CN105930210A - Method and device for calling MPI function - Google Patents
Method and device for calling MPI function Download PDFInfo
- Publication number
- CN105930210A CN105930210A CN201610229179.7A CN201610229179A CN105930210A CN 105930210 A CN105930210 A CN 105930210A CN 201610229179 A CN201610229179 A CN 201610229179A CN 105930210 A CN105930210 A CN 105930210A
- Authority
- CN
- China
- Prior art keywords
- mpi
- function
- call request
- request
- queue
- 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
- G06F9/00—Arrangements for program control, e.g. control units
- G06F9/06—Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
- G06F9/46—Multiprogramming arrangements
- G06F9/48—Program initiating; Program switching, e.g. by interrupt
- G06F9/4806—Task transfer initiation or dispatching
- G06F9/4843—Task transfer initiation or dispatching by program, e.g. task dispatcher, supervisor, operating system
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F9/00—Arrangements for program control, e.g. control units
- G06F9/06—Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
- G06F9/46—Multiprogramming arrangements
- G06F9/54—Interprogram communication
- G06F9/546—Message passing systems or structures, e.g. queues
Abstract
The application provides a method and device for calling the MPI function and relates to the technical field of testing. The device comprises a sequencing operation module, a calling execution module and a first preset module, wherein the sequencing operation module is used for conducting a uniform sequencing operation for all MPI calling requests when a computation server executes a task; the calling execution module is suitable for calling MPI functions corresponding to the MPI calling requests one by one for operation according to a sequence after the sequencing operation; the first preset module is suitable for constructing a queue and constructing a corresponding packaging function aiming at each MPI function in an OpenMPI library; when the computation server executes the task, if threads in the computation server send the respective MPI calling requests to call the MPI functions in the open MPI library, the queue and the packaging functions are used for conducting the uniform sequencing operation for all MPI calling requests. The MPI function calling method and the MPI function calling device have the beneficial effects that the MPI functions in the MPI library can be called by multiple threads, and in a cloud computing environment, the advantages of fine grit and high parallel performance of cloud computing can be fully exerted.
Description
Technical field
The application relates to technical field of measurement and test, particularly relates to a kind of MPI function calling method and device.
Background technology
Cloud computing (cloud computing: a kind of by computer network with service by the way of provide the most scalable
The computation schema of virtualized resource) kind of program can be divided into concurrent program without communication type, and also
Line program has communication type.For concurrent program without communication type, this calculating (commonly referred to operation)
Each process (the most each subtask), can at any time start, it is not necessary to and other subtasks are same
Shi Qidong, because being not required to communication between subtask, each subtask has processed the calculating of oneself just can be direct
Terminate.Therefore cloud computing platform scheduler can dispatch different subtasks at different time, processes this
The scheduling mode of dispatching requirement is referred to as task level scheduling.Communication type is had, due to this meter for concurrent program
The each subtask calculating (i.e. operation) is required for intercommunication, it is therefore desirable to wait whole subtask all to start
After, could start to calculate.Namely need cloud computing platform to be currently owned by starting all sons of this operation
After the resource of task, could be by this job initiation, otherwise, need other Activity Calculations by the time complete,
Discharge enough resources, this operation could be started.The scheduling mode processing this dispatching requirement is referred to as operation
Level scheduling.
MPI (Message Passing Interface, message passing interface；A kind of program message passing connects
Mouthful, provide the multilingual function library realizing one series interfaces simultaneously) it is conventional on cloud computing cluster
One of message passing mechanism, cloud computing program can use MPI to carry out transmission and the reception of data.
MPI standard defines one group of function, makes application program from a MPI process, message can be delivered to another
Individual MPI process.
There is communication type for concurrent program, prior art substantially uses OpenMPI carry out each meter
Communication between the process of operator node.OpenMPI is a kind of high-performance message passing library, is initially conduct
Merge technology and resource from other several projects (FT-MPI, LA-MPI, LAM/MPI, and
PACX-MPI), it is a realization of increasing income of MPI-2 standard, by some scientific research institutions together with enterprise
Exploitation and maintenance.Therefore, OpenMPI can obtain professional technique, industrial technology from high-performance community
With resource support, create best MPI library.OpenMPI be supplied to system and software vendor,
Program developer and research worker much facilitate.Easy of use, and run this in various operations
System, network interconnection, and a collection of/dispatching patcher.
OpenMPI the most only supports single-threaded calling, because not considering it it designs when
Applying to multi-thread environment, some key data structures do not add the protection of multi-thread access.If
Different threads calls the function of OpenMPI, these key data structures can be destroyed, cause
OpenMPI inter-process is made mistakes, and makes program occur abnormal.
In prior art, if use OpenMPI, each calculating node can only use single thread mode,
Or use strict program stage by stage.Wherein, pattern the most stage by stage, program life cycle can be divided
For several stages, each stage does different things.Calculating and network such as it is divided into receive and dispatch the stage, two
Stage will not be overlapping, i.e. during calculation stages, does not have network transmitting-receiving and calls, concentrate and data are all calculated
Complete (can calculate with multithreading)；And during the network transmitting-receiving stage, all computational threads are completely in resistance
Plug waiting state, is called OpenMPI function by single network receiving or sending thread, calculating is just tied
Fruit is sent on other machines, and receives their result of calculation from other machines.The most once more carry out
What calculating and network were received and dispatched replaces.
And in above two mode, single thread mode does not make full use of calculating node can support that multithreading enters
The advantage that row processes；
And strict pattern the most stage by stage, it is only suitable for traditional parallel computation environment.And under cloud computing environment,
Can have the disadvantage that
1, in cloud computing environment, generally require use multithreading and data are carried out more fine-grained cutting
Point, and calculating is got up the most parallel so that calculate and network can be carried out simultaneously, improve journey
Sequence energy.And strict program the most stage by stage calculates and network can not be carried out simultaneously, cloud can not be played completely
Calculate fine granularity, high parallel advantage.
2, programming person needs the moment to remember to use OpenMPI function at multiple threads, and this is right
For cloud computing programming person, it is easy to make mistakes, cause program that problem inconspicuous occurs.
Summary of the invention
In view of the above problems, it is proposed that the present invention is to provide one to overcome the problems referred to above or at least partly
Ground solves the MPI function call device of the problems referred to above and corresponding MPI function calling method.
According to one aspect of the present invention, it is provided that a kind of MPI function calling method, including:
For a calculation server in cloud computing, when described calculation server performs a task, if
Thread in described calculation server sends respective MPI call request and calls in OpenMPI storehouse
MPI function, carries out unified sorting operation by described each MPI call request；
By the order after described sorting operation, call the MPI letter that described MPI call request is corresponding one by one
Number operates；
A queue is built for OpenMPI storehouse；For each MPI function structure in OpenMPI storehouse
Build and encapsulate function accordingly；Described queue and described encapsulation function are for performing one when described calculation server
During task, if the thread in described calculation server sends respective MPI call request and calls OpenMPI
MPI function in storehouse, carries out unified sorting operation, wherein, institute by described each MPI call request
Stating encapsulation function please for the MPI call request of the MPI encapsulated for correspondence function is processed as first
Ask and put in aforementioned queue.
Optionally, described when the thread transmission respective MPI call request tune in described calculation server
During with MPI function in OpenMPI storehouse, described each MPI call request is carried out unified sequence behaviour
Work includes:
The MPI call request for a MPI function sent for each thread, described MPI function pair
MPI call request described in the encapsulation intercepting api calls answered, and it is processed as according to described MPI call request
One request, more described first request is put in described queue.
Optionally, MPI call request described in the encapsulation intercepting api calls that described MPI function is corresponding, according to
Described MPI call request is processed as the first request, more described first request is put into described queue includes:
By MPI call request described in described MPI encapsulation intercepting api calls；
The parameter of described MPI call request and the MPI function of correspondence are encapsulated in a structure；
Described structure is put into described queue as the first request.
Optionally, also include:
Build the life cycle receiving or sending thread equal to or more than described task life cycle；Described receipts
Hair line journey, for by the order after described sorting operation, calls described MPI call request corresponding one by one
MPI function operates.
Optionally, described by the order after described sorting operation, call described MPI call request one by one
Corresponding MPI function carries out operation and includes:
From described queue, MPI call request is sequentially read by described receiving or sending thread, and according to request
Content is called corresponding MPI function and is operated.
Optionally, the parameter list one of the parameter list of described encapsulation function and corresponding MPI function
Cause.
According to another aspect of the present invention, it is provided that a kind of MPI function call device, including:
Sorting operation module, is suitable to, for a calculation server in cloud computing, calculate service when described
When device performs a task, if the thread in described calculation server sends respective MPI call request and calls
MPI function in OpenMPI storehouse, carries out unified sorting operation by described each MPI call request；
Call execution module, the order after being suitable to by described sorting operation, call described MPI one by one and adjust
The MPI function corresponding with request operates；
First preset module, is suitable to build a queue for OpenMPI storehouse；For OpenMPI storehouse
In each MPI function build encapsulate function accordingly；Described queue and described encapsulation function are for working as institute
State calculation server when performing a task, if the thread in described calculation server sends respective MPI
Call request calls the MPI function in OpenMPI storehouse, described each MPI call request is unified
Sorting operation, wherein, described encapsulation function is for the MPI of MPI function that will encapsulate for correspondence
Call request is processed as the first request and puts in aforementioned queue.
Optionally, described sorting operation module includes:
First sequence operation module, is suitable to the MPI for a MPI function for each thread sends and adjusts
With request, MPI call request described in the encapsulation intercepting api calls that described MPI function is corresponding, and according to institute
State MPI call request and be processed as the first request, more described first request is put in described queue.
Optionally, described first sequence operation module includes:
Blocking module, is suitable to by MPI call request described in described MPI encapsulation intercepting api calls；
Structure package module, is suitable to the MPI letter of the parameter by described MPI call request and correspondence
Number is encapsulated in a structure；
Structure placement module, is suitable to as the first request, described structure is put into described queue.
Optionally, also include:
Second preset module, is suitable to build a life cycle equal to or more than described task life cycle
Receiving or sending thread；Described receiving or sending thread, for by the order after described sorting operation, calls described one by one
MPI function corresponding to MPI call request operates.
Optionally, call execution module described in include:
First calls execution module, is suitable to sequentially read MPI from described queue by described receiving or sending thread
Call request, and call corresponding MPI function according to request content and operate.
Optionally, the parameter list of described encapsulation function is consistent with the parameter list of corresponding MPI function.
MPI function calling method according to the present invention can be adjusted for the MPI of thread each in multithreading
With request, described each MPI call request is carried out unified sorting operation, thus solves prior art
In achieve and only support the single-threaded mechanism called for existing OpenMPI, for the application of multithreading
Environment, when multithreading calls the MPI function in OpenMPI storehouse, destroys the critical data knot of MPI function
The situation of structure, causes OpenMPI inter-process to be made mistakes, and makes program occur abnormal, and then can not be complete
Full performance cloud computing fine granularity, the problem of high parallel advantage, reached to make multithreading call MPI
Storehouse, in cloud computing environment, can give full play to cloud computing fine granularity, high parallel advantage beneficial effect.
Accompanying drawing explanation
By reading the detailed description of hereafter preferred implementation, various other advantage and benefit for this
Field those of ordinary skill will be clear from understanding.Accompanying drawing is only used for illustrating the purpose of preferred implementation,
And it is not considered as limitation of the present invention.And in whole accompanying drawing, be denoted by the same reference numerals
Identical parts.In the accompanying drawings:
Fig. 1 shows that one MPI function calling method the most of the present invention is implemented
The schematic flow sheet of example one；
Fig. 2 shows that one MPI function calling method the most of the present invention is implemented
The schematic flow sheet of example two；
Fig. 3 shows that one MPI function call device the most of the present invention is implemented
The structural representation of example one；And
Fig. 4 shows a kind of MPI function call device embodiment two
Structural representation.
Detailed description of the invention
It is more fully described the exemplary embodiment of the disclosure below with reference to accompanying drawings.Although accompanying drawing shows
The exemplary embodiment of the disclosure, it being understood, however, that may be realized in various forms the disclosure and not
Should be limited by embodiments set forth here.On the contrary, it is provided that these embodiments are able to more thoroughly
Understand the disclosure, and complete for the scope of the present disclosure can be conveyed to those skilled in the art.
With reference to Fig. 1, it illustrates the flow process signal of the present invention a kind of MPI function calling method embodiment one
Figure, specifically may include that
Step 110, for a calculation server in cloud computing, when described calculation server performs one
During task, if the thread in described calculation server sends respective MPI call request and calls OpenMPI
MPI function in storehouse, carries out unified sorting operation by described each MPI call request；
For the various MPI functions in the OpenMPI storehouse of prior art, only support single-threaded calling, because of
For not considering to be applied to multi-thread environment it designs when, some key data structures do not have
There is the protection adding multi-thread access.The public memory array storage running parameter of the most each MPI function,
If there is multiple thread request to call multiple MPI function simultaneously, the parameter of the most the plurality of thread can be simultaneously
Store in the memory array that this is common, then there may be conflict, cause key data structure to be destroyed.
And in the embodiment of the present invention, for each calculation server in cloud computing, for each calculation server
When using multithreading to perform operation (i.e. a task) having communication type, send MPI when there being thread
When call request calls the MPI function in OpenMPI storehouse, then by the MPI call request of these threads
It is ranked up operation.The MPI call request such as sent by each thread uses and synchronizes locking mechanisms, for
When the function in OpenMPI storehouse is called in multiple requests simultaneously, only only allow a thread in each moment
By (memory array) in key data structure public for parameter read-in.
Step 120, by the order after described sorting operation, calls described MPI call request corresponding one by one
MPI function operate.
After each MPI call request is carried out unified sorting operation, then for each MPI call request
Then there is sequencing, then the MPI call request after can calling sequence the most one by one operates.I.e.
The synchronous working mode making multithreading may conform to OpenMPI when calling the MPI function in OpenMPI storehouse
The single-threaded mechanism in storehouse.
With reference to Fig. 2, it illustrates the stream of currently preferred a kind of MPI function calling method embodiment two
Journey schematic diagram, specifically may include that
Step 200, builds a queue for OpenMPI storehouse；Each in OpenMPI storehouse
MPI function builds and encapsulates function accordingly；Described queue and described encapsulation function are for calculating clothes when described
When business device performs a task, if the thread in described calculation server sends respective MPI call request and adjusts
With the MPI function in OpenMPI storehouse, described each MPI call request is carried out unified sorting operation.
In the embodiment of the present invention, because each MPI function in OpenMPI storehouse public critical data knot
Structure carries out the actions such as the storage of parameter, therefore to avoid the occurrence of thread synchronization to call the situation of MPI function
Occur, for whole OpenMPI storehouse build a queue, and in order to make for each MPI function,
The multiple MPI call request that may be sent by multiple threads, all put in aforementioned queue, then for each
MPI function adds an encapsulation function, and this encapsulation function is for the MPI function that will encapsulate for correspondence
MPI call request is processed as the first request and puts in aforementioned queue.
Wherein, the parameter list of described encapsulation function is consistent with the parameter list of corresponding MPI function.
Being specially each OpenMPI function and provide an encapsulation function, for example, MPI_Send function carries
For an entitled MPI_ThreadSend function, the parameter list of MPI_ThreadSend and MPI_Send
Function keeps consistent.User does not directly invoke MPI_Send function, but passes through MPI_ThreadSend
Function, uses the sending function of OpenMPI.MPI_ThreadSend is by correspondence MPI_Send function
Call request be processed as the first request and put in described queue.
Step 210, builds the life cycle transmitting-receiving line equal to or more than described task life cycle
Journey；Described receiving or sending thread is for by the order after described sorting operation, and calling that described MPI calls one by one please
The MPI function asking corresponding operates.
In embodiments of the present invention, for by the first request in aforementioned queue is processed, creating one
Life cycle is more than or equal to the transmitting-receiving of described task life cycle (life cycle of the most aforementioned operation)
Thread, special the first request read in described queue.This thread reads one by one from queue in order
Request, and call corresponding OpenMPI function according to request content.Such as put into queue for aforementioned
For MPI_Send first request, this receiving or sending thread reads a MPI_Send's from queue
During request, the MPI_Send function just directly invoked in the first request performs.
In embodiments of the present invention, the life cycle that receiving or sending thread is optimum is equal to described task life cycle.
So, a queue is built aforementioned for OpenMPI storehouse；Every in OpenMPI storehouse
In the case of individual MPI function builds corresponding encapsulation function, then enter, when in described calculation server
Thread sends respective MPI call request when calling the MPI function in OpenMPI storehouse, by described respectively
MPI call request carries out unified sorting operation process:
Step 220, the MPI call request for a MPI function sent for each thread, described
MPI call request described in the encapsulation intercepting api calls that MPI function is corresponding, and according to described MPI call please
Ask first request that is processed as, more described first request is put in described queue.
Foregoing for the encapsulation function of each MPI function, its function is will to encapsulate for correspondence
The MPI call request of MPI function is processed as the first request and puts in aforementioned queue.So in this step,
When certain thread sends a MPI call request, during certain MPI function of request call, this MPI letter
Number for encapsulation function then intercept described MPI call request, then according to described MPI call request
It is processed as the first request, more described first request is put in described queue.
Further, the described MPI call request for a MPI function sent for each thread,
MPI call request described in the encapsulation intercepting api calls that described MPI function is corresponding, and adjust according to described MPI
It is processed as the first request with request, more described queue is put in described first request includes:
Step S221, by MPI call request described in described MPI encapsulation intercepting api calls；
As it was previously stated, the MPI for each MPI function encapsulates function, the parameter row of its encapsulation function
Table keeps consistent with actual MPI function.The call request for MPI function that so thread sends,
The encapsulation function of this MPI function the most packed intercepts, and thread does not directly invoke the MPI function of reality,
Also the key data structure that each MPI function is public would not directly be used.
The most aforementioned call request for MPI_Send, its encapsulation function MPI_ThreadSend is then
Call request for MPI_Send is intercepted.
Step S222, is encapsulated into one by the parameter of described MPI call request and the MPI function of correspondence
In structure；
By described encapsulation function by the parameter extraction in MPI call request out, then with corresponding MPI
Function encapsulates together to a structure, and this structure can be used as the first request.
Step S223, puts into described queue using described structure as the first request.
To put in described queue as the structure of the first request.
Receiving or sending thread based on aforementioned structure, described by the order after described sorting operation, call institute one by one
State MPI function corresponding to MPI call request to carry out operation and include:
Step 230, sequentially reads MPI call request by described receiving or sending thread from described queue, and
Call corresponding MPI function according to request content to operate.
Foregoing receiving or sending thread exclusively carries out network transmitting-receiving.This thread reads one from queue in order
Each and every one request, and call corresponding OpenMPI function according to request content.Such as this thread is from queue
When reading first request of a MPI_Send, just directly invoke MPI_Send function and perform (first
The parameter comprised in request performs as the parameter of MPI_Send).
It addition, in the embodiment of the present invention, the target MPI function for encapsulation is preferably non-obstruction MPI
Function.
The occupation mode of the MPI function of the OpenMPI of the embodiment of the present invention, it is provided that multithreading uses
The semanteme of OpenMPI.Owing to corresponding request is placed in queue rank, will not produce and adjust simultaneously
By the situation of OpenMPI function, therefore user has only to substitute real OpenMPI letter with encapsulation function
Number, it is possible to be used in multi-thread environment.I.e. can call OpenMPI at any time calculating when
Function, makes full use of CPU and network interface card resource.
It addition, multiple threads that the present invention can also be for a calculation server use conditional-variable mechanism right
The MPI call request of each thread carries out Synchronization Control.Such as each thread, one conditional-variable is set,
When the variable of a thread changes into true, then the MPI call request of this thread is allowed to go to call MPI
Function, when the variable of thread is false, then the MPI call request refusing this thread goes to call MPI
Function, i.e. blocks this thread.So, when there is multiple thread synchronization and sending MPI call request, first
The variable first only changing one of them thread is true, other threads then wait in line to send MPI calls please
Asking, after this thread completes, its variable changes false into, and the variable of during then notice waits thread changes
For true, then this thread can send MPI call request and conduct interviews.It addition, the present invention also can use
Semaphore Mechanism carries out Synchronization Control to the MPI call request of multithreading, i.e. for OpenMPI lab setting
Quantity i=n of thread can be called, when a thread in n is finished, then i-1, until subtracting simultaneously
It is 0, then the semaphore i=1 of the present invention, puts into a thread the most every time and send the visit of MPI call request
Asking MPI function, other MPI call request sent then wait, when i is kept to 0, the most again from wait
Selecting 1 in thread to put into, i becomes 1, then performs.Multithreading so can also be provided to use
The semanteme of OpenMPI, can call OpenMPI function at any time calculating when, make full use of CPU
With network interface card resource.
With reference to Fig. 3, it illustrates the structural representation of the present invention a kind of MPI function call device embodiment one,
Specifically may include that
Sorting operation module 310, is suitable to for a calculation server in cloud computing, when described calculating
When server performs a task, if the thread in described calculation server sends respective MPI call request
Call the MPI function in OpenMPI storehouse, described each MPI call request is carried out unified sequence behaviour
Make；
Call execution module 320, the order after being suitable to by described sorting operation, call described MPI one by one
MPI function corresponding to call request operates.
Multiple calculation server (calculating node) can be there is in the embodiment of the present invention in multiple cloud computings
Parallel situation, can include MPI function call device in each calculation server.
Optionally, also include:
First preset module, is suitable to build a queue for OpenMPI storehouse；For OpenMPI storehouse
In each MPI function build encapsulate function accordingly；Described queue and described encapsulation function are for working as institute
State calculation server when performing a task, if the thread in described calculation server sends respective MPI
Call request calls the MPI function in OpenMPI storehouse, described each MPI call request is unified
Sorting operation.
Optionally, described sorting operation module includes:
First sequence operation module, is suitable to the MPI for a MPI function for each thread sends and adjusts
With request, MPI call request described in the encapsulation intercepting api calls that described MPI function is corresponding, and according to institute
State MPI call request and be processed as the first request, more described first request is put in described queue.
Optionally, it is characterised in that described first sequence operation module includes:
Blocking module, is suitable to by MPI call request described in described MPI encapsulation intercepting api calls；
Structure package module, is suitable to the MPI letter of the parameter by described MPI call request and correspondence
Number is encapsulated in a structure；
Structure placement module, is suitable to as the first request, described structure is put into described queue.
Optionally, also include:
Second preset module, is suitable to build a life cycle equal to or more than described task life cycle
Receiving or sending thread；Described receiving or sending thread, for by the order after described sorting operation, calls described one by one
MPI function corresponding to MPI call request operates.
Optionally, call execution module described in include:
First calls execution module, is suitable to sequentially read MPI from described queue by described receiving or sending thread
Call request, and call corresponding MPI function according to request content and operate.
Wherein, the parameter list of described encapsulation function is consistent with the parameter list of corresponding MPI function.
With reference to Fig. 4, it illustrates the structural representation of the present invention a kind of MPI function call device embodiment two,
Specifically may include that
First preset module 400, is suitable to build a queue for OpenMPI storehouse；For OpenMPI
Each MPI function in storehouse builds and encapsulates function accordingly；Described queue and described encapsulation function are for working as
When described calculation server performs a task, if the thread in described calculation server sends respective MPI
Call request calls the MPI function in OpenMPI storehouse, described each MPI call request is unified
Sorting operation.
Second preset module 410, is suitable to build a life cycle equal to or more than described task life
The receiving or sending thread in cycle；Described receiving or sending thread, for by the order after described sorting operation, calls institute one by one
The MPI function stating MPI call request corresponding operates.
First sequence operation module 420, is suitable to the MPI for a MPI function sent for each thread
Call request, MPI call request described in the encapsulation intercepting api calls that described MPI function is corresponding, and according to
Described MPI call request is processed as the first request, more described first request is put in described queue.
First calls execution module 430, is suitable to sequentially be read from described queue by described receiving or sending thread
MPI call request, and call corresponding MPI function according to request content and operate.
Wherein, optionally, described first sequence operation module includes:
Blocking module, is suitable to by MPI call request described in described MPI encapsulation intercepting api calls；
Structure package module, is suitable to the MPI letter of the parameter by described MPI call request and correspondence
Number is encapsulated in a structure；
Structure placement module, is suitable to as the first request, described structure is put into described queue.
Wherein, the parameter list of described encapsulation function is consistent with the parameter list of corresponding MPI function.
Algorithm and display be not solid with any certain computer, virtual system or miscellaneous equipment provided herein
Have relevant.Various general-purpose systems can also be used together with based on teaching in this.As described above,
Construct the structure required by this kind of system to be apparent from.Additionally, the present invention is also not for any specific
Programming language.It is understood that, it is possible to use various programming languages realize the content of invention described herein,
And the description done language-specific above is the preferred forms in order to disclose the present invention.
In description mentioned herein, illustrate a large amount of detail.It is to be appreciated, however, that this
Inventive embodiment can be put into practice in the case of not having these details.In some instances, not
It is shown specifically known method, structure and technology, in order to do not obscure the understanding of this description.
Similarly, it will be appreciated that in order to simplify the disclosure and help understand in each inventive aspect one
Or multiple, above in the description of the exemplary embodiment of the present invention, each feature of the present invention is sometimes
It is grouped together in single embodiment, figure or descriptions thereof.But, should be by the disclosure
Method be construed to reflect an intention that i.e. the present invention for required protection require ratio in each claim
The middle more feature of feature be expressly recited.More precisely, as the following claims reflect
As, inventive aspect is all features less than single embodiment disclosed above.Therefore, it then follows
Claims of detailed description of the invention are thus expressly incorporated in this detailed description of the invention, the most each right
Requirement itself is all as the independent embodiment of the present invention.
Those skilled in the art are appreciated that and can carry out the module in the equipment in embodiment certainly
Change adaptively and they are arranged in one or more equipment different from this embodiment.Permissible
Module in embodiment or unit or assembly are combined into a module or unit or assembly, and in addition may be used
To put them into multiple submodule or subelement or sub-component.Except such feature and/or process or
Outside at least some in unit excludes each other, can use any combination that (this specification is included companion
With claim, summary and accompanying drawing) disclosed in all features and so disclosed any method or
All processes of person's equipment or unit are combined.Unless expressly stated otherwise, this specification (includes companion
With claim, summary and accompanying drawing) disclosed in each feature can by provide identical, equivalent or phase
Replace like the alternative features of purpose.
Although additionally, it will be appreciated by those of skill in the art that embodiments more described herein include it
Some feature included in its embodiment rather than further feature, but the group of the feature of different embodiment
Close and mean to be within the scope of the present invention and formed different embodiments.Such as, in following power
In profit claim, one of arbitrarily can mode making in any combination of embodiment required for protection
With.
The all parts embodiment of the present invention can realize with hardware, or to process at one or more
The software module run on device realizes, or realizes with combinations thereof.Those skilled in the art should
Understand, microprocessor or digital signal processor (DSP) can be used in practice to realize basis
A kind of some or all parts in the MPI function call device of the embodiment of the present invention some or
Repertoire.The present invention is also implemented as the part for performing method as described herein or complete
The equipment in portion or device program (such as, computer program and computer program).Such reality
The program of the existing present invention can store on a computer-readable medium, or can have one or more
The form of signal.Such signal can be downloaded from internet website and obtain, or on carrier signal
There is provided, or provide with any other form.
The present invention will be described rather than limits the invention to it should be noted above-described embodiment,
And those skilled in the art can design replacement without departing from the scope of the appended claims
Embodiment.In the claims, any reference marks that should not will be located between bracket is configured to right
The restriction required.Word " comprises " and does not excludes the presence of the element or step not arranged in the claims.Position
Word "a" or "an" before element does not excludes the presence of multiple such element.The present invention can
With by means of including the hardware of some different elements and realizing by means of properly programmed computer.
If in the unit claim listing equipment for drying, several in these devices can be by same
Individual hardware branch specifically embodies.Word first, second and third use do not indicate that any order.
Can be title by these word explanations.
The invention discloses A1, a kind of MPI function calling method, including:
For a calculation server in cloud computing, when described calculation server performs a task, if
Thread in described calculation server sends respective MPI call request and calls in OpenMPI storehouse
MPI function, carries out unified sorting operation by described each MPI call request；
By the order after described sorting operation, call the MPI letter that described MPI call request is corresponding one by one
Number operates；
A queue is built for OpenMPI storehouse；For each MPI function structure in OpenMPI storehouse
Build and encapsulate function accordingly；Described queue and described encapsulation function are for performing one when described calculation server
During task, if the thread in described calculation server sends respective MPI call request and calls OpenMPI
MPI function in storehouse, carries out unified sorting operation, wherein, institute by described each MPI call request
Stating encapsulation function please for the MPI call request of the MPI encapsulated for correspondence function is processed as first
Ask and put in aforementioned queue.
A2, method as described in A1, described send respective when the thread in described calculation server
When MPI call request calls the MPI function in OpenMPI storehouse, described each MPI call request is entered
The unified sorting operation of row includes:
The MPI call request for a MPI function sent for each thread, described MPI function pair
MPI call request described in the encapsulation intercepting api calls answered, and it is processed as according to described MPI call request
One request, more described first request is put in described queue.
A3, method as described in A2, MPI described in the encapsulation intercepting api calls that described MPI function is corresponding
Call request, is processed as the first request according to described MPI call request, more described first request is put into
Described queue includes:
By MPI call request described in described MPI encapsulation intercepting api calls；
The parameter of described MPI call request and the MPI function of correspondence are encapsulated in a structure；
Described structure is put into described queue as the first request.
A4, method as described in 2, also include:
Build the life cycle receiving or sending thread equal to or more than described task life cycle；Described receipts
Hair line journey, for by the order after described sorting operation, calls described MPI call request corresponding one by one
MPI function operates.
A5, method as described in A4, it is characterised in that described by the order after described sorting operation,
Call MPI function corresponding to described MPI call request one by one to carry out operation and include:
From described queue, MPI call request is sequentially read by described receiving or sending thread, and according to request
Content is called corresponding MPI function and is operated.
A6, method as described in A1 or A2,
The parameter list of described encapsulation function is consistent with the parameter list of corresponding MPI function.
The invention also discloses B7, a kind of MPI function call device, including:
Sorting operation module, is suitable to, for a calculation server in cloud computing, calculate service when described
When device performs a task, if the thread in described calculation server sends respective MPI call request and calls
MPI function in OpenMPI storehouse, carries out unified sorting operation by described each MPI call request；
Call execution module, the order after being suitable to by described sorting operation, call described MPI one by one and adjust
The MPI function corresponding with request operates；
First preset module, is suitable to build a queue for OpenMPI storehouse；For OpenMPI storehouse
In each MPI function build encapsulate function accordingly；Described queue and described encapsulation function are for working as institute
State calculation server when performing a task, if the thread in described calculation server sends respective MPI
Call request calls the MPI function in OpenMPI storehouse, described each MPI call request is unified
Sorting operation, wherein, described encapsulation function is for the MPI of MPI function that will encapsulate for correspondence
Call request is processed as the first request and puts in aforementioned queue.
B8, device as described in B 7, described sorting operation module includes:
First sequence operation module, is suitable to the MPI for a MPI function for each thread sends and adjusts
With request, MPI call request described in the encapsulation intercepting api calls that described MPI function is corresponding, and according to institute
State MPI call request and be processed as the first request, more described first request is put in described queue.
B9, device as described in B8, described first sequence operation module includes:
Blocking module, is suitable to by MPI call request described in described MPI encapsulation intercepting api calls；
Structure package module, is suitable to the MPI letter of the parameter by described MPI call request and correspondence
Number is encapsulated in a structure；
Structure placement module, is suitable to as the first request, described structure is put into described queue.
B10, device as described in B8, also include:
Second preset module, is suitable to build a life cycle equal to or more than described task life cycle
Receiving or sending thread；Described receiving or sending thread, for by the order after described sorting operation, calls described one by one
MPI function corresponding to MPI call request operates.
B11, device as described in B8, described in call execution module and include:
First calls execution module, is suitable to sequentially read MPI from described queue by described receiving or sending thread
Call request, and call corresponding MPI function according to request content and operate.
B12, device as described in B7,
The parameter list of described encapsulation function is consistent with the parameter list of corresponding MPI function.
Claims (10)
1. a MPI function calling method, it is characterised in that including:
For a calculation server in cloud computing, when described calculation server performs a task, if
Thread in described calculation server sends respective MPI call request and calls in OpenMPI storehouse
MPI function, carries out unified sorting operation by described each MPI call request；
By the order after described sorting operation, call the MPI letter that described MPI call request is corresponding one by one
Number operates；
A queue is built for OpenMPI storehouse；For each MPI function structure in OpenMPI storehouse
Build and encapsulate function accordingly；Described queue and described encapsulation function are for performing one when described calculation server
During task, if the thread in described calculation server sends respective MPI call request and calls OpenMPI
MPI function in storehouse, carries out unified sorting operation, wherein, institute by described each MPI call request
Stating encapsulation function please for the MPI call request of the MPI encapsulated for correspondence function is processed as first
Ask and put in aforementioned queue.
Method the most according to claim 1, it is characterised in that described when described calculation server
In thread send respective MPI call request when calling the MPI function in OpenMPI storehouse, by institute
State each MPI call request to carry out unified sorting operation and include:
The MPI call request for a MPI function sent for each thread, described MPI function pair
MPI call request described in the encapsulation intercepting api calls answered, and it is processed as according to described MPI call request
One request, more described first request is put in described queue.
Method the most according to claim 2, it is characterised in that the envelope that described MPI function is corresponding
MPI call request described in dress intercepting api calls, is processed as the first request according to described MPI call request,
Again described first request is put into described queue to include:
By MPI call request described in described MPI encapsulation intercepting api calls；
The parameter of described MPI call request and the MPI function of correspondence are encapsulated in a structure；
Described structure is put into described queue as the first request.
Method the most according to claim 2, it is characterised in that also include:
Build the life cycle receiving or sending thread equal to or more than described task life cycle；Described receipts
Hair line journey, for by the order after described sorting operation, calls described MPI call request corresponding one by one
MPI function operates.
Method the most according to claim 1 and 2, it is characterised in that
The parameter list of described encapsulation function is consistent with the parameter list of corresponding MPI function.
6. a MPI function call device, it is characterised in that including:
Sorting operation module, is suitable to, for a calculation server in cloud computing, calculate service when described
When device performs a task, if the thread in described calculation server sends respective MPI call request and calls
MPI function in OpenMPI storehouse, carries out unified sorting operation by described each MPI call request；
Call execution module, the order after being suitable to by described sorting operation, call described MPI one by one and adjust
The MPI function corresponding with request operates；
First preset module, is suitable to build a queue for OpenMPI storehouse；For OpenMPI storehouse
In each MPI function build encapsulate function accordingly；Described queue and described encapsulation function are for working as institute
State calculation server when performing a task, if the thread in described calculation server sends respective MPI
Call request calls the MPI function in OpenMPI storehouse, described each MPI call request is unified
Sorting operation, wherein, described encapsulation function is for the MPI of MPI function that will encapsulate for correspondence
Call request is processed as the first request and puts in aforementioned queue.
Device the most according to claim 7, it is characterised in that described sorting operation module includes:
First sequence operation module, is suitable to the MPI for a MPI function for each thread sends and adjusts
With request, MPI call request described in the encapsulation intercepting api calls that described MPI function is corresponding, and according to institute
State MPI call request and be processed as the first request, more described first request is put in described queue.
Device the most according to claim 8, it is characterised in that described first sequence operation module
Including:
Blocking module, is suitable to by MPI call request described in described MPI encapsulation intercepting api calls；
Structure package module, is suitable to the MPI letter of the parameter by described MPI call request and correspondence
Number is encapsulated in a structure；
Structure placement module, is suitable to as the first request, described structure is put into described queue.
Device the most according to claim 8, it is characterised in that also include:
Second preset module, is suitable to build a life cycle equal to or more than described task life cycle
Receiving or sending thread；Described receiving or sending thread, for by the order after described sorting operation, calls described one by one
MPI function corresponding to MPI call request operates.
Device the most according to claim 7, it is characterised in that
The parameter list of described encapsulation function is consistent with the parameter list of corresponding MPI function.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610229179.7A CN105930210B (en) | 2012-12-05 | 2012-12-05 | MPI function calling method and device |
Applications Claiming Priority (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201210518379.6A CN103019843B (en) | 2012-12-05 | 2012-12-05 | MPI function calling method and device |
| CN201610229179.7A CN105930210B (en) | 2012-12-05 | 2012-12-05 | MPI function calling method and device |
Related Parent Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201210518379.6A Division CN103019843B (en) | 2012-12-05 | 2012-12-05 | MPI function calling method and device |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN105930210A true CN105930210A (en) | 2016-09-07 |
| CN105930210B CN105930210B (en) | 2019-02-26 |
Family
ID=47968474
Family Applications (2)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201210518379.6A Active CN103019843B (en) | 2012-12-05 | 2012-12-05 | MPI function calling method and device |
| CN201610229179.7A Expired - Fee Related CN105930210B (en) | 2012-12-05 | 2012-12-05 | MPI function calling method and device |
Family Applications Before (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201210518379.6A Active CN103019843B (en) | 2012-12-05 | 2012-12-05 | MPI function calling method and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (2) | CN103019843B (en) |
Families Citing this family (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104281489B (en) * | 2013-07-12 | 2017-11-21 | 上海携程商务有限公司 | Multithreading requesting method and system under SOA framework |
| CN113395358B (en) * | 2021-08-16 | 2021-11-05 | 贝壳找房(北京)科技有限公司 | Network request execution method and execution system |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1289962A (en) * | 1999-09-23 | 2001-04-04 | 国际商业机器公司 | Establishment of multiple process spanned communication programme in multiple linear equation running environment |
| CN101196827A (en) * | 2007-12-28 | 2008-06-11 | 中国科学院计算技术研究所 | Parallel simulator and method |
| CN101833479A (en) * | 2010-04-16 | 2010-09-15 | 中国人民解放军国防科学技术大学 | MPI (Moldflow Plastics Insight) information scheduling method based on reinforcement learning under multi-network environment |
| CN101937367A (en) * | 2009-06-30 | 2011-01-05 | 英特尔公司 | The MPI source code program arrives the automatic conversion based on the program of MPI thread |
| US20120260261A1 (en) * | 2011-04-07 | 2012-10-11 | Microsoft Corporation | Asynchronous callback driven messaging request completion notification |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102707955A (en) * | 2012-05-18 | 2012-10-03 | 天津大学 | Method for realizing support vector machine by MPI programming and OpenMP programming |
-
2012
- 2012-12-05 CN CN201210518379.6A patent/CN103019843B/en active Active
- 2012-12-05 CN CN201610229179.7A patent/CN105930210B/en not_active Expired - Fee Related
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1289962A (en) * | 1999-09-23 | 2001-04-04 | 国际商业机器公司 | Establishment of multiple process spanned communication programme in multiple linear equation running environment |
| CN101196827A (en) * | 2007-12-28 | 2008-06-11 | 中国科学院计算技术研究所 | Parallel simulator and method |
| CN101937367A (en) * | 2009-06-30 | 2011-01-05 | 英特尔公司 | The MPI source code program arrives the automatic conversion based on the program of MPI thread |
| CN101833479A (en) * | 2010-04-16 | 2010-09-15 | 中国人民解放军国防科学技术大学 | MPI (Moldflow Plastics Insight) information scheduling method based on reinforcement learning under multi-network environment |
| US20120260261A1 (en) * | 2011-04-07 | 2012-10-11 | Microsoft Corporation | Asynchronous callback driven messaging request completion notification |
Also Published As
| Publication number | Publication date |
|---|---|
| CN105930210B (en) | 2019-02-26 |
| CN103019843B (en) | 2016-05-11 |
| CN103019843A (en) | 2013-04-03 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US11113782B2 (en) | Dynamic kernel slicing for VGPU sharing in serverless computing systems | |
| CN103069390B (en) | Method and system for re-scheduling workload in a hybrid computing environment | |
| CN106933669B (en) | Apparatus and method for data processing | |
| US9251103B2 (en) | Memory-access-resource management | |
| CN103999051B (en) | Strategy for tinter resource allocation in the minds of shader core | |
| CN104978228B (en) | A kind of dispatching method and device of distributed computing system | |
| Iordache et al. | Resilin: Elastic mapreduce over multiple clouds | |
| CN105045658A (en) | Method for realizing dynamic dispatching distribution of task by multi-core embedded DSP (Data Structure Processor) | |
| CN103595770B (en) | Method and device for achieving file downloading through SDK | |
| US20230393879A1 (en) | Coordinated Container Scheduling For Improved Resource Allocation In Virtual Computing Environment | |
| CN108196946A (en) | A kind of subregion multinuclear method of Mach | |
| CN103744716A (en) | Dynamic interrupt balanced mapping method based on current virtual central processing unit (VCPU) scheduling state | |
| CN116541134B (en) | Method and device for deploying containers in multi-architecture cluster | |
| CN101751288A (en) | Method, device and system applying process scheduler | |
| US8935699B1 (en) | CPU sharing techniques | |
| CN106250217A (en) | Synchronous dispatching method between a kind of many virtual processors and dispatching patcher thereof | |
| KR102052964B1 (en) | Method and system for scheduling computing | |
| US11886898B2 (en) | GPU-remoting latency aware virtual machine migration | |
| CN106383747A (en) | Method and device for scheduling computing resources | |
| CN103019844B (en) | A kind ofly support multithreading to call the method and apparatus of MPI function | |
| CN115686805A (en) | GPU resource sharing method and device, and GPU resource sharing scheduling method and device | |
| US9158601B2 (en) | Multithreaded event handling using partitioned event de-multiplexers | |
| CN105930210A (en) | Method and device for calling MPI function | |
| Leonenkov et al. | Introducing new backfill-based scheduler for slurm resource manager | |
| Shen et al. | Serpens: A high-performance serverless platform for nfv |
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 | ||
| CF01 | Termination of patent right due to non-payment of annual fee | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20190226 Termination date: 20211205 |