EA043837B1 - METHOD, DEVICE AND ELECTRONIC EQUIPMENT FOR DATA PROCESSING - Google Patents

Description

Field of technology

The present invention relates to the technical field of data processing and, in particular, to a method, apparatus and electronic equipment for data processing.

State of the art

Ocean circulation is the study of wind-induced sea currents and density currents resulting from uneven density distributions, the formation and distribution of cyclones in ocean circulation, the intensification of westerly ocean circulation, bends and changes in sea currents, the structure of current systems in the near-equatorial region, Antarctic circumpolar currents, oceanic thermohaline circulation, deep-sea circulation and its relationship with the main thermocline, connections between dispersive and convergent movements of sea water and uplifting currents and Langmuir circulation, etc., mesoscale eddies and their energy conversion, special flow phenomena such as drift ice, ocean response to wind load, etc., as well as circulation in coastal zones.

An ocean circulation model is a numerical model that quantitatively describes ocean circulation and its changes, reflecting the complex processes of ocean dynamics, physics, and interactions between them by establishing a system of mathematical physics equations. It discretizes the continuous movement of oceanic fluid, while the earth's space is divided to form a meridional, zonal and vertical three-dimensional grid structure, i.e. three-dimensional grid structure in the meridional, zonal and vertical directions, and then solve a set of differential equations using numerical integration.

As research into the ocean and climate change deepens, the ocean circulation model is gradually moving to higher resolution (the higher the resolution, the finer the grid), more physical processes (the number of equations increases in a set of equations), and faster computational speed. Currently, the ocean circulation model is studied using a secondary parallel structure of meridional-zonal horizontal partitioning and meridional (or zonal or vertical) cyclic parallel computation.

However, the horizontal resolution of the global ocean circulation model has been increased from 5° (about 500 km) to 0.1° (about 10 km), and each step in increasing its resolution will lead to a geometric increase in the need for computational resources and computational volume. Generally speaking, if the horizontal resolution increases by a factor of 10, then the computational volume increases by a factor of 1000, and as the number of physical processes increases, the computational volume also increases, so the above minor parallel structure limits the computational efficiency of the longitude-latitude-vertical ocean circulation model .

The essence of the invention

It is an object of the present invention to provide a method, apparatus and electronic equipment for data processing, which can realize the separation of the longitude-latitude-vertical ocean circulation model in three-dimensional space and improve the computational efficiency of the longitude-latitudinal-vertical ocean circulation model.

The first aspect of the invention relates to a method for processing data, the method being applied to an ocean circulation model scenario, which may include:

dividing the Earth into geographical areas in the meridional, zonal and vertical directions in accordance with a given spatial resolution, and obtaining a three-dimensional grid structure corresponding to the specified earth space and grid information of the corresponding structure; said grid information includes the geographic area to which the grid belongs and regional data within said geographic area;

dividing the grids in said grid structure in accordance with the configured number of computing nodes, on a grid plane corresponding to the meridional and zonal directions in said grid structure, and obtaining a subgrid structure after division corresponding to each computing node and regional data related to the corresponding structure;

relative to each computing node, based on a predetermined partitioning strategy, partitioning grids in said subgrid structure in a first direction or said vertical direction, and obtaining a secondary subgrid structure after the partition and area data related to the corresponding structure; and distributing regional data corresponding to said secondary subgrid structure to a main computing unit of said computing node; said first direction is said zonal direction or said vertical direction; said secondary subgrid structure includes a secondary subgrid structure corresponding to said main computing unit and at least one secondary subgrid structure corresponding to a computing subunit of said computing node;

based on said predetermined allocation strategy, controlling said main computing unit to allocate area data corresponding to the secondary secondary

- 1 043837 subgrid structure, on at least one specified computing subunit so that the specified main computing unit and at least one specified computing subunit, based on operational requirements, perform the calculation of operational data from the distributed area data; wherein said secondary secondary sub-grid structure is obtained by dividing grids in a secondary sub-grid structure corresponding to at least one said computing sub-unit, said main computing unit, on a grid plane corresponding to said second direction and said vertical direction, wherein said second direction is said zonal direction or the specified meridional direction.

In one particular case of implementation of the method, the main computing unit of the specified computing node is controlled to distribute the regional data of the secondary secondary subgrid structure to at least one computing subunit of the specified computing node, including:

relative to each computing sub-unit, controlling the main computing unit of said computing node to partition the grids in said secondary secondary sub-grid structure in a third direction according to the number of target processing units from said computing sub-units, obtaining a processed grid structure corresponding to each target processing unit and area data related to the corresponding structure, and distributing area data corresponding to the specified processed grid structure to the specified target processing unit so that the specified target processing unit performs calculation of the area data corresponding to the specified processed grid structure; said target processing unit is a processing unit with a vectorized instruction set function corresponding to the data type of said area data.

In a possible particular case of implementing the method, the main computing unit of the specified computing node exchanges data with the main computing unit in other computing nodes other than the specified computing node using a configured information transfer interface, wherein the specified method additionally includes:

transmitting regional data of the target grid transmitted by the main computing unit in other computing nodes to the main computing unit of the specified computing node; said target grid is a grid in which there is an overlap of geographic areas with a grid corresponding to area data that is distributed to said computing node;

distribution of regional data corresponding to the specified secondary subgrid structure to the main computing unit of the specified computing node, including:

distributing the region data of said secondary subgrid structure and the region data of the target grid transmitted by the main computing unit of said other computing node to the main computing unit of said computing node.

In a particular case of implementation of the method, the grids in the specified grid structure are divided in accordance with the configured number of computing nodes, on a grid plane corresponding to the meridional and zonal directions in the specified grid structure, and obtaining a subgrid structure after division corresponding to each computing node, and regional data related to the appropriate structure, including:

defining computed meshes in said grid structure based on the computed requirements;

evenly dividing the computational grids in said grid structure in accordance with the configured number of computational nodes, on a grid plane corresponding to the meridional and zonal directions in the specified grid structure, and obtaining a set of computational grids corresponding to each specified computational node;

arbitrarily dividing non-calculated grids in the specified grid structure, on a grid plane corresponding to the meridional and zonal directions, and obtaining a set of grids corresponding to the specified non-calculating grids;

based on the set of computed grids corresponding to each specified computational node, and the set of grids corresponding to the specified non-computable grids, determining the subgrid structure after partitioning corresponding to each specified computational node, and the area data of the corresponding structure.

In one possible implementation, the grid information further includes an area type of the geographic area to which the grid belongs, wherein the area type includes an ocean type;

If the specified operational computational requirements are the computational requirements of the target variable in the ocean circulation model, then determine the computational grids in the specified grid structure based on the computational requirements, including:

obtaining a region type corresponding to each grid in the specified grid structure;

defining grids whose specified area type is the specified ocean type as a computed grid.

- 2 043837

In a possible implementation of the method, data exchange between the specified main computing unit and the specified at least one computing subunit is carried out using direct memory access.

In the possible case of implementing the method based on the specified specified distribution strategy, the specified main computing unit is controlled to distribute the area data corresponding to the minor secondary subgrid structure to at least one specified computing subunit, including:

based on a predetermined two-dimensional separation strategy of the second direction and said vertical direction, controlling said main computing unit to separate grids in a secondary subgrid structure corresponding to at least one said computing subunit on a grid plane corresponding to said second direction and said vertical direction, and obtaining said minor secondary sub-grid structure and area data related to the corresponding structure, for distributing area data corresponding to the minor secondary sub-grid structure after being divided by said main computing unit to at least one said computing sub-unit.

A second aspect of the invention discloses a data processing apparatus, which may include:

an area division unit, which is used to divide the Earth into geographical areas in the meridional, zonal and vertical directions in accordance with a specified spatial resolution, and obtain a three-dimensional grid structure corresponding to the specified earth space and grid information of the corresponding structure; said grid information includes the geographic area to which the grid belongs and regional data within said geographic area;

a grid splitting unit, which is used to split grids according to the configured number of computing nodes, on a grid plane corresponding to the meridional and zonal directions in the specified grid structure, and obtain a subgrid structure corresponding to each computing node after the division, and regional data related to the corresponding structure;

and, with respect to each computing node, based on a given partitioning strategy, partitioning the grids in said subgrid structure in a first direction or in said vertical direction, and obtaining a secondary subgrid structure after the partition corresponding to the main computing unit in said computing node, and area data related to to the appropriate structure;

a data distribution unit, which is used to distribute area data corresponding to the specified secondary subgrid structure to the main computing unit of the corresponding computing node; said first direction is said zonal direction or said vertical direction; said secondary subgrid structure includes a secondary subgrid structure corresponding to said main computing unit and a secondary subgrid structure corresponding to at least one computing subunit of said computing node;

a control unit which, based on said predetermined distribution strategy, is used to control said main computing device to distribute area data corresponding to the minor secondary sub-grid structure to at least one said computing sub-unit so that said main computing unit and at least one the specified computing subunit, based on operational requirements, performed the calculation of operational data from the distributed area data; wherein said secondary secondary sub-grid structure is obtained by dividing grids in a secondary sub-grid structure corresponding to at least one said computing sub-unit, said main computing unit, on a grid plane corresponding to said second direction and said vertical direction, wherein said second direction is said zonal direction or the specified meridional direction.

In a particular case of the implementation of the device, the specified control unit, in particular with respect to each computing subunit, is used to control the main computing unit of the specified computing node to divide the grids in the specified secondary secondary subgrid structure in the third direction in accordance with the number of target processing blocks from the specified computing subblocks, obtaining a processed grid structure corresponding to each processing target block, and area data related to the corresponding structure, and distributing the area data corresponding to the specified processing grid structure to the specified processing target unit so that the specified processing target unit performs the calculation of the area data corresponding to the specified processed mesh structure; said target processing unit is a processing unit with a vectorized instruction set function corresponding to the data type of said area data.

- 3 043837

In a possible implementation of the device, the main computing unit of the specified computing node communicates with the main computing unit in other computing nodes other than the specified computing node using a configured information transfer interface, wherein the specified device further includes: a transmission unit;

the specified transmission unit is used to transmit the regional data of the target grid transmitted by the main computing unit in other computing nodes to the main computing unit of the specified computing node; said target grid is a grid in which there is an overlap of geographic areas with a grid corresponding to area data that is distributed to said computing node;

said data distribution unit is used, in particular, for distributing the area data of said secondary subgrid structure and the area data of the target grid transmitted by the main computing unit in said other computing node to the main computing unit of the corresponding computing node.

In one possible embodiment, said grid division unit is particularly used for:

defining computed meshes in said grid structure based on the computed requirements;

evenly dividing the calculated grids in the specified grid structure in accordance with the configured number of computing nodes, on a grid plane corresponding to the meridional and zonal directions in the specified grid structure, and obtaining a set of calculated grids corresponding to each specified computing node;

arbitrarily dividing non-calculated grids in the specified grid structure, on a grid plane corresponding to the meridional and zonal directions, and obtaining a set of grids corresponding to the specified non-calculating grids;

based on a set of computational grids corresponding to each specified computational node and a set of grids corresponding to the specified non-computational grids, a definition of the subgrid structure after partitioning corresponding to each specified computational node, and area data of the corresponding structure.

In one possible implementation, the grid information further includes an area type of the geographic area to which the grid belongs, wherein the area type includes an ocean type;

If the specified operational computational requirements are the computational requirements of the target variable in the ocean circulation model, then the specified grid division block is, in particular, additionally applied to:

obtaining a region type corresponding to each grid in said grid structure;

defines grids whose specified area type is the specified ocean type as a computed grid.

In one possible implementation, communication between said main computing unit and said at least one computing subunit is accomplished using direct memory access.

In one possible implementation, said control unit is additionally used in particular for:

based on a predetermined strategy of two-dimensional separation of the second direction and the specified vertical direction, controlling the specified main computing unit to separate the grids in a secondary subgrid structure corresponding to at least one specified computing subunit on a grid plane corresponding to the specified second direction and the specified vertical direction, and obtaining said minor secondary sub-grid structure and area data related to the corresponding structure, for distributing area data corresponding to the minor secondary sub-grid structure after being divided by said main computing unit to at least one said computing sub-unit.

A third aspect of the invention is an electronic device that includes a processor, a communication interface, a memory and a communication bus, wherein communication between the processor, the communication interface and the memory is effected through a communication bus;

memory is used to store computer programs;

the processor is used to implement the steps of any method in the first aspect while executing a program stored in the memory.

In a fourth aspect, there is provided a computer-readable storage medium that contains a computer program, wherein said computer program implements the steps of any method in the first aspect described above when executed by a processor.

In the data processing method provided by an embodiment of the present invention, firstly, three-level parallel earth space computation is achieved by dividing the three-dimensional grid structure corresponding to the earth space into three layers, and further achieves four-level parallel earth space computation by dividing the fourth layer in each computing layer. subblock, which increases the efficiency of calculating the longitude-width-vertical ocean circulation model.

Description of drawings

To more clearly explain the technical solutions of the embodiments of the present invention, the following is a brief description of the necessary accompanying drawings used in the embodiments of the present invention. It should be understood that the following accompanying drawings show only some embodiments of the present invention and therefore should not be construed as a limitation on scope, and that other related drawings may also be derived from these accompanying drawings without any creative work for those skilled in the art.

Fig. 1 is a schematic diagram of the structure of a new generation Shenwei multi-core processor provided by an embodiment of the present invention.

Fig. 2 is a schematic drawing of a parallel processing procedure in an ocean circulation model provided by an embodiment of the present invention.

Fig. 3 is a schematic flow diagram of a data processing method provided by an embodiment of the present invention.

Fig. 4 is a schematic drawing of the results of strong scalability tests provided by an embodiment of the present invention.

Fig. 5 is a schematic drawing of the results of weak scalability tests provided by an embodiment of the present invention.

Fig. 6 is a schematic diagram of the structure of a data processing apparatus provided by an embodiment of the present invention.

Fig. 7 is a schematic diagram of the structure of electronic equipment provided by an embodiment of the present invention.

Specific methods for carrying out the invention

The technical solutions in the embodiments of the present invention are clearly and fully described below in conjunction with the accompanying drawings of the embodiments of the present invention. It will be appreciated that the described embodiments are only a portion of the embodiments of the present invention and not all of them. Based on the embodiments of the present invention, all other embodiments obtained without creative work by those skilled in the art fall within the scope of protection of the present invention.

To facilitate understanding, an explanation of the terms used in the embodiments of the invention is provided below.

The new generation Shenwei multi-core processor (an upgraded version of the SW26010 model) integrates 6 Core Groups (CGs) (or compute nodes), which are connected via a Loop network, with each CG compute node containing one synchronous dynamic memory with random access and double data rate (Double Data Rate) with a memory interface, one computing control core (Management Processing Element, MPE) (or main computing unit) and one computing core 8 * 8 (Computing Processing Element, CPE) ( or computing subunit) which are connected via a ring network, as shown in FIG. 1. This processor has a 512-bit instruction set in the form of a Single Instruction Multiple Data (SIMD) instruction that can compute 8 double precision numbers simultaneously, so this processor has at least one processing unit with a vectorized instruction set function.

In this case, the main computing unit MPE has an independent operating system, which is mainly used for scheduling, distribution, control, communication, 10 and calculation, and the computing subunit CPE is mainly used for computing. The CPE computing subblocks in the CG computing node are located in the form of an 8*8 array. The CPE compute units interact with each other and with the main MPE compute unit, connecting through a network within the array.

Data exchange between the main computing units of the MPE is carried out using the Massage Passing Interface (MPI) method; Data exchange between any two CPE compute subunits in an array of CPE compute subunits can be carried out using one of such communication methods as: remote memory access (RMA), P2P and aggregated communication. Data exchange between the main computing unit MPE and the computing subunit CPE is carried out via Direct Memory Access (DMA).

The data processing method provided by embodiments of the present invention can be applied to an ocean circulation model learning scenario, and in particular can be applied to the above-mentioned Shenwei next-generation multi-core processor to implement ocean circulation model learning. The data processing method can obtain real-time three-dimensional ocean values such as ocean temperature, salinity, dynamic sea surface height, ocean currents and sea ice density, providing computational support for

- 5 043837 studying the ocean circulation model.

It should be noted that the data processing method provided by embodiments of the present invention may also be applied to other processors where multiple computing units are present, which is not limited to the present invention.

As shown in FIG. 2, in the data processing method provided by the embodiment of the present invention, in combination with the characteristics of the ocean circulation model, the first-level parallel calculation of the ocean circulation model is mainly developed - based on the meridional-zonal horizontal spatial partition by the main computing unit, the second-level parallel calculation of the circulation model of the ocean - based on the meridional-zonal horizontal spatial partition by the main computing unit and the computational subunit, and parallel calculations of the third level of the ocean circulation model - based on the meridional or zonal-vertical horizontal spatial partition by the computational subblocks; in the future, parallel calculations of the fourth level can be included - on based on the meridional or zonal spatial partitioning of a set of vectorized instructions, which makes it possible to implement a three-dimensional spatial partition of the ocean circulation model along longitude-latitude-vertical, and to increase the efficiency of calculating the ocean circulation model along longitude-latitude-vertical.

Where longitude refers to the meridional direction based on earth space, for example: x-axis direction; latitude refers to the zonal direction based on the earth's space, for example: y-axis direction; and vertical refers to a vertical direction based on earth space, such as the z-axis direction, and n is the number of x-axis directions.

Preferred embodiments of the present invention are described below in conjunction with the accompanying descriptive drawings, it being understood that the preferred embodiments described herein are intended to describe and explain the invention only and are not intended to limit the invention, and that the embodiments and features in the embodiments of the present invention can be combined with each other if they do not contradict each other.

Fig. 3 is a schematic flow diagram of a data processing method provided in one embodiment of the present invention. As shown in FIG. 3, this method may include the following.

Step S310: Divide the Earth into geographic areas in the meridional, zonal and vertical directions in accordance with a predetermined spatial resolution, obtaining the corresponding three-dimensional grid structure of the Earth and the corresponding grid structure information.

The specified spatial resolution determines the size of the divided grid; the higher the resolution, the finer the grid, i.e. the smaller the mesh size.

In a particular implementation, in combination with FIG. 2, the Earth is divided into geographical areas according to a given spatial resolution in the meridional, zonal and vertical directions, obtaining a three-dimensional grid structure corresponding to the Earth. Moreover, each geographic area is a grid in this grid structure.

Because a geographic area has a geographic location, the corresponding grid also has the same geographic location, and the corresponding grid also has area data that exists within that geographic area.

For example, the target geographic area is a maritime area belonging to country A on a plane corresponding to the meridional and zonal directions, the area being an area surrounded by geographic location point a, geographic location point b, geographic location point c, and geographic location point d , defined in the meridional and zonal directions, i.e. the dimensions of the grid corresponding to this target geographical area, on a plane corresponding to the meridional and zonal directions, represent the sea area belonging to country A, in this case this grid contains regional data of this ocean area, such as ocean temperature, salinity, dynamic sea surface height , sea currents, sea ice concentration, etc.

However, if the research requirement of an ocean circulation model is to study ocean temperature, the ocean temperature data from the regional data is the operational data contained in this grid.

For example, take the ocean circulation model as an example to study ocean temperature, in this case the operational data is ocean temperature. Given constant size of the target geographic area, ocean temperature gradually decreases as ocean depth (i.e., vertical altitude) increases, so the grid corresponding to the target geographic area includes ocean temperatures corresponding to different vertical altitudes.

That is, the grid information of each grid includes the geographic area to which the grid belongs and operational data within the corresponding geographic area.

Step S320: divide the grids into a grid structure according to the configured number of computing nodes, on a grid plane corresponding to meridional and zonal

- 6 043837 directions in the grid structure, obtaining the subgrid structure after division corresponding to each computing node, and the area data related to the corresponding structure.

In combination with fig. 2, since the grid scales along the x and y axes are much larger than those along the z axis, to reduce the computational volume of this grid structure, you can first separate the grids in the grid structure in accordance with the meridional-zonal plane of the grid, i.e. separate meshes in a grid structure along the x and y axes, i.e. achieve the decomposition of regional data from different geographical areas, and distribute the obtained by dividing the grid into different main computing units, i.e. to different MPI processes, allowing different main computational units to perform parallel computation of distributed domain data of different grids.

In a specific implementation, based on the obtained three-dimensional grid structure, the grids in the grid structure are divided in accordance with the configured number of computing nodes, on the corresponding grid plane corresponding to the meridional and zonal directions, i.e. by dividing, a sub-grid structure with the same configured number as the computing nodes is obtained, thereby obtaining area data corresponding to each sub-grid structure, and distributing each sub-grid structure obtained by dividing and area data related to the corresponding structure to each computing node.

In the following, since the area data of some grids in the resulting 3D grid structure may not include operational data, these grids may be referred to as invalid grids, i.e. grids that are not relevant to the operational computational requirements, so in order to obtain accurate data of the target variables under the operational computational requirements, it is necessary to first define the computational grids in the grid structure that are relevant to the target variables, in particular the following.

If the operational computational requirements are the computational requirements for the target variable in the ocean circulation model, then a domain type corresponding to each grid in the grid structure is obtained; where area type is the regional type of the geographic area to which the grid belongs. The regional type of geographic area may include a sea area, a land area, a plain area, a mountain forest area, and so on. The mesh with marine type as the area type is then defined as a calculated mesh.

In some embodiments, to achieve load balancing on the compute nodes and increase computation speed, once the compute meshes are defined, the compute meshes may be divided equally into the meshes according to the configured number of compute nodes, on a mesh plane corresponding to the meridional and zonal directions in the mesh structure, yielding a set of computed grids corresponding to each computational node.

At that time, non-calculated grids (ie, invalid grids) in the grid structure are arbitrarily divided on a grid plane corresponding to the meridional and zonal directions in said grid structure, obtaining a set of grids corresponding to the non-calculated grids.

Based on the set of computed meshes corresponding to each computational node and the set of meshes corresponding to the non-computable meshes, the subgrid structure after division corresponding to each computational node and the area data of the corresponding structure are determined.

In the above-described embodiment, it is only possible to perform domain data computation on computed grids rather than performing domain data computation on invalid grids, so that the computational capacity of each computation node is determined only by the number of computed grids, so that the computed grids are equally divided according to the number of computation nodes , thus the number of distributed computational grids per computing node is balanced, thus achieving load balancing of the computing nodes.

It should be noted that in the above implementation method, it only needs to consider whether the number of distributed computed grids on each computing node is balanced, and does not need to consider the number of distributed invalid grids on it, for example, all invalid grids can be distributed to one entire computing node or to several computing nodes randomly, i.e. the method of distribution is not limited.

Step S330: With respect to each computing node, based on a predetermined division strategy, the grids in the subgrid structure are divided in the first direction, obtaining a secondary subgrid structure after division and area data related to the corresponding structure; and distributing area data corresponding to the secondary subgrid structure to this main computing unit.

Where the first direction is the zonal or vertical direction.

In a specific implementation, to achieve load balancing of the main computing unit and at least one computing subunit in each computing node, a partitioning strategy may be set according to the configuration of the number of computing units in the computing node, for example, a partitioning strategy according to the number of computing units.

Separate the meshes in the subgrid structure in the zonal direction or vertical direction based on the specified division strategy, obtaining a secondary subgrid structure after division

- 7 043837 and regional data related to the relevant structure; and distributing the area data corresponding to the secondary subgrid structure to the main computing unit. The secondary subgrid structure includes a secondary subgrid structure corresponding to the main computing unit and a secondary subgrid structure corresponding to at least one computing subunit of the computing node.

In some embodiments, a main compute unit at a different compute node than the current compute node transmits the target grid region data to the main compute unit at the given compute node in a data transfer mode via an MPI communication interface. Where the target grid is a grid in which there is an overlap of the geographic area with the grid corresponding to the distributed area data per computing node, i.e. in a given grid corresponding to a computing node, there is an overlap of a geographic area with the target grid, and this overlap may be in the form of a partial overlap of two geographic areas.

Thus, after receiving the secondary subgrid structure after division and the area data related to the corresponding structure, the area data of the secondary subgrid structure and the area data of the target grid transmitted by the main computing unit of another computing node are distributed to the main computing unit of that computing node.

In the above embodiment, when the main computing unit is responsible for exchanging data between main computing units in different computing nodes, its communication with another main computing unit is carried out in MPI mode.

Step S340: Based on the predetermined distribution strategy, the main computing unit is controlled to distribute the area data corresponding to the minor secondary subgrid structure to at least one computing subunit.

Where the second direction is the zonal or meridional direction. In terms of a given distribution strategy, based on the memory capacity of the empirical value device or the computing subunit, it is possible to set a two-dimensional division strategy of the second direction with the vertical direction, i.e. a strategy for two-dimensional separation of the zonal direction or the meridional direction with the vertical direction; or a division strategy according to the number of computing units, which may be particularly set according to the actual situation, and the present invention is not limited to this.

Taking the two-dimensional separation strategy of the meridional and vertical directions as an example, this two-dimensional separation strategy refers to the strategy that keeps the meridional direction grid structure unchanged, and separates the grids into the corresponding grid structure, zonal and vertical directions according to the specified two-dimensional value (e.g. two-dimensional 3x4 value) on a grid plane corresponding to the zonal and vertical directions.

In a specific implementation, taking the predetermined distribution strategy as an example of the predetermined two-dimensional division strategy in the second direction and the vertical direction, based on the predetermined two-dimensional division strategy corresponding to the second direction and the vertical direction, the main computing unit in the computing node for dividing the grids in the secondary subgrid structure is controlled , corresponding to at least one computational sub-unit, on a grid plane corresponding to the second direction (i.e., zonal or meridional) and the vertical direction, obtaining the secondary secondary sub-grid structure and area data related to the corresponding structure, and after the main computation unit distributes the area of the data corresponding to the secondary sub-grid structure after being divided into at least one computing sub-unit, the main computing unit and the at least one computing sub-unit perform calculation of operational data from the distributed area data based on operational requirements.

Subsequently, since the data is calculated in each computing sub-unit in a parallel data vectorization manner, with respect to each computing sub-unit, the main computing unit in the computing node is controlled to divide the grids in the secondary sub-grid structure in the third direction, according to the number of target processing blocks from the computing sub-blocks, obtaining a processed a grid structure corresponding to each target processing unit and area data related to the corresponding structure, and distributing area data corresponding to the grid structure being processed to the target processing unit so that the target processing unit performs calculation of the area data corresponding to the grid structure being processed.

This target processing block is a processing block with a vectorized instruction set function corresponding to the area data type.

In the above implementation, since the compute subunits cannot communicate between compute nodes, when the main compute unit is responsible for the fine-grained grid division of the compute subunits, its data communication with the compute units

- 8 043837 is carried out using DMA; Data communication between computing subunits in the same computing node is carried out using RMA. Therefore, in combination with the previous step, data transmission in an embodiment of the present invention can be performed using the DMA+MPI+RMA method.

In one example, after obtaining a three-dimensional grid structure of 752*502*65 (longitude-latitude-vertical) corresponding to the Earth and grid information of the corresponding structure, parallel data processing in an embodiment of the present invention can be carried out using four-level partitioning, in particular the following.

Level 1. Let the new generation Shenwei multi-core processor have 128 computing nodes, divide 128 computing nodes into a column of a 16*8 two-dimensional matrix corresponding to a 752*502 grid plane corresponding to the meridional and zonal directions, and divide the grids in the grid structure, obtaining approximately 49*65 grids (including the number of boundary elements) for which each computational node is responsible, on 752*502 grid planes corresponding to the meridional and zonal directions, so a subgrid structure consisting of 49*65*65 can actually be distributed to each group of cores grids, in which case data transfer between computing nodes is performed using the MPI communication mechanism.

Level 2. Since the computing node includes a main computing unit and several computing subunits, the grids in the 49*65*65 subgrid structure can be divided in the vertical direction in a 1+64 vertical mode according to the number of computing units in the computing node, obtaining a secondary subgrid structure 49*65*1, for which the main computing unit is responsible, and regional data related to the corresponding structure, as well as a subgrid structure 49*65*64, for which several computing subblocks are responsible, and regional data related to the corresponding structure .

Or, they divide the grids in a 49*65*65 subgrid structure in the zonal direction in 1+64 latitude mode in accordance with the number of computing units in the computing node, obtaining a secondary 49*1*65 subgrid structure, for which the main computing unit is responsible, and regional data related to the corresponding structure, as well as a 49*64*65 subgrid structure, for which several computational subblocks are responsible, and regional data related to the corresponding structure.

After the division, all regional data corresponding to the secondary subgrid structures obtained by division are distributed to the main computing unit to ensure that the main computing unit distributes the grid data to the computing subblocks for which it is responsible.

Level 3: Based on the memory capacity of the computing subunit, a 16*2 two-dimensional partition strategy is set in the second direction and the vertical direction. The method for implementing the third level is described in detail using the example that the subgrid structure, for which several computational subblocks obtained from the second level are responsible, is 49*65*64, and the second direction is the zonal direction:

The grids are divided into a secondary subgrid structure of 49*65*64, on a grid plane corresponding to the zonal and vertical directions, in accordance with the two-dimensional division strategy of 16*2 in the zonal direction and the vertical direction, obtaining a secondary secondary subgrid structure of 49*32*4 (or 49* 33*4) and regional data related to the corresponding structure.

Level 4. Separate the grids in the secondary secondary subgrid structure 49*32*4 (or 49*33*4) in the meridional direction in accordance with the number of target processing grids from the computational subblocks, obtaining a processed grid structure corresponding to each target processing block, and regional data related to the relevant structure. For example, if the number of processing target grids is 49, then the grid structure corresponding to each processing target block is 1*32*4 (or 1*33*4), which further increases the possibility of parallel data computation.

In the data processing method provided by an embodiment of the present invention, firstly, three-level parallel earth space calculation is achieved by three-level division of the three-dimensional grid structure corresponding to the earth space, and further, four-level parallel earth space calculation is achieved by three-level division in each computing sub-unit, which improves the efficiency calculations of the ocean circulation model in longitude-width-vertical.

Subsequently, as a result of applying the present invention to the NEMO ocean circulation model, a new version of the swNEMO new generation Shenwei multi-core processor was developed, and NEMO four-level parallelization was implemented on a supercomputer consisting of Shenwei new generation multi-core processors. Then, by selecting benchmarks called GYRE-PISCES, swNEMO performance testing is carried out.

The modeled area GYRE-PISCES is a closed rectangular pool

- 9 043837 on the beta plane, which is 3180 km long, 2120 km wide and 4 km deep, with a center of 30° N and a rotation of 45°. This standard represents the ideal North Atlantic or North Pacific basin. In addition, considering the influence of the third level grid structure in the global mode on the periodic conditions of the eastern and western oceans and the Arctic folding of the global ocean, the BENCH option is activated in the benchmarks to include these periodic conditions, and the coupling mode of the third level grid structure for the global ocean is reproduced. In terms of computational costs and computational properties, it is equivalent to a global ocean with a third-level grid structure and the same number of grids. Below, the horizontal resolution is the horizontal resolution equivalent to the global ocean after resampling.

3 experiments were designed with horizontal resolutions of 2 km, 1 km and 500 m to perform the strong scalability analysis. Moreover, each experiment consisted of 8 different parallel ratios. Speedup is the division of clock times of various ratios by the base entry of the smallest ratio with 2,129,920 compute nodes. Experiments with 8 resolutions are designed to analyze weak scalability. All experiments were carried out during 1 day of simulation without I/O.

Analysis of tests for strong scalability: experiments were carried out at resolutions of 2 km, 1 km and 500 m, increasing the number of computing nodes (Number of cores) from 2,129,9200 to 27,988,480. In the ocean circulation model, the resolution uniquely corresponds to the number of grids, with resolutions of 2 km, 1 km and 500 m corresponding to such numbers of grids on the global horizontal plane (latitude-longitude) as: 24002x16002, 43502x29002 and 82502x55002, respectively, and the number of vertical layer grids of 128 is the usual specification of an arithmetic example, so the number of grids corresponding to the 2 km, 1 km and 500 m resolutions is 24002x16002x128, 43502x29002x128 and 82502x55002x128 respectively. Setting 2,129,920 compute nodes as the strong scaling benchmark, the final parallel efficiency is 74.18%, 83.40%, and 99.29%, respectively, as shown in FIG. 4. It can be seen that all three resolutions have very good scalability, and the hyperscale speedup remains close to linear, in the case of testing at the highest resolution of 500m, the parallel efficiency can still reach up to 99.29% using 27,988,480 compute nodes.

Weak scalability test analysis: Weak scalability tests are performed at resolutions of 9 km, 7 km, 4.5 km, 3.5 km, 2.5 km, 2.0 km, 1.2 km and 1.0 km.

In weak scalability analysis: theoretically, the floating point efficiency of a single compute node remains the same on average when processing the same number of grids within a node. Thus, when analyzing weak scalability, it is necessary to take into account trends in two aspects: 1. changes in the total number of grids as the computing node progresses; 2. change in overall computing efficiency over the course of a computing node.

As shown in FIG. 5, the abscissa represents the number of computing nodes, and the ordinate represents the incremental dependence of the total number of grids on the computing node, and from FIG. Figure 5 shows that the change in the total number of grids (computed grid points number) depending on the number of computational nodes (Number of cores) has an almost linear trend, i.e. the number of meshes processed by a single node (i.e., the slope of the line graph) is almost constant, and the percentage values in Fig. 5 represents the overall computational efficiency at this scale, and the computational efficiency values in FIG. 5 represent that the overall computational efficiency under different computational scales is also almost linear, taking 9 km as the test benchmark, the theoretical constant computational efficiency under weak scaling is 2.03/1000, as shown in Fig. 5, the computational efficiency at a resolution increased to 1 km is 1.99/1000, which is 0.04/1000 lower than the theoretical value, while the reduction is less than 2%. The results show that the performance is stable across different resolutions, and the computational efficiency at 1 km resolution remains at 1.99/1000, with a trend close to the baseline indicating that swNEMO has excellent weak scalability, as shown in Fig. 5.

According to the method described above, an embodiment of the present invention also provides a data processing apparatus as shown in FIG. 6, which includes: an area division unit 610, a grid division unit 620, a data distribution unit 630, and a control unit 640.

The area division unit 610 is used to divide the Earth into geographic areas in the meridional, zonal and vertical directions in accordance with a specified spatial resolution, and obtain a three-dimensional grid structure corresponding to the specified earth space and grid information of the corresponding structure; said grid information includes the geographic area to which the grid belongs and regional data within said geographic area.

The mesh separator 620 is used to separate the meshes in the specified mesh structure into

- 10 043837 according to the configured number of computing nodes, on a grid plane corresponding to the meridional and zonal directions in the specified grid structure, and obtaining a subgrid structure after division corresponding to each computing node, and regional data related to the corresponding structure.

And, with respect to each computing node, based on a given partitioning strategy, partitioning the meshes in said subgrid structure in a first direction or said vertical direction, and obtaining the subgrid structure after partitioning and area data related to the corresponding structure.

The data distribution unit 630 is used to distribute the area data corresponding to the specified secondary subgrid structure to the main computing unit of the corresponding computing node; said first direction is said zonal direction or said vertical direction; said secondary subgrid structure includes a secondary subgrid structure corresponding to said main computing unit and a secondary subgrid structure corresponding to at least one computing subunit of said computing node.

The control unit 640 is applied, based on said predetermined distribution strategy, to control said main computing device to distribute area data corresponding to the minor secondary subgrid structure to at least one specified computing subunit so that said main computing unit and at least one specified the computing subunit, based on operational requirements, performed the calculation of operational data from the distributed area data; wherein said secondary secondary sub-grid structure is obtained by dividing grids in a secondary sub-grid structure corresponding to at least one said computing sub-unit, said main computing unit, on a grid plane corresponding to said second direction and said vertical direction, wherein said second direction is said zonal direction or the specified meridional direction.

In one possible implementation, control unit 640 is specifically applied with respect to each computing subunit to control the main computing unit of said computing node to partition the grids in said secondary secondary subgrid structure in a third direction in accordance with the number of target processing units from said computing subunits, obtaining a processed grid structure corresponding to each processing target unit, and area data related to the corresponding structure, and distributing area data corresponding to the specified processing grid structure to the specified processing target unit so that the specified processing target unit performs calculation of the area data corresponding to the specified processing grid structure; said target processing unit is a processing unit with a vectorized instruction set function corresponding to the data type of said area data.

In one possible implementation, the main computing unit of the specified computing node communicates with the main computing unit in other computing nodes other than the specified computing node using a configured communication interface;

said device further includes: a transmission unit 650;

the transmission unit 650 is used to transmit the target grid area data transmitted by the main computing unit in other computing nodes to the main computing unit of the specified computing node; said target grid is a grid in which there is an overlap of geographic areas with a grid corresponding to area data that is distributed to said computing node;

The data distribution unit 630 is used, in particular, to distribute the area data of the specified secondary subgrid structure and the area data of the target grid transmitted by the main computing unit in the specified other computing node to the main computing unit of the corresponding computing node.

In one possible implementation, grid partitioning unit 620 is specifically used for the following.

Defines computed meshes in a specified grid structure based on computed requirements.

Equally dividing the computational grids in the specified grid structure, on a grid plane corresponding to the meridional and zonal directions in the specified grid structure, in accordance with the configured number of computational nodes, and obtaining a set of computational grids corresponding to each specified computational node.

Arbitrarily dividing non-calculated grids in the specified grid structure, on a grid plane corresponding to the meridional and zonal directions, and obtaining a set of grids corresponding to the specified non-calculating grids.

Based on a set of computational meshes corresponding to each specified computational

- 11 043837 node, and a set of grids corresponding to the specified non-computable grids, a definition of the subgrid structure after division corresponding to each specified computational node, and the area data of the corresponding structure.

In one possible implementation, the grid information further includes an area type of the geographic area to which the grid belongs, wherein the area type includes an ocean type;

If the specified operational computational requirements are the computational requirements of a target variable in an ocean circulation model, then the grid partitioning block 620 is specifically further applied to:

obtaining a region type corresponding to each grid in said grid structure;

defines grids whose specified area type is the specified ocean type as a computed grid.

In one possible implementation, communication between said main computing unit and said at least one computing subunit is accomplished using direct memory access.

In one possible implementation, control unit 640 is specifically further used to:

based on a predetermined strategy of two-dimensional separation of the second direction and the specified vertical direction, controlling the specified main computing unit to separate the grids in a secondary subgrid structure corresponding to at least one specified computing subunit on a grid plane corresponding to the specified second direction and the specified vertical direction, and obtaining said minor secondary sub-grid structure and area data related to the corresponding structure, for distributing the area data corresponding to the minor secondary sub-grid structure after being divided by said main computing unit to at least one said computing sub-unit.

The functions of each functional block of the data processing device provided by the above embodiment of the present invention can be achieved through the steps of each method described above, and therefore the specific working process and beneficial effects of each block in the data processing device provided by the embodiment of the present invention will not be repeated here. .

An embodiment of the present invention also provides an electronic device as shown in FIG. 7, including a processor 710, a communications interface 720, a memory 730, and a communications bus 740, wherein communication between the processor 710, communications interface 720, and memory 730 occurs via communications bus 740.

The memory 730 is adopted for storing computer programs.

Processor 710 is adopted to perform the following steps when executing a program stored in memory 730:

dividing the Earth into geographical areas in the meridional, zonal and vertical directions in accordance with a given spatial resolution, and obtaining a three-dimensional grid structure corresponding to the specified earth space and grid structure information; said grid information includes the geographic area to which the grid belongs and regional data within said geographic area;

dividing the grids in said grid structure in accordance with the configured number of computing nodes, on a grid plane corresponding to the meridional and zonal directions in the specified grid structure, and obtaining a subgrid structure after division corresponding to each computing node, and regional data related to the corresponding structure;

relative to each computing node, based on a predetermined partitioning strategy, partitioning grids in said subgrid structure in a first direction or said vertical direction, and obtaining a secondary subgrid structure after the partition and area data related to the corresponding structure; and distributing regional data corresponding to said secondary subgrid structure to a main computing unit of said computing node; said first direction is said zonal direction or said vertical direction; said secondary subgrid structure includes a secondary subgrid structure corresponding to said main computing unit and at least one secondary subgrid structure corresponding to a computing subunit of said computing node;

based on said predetermined allocation strategy, controlling said main computing unit to distribute the area data corresponding to the minor secondary subgrid structure to at least one said computing subunit so that said main computing unit and at least one said computing subunit, based on operational requirements, performed calculations of operational data from distributed regional data; wherein said secondary secondary subgrid structure is obtained

- 12 043837 by dividing the grids in a secondary sub-grid structure corresponding to at least one specified computing sub-unit by said main computing block on a grid plane corresponding to said second direction and said vertical direction, wherein said second direction is said zonal direction or said meridional direction.

In one possible implementation, the main computing unit of said computing node is controlled to distribute regional data of a secondary secondary subgrid structure to at least one computing subunit of said computing node, including, with respect to each computing subunit, control of the main computing unit of said computing node to partition grids in said secondary secondary sub-grid structure in the third direction in accordance with the number of target processing units in the specified computing sub-unit, obtaining a processed grid structure corresponding to each target processing unit and area data related to the corresponding structure, and distributing area data corresponding to said processing grid structure to said a target processing unit for causing said target processing unit to perform calculation of area data corresponding to said grid structure to be processed; said target processing unit is a processing unit with a vectorized instruction set function corresponding to the data type of said area data.

In one possible implementation, the main computing unit of the specified computing node communicates with the main computing unit in other computing nodes other than the specified computing node using a configured communication interface, wherein the method further includes transmitting regional data of the target grid transmitted to the main computing block in other computing nodes, to the main computing block of the specified computing node; said target grid is a grid in which there is an overlap of geographic areas with a grid corresponding to area data that is distributed to said computing node;

distribution of regional data corresponding to the specified secondary subgrid structure to the main computing unit of the specified computing node, including:

distributing the region data of said secondary subgrid structure and the region data of the target grid transmitted by the main computing unit of said other computing node to the main computing unit of said computing node.

In one possible implementation, the grid within said grid structure is divided according to a configured number of compute nodes on a grid plane corresponding to the meridional and zonal directions in said grid structure, resulting in a post-divided subgrid structure corresponding to each compute node and area data associated with the corresponding structure, including:

defining computed meshes in said grid structure based on the computed requirements;

evenly dividing the computational grids in said grid structure in accordance with the configured number of computational nodes, on a grid plane corresponding to the meridional and zonal directions in the specified grid structure, and obtaining a set of computational grids corresponding to each specified computational node;

arbitrarily dividing non-calculated grids in the specified grid structure, on a grid plane corresponding to the meridional and zonal directions, and obtaining a set of grids corresponding to the specified non-calculating grids;

based on the set of computed grids corresponding to each specified computational node, and the set of grids corresponding to the specified non-computable grids, determining the subgrid structure after partitioning corresponding to each specified computational node, and the area data of the corresponding structure.

In one possible implementation, the grid information further includes an area type of the geographic area to which the grid belongs, wherein the area type includes an ocean type;

If the specified operational computational requirements are the computational requirements of the target variable in the ocean circulation model, then determine the computational grids in the specified grid structure based on the computational requirements, including:

obtaining a region type corresponding to each grid in the specified grid structure;

defining grids whose specified area type is the specified ocean type as a computed grid.

In one possible implementation, communication between said main computing unit and said at least one computing subunit is accomplished using direct memory access.

In one possible implementation, based on said specified allocation strategy, said main computing unit is controlled to allocate area data corresponding to

- 13 043837 minor secondary subgrid structure, for at least one specified computing subblock, including:

based on a predetermined two-dimensional separation strategy of the second direction and said vertical direction, controlling said main computing unit to separate grids in a secondary subgrid structure corresponding to at least one said computing subunit on a grid plane corresponding to said second direction and said vertical direction, and obtaining said minor secondary sub-grid structure and area data related to the corresponding structure, for distributing area data corresponding to the minor secondary sub-grid structure after being divided by said main computing unit to at least one said computing sub-unit.

The above communication bus may be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus or the like. This communication bus may be of various types, such as an address bus, a data bus, a control bus, and the like. For ease of presentation in the drawings, it is represented by only one thick line, but this does not mean that there is only one piece of tire or one type of tire.

A communication interface is used for communication between the above-mentioned electronic device and another device.

The memory may include random access memory (RAM), as well as non-volatile memory (NVM), such as at least one disk memory. Alternatively, the memory may also be at least one storage device located at a distance from the aforementioned processor.

The above-mentioned processor may be a general purpose processor including a Central Processing Unit (CPU), a Network Processor (NP), etc.; may also be a Digital Signal Processing (DSP), Application Specific Integrated Circuit (ASIC), Field-Programmable Gate Array (FPGA), or other programmable logic device, discrete gate or transistor logical device, discrete hardware node.

Since implementation methods for solving the problem of each element of the electronic device in the above embodiment of the invention and beneficial effects can be achieved by referring to the steps of the embodiment shown in FIG. 3, the specific operating process and beneficial effects of the electronic device provided by the embodiment of the present invention are not repeated here.

In another embodiment of the present invention, there is also provided a computer-readable storage medium, wherein instructions stored on the computer-readable storage medium that, when operated on a computer, cause the computer to perform any of the data processing methods of the above embodiment.

In another embodiment of the present invention, there is also provided a computer program product including instructions that, when operated on a computer, cause the computer to perform any of the data processing methods of the above embodiment.

Those skilled in the art will understand that an embodiment of the embodiments of the present invention may be in the form of a method, system, or computer program product. Thus, an embodiment of the present invention may be taken in the form of a full hardware embodiment, a full software embodiment, or an embodiment combining software and hardware aspects. Moreover, an embodiment of the present invention may be taken in the form of a computer-usable software product implemented on one or more computer storage media (including, but not limited to, disk memory, CD-ROM, optical memory, etc.), containing computer-used program code.

The computer program instructions in an embodiment of the invention are described in accordance with a process diagram and/or a flow diagram of a method, apparatus (system) and computer program product in an embodiment of embodiments of the present invention. It should be understood that each process and/or block from the process diagram and/or flowchart and/or a combination of processes and/or blocks from the flowchart and/or flowchart can be implemented using computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, a special purpose computer, an embedded processor, or other programmable data processing device to create a machine such that the instructions executed by the processor of the computer or other programmable data processing device create a device to implement the designated functions of one or several processes from a process diagram and/or one or more blocks from a flow diagram.

These computer program instructions may also be stored in a computer readable

-

Claims (9)

memory capable of putting a computer or other programmable data processing device into operation in a specified mode such that instructions stored in the computer-readable memory create an article with a command device, wherein the command device implements the function of one or more processes, a process block and/or one or more blocks from the flowchart. These computer program instructions may also be loaded onto a computer or other programmable processing device such that a series of operational steps are performed on the computer or other programmable device to create computer implementable processing such that the instructions executed on the computer or other programmable device provide instructions for implementing the steps of an assigned function of one or more processes from a process diagram and/or one or more blocks from a flow diagram. Although preferred embodiments of the present invention have been described, further changes and modifications to these embodiments may be made by those skilled in the art once the basic inventive concepts become known. Therefore, the appended claims are intended to be construed as including the preferred embodiments of the invention and all changes and modifications that come within the scope of the embodiments of the present invention. It will be appreciated that those skilled in the art may make various changes and variations to an embodiment of the embodiments of the present invention without departing from the spirit and scope of the embodiments in the embodiments of the present invention. Thus, if these changes and variations with respect to an embodiment of the embodiments of the present invention come within the scope of the claims of the embodiments of the present invention and their technical equivalents, it is intended that the embodiments of the present invention also include these changes and variations. CLAIM 1. A method of processing data to obtain three-dimensional ocean values in real time, characterized in that the said method includes dividing the Earth into geographical areas in the meridional, zonal and vertical directions in accordance with a given spatial resolution and obtaining a three-dimensional grid structure corresponding to the specified earth space, and information about the corresponding grid structure; said grid information includes the geographic area to which the grid belongs and regional data within said geographic area, wherein the data from the regional data is operational data within the corresponding geographic area; dividing grids in said grid structure, on a grid plane corresponding to the meridional and zonal directions in said grid structure, in accordance with the configured number of computing nodes, and obtaining a subgrid structure after division corresponding to each computing node, and area data related to the corresponding structure; with respect to each computing node, based on a predetermined partitioning strategy, performing partitioning of grids in said subgrid structure in a first direction or said vertical direction and obtaining a secondary subgrid structure after partitioning and area data related to the corresponding structure; and distributing regional data corresponding to said secondary subgrid structure to a main computing unit of said computing node; said first direction is said zonal direction or said vertical direction; said secondary subgrid structure includes a secondary subgrid structure corresponding to said main computing unit and a secondary subgrid structure corresponding to at least one computing subunit of said computing node; based on said predetermined allocation strategy, controlling said main computing unit to distribute the area data corresponding to the minor secondary subgrid structure to at least one said computing subunit so that said main computing unit and at least one said computing subunit, based on operational requirements, performed calculations of operational data from distributed regional data; wherein said secondary secondary subgrid structure is obtained by dividing grids in a secondary subgrid structure corresponding to at least one said computing subunit by said main computing unit on a grid plane corresponding to said second direction and said vertical direction, wherein said second direction is said zonal direction or the indicated meridional direction. 2. The method according to claim 1, characterized in that the main computing unit is controlled - 15 043837 of a specified computing node for distributing regional data of a secondary secondary subgrid structure to at least one computing subunit of the specified computing node, including: relative to each computing sub-unit, controlling the main computing unit of said computing node to partition the grids in said secondary secondary sub-grid structure in a third direction according to the number of target processing units in said computing sub-unit, obtaining a processed grid structure corresponding to each target processing unit and area data related to the corresponding structure, and distributing area data corresponding to the specified processed grid structure to the specified target processing unit so that the specified target processing unit performs calculation of the area data corresponding to the specified processed grid structure; said target processing unit is a processing unit with a vectorized instruction set function corresponding to the data type of said area data. 3. The method according to claim 1, characterized in that the main computing unit of the specified computing node communicates with the main computing unit in other computing nodes other than the specified computing node using a configured communication interface, wherein the method further includes: transmitting regional data of the target grid transmitted by the main computing unit in other computing nodes to the main computing unit of the specified computing node; said target grid is a grid in which there is an overlap of geographic areas with a grid corresponding to area data that is distributed to said computing node; distributing area data corresponding to said subgrid structure to a main computing unit of said computing node, including distributing area data of said subgrid structure and area data of a target grid transmitted by the main computing unit of said other computing node to the main computing unit of said computing node. 4. The method according to claim 1, characterized in that the grids in the specified grid structure are divided in accordance with the configured number of computing nodes, on a grid plane corresponding to the meridional and zonal directions in the specified grid structure, obtaining a subgrid structure after the division corresponding to each computing node , and area data related to the corresponding structure, including the definition of calculated grids in the specified grid structure based on the calculated requirements; evenly dividing the computational grids in said grid structure in accordance with the configured number of computational nodes, on a grid plane corresponding to the meridional and zonal directions in the specified grid structure, and obtaining a set of computational grids corresponding to each specified computational node; arbitrarily dividing non-calculated grids in the specified grid structure, on a grid plane corresponding to the meridional and zonal directions, and obtaining a set of grids corresponding to the specified non-calculating grids; based on the set of computed grids corresponding to each specified computational node, and the set of grids corresponding to the specified non-computable grids, determining the subgrid structure after partitioning corresponding to each specified computational node, and the area data of the corresponding structure. 5. The method according to claim 4, characterized in that the grid information further includes an area type of the geographic area to which the specified grid belongs, and the specified area type includes an ocean type; if said operational computational requirements are the computational requirements of a target variable in an ocean circulation model, then determining computational grids in said grid structure based on the computational requirements, including obtaining a domain type corresponding to each grid in said grid structure; defining grids whose specified area type is the specified ocean type as a computed grid. 6. The method according to claim 1, characterized in that data exchange between said main computing unit and said at least one computing subunit is carried out using direct memory access. 7. The method according to claim 1 or 2, characterized in that, on the basis of the specified specified distribution strategy, the specified main computing unit is controlled to distribute the area data corresponding to the minor secondary subgrid structure to at least one specified computing subunit, including, based on the specified strategies for two-dimensional separation of the second direction and the specified vertical direction, controlling the specified main computing unit for separating the grids in a secondary subgrid structure corresponding to at least one specified computing subunit, on a grid plane corresponding - 16 043837 to the specified second direction and the specified vertical direction, and obtaining the specified minor secondary subgrid structure and the area data related to the corresponding structure, for distributing the area data corresponding to the minor secondary subgrid structure after dividing, the specified main computing unit, into at least one specified computing subunit. 8. A device for processing data to obtain three-dimensional ocean values in real time, characterized in that the device includes a regional division unit, which is configured to be used to divide the Earth into geographical areas in the meridional, zonal and vertical directions in accordance with a given spatial resolution, and obtaining a three-dimensional grid structure corresponding to the specified earth space, and grid information of the corresponding structure; said grid information includes the geographic area to which the grid belongs and regional data within said geographic area, wherein the data from the regional data is operational data within the corresponding geographic area; a grid division unit, which is configured to be used for dividing the grids in the specified structure in accordance with the configured number of computing nodes, on a grid plane corresponding to the meridional and zonal directions in the specified grid structure, and obtaining a subgrid structure corresponding to each computing node after the division, and regional data related to the relevant structure; and, with respect to each computing node, based on a predetermined partitioning strategy, dividing the grids in said subgrid structure in a first direction or in said vertical direction and obtaining a secondary subgrid structure after the partition corresponding to the main computing unit in said computing node, and area data related to appropriate structure; a data distribution unit, which is configured to be used for distributing regional data corresponding to the specified secondary subgrid structure to the main computing unit of the corresponding computing node; said first direction is said zonal direction or said vertical direction; said secondary subgrid structure includes a secondary subgrid structure corresponding to said main computing unit and a secondary subgrid structure corresponding to at least one computing subunit of said computing node; a control unit which, based on said predetermined distribution strategy, is configured to be used to control said main computing device to distribute area data corresponding to the minor secondary sub-grid structure to at least one said computing sub-unit so that said main computing unit and at least one specified computing subunit, based on operational requirements, performed the calculation of operational data from the distributed area data; wherein said secondary secondary sub-grid structure is obtained by dividing grids in a secondary sub-grid structure corresponding to at least one said computing sub-unit, said main computing unit, on a grid plane corresponding to said second direction and said vertical direction, wherein said second direction is said zonal direction or the specified meridional direction. 9. An electronic device, characterized in that said electronic device includes a processor, a communication interface, a memory and a communication bus, wherein communication between the processor, the communication interface and the memory is effected through the communication bus; memory provides the ability to use it to store computer programs; the processor is configured to be used to implement any step of the method according to any one of claims 1 to 7 when executing a program stored in memory. -