CN105335215B - A cloud computing-based Monte Carlo simulation acceleration method and system - Google Patents
A cloud computing-based Monte Carlo simulation acceleration method and system Download PDFInfo
- Publication number
- CN105335215B CN105335215B CN201510885304.5A CN201510885304A CN105335215B CN 105335215 B CN105335215 B CN 105335215B CN 201510885304 A CN201510885304 A CN 201510885304A CN 105335215 B CN105335215 B CN 105335215B
- Authority
- CN
- China
- Prior art keywords
- hadoop
- monte carlo
- program
- simulation
- cloud
- 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
- 238000000342 Monte Carlo simulation Methods 0.000 title claims abstract description 58
- 238000000034 method Methods 0.000 title claims abstract description 57
- 230000001133 acceleration Effects 0.000 title claims description 21
- 238000004519 manufacturing process Methods 0.000 claims abstract description 22
- 238000004088 simulation Methods 0.000 claims description 42
- 238000004364 calculation method Methods 0.000 claims description 26
- 230000005540 biological transmission Effects 0.000 claims description 6
- 230000008859 change Effects 0.000 claims description 4
- 239000003638 chemical reducing agent Substances 0.000 claims description 4
- 238000004891 communication Methods 0.000 claims description 4
- 238000012544 monitoring process Methods 0.000 claims description 4
- 238000009826 distribution Methods 0.000 abstract description 11
- 238000009434 installation Methods 0.000 abstract description 5
- 230000006870 function Effects 0.000 description 29
- 238000005516 engineering process Methods 0.000 description 10
- 238000003860 storage Methods 0.000 description 7
- 230000008569 process Effects 0.000 description 6
- 238000004422 calculation algorithm Methods 0.000 description 5
- 230000008901 benefit Effects 0.000 description 4
- 230000000694 effects Effects 0.000 description 4
- 239000002245 particle Substances 0.000 description 4
- 238000012545 processing Methods 0.000 description 4
- 230000032258 transport Effects 0.000 description 4
- 238000010586 diagram Methods 0.000 description 3
- 230000018109 developmental process Effects 0.000 description 2
- 238000002474 experimental method Methods 0.000 description 2
- 230000003993 interaction Effects 0.000 description 2
- 230000002452 interceptive effect Effects 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000003325 tomography Methods 0.000 description 2
- 230000006399 behavior Effects 0.000 description 1
- 210000004556 brain Anatomy 0.000 description 1
- 238000005314 correlation function Methods 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 229940079593 drug Drugs 0.000 description 1
- 239000003814 drug Substances 0.000 description 1
- 239000004744 fabric Substances 0.000 description 1
- 230000004927 fusion Effects 0.000 description 1
- 230000010354 integration Effects 0.000 description 1
- 230000008140 language development Effects 0.000 description 1
- 238000012423 maintenance Methods 0.000 description 1
- 238000007726 management method Methods 0.000 description 1
- 230000035800 maturation Effects 0.000 description 1
- 230000005658 nuclear physics Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
- 238000000528 statistical test Methods 0.000 description 1
- 239000011800 void material Substances 0.000 description 1
Landscapes
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The present invention relates to a kind of Monte Carlo simulation accelerated method and system based on cloud computing, comprising: install Hadoop and Monte Carlo software on the local computer, and configure Hadoop and operate in pseudo- distribution pattern；MapReduce program is write on the local computer, and makes emulation input text；Fabrication and installation have the machine images of Hadoop and Monte Carlo software beyond the clouds, instantiate a certain number of virtual servers using the machine images of production, and configure the Hadoop on all virtual servers in cloud and operate in super distributed mode, form Hadoop cluster；Local MapReduce program and emulation input text are uploaded to virtual server, and run MapReduce on Hadoop cluster beyond the clouds, carries out the distributed computing of Monte Carlo simulation.The present invention can flexible choice virtual server quantity and configuration, each accessible network place can be used.
Description
Technical field
The invention belongs to field of cloud computer technology more particularly to a kind of Monte Carlo simulation accelerated methods based on cloud computing
And system.
Background technique
Monte Carlo (Monte Carlo, MC) method is also known as random sampling skill or statistical test method, with prevailing value
Calculation method has very big difference, it is a kind of method based on Probability Statistics Theory.Monte carlo method, which can compare, forces
The characteristics of very describing things and Physical Experiment process solve the problems, such as that some numerical methods are insoluble, thus this method has
Be widely applied field.
The advantages of monte carlo method includes: that can more realistically describe the things with random nature and physics
Experimentation；It is limited by geometrical condition small；Convergence rate is unrelated with the dimension of problem；With calculating multiple schemes and multiple simultaneously
The ability of unknown quantity；Error is easy to determine；Program structure is simple, it is easy to accomplish.These advantages make its application range increasingly
Extensively.The main application of this method include: PARTICLE TRANSPORT FROM problem, Statistical Physics, typical mathematical problem, vacuum technique, laser technology,
Medicine, biology, mine locating etc..Wherein in PARTICLE TRANSPORT FROM problem application mainly cover In Experimental Nuclear Physics, reactor physics,
High-energy physics etc..
The shortcomings that monte carlo method, is also than more prominent, including convergence rate is slow；Error has probability；In PARTICLE TRANSPORT FROM
In problem, calculated result is related with system size.When solving the problems, such as more complex (or model) using Monte Carlo simulation, convergence
Slow-footed problem seems especially prominent.In addition, in order to improve simulation accuracy (i.e. with really solve approximation ratio), Monte Carlo
Method requires the quantity of random number to reach million or even ten million rank.Slower convergence rate and huge random number quantity make
The calculation amount for obtaining Monte Carlo simulation surges, time-consuming serious, which has limited this method in the high field of requirement of real-time (as radiated
Treatment planning) in application.
In monte carlo method, since the random process of each random number (or particle) in a model is independent mutually, can
To carry out parallel computation.General each Monte Carlo program all has parallel computation characteristic, i.e., decomposes calculating task, and obtaining can be with
The subtask individually calculated gives and obtains son after computing unit is calculated as a result, carrying out the merging of result again.Monte Carlo is imitative
True accelerated method is carried out according to this characteristic.
Existing Monte Carlo simulation accelerated method includes:
(1) the Monte Carlo simulation accelerated method based on CPU cluster
In high-performance computing sector, CPU cluster calculating method is carried out more early.The supercomputer of domestic contrast maturation is most
Using hundreds of CPU array.MPI is most important multiple programming realization means at present, it defines one group with removable
The programming interface of plant property, therefore programmer only needs to design parallel algorithm, calls the correlation function in MPI library that can realize
Operation on multiple computing units.Such as the Lu Yune of Institutes Of Technology Of Changsha, " microcomputer cluster system is based in the paper that it is delivered
MPI parallel computation " in, author realizes the acceleration that PI value is sought using Monte Carlo integration method by MPI parallel Programming
It calculates.Experiment shows: parallel Programming solve using multicomputer cluster the calculating speed for relatively using single computer
It is fast.The for another example Fu You of University Of Science and Technology Of Shandong, in the paper that it is delivered, " low density gas Straight simulation monte carlo method is interactive
Parallelizingsystem research and realization " in, 8 node group systems are used, low density gas Straight simulation Monte Carlo side is realized
Method interactive parallelization calculates.Author will be converted into the parallel source under MPI environment based on 77 serial D SMC source program of Fortran
Program achieves preferable acceleration effect.
(2) it is based on the Monte Carlo simulation accelerated method of GPU (cluster)
GPU full name in English is Graphic Processing Unit, and translator of Chinese is " graphics processor ".GPU is equivalent to
" brain " of video card is special image core processor.The maximum characteristic of GPU is exactly to possess powerful operational capability, even
It is eager to excel than the operational capability of CPU many times larger.GPU free time is effectively utilized, the potential for playing GPU becomes industry concern
Focus, that is, so-called " GPU hardware acceleration ".High speed, concurrency and the programmable functions of GPU are other than image procossing
General-purpose computations provide good operation platform, make GPU algebraic manipulation, Solving Partial Differential Equations and in terms of
There is preferable application.CUDA (Compute Unified Device Architecture) is that video card manufacturer NVIDIA is released
A kind of universal parallel computing architecture, at present using wide.CUDA is contained inside instruction set architecture (ISA) and GPU
C language can be used to write program in parallel computation engine, developer, can be on the GPU for supporting CUDA with very-high performance
Operation.Such as the national great waves doctor of the Central China University of Science and Technology, in its doctor's thesis " steady-state fluorescence molecular tomography restructing algorithm
Theoretical and experimental study " in, propose it is a kind of using GPU cluster accelerate the steady-state fluorescence molecule based on monte carlo method at
As (Fluorescence Molecular Tomography, FMT) restructing algorithm.This method utilizes 3 outfits in local area network
The computer of NVIDIA video card sets up GPU cluster by Message Passing Interface (MPI) technology, will be total
Calculating task is evenly distributed in 3 calculate nodes, realizes multiple GPU parallel computations.Wherein 3 computers are equipped with
NVIDIA video card is G200, and GPU program is write based on CUDA.Author is accelerated using GPU cluster, is solved the side MC
Method achieves preferable acceleration effect for huge time loss problem present in FMT reconstruct.The for another example Central China University of Science and Technology
He Yongxiang is realized in the paper " the efficient GPU parallel computation of aerodynamics direct simulation Monte Carlo " that it is delivered
The parallel computation of the direct Monte Carlo emulation of aerodynamics based on CUDA.Author uses NVIDA Tesla C2075
GPU has carried out the Acceleration study of single GPU, double GPU and more GPU, and opposite CPU is calculated, and achieves good acceleration effect, and guarantee
Computational accuracy.
Existing Monte Carlo simulation accelerated method suffers a disadvantage in that
(1) when accelerating program parallel using language developments GPU such as CUDA, developer needs to have parallel thinking, exploitation
Program be difficult to debug and to continue to optimize and could obtain preferable acceleration effect.
(2) it purchases supercomputer and realizes CPU cluster, it is with high costs；And the Small-sized C PU collection built using multiple computers
Group, computing capability are limited.Carry out parallel program development using MPI technology, need voluntarily to carry out memory, thread management, difficulty compared with
Greatly.In addition, local computer cluster needs constantly maintenance, increase personnel cost.
Summary of the invention
The present invention provides a kind of Monte Carlo simulation accelerated method and system based on cloud computing, it is intended at least certain
One of above-mentioned technical problem in the prior art is solved in degree.
Implementation of the present invention is as follows, a kind of Monte Carlo simulation accelerated method based on cloud computing, comprising the following steps:
Step a: Hadoop and Monte Carlo software are installed on the local computer, and configures Hadoop and operates in pseudo- distribution
Mode；
Step b: writing MapReduce program on the local computer, and makes emulation input text；
Step c: fabrication and installation have the machine images of Hadoop and Monte Carlo software beyond the clouds, utilize the machine mirror of production
As instantiating a certain number of virtual servers, and the Hadoop configured on all virtual servers in cloud operates in super distributed mould
Formula forms Hadoop cluster；
Step d: local MapReduce program and emulation input text are uploaded to virtual server, and beyond the clouds
MapReduce is run on Hadoop cluster, carries out the distributed computing of Monte Carlo simulation.
The technical solution that the embodiment of the present invention is taken further include: described to write MapReduce program in the step b
It specifically includes:
Step b1: writing map program, and program successively includes inputting the simulation calculation task that reads from standard, calling Meng Teka
Lip river program carries out simulation calculation, standard output is written in calculated result in the form of key-value pair；
Step b2: writing reduce program, and program successively includes reading the simulation calculation with same keys from standard input
As a result, calculated result is merged, standard output is written into combined result in the form of key-value pair；
Step b3: writing Hadoop Streaming operation procedure, and program includes the input and output of map and reduce program
Stream format, Map and Reduce number of tasks, input text title, outgoing route, mapper and reducer filename and upload text
Part path.
The technical solution that the embodiment of the present invention is taken further include: in the step b, the production emulation input text tool
Body includes:
Step b4: if the input of Monte Carlo software is random number, random number needed for just generating actual emulation；If defeated
Entering is program file, program file needed for just generating actual emulation；
Step b5: according to the scale of parallel computation to be carried out, being grouped random number or decompose to program file,
Every group of random number or the corresponding parallel artificial of each subprogram file；
Step b6: the path of random number or program file is entered into a text by row write, as input file.
The technical solution that the embodiment of the present invention is taken further include: the step b further include: run on the local computer
Hadoop Streaming, debugging MapReduce program and verifying emulation input text.
The technical solution that the embodiment of the present invention is taken further include: in the step c, all virtual clothes in configuration cloud
Hadoop on business device operates in super distributed mode and specifically includes: select respectively one virtual server as Master with
Secondary NameNode, remaining virtual server is as Worker；In any virtual server of local computer or cloud
On, using SSH communications protocol, successively according to virtual server Type Change Hadoop configuration file, and transmit them to phase
On the virtual server answered, the configuration file of original position is replaced；Hadoop initialization behaviour is carried out on Master virtual server
Make, Hadoop is made to operate in complete distribution pattern, forms Hadoop cluster.
The technical solution that the embodiment of the present invention is taken further include: in the step d, the operation cloud MapReduce
Carry out the distributed computing of Monte Carlo simulation specifically: operation Hadoop Streaming operation procedure, MapReduce are automatic
Map program and reduce program are operated on different Worker virtual servers, forms Map task and Reduce task；
In Map task, map program realizes reading, simulation calculation, the intermediate result output of Monte Carlo simulation task；In Reduce
In task, the reading of result, the merging of result and output between the realization of reduce program；The monitoring page provided by Hadoop
Cluster operating status is monitored.
The technical solution that the embodiment of the present invention is taken further include: the step d further include: after simulation calculation, by cloud
End simulation result is downloaded to local computer.
A kind of another technical solution that the embodiment of the present invention is taken are as follows: Monte Carlo simulation acceleration system based on cloud computing
System, including pattern configurations module, function write module, text writing module, cluster configuration module, data transmission module and emulation
Computing module；
The pattern configurations module configures Hadoop for installing Hadoop and Monte Carlo software on the local computer
It operates under pseudo- distribution pattern；
The function writes module for writing the MapReduce journey for Monte Carlo simulation on the local computer
Sequence；
The text writing module is for production emulation input text on the local computer；
The mirror image production module has the machine images of Hadoop and Monte Carlo software for fabrication and installation beyond the clouds, and
A certain number of virtual servers are instantiated using the machine images of production；
The Hadoop that the cluster configuration module is used to configure on all virtual servers in cloud operates in super distributed mode,
Form Hadoop cluster；
The data transmission module is used to the MapReduce program of local computer and emulation input text being uploaded to void
Quasi- server；
The simulation algorithm model runs MapReduce on Hadoop cluster beyond the clouds, carries out Monte Carlo simulation
Distributed computing.
The technical solution that the embodiment of the present invention is taken further include: it further include function debugging module, the function debugging module
For running Hadoop Streaming, debugging MapReduce program and verifying emulation input text on the local computer.
The technical solution that the embodiment of the present invention is taken further include: it further include data download module, the data download module
For after simulation calculation, cloud simulation result to be downloaded to local computer.
What the Monte Carlo simulation accelerated method based on cloud computing and system of the embodiment of the present invention were provided using cloud platform
Virtual server builds Hadoop cluster, and MapReduce frame is relied on to realize distributed computing, and user only needs customized
The processes such as the calling to Monte Carlo program, intermediate result processing are realized in map, reduce function, in Hadoop cluster
Operation can be realized the emulation of the Monte Carlo based on distributed computing framework MapReduce and accelerate.The present invention is real beyond the clouds
Existing calculating, can be with the quantity and configuration of flexible choice virtual server, and the mode charged on time guarantees that cost is controllable, and every
The place of a accessible network can be used.
Detailed description of the invention
Fig. 1 is the flow chart of the Monte Carlo simulation accelerated method based on cloud computing of the embodiment of the present invention；
Fig. 2 is the flow chart of the method for writing MapReduce program on the local computer of the embodiment of the present invention；
Fig. 3 is the flow chart of the method for the emulation input text of production on the local computer of the embodiment of the present invention；
Fig. 4 is cloud computing service form schematic diagram；
Fig. 5 is the structural schematic diagram of the Monte Carlo simulation acceleration system based on cloud computing of the embodiment of the present invention.
Specific embodiment
In order to make the objectives, technical solutions, and advantages of the present invention clearer, with reference to the accompanying drawings and embodiments, right
The present invention is further elaborated.It should be appreciated that the specific embodiments described herein are merely illustrative of the present invention, and
It is not used in the restriction present invention.
Referring to Fig. 1, being the flow chart of the Monte Carlo simulation accelerated method based on cloud computing of the embodiment of the present invention.This
The Monte Carlo simulation accelerated methods based on cloud computing of inventive embodiments the following steps are included:
Step 100: Hadoop and Monte Carlo software being installed on the local computer, configuration Hadoop operates in pseudo- distribution
Under mode；
In step 100, Hadoop is the distributed computing architecture developed by Apache foundation, is a kind of open source
Software.User can develop distributed program using Hadoop, make full use of without understanding the details of the distributed bottom layer
The power of cluster carries out high speed computing and storage.The core of Hadoop includes HDFS (Hadoop Distributed File
System, Hadoop distributed file system) and MapReduce frame, HDFS provide storage, MapReduce for mass data
Frame provides calculating for mass data.
Hadoop includes three kinds of operational modes:
(1) single cpu mode (Standalone Mode)
Single cpu mode is the default mode of Hadoop, and Hadoop can operate in local completely.Because not needing to save with other
Point interaction, single cpu mode just do not use HDFS, are not loaded with the finger daemon of any Hadoop yet.The mode is mainly used for exploitation and adjusts
Try the application logic of MapReduce frame.
(2) pseudo- distribution pattern (Pseudo-Distributed Mode)
Pseudo- distribution pattern is that Hadoop is operated on " single node cluster ", wherein all finger daemons all operate in it is same
On platform machine, HDFS is also to establish in local file system.The mode increases code debugging function on single cpu mode,
You is allowed to check memory service condition, HDFS input and output and other finger daemon interactions.It has debugged in this mode
Program can operate in without modification under super distributed mode.
(3) super distributed mode (Fully Distributed Mode)
Hadoop finger daemon operates on a cluster, realizes real distributed computing and storage.Wherein guard into
On different machines, HDFS is also to establish on different machines to Cheng Yunhang.There are three types of role, master for machine in cluster
Machine is responsible for the scheduling of task, and Secondary NameNode machine is responsible for the backup to critical data on master machine, and
Worker machine undertakes data processing task.
Step 200: writing the MapReduce program for Monte Carlo simulation on the local computer；
It is writing on the local computer for the embodiment of the present invention referring to Figure 2 together to clearly illustrate step 200
The flow chart of the method for MapReduce program.The side for writing MapReduce program on the local computer of the embodiment of the present invention
Method the following steps are included:
Step 201: writing map program, program successively includes reading simulation calculation task from standard input (stdin), adjusting
Simulation calculation is carried out with Monte Carlo program, marks calculated result with the form write-in of key-value pair (Key-Value-Pair, KVP)
Quasi- output (stdout)；
Step 202: writing reduce program, program successively includes that the emulation meter with same keys (Key) is read from stdin
Result is calculated, calculated result is merged, stdout is written into combined result in the form of KVP；
Step 203: writing Hadoop Streaming operation procedure, program includes that the input of map and reduce program is defeated
Out flow (stream) format, Map and Reduce task (task) number, input text title, outgoing route, mapper and
Reducer filename, upload file path etc..
Step 300: production emulation input text on the local computer；
It is the production on the local computer of the embodiment of the present invention also referring to Fig. 3 to clearly illustrate step 300
The flow chart of the method for emulation input text.The method of the emulation input text of production on the local computer of the embodiment of the present invention
The following steps are included:
Step 301: if the input of Monte Carlo software is random number, random number needed for just generating actual emulation；If
Input is program file, program file needed for just generating actual emulation；
Step 302: according to the scale (Thread Count) of parallel computation to be carried out, random number being grouped or to program file
It is decomposed, every group of random number or the corresponding parallel artificial of each subprogram file；
Step 303: the path of random number or program file being entered into (text) text by row write, as input file.
Step 400: running Hadoop Streaming on the local computer, debugging MapReduce program and verifying are imitative
True input text；
In step 400, when realizing distributed computing using Hadoop, need to develop MapReduce application program.It is most simple
Single MapReudce application program contains at least one map function, a reduce function and a main function.Map and
The general format that reduce function follows is:
map:(k1,v1)→list(k2,v2)
reduce:(k2,list(v2))→list(k3,v3)
Wherein, each member that map function receives one group of data and is converted into a key/value to list, in input domain
The corresponding key/value pair of element；Reduce function receives the list of map function generation, then (raw for each key according to their key
At a key/value to) reduce key/value to list.
During the entire process of writing map and reduce function, input data is to come from bottom distributed file system HDFS,
Intermediate data is placed on local file system, and final output data are write-in bottom distributed file system HDFS.
Step 500: fabrication and installation have the machine images of Hadoop and Monte Carlo software beyond the clouds, according to calculate needs,
Selected hardware configuration instantiates a certain number of virtual servers using the machine images of production；
In step 500, cloud computing (Cloud Computing) is grid computing (Grid Computing), distribution
Calculate (Distributed Computing), parallel computation (Parallel Computing), effectiveness calculating (Utility
Computing), network storage (Network Storage Technologies), virtualization (Virtualization), load
The product of traditional computers technologies such as balanced (Load Balance) and network technical development fusion.It is intended to through network more
The computational entity of a advantage of lower cost is integrated into the perfect system with powerful calculating ability, services by software
(SaaS), platform services (PaaS) and basis instrument and services advanced business models such as (IaaS) powerful computing capability
It is distributed in terminal user's hand.Currently, the main services form of cloud computing include: SaaS (Software as aService, it is soft
Part services), PaaS (Platform as a Service, platform i.e. service) and IaaS (Infrastructure as a
Service, infrastructure services), specifically as shown in figure 4, being cloud computing service form schematic diagram.The present invention is in cloud platform
IaaS builds Hadoop cluster, and the distributed computing of Monte Carlo simulation is realized using the MapReduce frame in Hadoop.
IaaS is supplied to client as metering service i.e. " cloud " infrastructure being made of multiple servers.It is by memory, I/O
Equipment, storage and computing capability are integrated into storage resource required for a virtual resource pool provides for entire industry and virtual
Change the service such as server.The cloud service provider of IaaS type can provide the host of various configurations, and user is it is required to determine that machine
Device mirror image and host hardware configuration.
Machine images provided by cloud service provider are not able to satisfy exploitation demand generally, need further configuration surroundings.
The machine images that the embodiment of the present invention makes beyond the clouds need to install Monte Carlo and Hadoop software, and with local computing equipment
There are identical configuration surroundings, the program debugged on the local computer in this way can directly be run beyond the clouds.Utilize production
Machine images can instantiate any number of virtual server (also known as " node " or " example instance "), avoid to big
Measure the cumbersome of the virtual server configuration surroundings one by one of instantiation.
Step 600: the Hadoop on the configuration all virtual servers in cloud operates in super distributed mode, forms Hadoop collection
Group；
In step 600, the Hadoop on the configuration all virtual servers in cloud operates in super distributed mode and specifically includes:
Select a node as Master and Secondary NameNode respectively, remaining node is as Worker；In local computing
On machine or cloud any node, using SSH communications protocol, successively change Hadoop configuration file according to node type, and by it
Be transmitted on corresponding node, replace the configuration file of original position；Hadoop initialization operation is carried out on Master node,
So that Hadoop is operated in complete distribution pattern, forms Hadoop cluster.
Step 700: the MapReduce program of local computer and emulation input text are uploaded to Master Virtual Service
Device；
Step 800: running MapReduce on Hadoop cluster beyond the clouds, carry out the distributed computing of Monte Carlo simulation；
In step 800, the operation cloud MapReduce carries out the distributed computing of Monte Carlo simulation specifically:
Hadoop Streaming operation procedure is run, MapReduce automatically operates in map program and reduce program different
On Worker node, Map task and Reduce task are formed；In Map task, map program realizes Monte Carlo simulation task
Reading, simulation calculation, intermediate result output；In Reduce task, the reading of result, result between the realization of reduce program
Merging and output；Cluster operating status is monitored by the monitoring page that Hadoop is provided.
Step 900: after simulation calculation, cloud simulation result being downloaded to local computer.
Referring to Fig. 5, being the structural representation of the Monte Carlo simulation acceleration system based on cloud computing of the embodiment of the present invention
Figure.The Monte Carlo simulation acceleration system based on cloud computing of the embodiment of the present invention includes that pattern configurations module, function write mould
Block, text writing module, function debugging module, mirror image make module, cluster configuration module, data transmission module, simulation calculation
Module and data download module；
Pattern configurations module for installing Hadoop and Monte Carlo software, configuration Hadoop operation on the local computer
Under pseudo- distribution pattern；
Function writes module for writing the MapReduce program for Monte Carlo simulation on the local computer；Its
In, it includes: to write map program that function, which writes the method that module writes MapReduce program, and program successively includes inputting from standard
(stdin) read simulation calculation task, call Monte Carlo program carry out simulation calculation, by calculated result with key-value pair (Key-
Value-Pair, KVP) form be written standard output (stdout)；Reduce program is write, program successively includes from stdin
Read the simulation result with same keys (Key), calculated result merged, by combined result in the form of KVP
Stdout is written；Hadoop Streaming operation procedure is write, program includes the iostream of map and reduce program
(stream) format, Map and Reduce task (task) number, input text title, outgoing route, mapper and reducer text
Part name, upload file path etc..
Text writing module is for production emulation input text on the local computer；Wherein, text writing module making
If the input that the method for emulation input text includes: Monte Carlo software is random number, just generate random needed for actual emulation
Number；If input is program file, program file needed for just generating actual emulation；According to the scale (line of parallel computation to be carried out
Number of passes), random number is grouped or program file is decomposed, every group of random number or each subprogram file are one corresponding
Parallel artificial；The path of random number or program file is entered into (text) text by row write, as input file.
Function debugging module debugs MapReduce program for running Hadoop Streaming on the local computer
Input text is emulated with verifying；Wherein, when realizing distributed computing using Hadoop, need to develop MapReduce application program.
Simplest MapReudce application program contains at least one map function, a reduce function and a main function.map
The general format followed with reduce function is:
map:(k1,v1)→list(k2,v2)
reduce:(k2,list(v2))→list(k3,v3)
Wherein, each member that map function receives one group of data and is converted into a key/value to list, in input domain
The corresponding key/value pair of element；Reduce function receives the list of map function generation, then (raw for each key according to their key
At a key/value to) reduce key/value to list.
During the entire process of writing map and reduce function, input data is to come from bottom distributed file system HDFS,
Intermediate data is placed on local file system, and final output data are write-in bottom distributed file system HDFS.
Mirror image production module has the machine images of Hadoop and Monte Carlo software for fabrication and installation beyond the clouds, according to meter
It calculates and needs, selected hardware configuration instantiates a certain number of virtual servers using the machine images of production；
The Hadoop that cluster configuration module is used to configure on all virtual servers in cloud operates in super distributed mode, is formed
Hadoop cluster；Wherein, the Hadoop on the configuration all virtual servers in cloud operates in super distributed mode and specifically includes: respectively
Select a node as Master and Secondary NameNode, remaining node is as Worker；In local computer or
In any node of cloud, using SSH communications protocol, Hadoop configuration file successively is changed according to node type, and they are passed
It transports on corresponding node, replaces the configuration file of original position；Hadoop initialization operation is carried out on Master node, is made
Hadoop operates in complete distribution pattern, forms Hadoop cluster.
Data transmission module is used to the MapReduce program of local computer and emulation input text being uploaded to Master
Virtual server；
Simulation algorithm model runs MapReduce on Hadoop cluster beyond the clouds, carries out point of Monte Carlo simulation
Cloth calculates；Wherein, the operation cloud MapReduce carries out the distributed computing of Monte Carlo simulation specifically: operation
Map program and reduce program are operated in different Worker automatically by Hadoop Streaming operation procedure, MapReduce
On node, Map task and Reduce task are formed；In Map task, the reading of map program realization Monte Carlo simulation task,
Simulation calculation, intermediate result output；In Reduce task, the reading of result, the merging of result between the realization of reduce program
And output；Cluster operating status is monitored by the monitoring page that Hadoop is provided.
Data download module is used for after simulation calculation, and cloud simulation result is downloaded to local computer.
What the Monte Carlo simulation accelerated method based on cloud computing and system of the embodiment of the present invention were provided using cloud platform
Virtual server builds Hadoop cluster, and MapReduce frame is relied on to realize distributed computing, and user only needs customized
The processes such as the calling to Monte Carlo program, intermediate result processing are realized in map, reduce function, in Hadoop cluster
Operation can be realized the Monte Carlo simulation based on distributed computing framework MapReduce and accelerate.The present invention is to realize beyond the clouds
Calculating, can be with the quantity and configuration of flexible choice virtual server, the mode charged on time guarantees that cost is controllable, and each
The place that can access network can be used.
The foregoing is merely illustrative of the preferred embodiments of the present invention, is not intended to limit the invention, all in essence of the invention
Made any modifications, equivalent replacements, and improvements etc., should all be included in the protection scope of the present invention within mind and principle.
Claims (10)
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510885304.5A CN105335215B (en) | 2015-12-05 | 2015-12-05 | A cloud computing-based Monte Carlo simulation acceleration method and system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510885304.5A CN105335215B (en) | 2015-12-05 | 2015-12-05 | A cloud computing-based Monte Carlo simulation acceleration method and system |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN105335215A CN105335215A (en) | 2016-02-17 |
| CN105335215B true CN105335215B (en) | 2019-02-05 |
Family
ID=55285774
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201510885304.5A Active CN105335215B (en) | 2015-12-05 | 2015-12-05 | A cloud computing-based Monte Carlo simulation acceleration method and system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN105335215B (en) |
Families Citing this family (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105740573B (en) * | 2016-03-02 | 2019-10-11 | 苏州网颢信息科技有限公司 | A kind of two-step Monte-carlo Simulation Method calculated for dose radiation |
| CN107172650B (en) * | 2016-03-08 | 2022-03-25 | 中兴通讯股份有限公司 | A simulation method and system for a large-scale complex wireless communication system |
| CN105933154A (en) * | 2016-04-28 | 2016-09-07 | 安徽四创电子股份有限公司 | Management method of cloud calculation resources |
| CN106951324B (en) * | 2017-03-10 | 2021-03-02 | 广东恒聚医疗科技有限公司 | Parallel operation system and method for rapid FLUKA simulation |
| US10147103B2 (en) | 2017-03-24 | 2018-12-04 | International Business Machines Corproation | System and method for a scalable recommender system using massively parallel processors |
| CN109729121B (en) * | 2017-10-31 | 2022-05-06 | 阿里巴巴集团控股有限公司 | Cloud storage system and method for realizing custom data processing in cloud storage system |
| CN110302475B (en) * | 2018-03-20 | 2021-02-19 | 北京连心医疗科技有限公司 | Cloud Monte Carlo dose verification analysis method, equipment and storage medium |
| US10928297B2 (en) | 2019-01-09 | 2021-02-23 | University Of Washington | Method for determining detection angle of optical particle sizer |
| CN109978171B (en) * | 2019-02-26 | 2023-10-10 | 南京航空航天大学 | Grover quantum simulation algorithm optimization method based on cloud computing |
| CN111724451B (en) * | 2020-06-09 | 2024-07-16 | 中国科学院苏州生物医学工程技术研究所 | Tomographic image reconstruction acceleration method, system, terminal and storage medium based on cloud computing |
| CN112001108B (en) * | 2020-07-08 | 2024-02-02 | 中国人民解放军战略支援部队信息工程大学 | Cone beam CT Monte Carlo simulation cluster parallel acceleration method and system |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20130238621A1 (en) * | 2012-03-06 | 2013-09-12 | Microsoft Corporation | Entity Augmentation Service from Latent Relational Data |
| CN103488775A (en) * | 2013-09-29 | 2014-01-01 | 中国科学院信息工程研究所 | Computing system and computing method for big data processing |
-
2015
- 2015-12-05 CN CN201510885304.5A patent/CN105335215B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20130238621A1 (en) * | 2012-03-06 | 2013-09-12 | Microsoft Corporation | Entity Augmentation Service from Latent Relational Data |
| CN103488775A (en) * | 2013-09-29 | 2014-01-01 | 中国科学院信息工程研究所 | Computing system and computing method for big data processing |
Non-Patent Citations (1)
| Title |
|---|
| "基于Hadoop云平台的并行数据挖掘方法";杨来等;《系统仿真学报》;20130531;第25卷(第5期);全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN105335215A (en) | 2016-02-17 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN105335215B (en) | A cloud computing-based Monte Carlo simulation acceleration method and system | |
| Dreher et al. | A flexible framework for asynchronous in situ and in transit analytics for scientific simulations | |
| Cabarle et al. | A spiking neural P system simulator based on CUDA | |
| Huang et al. | Blockemulator: An emulator enabling to test blockchain sharding protocols | |
| Valencia-Cabrera et al. | Simulation challenges in membrane computing | |
| CN106168993A (en) | Electrical network real-time simulation analysis platform | |
| Li et al. | SGL: towards a bridging model for heterogeneous hierarchical platforms | |
| Sahebi et al. | Distributed large-scale graph processing on FPGAs | |
| Wei et al. | LICOM3-CUDA: a GPU version of LASG/IAP climate system ocean model version 3 based on CUDA | |
| Mei et al. | Helix: Serving large language models over heterogeneous gpus and network via max-flow | |
| Han et al. | Bigdatabench-mt: A benchmark tool for generating realistic mixed data center workloads | |
| Ono et al. | Data centric framework for large-scale high-performance parallel computation | |
| DeRose et al. | Relative debugging for a highly parallel hybrid computer system | |
| He et al. | Scalability and efficiency challenges for the exascale supercomputing system: practice of a parallel supporting environment on the Sunway exascale prototype system | |
| Ma et al. | DVM: Towards a datacenter-scale virtual machine | |
| Soytürk et al. | Monitoring collective communication among GPUs | |
| Walker et al. | Composing and executing parallel data-flow graphs with shell pipes | |
| Li et al. | Research and application on cloud simulation | |
| Wu et al. | Parallel artificial neural network using CUDA-enabled GPU for extracting hydraulic domain knowledge of large water distribution systems | |
| Hammond et al. | Predictive simulation of HPC applications | |
| Zehe | Cloud Simulation for Large-Scale Agent-Based Traffic Simulations | |
| Krol et al. | Solving PDEs in modern multiphysics simulation software | |
| Wang | [Retracted] Heterogeneous Cluster Application Communication Optimization and Computer Big Data Management | |
| Tadvin et al. | HELICSAuto: Automating the Development of Cyber-Physical Co-Simulation Framework for Smart Grids | |
| Tennander | Energy Consumption of Micro Frontends: A comparison of micro frontends and single-page applications |
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 |