CN108132872A - Based on the parallel super GRAPES system optimization methods for calculating grid cloud platform - Google Patents
Based on the parallel super GRAPES system optimization methods for calculating grid cloud platform Download PDFInfo
- Publication number
- CN108132872A CN108132872A CN201810021292.5A CN201810021292A CN108132872A CN 108132872 A CN108132872 A CN 108132872A CN 201810021292 A CN201810021292 A CN 201810021292A CN 108132872 A CN108132872 A CN 108132872A
- Authority
- CN
- China
- Prior art keywords
- function
- grapes
- processes
- test
- analysis
- 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
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F11/00—Error detection; Error correction; Monitoring
- G06F11/30—Monitoring
- G06F11/34—Recording or statistical evaluation of computer activity, e.g. of down time, of input/output operation ; Recording or statistical evaluation of user activity, e.g. usability assessment
- G06F11/3452—Performance evaluation by statistical analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F11/00—Error detection; Error correction; Monitoring
- G06F11/30—Monitoring
- G06F11/34—Recording or statistical evaluation of computer activity, e.g. of down time, of input/output operation ; Recording or statistical evaluation of user activity, e.g. usability assessment
- G06F11/3409—Recording or statistical evaluation of computer activity, e.g. of down time, of input/output operation ; Recording or statistical evaluation of user activity, e.g. usability assessment for performance assessment
- G06F11/3419—Recording or statistical evaluation of computer activity, e.g. of down time, of input/output operation ; Recording or statistical evaluation of user activity, e.g. usability assessment for performance assessment by assessing time
- G06F11/3423—Recording or statistical evaluation of computer activity, e.g. of down time, of input/output operation ; Recording or statistical evaluation of user activity, e.g. usability assessment for performance assessment by assessing time where the assessed time is active or idle time
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F11/00—Error detection; Error correction; Monitoring
- G06F11/36—Preventing errors by testing or debugging software
- G06F11/3668—Software testing
- G06F11/3672—Test management
- G06F11/3688—Test management for test execution, e.g. scheduling of test suites
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F11/00—Error detection; Error correction; Monitoring
- G06F11/36—Preventing errors by testing or debugging software
- G06F11/3668—Software testing
- G06F11/3672—Test management
- G06F11/3692—Test management for test results analysis
Abstract
The present invention relates to a kind of based on the parallel super GRAPES system optimization methods for calculating grid cloud platform, including：S1 test data set and operating system) are loaded into, carries out system level testing, communication stage test and the test of function grade respectively, including：S1.1) system level testing；S1.2) communication stage is tested；S1.3) function grade is tested：The function of calling is monitored, obtains the operation characteristic of function.S2 test result analysis) is carried out according to derived system features file, including：S2.1) system test result is analyzed；S2.2) MPI communication stages test result analysis；S2.3) function grade test result analysis.S3 processing) is optimized according to analysis result, optimization processing includes：Vectorization, load balancing substitute the function in GRAPES_GFS using library function.The present invention solves optimization problems of the Grapes on parallel super calculation grid platform, improves running efficiency of system.
Description
Technical field
The present invention relates to process meteorological data field more particularly to a kind of GRAPES based on parallel super calculation grid cloud platform
System optimization method.
Background technology
By long-term development, the scientific basic and technical method of numerical weather forecast comparative maturity, are to make day
The most important Scientific Approaches of gas forecast, and there is unique advantage in terms of the forecast of Extreme Weather Events.Weather numerical value is pre-
Report is other than it will consider the contribution of air motion of various scales comprehensively, it is necessary to mutual in view of air and the other ring layers of the earth
Effect, therefore the data volume for participating in calculating is very huge, while believable mid-term Numerical Weather timeliness further extends, and is based on
Numerical forecast requirement of real-time is higher and higher, improves GRAPES calculating speeds with regard to increasingly important.
GRAPES systematic researches exploitation is from Data Assimilation, Forecast Mode dynamical frame, Atmospheric dynamics and calculating
Machine supports what four direction carried out, this four aspect achievements in research are integrated into：Region mesoscale assimilation system and the whole world assimilation with
Two class forecast system of mid-range forecast, wherein whole world medium-range numerical weather forecast system GRAPES-GFS is numerical forecast business body
The core of system not only provided boundary condition and background information for region Mesoscale Numerical Forecast System (GRAPES-MESO), but also is complete
The basis of ball DATA PROCESSING IN ENSEMBLE PREDICTION SYSTEM.At present, GRAPES-GFS can carry out the weather situation of 10 days in global range and precipitation event pre-
It surveys.
The currently running environment of GRAPES systems is the martial prowess 4000A (Sunway based on IBM POWER processor architectures
4000A).Martial prowess 4000A completes hardware installation and after 2011 complete to expand in part of in September, 2009, and GRAPES systems just exist
It is run on this parallel tables.Martial prowess 4000A computing subsystem peak computationals ability is 15.75TFLOPS, containing processor 42,
Every has 4 child nodes, and 24 core CPU are configured in each child node, share 36GB memories.Memory node 16, it is complete to provide 128TB
Office's memory capacity.Martial prowess 4000A systems use the low power consumption processor of Intel mainstreams, and each calculate node contains 2
The 4 core Xeons (Intel Xeon Nehalem X5570/2.93GHz) of 2.93GHz, the memory that single cpu caryogamy is put
For 4.5GB；Node, which interconnects, uses Infiniband networks, and the two-way bandwidth of High speed network between node is 80Gbps.
However, with the continuously improving of numerical weather prediction model, the continuous improvement of resolution ratio and forecast gradually walk to
Extended peroid is forecast and the development of Short-term Forecast, and GRAPES systems run on martial prowess 4000A platforms, computing capability to platform and
The demand of resolution ratio is also constantly expanding, therefore GRAPES systems are run under existing running environment, it may appear that because of calculation amount
Increased dramatically, computing capability is limited and the problem of causing the computational efficiencies such as calculating speed is slack-off, calculating cycle is elongated low；It is another
Aspect, GRAPES patterns are a mesh point modes, are drawn on the parallel decomposition of zoning using general horizontal grid
Point, and in order to improve the accuracy of forecast, calculating grid can divide less and less, and corresponding calculation amount can increase.Likewise,
Because the computing capability of martial prowess 4000A platforms is limited, this can lead to the decline of computational accuracy.Computational efficiency is low and computational accuracy
Declining the accuracy that can directly result in the forecast of GRAPES system values and timeliness cannot ensure, and the timeliness and essence forecast
True property is the most important demand and feature of meteorological field application.In addition, as GRAPES systems are run on this parallel tables
The passage of time, user volume are also increasing, and the resource management system of former running environment cannot be tracked dynamically, reflected in real time
User is to the service condition of computing resource, it is impossible to implement resource control scheme in time, it can not be with a kind of unitized quantization hand
The usage amount of the quantity of segment description resource, accurate record and control user resources.
In conclusion martial prowess 4000A cannot meet GRAPES system values forecast system for computing capability, calculate
The demand of precision and resource reasonable distribution.Solve the problems, such as existing platform exist just must to using INTEL X86-baseds as
It is transplanted on the High-Performance Computing Cluster platform of a new generation of core.Positioned at Guangzhou, country surpasses the Milky Way two at calculation center and disclosure satisfy that
For computing capability caused by GRAPES system upgrades and mode conversion, the need of computational accuracy and reasonable distribution related resource
It asks.Therefore it needs to provide corresponding optimization method, solves Grapes in the parallel super optimization problem calculated on grid platform.
Invention content
For the deficiency of the prior art, the present invention proposes a kind of based on the parallel super GRAPES systems for calculating grid cloud platform
Optimization method includes the following steps：
S1 test data set and operating system) are loaded into, system level testing, communication stage test and function grade is carried out respectively and surveys
Examination, including：
S1.1) system level testing：The weather condition that 0.1 ° of resolution ratio example is used to forecast on GRAPES_GFS 24 hours,
Every 6 hours export a modvar, carry out performance using 2048 processes, 4096 processes, 8192 processes and 12000 processes respectively
Extension test；
S1.2) communication stage is tested：Communication conditions when being run to program are monitored, and are forecast using 0.1 ° of resolution ratio example
The weather condition of 24 hours, every 6 hours are exported a modvar, are tested using 2048 processes；
S1.3) function grade is tested：The function called in pattern is monitored, obtains the operation characteristic of function, is used
0.1 ° of resolution ratio example forecasts the weather condition of 24 hours, and every 6 hours are exported a modvar, surveyed using 8192 processes
Examination；
S2 test result analysis) is carried out according to derived system features file, including：
S2.1) carry out system test result analysis, analysis GRAPES_GFS patterns respectively 2048 processes, 4096 processes,
Operation characteristic under 8192 processes and 12000 process scales, the use of the operation characteristic including cpu resource, Internet resources
Use, the use of memory source and the use of disk；
S2.2 MPI communication stage test result analysis) is carried out, analyzes MP1 call duration times accounting and GRAPES logical process
Event accounting；
S2.3) into line function grade test result analysis, analysis of central issue, analysis carry out the function in GRAPES by software
The most function of holding time；
S3 processing) is optimized according to the analysis result of step S2, optimization processing includes：Vectorization, load balancing, use
Library function substitutes the function in GRAPES_GFS.
According to a preferred embodiment, step S2.2 is further included：It analyzes the distribution of call duration time and is related to global scope
Time accounting.
According to a preferred embodiment, in step s3, ff2 functions highest to usage time accounting optimize,
Maximum, the minimum value that the mode recycled in source program is found a function are replaced using library function.
The invention has the advantages that：
The present invention solves optimization problems of the Grapes on parallel super calculation grid platform, improves running efficiency of system.
No. two superpower computing capabilitys of the operation demand of GRAPES system super larges and computational accuracy demand and the Milky Way just mutually suit, will
The function of GRAPES is realized by the Milky Way two, can promote the speed of service and computational accuracy of GRAPES systems, being capable of big model
Exclosure improves the accuracy of numerical weather forecast, realizes the real-time weather conditions forecast in a wide range of area.
Description of the drawings
Fig. 1 illustrates the flow chart of the present invention；
Fig. 2 shows operation characteristic figures of the GRAPES_GFS under different process scales；
Fig. 3 shows that MPI communication stages test result counts schematic diagram；
Fig. 4 shows the time accounting change schematic diagram of integrate function calls.
Specific embodiment
It is described in detail below in conjunction with the accompanying drawings.
Understand to make the object, technical solutions and advantages of the present invention clearer, With reference to embodiment and join
According to attached drawing, the present invention is described in more detail.It should be understood that these descriptions are merely illustrative, and it is not intended to limit this hair
Bright range.In addition, in the following description, the description to known features and technology is omitted, to avoid this is unnecessarily obscured
The concept of invention.
Test platform and monitoring instrument：This test job is to calculate Guangzhou center " Milky Way two " in national Super
Upper progress, the details of test environment are as shown in the table：
In order to help that whole analysis is carried out to GRAPES systems, whole system operation conditions is monitored, acquisition system
Every operating index of system.It is main to have used following 4 tools：
VTune
The operation data of acquisition system realizes the monitoring of system.Each section can be directly obtained by collected data
The operation conditions of program can also obtain the run time of system and the run time of every section of function, can therefrom find system
The run time of each function, is targetedly analyzed.
Paramoon
Application operation feature extractor, by monitoring the clothes such as cluster management/login node, calculate node, I/O node in real time
It is special to provide the operation that application software in cluster system changes over time for processor, memory, network and the storage performance data of business device
Sign.
ParaTune
Application operation feature analyzer, can analyze the .para application operation tag files of Paramon generations, and display should
With the performance data of processor, memory, network and disk in each node during operation, group of planes application operation process is reconstructed, efficiently, accurate
The operation characteristic of application really is described.
Step S1：Test data set and operating system are loaded into, carries out system level testing, communication stage test and function grade respectively
Test.
Step S1.1：System level testing.
System level testing is the situation of the operation of test system on the whole, and 0.1 ° of resolution ratio is used on GRAPES_GFS
Example, the forecast weather condition of 24 hours, every 6 hours export a modvar, respectively using 2048 processes, 4096 processes,
8192 processes, 12000 processes carry out behavior extension.
Step S1.2：Communication stage is tested.
Communication stage test is that communication conditions when being run to program are monitored, and uses 0.1 ° of resolution ratio example, forecast 24
The weather condition of hour, every 6 hours are exported a modvar, are tested using 2048 processes.
Step S1.3：Function grade is tested.
The test of function grade is that the function called in pattern is monitored, and obtains the operation characteristic of function, uses 0.1 ° point
Resolution example, the weather condition of forecast 24 hours, every 6 hours are exported a modvar, are tested using 8192 processes.
Step S2：Test result analysis.
Step S2.1：System level testing interpretation of result.
Observe GRAPES_GFS patterns operation characteristic overall condition, respectively 2048 processes, 4096 processes, 8192 into
Under journey, 12000 process scales.Aforementioned operating condition includes the use of cpu resource, the use of Internet resources, memory source make
With and disk use.
Step S2.2：MPI communication stage test result analysis.
Analyze the run time distribution situation of entire program, observation MPI call duration times accounting, GRAPES logical process events
Accounting.The distribution situation of further analysis call duration time, the time accounting for being related to global scope.
Step S2.3：Function grade test result analysis.
Analysis of central issue, the most function of analysis holding time carry out the function in GRAPES by software.Further,
Hot spot function concrete function is analyzed, so as to find the mode that may optimize.
Step S3：Processing is optimized according to the analysis result of step S2.
In the work step of front, we have done the test job of different stage, and its object is to find GRAPES_
The weak link of GFS patterns is optimized by targetedly method, improves operational efficiency.It is main according to test result
The optimization means of use include three kinds：Vectorization, load balancing substitute the function in GRAPES_GFS using library function.
Due to the complexity of GRAPES systems in itself, Optimization Work is an extremely complex process, has above been carried
Hotspot's distribution into GRAPES_GFS is on multiple functions, therefore the highest ff2 functions of usage time accounting are made in optimization
It is optimized for example.Further, the hot spot in ff2 functions is concentrated mainly in the calculating of maximin, optimizes work
Work is that maximum, the minimum value that the mode that will be recycled in source program is found a function are replaced using library function
For a better understanding of the present invention, step S2 and step S3 is carried out with reference to specific embodiment further
Description, it should be noted that the embodiment content is to understand for convenience, cannot form the limit to the scope of the present invention
System.
Step S2.1：System level testing interpretation of result.
Be illustrated in figure 2 the overall condition of the operation characteristic of GRAPES_GFS patterns under different processes, which show
Operating condition under 2048 processes, 4096 processes, 8192 processes, 12000 process scales, specifically includes：The use of cpu resource,
The use of Internet resources, the use of memory source and the use of disk.
As can be seen that the utilization rate of CPU, close to 100%, the operation ratio of CPU (sywa) %d under systematic thinking way is smaller,
Illustrate that CPU most times are all spent on processing user program.(cycles perinstruction, 0.25) theory is to CPI
It is 0.6 to be worth, and shows that instruction execution efficiency is higher.The percentage of the Gflpps of local runtime is 6%, from test result,
Gflops is only relatively low in 1% or so, Gflops values, illustrates that the efficiency of Floating-point Computation is relatively low.The Milky Way two supports AVX instructions,
The GRAPES indexs are that the percentage of 0%, VEC is only 3% or so, it may be said that the main reason for bright floating number computational efficiency is relatively low
One of be that vectorization degree is not high.
From peak information as can be seen that system resource is also not up to saturation, compare using 2048 processes and 12000 into
That Cheng Jinhang is tested as a result, number of processes increases, express network bandwidth peak reduces, and network bandwidth resources do not reach saturation,
The usage amount that same number of processes increases memory is only 1/3rd of peak value, and memory source does not also reach saturation.From figure
In it can also be seen that Gflops peak values only reach 7.74%, only account for the half of peak value, still have greatly improved space.
Step S2.2：MPI communication stage test result analysis.
As shown in Figure 3, it can be seen that the run time distribution situation of entire program, wherein MPI call duration times accounting
68.1%, GRAPES logical process time accounting 31.8%, about 1/3rd time are all consumed in communication above.Further
Analysis, the distribution situation of call duration time, be related to global scope MPI_Barrier, MPI_Allgather simultaneously operating occupy
It is more, account for about the 27% of entire program runtime.
Further analyze the time accounting of the time accounting situation, on the whole each process of each process of each process
Than more uniform, the time loss that process that wherein process number is 1416~1439 calculates is close to 2 times of other processes, MPI_
It is relatively low that Sendrecv communications take ratio.Other statistics processes：Computing module User_Code, MPI_Barrier, MPI_
Sendrecv, which takes, is presented wavy cyclically-varying, and in 1600 process sections, computing module User_Code fluctuations are on 33% left side
The right side, read-me make corresponding adjustment in terms of load balancing.
Step S2.3：Function grade test result analysis.
Integrate function calls are the major parts of grapes logical process, to 8192 processes, in the uniform time
6 processes are randomly selected at interval, count the time accounting variation of different process integrate function calls.
As shown in figure 4, the time scale of Integrate function calls is different between different processes.Process number it is bigger than normal into
Cheng Zhong：The time accounting of colm_init reduces sharply, the increase of solver_grapes accountings, in addition colm_init, med_before_
The amplitude of variation of solve_io, MPI_BARRIER between different processes relatively stablize, solver_grapes function calls when
Between ratio change among varying processes it is obvious.Solver_grapes function calls are further analyzed, it is right
8192 processes, evenly spaced to randomly select 6 processes, there is accounting in main hot spot solve_helmholts and prm_3d
Alternately change, the variations such as radiation_driver, upstream_interp_jin are more steady.
Next the hot spot function of GRAPES programs according to itself is taken and be ranked up, wherein holding time is at most
Ff2 functions account for about full-time 3%.Ff2 functions are further analyzed, the function is in the operation of upper strata function loops
Repeatedly called, the hot spot of function concentrates on the maximum value and minimum value of evaluation.
Step S3：
The hotspot's distribution in GRAPES_GFS is mentioned above on multiple functions, the usage time accounting highest in optimization
Ff2 functions optimized as example.Hot spot in ff2 functions is concentrated mainly in the calculating of maximin, optimization
Work is that the maximin that the mode that will be recycled in source program is found a function is replaced using library function, optimizes front and rear code such as
Under：
A1=f (1)
B1=f (1)
Before optimization：
Do i=2,32
A1=amax1 (a1, f (i))
B1=amin1 (b1, f (i))
End do
After optimization:
Do i=2,32
A1m=maxval (f)
B1m=minval (f)
End do
Operation exports result：
Before optimization
A1=98.94819
B1=2.946837
RES_max=98.9489059448242
Total_time=59.90730
After optimization
A1m=98.94819
B1m=2.946837
RES_max=98.9489059448242
Total_time=24.00490
It recycles and performs 1000*1000*1000 times on former ff2 functional foundations.The time is 59.0730 before optimization；Optimization
The time is 24.0049 afterwards, speed-up ratio 2.4956.From the point of view of optimum results, the effect of optimization is fairly obvious, has and improves operation
The advantages of efficiency.
The present invention is based on GRAPES test results, it is proposed that the optimization method of hot spot function in GRAPES_GFS patterns.This
Outside, entire GRAPES systems are further expanded to, improve running efficiency of system, solve Grapes in parallel super calculation grid
Optimization problem on platform.
It should be noted that above-mentioned specific embodiment is exemplary, those skilled in the art can disclose in the present invention
Various solutions are found out under the inspiration of content, and these solutions also belong to disclosure of the invention range and fall into this hair
Within bright protection domain.It will be understood by those skilled in the art that description of the invention and its attached drawing are illustrative and are not
Form limitations on claims.Protection scope of the present invention is limited by claim and its equivalent.
Claims (3)
- It is 1. a kind of based on the parallel super GRAPES system optimization methods for calculating grid cloud platform, which is characterized in that include the following steps：S1 test data set and operating system) are loaded into, carries out system level testing, communication stage test and the test of function grade respectively, Including：S1.1) system level testing：The weather condition that 0.1 ° of resolution ratio example is used to forecast on GRAPES_GFS 24 hours, every 6 The modvar of output in a hour, carries out performance expansion using 2048 processes, 4096 processes, 8192 processes and 12000 processes respectively Exhibition test；S1.2) communication stage is tested：Communication conditions when being run to program are monitored, and forecast that 24 is small using 0.1 ° of resolution ratio example When weather condition, every 6 hours are exported a modvar, are tested using 2048 processes；S1.3) function grade is tested：The function called in pattern is monitored, the operation characteristic of function is obtained, uses 0.1 ° point Resolution example forecasts the weather condition of 24 hours, and every 6 hours are exported a modvar, tested using 8192 processes；S2 test result analysis) is carried out according to derived system features file, including：S2.1 system test result analysis) is carried out, analysis GRAPES_GFS patterns are respectively in 2048 processes, 4096 processes, 8192 Operation characteristic under process and 12000 process scales, the use of the operation characteristic including cpu resource, the use of Internet resources, The use of memory source and the use of disk；S2.2 MPI communication stage test result analysis) is carried out, analyzes MP1 call duration times accounting and GRAPES logical process events Accounting；S2.3) into line function grade test result analysis, analysis of central issue carries out the function in GRAPES by software, analysis occupies Time most function；S3 processing) is optimized according to the analysis result of step S2, optimization processing includes：Vectorization, uses library letter at load balancing Number substitutes the function in GRAPES_GFS.
- 2. the method as described in claim 1, which is characterized in that step S2.2 is further included：It analyzes the distribution of call duration time and relates to And the time accounting of global scope.
- 3. method as claimed in claim 2, which is characterized in that in step s3, ff2 functions highest to usage time accounting It optimizes, maximum, the minimum value that the mode recycled in source program is found a function are replaced using library function.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810021292.5A CN108132872B (en) | 2018-01-10 | 2018-01-10 | GRAPES (GRAPES) system optimization method based on parallel supercomputing grid cloud platform |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810021292.5A CN108132872B (en) | 2018-01-10 | 2018-01-10 | GRAPES (GRAPES) system optimization method based on parallel supercomputing grid cloud platform |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108132872A true CN108132872A (en) | 2018-06-08 |
| CN108132872B CN108132872B (en) | 2020-04-03 |
Family
ID=62399637
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810021292.5A Active CN108132872B (en) | 2018-01-10 | 2018-01-10 | GRAPES (GRAPES) system optimization method based on parallel supercomputing grid cloud platform |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108132872B (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109101397A (en) * | 2018-08-01 | 2018-12-28 | 武汉索雅信息技术有限公司 | High-Performance Computing Cluster monitoring method, unit and storage medium |
| CN111506442A (en) * | 2020-04-16 | 2020-08-07 | 艾普阳科技（深圳）有限公司 | Local procedure calling method, device, equipment and medium |
| CN113886251A (en) * | 2021-09-30 | 2022-01-04 | 四川大学 | Hot spot function determination method based on thermodynamic diagram |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1588323A (en) * | 2004-09-23 | 2005-03-02 | 上海交通大学 | Parallel program visuable debugging method |
| CN102141962A (en) * | 2011-04-07 | 2011-08-03 | 北京航空航天大学 | Safety distributed test framework system and test method thereof |
| US20130006569A1 (en) * | 2010-03-05 | 2013-01-03 | Nec Corporation | Control policy adjusting apparatus, method of adjusting control policy, and program |
| CN106776289A (en) * | 2016-11-24 | 2017-05-31 | 山东交通学院 | Multitask self adaptation cloud method of testing |
-
2018
- 2018-01-10 CN CN201810021292.5A patent/CN108132872B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1588323A (en) * | 2004-09-23 | 2005-03-02 | 上海交通大学 | Parallel program visuable debugging method |
| US20130006569A1 (en) * | 2010-03-05 | 2013-01-03 | Nec Corporation | Control policy adjusting apparatus, method of adjusting control policy, and program |
| CN102141962A (en) * | 2011-04-07 | 2011-08-03 | 北京航空航天大学 | Safety distributed test framework system and test method thereof |
| CN106776289A (en) * | 2016-11-24 | 2017-05-31 | 山东交通学院 | Multitask self adaptation cloud method of testing |
Non-Patent Citations (2)
| Title |
|---|
| 刘钊: "基于国产高性能计算机的GRAPES性能优化研究", 《中国优秀硕士学位论文全文数据库》 * |
| 李婵娟: "高性能计算综合评测框架的设计和实现", 《中国优秀硕士学位论文全文数据库》 * |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109101397A (en) * | 2018-08-01 | 2018-12-28 | 武汉索雅信息技术有限公司 | High-Performance Computing Cluster monitoring method, unit and storage medium |
| CN111506442A (en) * | 2020-04-16 | 2020-08-07 | 艾普阳科技（深圳）有限公司 | Local procedure calling method, device, equipment and medium |
| CN111506442B (en) * | 2020-04-16 | 2023-05-09 | 艾普阳科技（深圳）有限公司 | Local procedure call method, device, equipment and medium |
| CN113886251A (en) * | 2021-09-30 | 2022-01-04 | 四川大学 | Hot spot function determination method based on thermodynamic diagram |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108132872B (en) | 2020-04-03 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Xu et al. | Graph processing on GPUs: Where are the bottlenecks? | |
| Qin et al. | Parallelizing flow-accumulation calculations on graphics processing units—From iterative DEM preprocessing algorithm to recursive multiple-flow-direction algorithm | |
| CN102854968B (en) | Real-time energy consumption metering method of virtual machine | |
| CN102323957B (en) | Distributed parallel Skyline query method based on vertical dividing mode | |
| CN102214086A (en) | General-purpose parallel acceleration algorithm based on multi-core processor | |
| CN104598565B (en) | A kind of K mean value large-scale data clustering methods based on stochastic gradient descent algorithm | |
| CN108132872A (en) | Based on the parallel super GRAPES system optimization methods for calculating grid cloud platform | |
| Guo et al. | A container scheduling strategy based on neighborhood division in micro service | |
| Pilla et al. | A topology-aware load balancing algorithm for clustered hierarchical multi-core machines | |
| Liao et al. | Long-term generation scheduling of hydropower system using multi-core parallelization of particle swarm optimization | |
| CN103309889A (en) | Method for realizing of real-time data parallel compression by utilizing GPU (Graphic processing unit) cooperative computing | |
| Huo et al. | An improved multi-cores parallel artificial Bee colony optimization algorithm for parameters calibration of hydrological model | |
| Loghin et al. | On understanding time, energy and cost performance of wimpy heterogeneous systems for edge computing | |
| Zhang et al. | Using big data computing framework and parallelized PSO algorithm to construct the reservoir dispatching rule optimization | |
| CN105700998A (en) | Method and device for monitoring and analyzing performance of parallel programs | |
| Zhang et al. | Estimating power consumption of containers and virtual machines in data centers | |
| Anand et al. | The odd one out: Energy is not like other metrics | |
| Gu et al. | Adaptive online cache capacity optimization via lightweight working set size estimation at scale | |
| Zhong et al. | The cloud computing load forecasting algorithm based on wavelet support vector machine | |
| Niu et al. | Parallel grid-based density peak clustering of big trajectory data | |
| CN204066111U (en) | A kind of quick retrieval system of magnanimity electric-power metering data | |
| CN112784435B (en) | GPU real-time power modeling method based on performance event counting and temperature | |
| Bermejo et al. | On the linearity of performance and energy at virtual machine consolidation: the cis2 index for cpu workload in server saturation | |
| CN104536938A (en) | Method and system for calculating earthquake pre-stack time migration | |
| CN108304549A (en) | A kind of big data Intelligent processing 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 |