CN107180153A - The method and system of full waveform inversion is realized using MPI - Google Patents
The method and system of full waveform inversion is realized using MPI Download PDFInfo
- Publication number
- CN107180153A CN107180153A CN201610141098.1A CN201610141098A CN107180153A CN 107180153 A CN107180153 A CN 107180153A CN 201610141098 A CN201610141098 A CN 201610141098A CN 107180153 A CN107180153 A CN 107180153A
- Authority
- CN
- China
- Prior art keywords
- speed
- concurrent processes
- task
- mpi
- calculation
- 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.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16Z—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS, NOT OTHERWISE PROVIDED FOR
- G16Z99/00—Subject matter not provided for in other main groups of this subclass
Landscapes
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses the method and system that full waveform inversion is realized using MPI.In invention, each MPI concurrent processes in multiple MPI concurrent processes are accordingly to be regarded as host process, each MPI concurrent processes, which are set to, to be obtained task that the MPI concurrent processes currently perform and the geological data for performing the required by task by accessing share dish and is back to corresponding result of calculation file by the result of calculation that the task obtains is performed, and different MPI concurrent processes are configured to not communicate mutually.Using the present invention, the dynamic change of hardware device can be supported in full waveform inversion, enables to that Network Computing Architecture is more stable, heterogeneous device can be supported, being easy to reduce hardware cost, in addition using present invention also eliminates limitation of the network service bottleneck to full waveform inversion efficiency between different MPI concurrent processes.
Description
Technical field
The present invention relates to imaging of seismic data field, more particularly, to the method and system that full waveform inversion (FWI) is realized using MPI.
Background technology
Full waveform inversion is the key and cutting edge technology that seismic wave is modeled as in, because its amount of calculation is very huge, it usually needs rely on the parallel computation of large-scale high-performance computer cluster to realize.Run on large-scale high-performance computer cluster, the single project operation time generally reaches a few weeks or months.Therefore the parallel computation of full waveform inversion how is realized, and enables full waveform inversion project to overcome the hardware fault of computer cluster, while realizing efficient parallel computation, the key difficulties of industrialization full waveform inversion technology are realized as people.
Disclosed full waveform inversion parallel computation implementations all at present, are all realized using MPI (Message Passing Interface) technologies.MPI Parallel Implementations, are typically a number of from node real-time collaborative Parallel Implementation by host node band.The gradient calculation and step size computation of order execution single-shot are configured as from node as shown in figure 1, each, and is superimposed the gradient of many big guns and determines that final step-length is performed in the master node according to the step-length of many big guns.The whole process calculated in full waveform inversion as shown in Figure 1, host node exchanges data between node, from there will be real-time network communication between node.
In actual applications, there are many notable defects in this method:1) the dynamic increase of hardware device is not supported and is reduced:MPI concurrent jobs are once sent, it cannot add or reduce the computer node for participating in calculating, and it is generally configured with a large amount of computer nodes in large data processing center, there are a large amount of job runs simultaneously, available computer node Number dynamics are caused to change greatly, only support dynamic device increase and reduce, making full use of for resource could be realized;2) stability is poor:During using MPI technologies, a large amount of computer nodes are calculated and interrelated simultaneously, and the calculating time of usual full waveform inversion technology is long (a few week to some months), once hardware fault occurs in some node, whole calculate fails immediately;3) heterogeneous device is not supported:MPI technologies can not operate in the computer cluster of isomery, and all computer nodes of MPI technical requirements must be identical, if computer node is different, and it is run by the efficiency of most slow computer node.And in production application, because the time that computer node is purchased is different, quickly, inevitably lead to computer cluster is made up of simultaneous computer hardware update speed different computer nodes, therefore be not suitable for MPI parallel computation;4) hardware cost is high:MPI Parallel Implementations are very high to the network speed between the node of computer cluster and stability requirement, people are forced using Infiniband and other equipment to improve network communication efficiency, people are forced and are improved using more expensive hardware the stability of system, these are obviously improved the buying deployment cost of high-performance calculation machine equipment without doubt;5) Large-scale parallel computing inefficiency:The parallel efficiency of MPI technologies increases and reduced with computer node number, once computer node number is more than 10, the bottleneck effect that MPI is lifted to computational efficiency just has embodiment.
Therefore, although current people generally use above-mentioned MPI parallel methods in the implementation process of full waveform inversion technology, but because it has drawbacks described above, seriously constrain the further development of full waveform inversion technology.
The content of the invention
The present invention proposes the system that a kind of use MPI that can overcome drawbacks described above realizes full waveform inversion.The invention also provides corresponding method.
According to an aspect of the invention, it is proposed that a kind of method that use MPI realizes full waveform inversion, this method includes:Each MPI concurrent processes in multiple MPI concurrent processes are set to and obtains task that the MPI concurrent processes currently perform and the geological data for performing the required by task by accessing share dish and is back to corresponding result of calculation file by the result of calculation that the task obtains is performed, and different MPI concurrent processes are configured to not communicate mutually.Wherein, the share dish are used to store shared status file and shared data body, task, the storage path of the result of calculation file of the current task of each MPI concurrent processes that maximum cycle, current cycle time, each MPI concurrent processes of full waveform inversion are currently performed are stored in the shared status file, the shared data body includes the geological data needed for full waveform inversion.
According to another aspect of the invention, it is proposed that a kind of system that use MPI realizes full waveform inversion, the system includes share dish and multiple MPI concurrent processes, wherein;The share dish are used to store shared status file and shared data body, task, the storage path of the result of calculation file of the current task of each MPI concurrent processes that maximum cycle, current cycle time, each MPI concurrent processes of full waveform inversion are currently performed are stored in the shared status file, the shared data body includes the geological data needed for full waveform inversion;Each MPI concurrent processes in multiple MPI concurrent processes are both configured to obtain task that the MPI concurrent processes currently perform and the geological data for performing the required by task by accessing share dish and are back to corresponding result of calculation file by the result of calculation that the task obtains is performed, and different MPI concurrent processes are configured as not communicating mutually.
Each aspect of the present invention, each MPI concurrent processes are considered as host process, each MPI concurrent processes obtain its being performed for task by way of accessing share dish and carry out data interaction, do not communicated mutually between different MPI concurrent processes, so as to support the dynamic change of hardware device in full waveform inversion, enable to that Network Computing Architecture is more stable, heterogeneous device can be supported, be easy to reduction hardware cost, in addition using present invention also eliminates limitation of the network service bottleneck to full waveform inversion efficiency between different MPI concurrent processes.
Brief description of the drawings
By the way that exemplary embodiment of the invention is described in more detail with reference to accompanying drawing, above-mentioned and other purpose, the feature and advantage of the present invention will be apparent, wherein, in exemplary embodiment of the invention, identical reference number typically represents same parts.
Fig. 1 shows the schematic diagram for realizing full waveform inversion using MPI in the prior art.
Fig. 2 shows that the full waveform inversion Distributed Parallel Computing of example according to an embodiment of the invention realizes schematic diagram.
Fig. 3 shows that use MPI according to an embodiment of the invention realizes the schematic diagram of the system of full waveform inversion.
Embodiment
The preferred embodiment of the present invention is more fully described below with reference to accompanying drawings.Although showing the preferred embodiment of the present invention in accompanying drawing, however, it is to be appreciated that may be realized in various forms the present invention without that should be limited by embodiments set forth herein.On the contrary, thesing embodiments are provided so that the present invention is more thorough and complete, and those skilled in the art can be will fully convey the scope of the invention to.
Embodiment 1
The invention discloses a kind of method that use MPI realizes full waveform inversion.This method carries out full waveform inversion including the use of share dish and multiple MPI concurrent processes.Each MPI concurrent processes in multiple MPI concurrent processes are set to and obtains task that the MPI concurrent processes currently perform and the geological data for performing the required by task by accessing share dish and is back to corresponding result of calculation file by the result of calculation that the task obtains is performed, and different MPI concurrent processes are configured to not communicate mutually.Wherein, the share dish are used to store shared status file and shared data body, task, the storage path of the result of calculation file of the current task of each MPI concurrent processes that maximum cycle, current cycle time, each MPI concurrent processes of full waveform inversion are currently performed are stored in the shared status file, the shared data body includes the geological data needed for full waveform inversion.
In above-described embodiment, each MPI concurrent processes are considered as host process, each MPI concurrent processes obtain its being performed for task by way of accessing share dish and carry out data interaction, do not communicated mutually between different MPI concurrent processes, so as to support the dynamic change of hardware device in full waveform inversion, enable to that Network Computing Architecture is more stable, heterogeneous device can be supported, be easy to reduction hardware cost, in addition using present invention also eliminates limitation of the network service bottleneck to full waveform inversion efficiency between different MPI concurrent processes.
The task can include:Gradient calculation, step size computation, renewal speed.The gradient calculation can include:The speed that single-shot can be calculated updates gradient, can be superimposed the speed renewal gradient and current interim total gradient of calculated single-shot to obtain new interim total gradient, and can return to new interim total gradient to corresponding result of calculation file.The step size computation can include:Can calculate single-shot speed update step-length, can the speed based on the single-shot calculated step-length and current interim total step-length is updated to obtain new interim total step-length, and new interim total step-length can be returned to corresponding result of calculation file.The renewal speed includes the speed renewal amount based on this circulation obtained from share dish and calculates the speed after single-shot updates, and returns to the speed after single-shot updates to corresponding result of calculation file.As known to those skilled in the art, the speed renewal amount of this circulation can be obtained based on the total step-length obtained after all big guns in the final total gradient and traversal work area obtained after all big guns in traversal work area.
After all big guns traversal in for work area performs the gradient calculation, task that can be by the current execution of part or all of MPI concurrent processes in the shared status file is rewritten as step size computation;After all big guns traversal in for work area performs the step size computation, the task of the current execution of part or all of MPI concurrent processes is rewritten as renewal speed in the shared status file.All big guns i.e. in work area are traveled through after the gradient calculation task, then travel through the step size computation task to all big guns in work area;All big guns in work area are traveled through after the step size computation task, then travel through the renewal speed task to all big guns in work area.
When meeting following condition for the moment, full waveform inversion can be terminated:When the speed renewal amount of this circulation is less than given threshold, when current cycle time reaches the maximum cycle;If being unsatisfactory for, it can start to circulate next time using the speed after renewal as new initial velocity model.
The form that multiple MPI concurrent processes can be locked by document of agreement accesses the shared status file and shared data body in the share dish.Document of agreement lock is a kind of file read-write mechanism, only allows a MPI concurrent process to access same file at any time, it can be ensured that the uniqueness and correctness of FileVersion.
Fig. 2 shows that the full waveform inversion Distributed Parallel Computing of example according to an embodiment of the invention realizes schematic diagram.
Embodiment 2
Fig. 3 shows that use MPI according to an embodiment of the invention realizes the schematic diagram of the system of full waveform inversion.In the present embodiment, the system includes:Share dish 101 and MPI concurrent processes 201,202,203.Share dish 101 are used to store shared status file and shared data body, task, the storage path of the result of calculation file of the current task of each MPI concurrent processes that maximum cycle, current cycle time, each MPI concurrent processes of full waveform inversion are currently performed are stored in the shared status file, the shared data body includes the geological data needed for full waveform inversion.MPI concurrent processes 201,202,203 are both configured to obtain the MPI concurrent processes currently the performing of the task and the geological data for performing the required by task by accessing share dish 101 and are back to corresponding result of calculation file by the result of calculation that the task obtains is performed.MPI concurrent processes 201,202,203 are configured as not communicating mutually.3 MPI concurrent processes 201,202,203 are illustrate only in Fig. 3 as an example, it will be appreciated by those skilled in the art that it is only used for schematically illustrating, being not used in the quantity of limitation MPI concurrent processes.
The task can include:Gradient calculation, step size computation, renewal speed.The gradient calculation can include:It can calculate that the speed of single-shot updates gradient, the speed that can be superimposed calculated single-shot updates gradient and current interim total gradient to obtain new interim total gradient and can return to new interim total gradient to corresponding result of calculation file.The step size computation can include:Can calculate single-shot speed update step-length, can the speed based on the single-shot calculated step-length and current interim total step-length is updated to obtain new interim total step-length and new interim total step-length can be returned to corresponding result of calculation file.The renewal speed can include the speed renewal amount based on this circulation obtained from share dish and calculate the speed after single-shot updates and return to the speed after single-shot updates to corresponding result of calculation file.As known to those skilled in the art, the speed renewal amount of this circulation can be obtained based on the total step-length obtained after all big guns in the final total gradient and traversal work area obtained after all big guns in traversal work area.
After all big guns traversal in for work area performs the gradient calculation, the task of the current execution of the part or all of MPI concurrent processes in the shared status file can be rewritten as step size computation;After all big guns traversal in for work area performs the step size computation, the task of the current execution of the part or all of MPI concurrent processes in the shared status file can be rewritten as renewal speed.
When meeting following condition for the moment, full waveform inversion can be terminated:When the speed renewal amount of this circulation is less than given threshold, when current cycle time reaches the maximum cycle;If being unsatisfactory for, it can start to circulate next time using the speed after renewal as new initial velocity model.
The form that multiple MPI concurrent processes can be locked by document of agreement accesses the shared status file and shared data body in the share dish.Document of agreement lock is a kind of file read-write mechanism, only allows a MPI concurrent process to access same file at any time, it can be ensured that the uniqueness and correctness of FileVersion.
There is no close dependence according between this distributed MPI concurrent processes of the present invention, therefore have the following advantages that in actual applications:1) support the dynamic increase of hardware device and reduce:The node for participating in calculating dynamically can increase and reduce, and each node/process can be by shared status file/shared data body come adjustment work point of penetration;2) stability is good:Because computer node is without interrelated, in the calculating of prolonged full waveform inversion, interrupted even if part of nodes breaks down, whole calculating operation still may proceed to;3) heterogeneous device is supported:The node of different computational efficiencies can update shared status file, voluntarily Adjustable calculation amount, will not intervene the calculating of other nodes;4) hardware cost is low:Parallel computation than relatively low, therefore can use the hardware device of low cost to the requirement such as stability, network service of system;5) parallel efficiency calculation is not influenceed by interstitial content:Because without the network communication efficiency bottle neck between node, increase node will not bring extra computational efficiency to lose.
Contrasted table 1 below illustrates the parallel full waveform inversions of the MPI of prior art and according to the key technical indexes of the parallel full waveform inversions of MPI of the present invention.
Technical indicator | Prior art | The present invention |
Dynamic increase and decrease calculate node | Can not | Energy |
Part of nodes fault interrupting | Overall operation is interrupted | Overall operation continues |
Support Heterogeneous Computing node | Can not | Energy |
Node hardware performance requirement | It is high | It is low |
Parallel efficiency and average single-unit point efficiency | Node increases, efficiency reduction | Node increases, and efficiency is constant |
It is described above various embodiments of the present invention, described above is exemplary, and non-exclusive, and is also not necessarily limited to disclosed each embodiment.In the case of without departing from the scope and spirit of illustrated each embodiment, many modifications and changes will be apparent from for those skilled in the art.The selection of term used herein, it is intended to best explain the principle, practical application or the improvement to the technology in market of each embodiment, or other those of ordinary skill of the art is understood that each embodiment disclosed herein.
Claims (10)
1. a kind of method that use MPI realizes full waveform inversion, this method includes;
Each MPI concurrent processes in multiple MPI concurrent processes are set to and obtains task that the MPI concurrent processes currently perform and the geological data for performing the required by task by accessing share dish and is back to corresponding result of calculation file by the result of calculation that the task obtains is performed, and different MPI concurrent processes are configured to not communicate mutually;Wherein,
The share dish are used to store shared status file and shared data body, task, the storage path of the result of calculation file of the current task of each MPI concurrent processes that maximum cycle, current cycle time, each MPI concurrent processes of full waveform inversion are currently performed are stored in the shared status file, the shared data body includes the geological data needed for full waveform inversion.
2. according to the method described in claim 1, wherein, the task includes:Gradient calculation, step size computation, renewal speed;
The gradient calculation includes:The speed for calculating single-shot updates gradient, is superimposed the speed renewal gradient and current interim total gradient of calculated single-shot to obtain new interim total gradient, and returns to new interim total gradient to corresponding result of calculation file;
The step size computation includes:The speed for calculating single-shot updates step-length, the speed renewal step-length based on the single-shot calculated and current interim total step-length to obtain new interim total step-length, and returns to new interim total step-length to corresponding result of calculation file;
The renewal speed includes the speed renewal amount based on this circulation obtained from share dish and calculates the speed after single-shot updates, and returns to the speed after single-shot updates to corresponding result of calculation file.
3. method according to claim 2, wherein,
After all big guns traversal in for work area performs the gradient calculation, the task by the current execution of part or all of MPI concurrent processes in the shared status file is rewritten as step size computation;
After all big guns traversal in for work area performs the step size computation, the task by the current execution of part or all of MPI concurrent processes in the shared status file is rewritten as renewal speed.
4. according to the method described in claim 1, wherein,
When meeting following condition for the moment, terminate full waveform inversion:When the speed renewal amount of this circulation is less than given threshold, when current cycle time reaches the maximum cycle;
If being unsatisfactory for, using the speed after renewal as new initial velocity model, start to circulate next time.
5. according to the method described in claim 1, wherein, the form that the multiple MPI concurrent processes are locked by document of agreement accesses the shared status file and shared data body in the share dish.
6. a kind of system that use MPI realizes full waveform inversion, the system includes share dish and multiple MPI concurrent processes, wherein;
The share dish are used to store shared status file and shared data body, task, the storage path of the result of calculation file of the current task of each MPI concurrent processes that maximum cycle, current cycle time, each MPI concurrent processes of full waveform inversion are currently performed are stored in the shared status file, the shared data body includes the geological data needed for full waveform inversion;
Each MPI concurrent processes in multiple MPI concurrent processes are both configured to obtain task that the MPI concurrent processes currently perform and the geological data for performing the required by task by accessing share dish and are back to corresponding result of calculation file by the result of calculation that the task obtains is performed, and different MPI concurrent processes are configured as not communicating mutually.
7. system according to claim 6, wherein, the task includes:Gradient calculation, step size computation, renewal speed;
The gradient calculation includes:The speed for calculating single-shot updates gradient, is superimposed the speed renewal gradient and current interim total gradient of calculated single-shot to obtain new interim total gradient, and returns to new interim total gradient to corresponding result of calculation file;
The step size computation includes:The speed for calculating single-shot updates step-length, the speed renewal step-length based on the single-shot calculated and current interim total step-length to obtain new interim total step-length, and returns to new interim total step-length to corresponding result of calculation file;
The renewal speed includes the speed renewal amount based on this circulation obtained from share dish and calculates the speed after single-shot updates, and returns to the speed after single-shot updates to corresponding result of calculation file.
8. system according to claim 7, wherein,
After all big guns traversal in for work area performs the gradient calculation, the task of the current execution of the part or all of MPI concurrent processes in the shared status file is rewritten as step size computation;
After all big guns traversal in for work area performs the step size computation, the task of the current execution of the part or all of MPI concurrent processes in the shared status file is rewritten as renewal speed.
9. system according to claim 6, wherein,
When meeting following condition for the moment, terminate full waveform inversion:When the speed renewal amount of this circulation is less than given threshold, when current cycle time reaches the maximum cycle;
If being unsatisfactory for, using the speed after renewal as new initial velocity model, start to circulate next time.
10. system according to claim 6, wherein, the form that the multiple MPI concurrent processes are locked by document of agreement accesses shared status file and shared data body in the share dish.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610141098.1A CN107180153A (en) | 2016-03-11 | 2016-03-11 | The method and system of full waveform inversion is realized using MPI |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610141098.1A CN107180153A (en) | 2016-03-11 | 2016-03-11 | The method and system of full waveform inversion is realized using MPI |
Publications (1)
Publication Number | Publication Date |
---|---|
CN107180153A true CN107180153A (en) | 2017-09-19 |
Family
ID=59830845
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201610141098.1A Pending CN107180153A (en) | 2016-03-11 | 2016-03-11 | The method and system of full waveform inversion is realized using MPI |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN107180153A (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109344135A (en) * | 2018-10-18 | 2019-02-15 | 中国海洋石油集团有限公司 | A kind of parallel seismic processing job scheduling method of the file lock of automatic load balancing |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20100161911A1 (en) * | 2007-05-31 | 2010-06-24 | Eric Li | Method and apparatus for mpi program optimization |
CN102902514A (en) * | 2012-09-07 | 2013-01-30 | 西安交通大学 | Large-scale parallel processing method of moving particle semi-implicit method |
CN103076627A (en) * | 2011-10-26 | 2013-05-01 | 中国石油化工股份有限公司 | Smoothing optimization method of velocity model |
CN104463010A (en) * | 2014-10-31 | 2015-03-25 | 华为技术有限公司 | File lock implementation method and device |
-
2016
- 2016-03-11 CN CN201610141098.1A patent/CN107180153A/en active Pending
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20100161911A1 (en) * | 2007-05-31 | 2010-06-24 | Eric Li | Method and apparatus for mpi program optimization |
CN103076627A (en) * | 2011-10-26 | 2013-05-01 | 中国石油化工股份有限公司 | Smoothing optimization method of velocity model |
CN102902514A (en) * | 2012-09-07 | 2013-01-30 | 西安交通大学 | Large-scale parallel processing method of moving particle semi-implicit method |
CN104463010A (en) * | 2014-10-31 | 2015-03-25 | 华为技术有限公司 | File lock implementation method and device |
Non-Patent Citations (1)
Title |
---|
张文生等: "频率多尺度全波形速度反演", 《地球物理学报》 * |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109344135A (en) * | 2018-10-18 | 2019-02-15 | 中国海洋石油集团有限公司 | A kind of parallel seismic processing job scheduling method of the file lock of automatic load balancing |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
US11514045B2 (en) | Structured cluster execution for data streams | |
Liu et al. | Adaptive asynchronous federated learning in resource-constrained edge computing | |
EP3014444B1 (en) | Computing connected components in large graphs | |
CN106156159B (en) | A kind of table connection processing method, device and cloud computing system | |
US10209908B2 (en) | Optimization of in-memory data grid placement | |
EP3251034B1 (en) | Query optimization adaptive to system memory load for parallel database systems | |
US9152669B2 (en) | System and method for distributed SQL join processing in shared-nothing relational database clusters using stationary tables | |
US9479449B2 (en) | Workload partitioning among heterogeneous processing nodes | |
US9576026B2 (en) | System and method for distributed SQL join processing in shared-nothing relational database clusters using self directed data streams | |
US10127281B2 (en) | Dynamic hash table size estimation during database aggregation processing | |
CN102929989B (en) | The load-balancing method of a kind of geographical spatial data on cloud computing platform | |
CN110659278A (en) | Graph data distributed processing system based on CPU-GPU heterogeneous architecture | |
CN110222029A (en) | A kind of big data multidimensional analysis computational efficiency method for improving and system | |
CN105045871A (en) | Data aggregation query method and apparatus | |
US20160337442A1 (en) | Scheduled network communication for efficient re-partitioning of data | |
WO2017088666A1 (en) | Data storage method and coordinator node | |
CN107070645A (en) | Compare the method and system of the data of tables of data | |
CN104834709B (en) | A kind of parallel cosine mode method for digging based on load balancing | |
CN101800768A (en) | Gridding data transcription generation method based on storage alliance subset partition | |
CN111522811A (en) | Database processing method and device, storage medium and terminal | |
US10402385B1 (en) | Database live reindex | |
CN107180153A (en) | The method and system of full waveform inversion is realized using MPI | |
CN107203437A (en) | The methods, devices and systems for preventing internal storage data from losing | |
CN106302778A (en) | A kind of distributed flow process automotive engine system | |
CN104331336B (en) | Be matched with the multilayer nest balancing method of loads of high-performance computer structure |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20170919 |
|
RJ01 | Rejection of invention patent application after publication |