method and node | |
| CN109815537B (en) | High-flux material simulation calculation optimization method based on time prediction | |
| Wang et al. | Predator—An experience guided configuration optimizer for Hadoop MapReduce | |
| CN107908536B (en) | Performance evaluation method and system for GPU application in CPU-GPU heterogeneous environment | |
| CN104750780B (en) | A kind of Hadoop configuration parameter optimization methods based on statistical analysis | |
| WO2016025924A1 (en) | Systems and methods for auto-scaling a big data system | |
| CN109522428B (en) | External memory access method of graph computing system based on index positioning | |
| CN106446134B (en) | Local multi-query optimization method based on predicate specification and cost estimation | |
| CN102799486A (en) | Data sampling and partitioning method for MapReduce system | |
| CN102855153B (en) | Towards the stream compile optimization method of chip polycaryon processor | |
| CN109635473B (en) | Heuristic high-flux material simulation calculation optimization method | |
| CN106371924B (en) | A kind of method for scheduling task minimizing MapReduce cluster energy consumption | |
| CN105468439A (en) | Adaptive parallel algorithm for traversing neighbors in fixed radius under CPU-GPU (Central Processing Unit-Graphic Processing Unit) heterogeneous framework | |
| JP6694447B2 (en) | Big data calculation method and system, program, and recording medium | |
| CN104408106A (en) | Scheduling method for big data inquiry in distributed file system | |
| Schlag et al. | Scalable edge partitioning | |
| Xu et al. | Distributed maximal clique computation and management | |
| CN106202092A (en) | The method and system that data process | |
| CN105654187A (en) | Grid binary tree method of control system midpoint locating method | |
| CN105205052A (en) | Method and device for mining data | |
| CN105867998A (en) | Virtual machine cluster deployment algorithm | |
| CN106502790A (en) | A kind of task distribution optimization method based on data distribution | |
| CN111553834A (en) | Concurrent graph data preprocessing method based on FPGA | |
| CN111078380A (en) | Multi-target task scheduling method and system |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |