Spatial image storage and calculation method and equipment based on spatial anti-affinity distribution and storage and calculation integration

Abstract

The invention discloses a spatial image storage and calculation method and equipment based on spatial anti-affinity distribution and storage and calculation integration, and belongs to the field of spatial image storage and calculation. Firstly, a storage and calculation integrated system architecture is constructed, the architecture adopts a two-layer structure combining a master control node and distributed child nodes, and the storage and calculation integrated system architecture is constructed; the master control node uniformly controls the parallel operation and data interaction of each distributed child node; thirdly, storing image data on each distributed sub-node, and synchronously deploying calculation software, so that each sub-node not only undertakes a data reading task, but also takes charge of a data processing task, thereby constructing and forming a storage and calculation integrated system cluster; and finally, based on the storage and calculation integrated system cluster, executing image file parallel processing and storage during data storage. During a tile request, executing direct parallel reading and nearby parallel computing of small data in the stored image file on each storage node; according to the invention, the space image slicing-free application efficiency is greatly improved.

Description

Technical Field

[0001] This invention relates to the field of spatial image storage and computing, and more specifically, to a spatial image storage and computing method and device based on spatial anti-affinity distribution and integrated storage and computing. Background Technology

[0002] Spatial imagery typically involves massive amounts of data, making direct web-based applications difficult. According to the current WMTS standard, it requires static or dynamic tile tiling for application. To avoid the problems of excessive tiling time, massive redundant storage of tiled data, and inability to implement dynamic applications caused by large-scale static tiling of spatial images, tile-free applications with dynamic real-time tile construction are gradually becoming a trend. Therefore, the efficient reading and processing of small image data patches becomes crucial for spatial imagery application technology.

[0003] Currently, the industry generally uses an architecture based on object-oriented distributed storage clusters and independent computing nodes to support image slicing-free applications, where storage and computing are separated, such as... Figure 1 As shown. For example, image cloud storage based on AWS S3 is currently the mainstream large-scale spatial image storage solution. Based on the object storage model, it uses files as storage objects. Each object is divided into blocks of a fixed size and then distributed for storage. When reading a file, the efficiency of reading the entire file is improved by reading each block in parallel. Generally, it requires expensive high-performance storage, network, and computing equipment to achieve high efficiency in slice-free applications. This solution has the following drawbacks in slice-free applications: 1) Adaptability issues: Distributed storage and parallel reading solutions are designed for efficient reading of the entire file, and are not suitable for applications that only need to read a small chunk of data within a file without slicing. For example, tiles are generally much smaller than blocks, so reading tile data cannot be made more efficient through parallel reading.

[0004] 2) Data retrieval efficiency issues: Because a single file is stored in blocks, and the spatial coverage of a block is much smaller than that of the file, the probability of a tile involving multiple blocks is much higher than the probability of it involving multiple files. For example, for a 256... A 256-pixel tile, with a size of L For an L-pixel image, the probability that a tile intersects with its edge is P = 1 - ((L - 256) / L). ^2 It can be seen that as the image range decreases (the L value decreases), the probability of the tile intersecting with the edge increases exponentially. That is, the probability of a tile involving multiple blocks increases significantly. However, the block division does not consider the spatial anti-affinity mechanism, which greatly increases the probability that data needs to be read from the same storage address multiple times to generate a tile, resulting in a decrease in reading efficiency by a percentage V=1 / (1+P).

[0005] 3) Computational efficiency issues: All the read data is handed over to the computing nodes for processing, which puts enormous pressure on the computing nodes under high concurrency.

[0006] 4) Cost issues: The above solutions can only improve the efficiency of slice-free applications by relying on expensive high-performance storage, computing and transmission equipment, which results in huge application costs. Moreover, the efficiency improvement is limited by the current level of related technology development. Summary of the Invention

[0007] The purpose of this invention is to overcome the shortcomings of the prior art and provide a spatial image storage and computing method and device based on spatial anti-affinity distribution and storage-computing integration. By taking into account the characteristics of spatial image slicing-free applications, it fully leverages the advantages of distributed storage and proximity parallel reading and parallel computing.

[0008] The objective of this invention is achieved through the following solution: A spatial image storage and computation method based on spatial anti-affinity distribution and integrated storage and computation includes the following steps: First, a system architecture integrating storage and computing is constructed. This architecture adopts a two-layer structure combining a master node and distributed child nodes. The master node uniformly controls the parallel operation and data interaction of each distributed child node. Next, image data is stored on each distributed child node, and computing software is deployed synchronously, so that each child node undertakes both data reading and data processing tasks, thereby building a system cluster that integrates storage and computing. Finally, based on this integrated storage and computing system cluster, image files are processed and stored in parallel during data storage; when a tile is requested, small blocks of data within the image files stored on each storage node are read directly in parallel and computed in parallel with the nearest available data.

[0009] Furthermore, when data is entered into the database, image files are processed and entered into the database in parallel. When a tile is requested, small blocks of data within the image files to be entered into the database on each storage node are read directly in parallel and computed in parallel with the nearest neighbor. This is specifically implemented based on an image distributed storage model with anti-affinity in storage address space. The storage model uses files as storage units and achieves anti-affinity distribution in storage address space by not distributing spatially adjacent images to the same storage address.

[0010] Furthermore, the master control node is only responsible for task assignment and calculation of the distributed storage address of the image file for spatial anti-affinity; the data processing and writing tasks are completed in parallel by each storage sub-node.

[0011] Furthermore, the two-layer structure combining a master control node and distributed child nodes is adopted, with the master control node uniformly controlling the parallel operations and data interactions of each distributed child node, specifically including the following sub-steps: After the master node software deploys sub-node software on interconnected storage servers as needed, the master node software performs image data database creation, image database and metadata database management, service construction and publishing, sub-node management and data communication management; the sub-node software performs operation command reception, data processing, local or remote data reading and writing and result return.

[0012] Furthermore, the direct parallel reading and nearest parallel computation of small blocks of data within the image files stored on each storage node specifically includes the following sub-steps: The master node calculates the file address based on the request and issues the task to one or more storage child nodes storing the file. Each storage child node performs direct reading and local computation processing of small blocks of data within the local file in parallel according to the master node's instructions. When a tile involves multiple files, based on spatial anti-affinity distribution, for clusters with more than four child nodes, each request performs only one parallel read and parallel computation processing. When it does not involve multiple files, the storage child node receiving the task independently completes one read and computation processing of the entire tile and returns it to the master node or directly to the requester.

[0013] Furthermore, the process of performing parallel image file processing and storage during data storage, and the direct parallel reading and nearest parallel computation of small blocks of data within the image files stored on each storage node during tile requests, specifically includes the following sub-steps: The master node sends the data processing task and storage address to the storage server based on the spatial range of the image file to be imported and the storage address that is not compatible with the computational space of the metadata database. After the child node completes the task, it notifies the master node. The master node receives application requests, performs spatial awareness based on the requested spatial range, and obtains the file storage addresses on one or more child nodes. If the request involves only a single file, it sends data reading and processing instructions to the corresponding child node. If the request involves multiple files, it simultaneously sends data processing, data storage, or data reading and processing instructions to the relevant child nodes. After receiving the instructions, the child nodes, based on the tile range, storage address, and internal index of the image file, bypass the operating system's file directory system, directly read and process small blocks of image data, and determine whether to return the data to the master node or directly to the requester based on the instruction content, the result, and the tile matching. The master node determines the next operation based on the returned result: if the returned result indicates that the tile has been sent to the request, it continues to send the next instruction; if the returned result indicates that multiple small image data blocks have been assembled, the tile is returned to the request, and then the next instruction is sent according to the request; if the returned result indicates that the data import is complete, the next data import task is sent.

[0014] Furthermore, based on the storage address and the internal index of the image file, the system bypasses the operating system's file directory structure and directly reads small blocks of image data from the local disk or the corresponding virtual disk in the cloud.

[0015] Furthermore, the process of directly reading and processing small image data blocks specifically includes: after directly reading the small image data blocks, performing projection calculations, coordinate transformations, constructing pyramids, converting image data formats, constructing internal indexes for image files, and processing data storage.

[0016] Furthermore, the computational processing instructions specifically include data resampling, projection calculation, coordinate transformation, and clipping processing instructions.

[0017] A spatial image storage and computing device based on spatial anti-affinity distribution and in-memory computing integration includes system equipment and software system. The system equipment consists of a main control unit and a storage unit based on an in-memory computing integration architecture. The main control unit stores the main control node software, and the storage unit stores both data and data read / write and processing software. The software system executes the spatial image storage and computing method based on spatial anti-affinity distribution and in-memory computing integration as described above.

[0018] The beneficial effects of this invention include: (1) This invention proposes a new storage-computing integrated system architecture. Specifically, image data is stored on distributed sub-nodes, while reading and processing software are deployed at the same time to build a system cluster integrating storage and computing. This fully utilizes the computing power of storage nodes to achieve direct parallel reading and local parallel computing of small blocks of data within image files. When a single tile does not involve multiple files, the processing result can be directly returned to the request, reducing data transmission and improving system efficiency.

[0019] (2) This invention proposes an image distributed storage model with anti-affinity storage address space, which improves the parallel reading efficiency of small blocks of data within a file. Specifically, using files as storage units and adhering to the principle that spatially adjacent images cannot be distributed to the same storage address, an anti-affinity distribution of storage address space is achieved. This makes it easier for a cluster of more than four child nodes to obtain all the data for constructing a tile in a single parallel read when a tile involves multiple storage units. This improves the problem of multiple parallel reads caused by the existing object storage using blocks as storage units, thus improving data reading efficiency.

[0020] (3) This invention proposes a spatial image storage and computing method based on spatial anti-affinity distribution and storage-computing integration. It also proposes a collaborative computing architecture that supports the practical application of system architecture and storage model, which can reduce the latency caused by a large amount of data transmission and improve the efficiency of the entire system's storage and computing. Attached Figure Description

[0021] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0022] Figure 1 This is a diagram of the current mainstream spatial image computing and storage system architecture; Figure 2 This is a diagram of the spatial image storage and computing architecture of the method in an embodiment of the present invention. Detailed Implementation

[0023] All features disclosed in all embodiments of this specification, or steps in all methods or processes implied in the disclosure, may be combined and / or extended or replaced in any way, except for mutually exclusive features and / or steps.

[0024] In view of the problems in the background, the inventors of this invention, after further creative thinking, believe that: Figure 1 The diagram shows the current mainstream spatial image computing and storage system architecture, which is a complex three-layer structure. During data database construction, the compute nodes are responsible for all data processing, including coordinate transformation, projection processing, pyramid construction, data scheduling, and bucket node management, resulting in low efficiency. Bucket nodes are responsible for data partitioning and distribution, typically using hashing or random distribution, without considering the spatial adjacency of files or blocks. Block nodes are responsible for data storage. During tile requests, the compute nodes handle remote data reading, transmission control, and data processing, lacking parallel computing support and experiencing immense pressure. Since the tile range is much smaller than the block range, the ability to read the entire file in parallel is largely wasted, making it unsuitable for image tiling-free applications. Even when multiple blocks are involved, the lack of spatial non-adjacency distribution support makes it difficult to achieve a single parallel read per request, resulting in low data reading efficiency. Significant data transmission latency further impacts the overall system efficiency.

[0025] Therefore, in a preferred embodiment, this invention specifically relates to the fields of spatial information technology, distributed computing, and data storage, proposing a spatial image storage and computing scheme based on spatial anti-affinity distribution and integrated storage and computing, particularly involving methods and devices related to large-scale spatial image data distributed storage and high-efficiency dynamic tile construction. Specifically, in the inventive concept, considering the characteristic of tile-free applications primarily involving the reading and processing of small blocks of data within file objects, an integrated storage and computing system architecture is first proposed to improve upon the shortcomings of existing cloud storage architectures. This architecture includes a master control node and distributed sub-nodes, with the master control node interacting with the distributed sub-nodes to form a two-layer architecture, which is simpler than the existing three-layer architecture. Furthermore, image data is stored on the distributed sub-nodes, while reading and processing software is deployed simultaneously, constructing a system cluster integrating storage and computing. This fully utilizes the computing power of the storage nodes to achieve direct parallel reading and local parallel computing of small blocks of data within image files. When a single tile does not involve multiple files, the processing result can be directly returned to the request, thereby reducing data transmission and improving system efficiency.

[0026] Furthermore, based on the aforementioned in-memory computing system architecture, another embodiment proposes an image distributed storage model with anti-affinity storage address space, aiming to improve the parallel reading efficiency of small blocks of data within a file. This model uses files as storage units and adheres to the principle that spatially adjacent images cannot be distributed to the same storage address, achieving anti-affinity distribution of storage address space. This facilitates the acquisition of all data constructing a tile in a single parallel read when a tile involves multiple storage units, thereby improving the problem of multiple parallel reads of small blocks of data within a file caused by existing object storage using block-based storage units, and enhancing data reading efficiency.

[0027] In a further inventive concept, based on the system architecture and storage model proposed above, other embodiments further propose a collaborative computing software architecture method to support more practical applications, which includes the following: The master node software deploys sub-node software on interconnected storage servers as needed, integrating the storage server into the system cluster to build an easily scalable two-tier architecture cluster that integrates storage and computing.

[0028] The master node software is used to perform tasks such as image data database construction, image database and metadata database management, service construction and application, sub-node management, and data communication management; the sub-node software is used to perform tasks related to the master node, such as receiving operation instructions, data processing, local or remote data reading and writing, and returning results.

[0029] When building the database, the master node sends the data processing task and storage address to the storage server (child node) based on the spatial range of the image files to be imported and the storage address that is not compatible with the metadata database computing space. After the child node completes the task, it notifies the master node.

[0030] When a service application is applied, the master node receives the application request, performs spatial awareness based on the requested spatial range, and obtains the relevant file storage addresses on one or more child nodes. If the request involves only a single file, it sends data reading and processing instructions to the corresponding child node. If the request involves multiple files, it simultaneously sends data processing, data storage, or data reading and processing instructions to the relevant child nodes. After receiving the instructions, the child nodes, based on the tile range, storage address, and internal index of the image file, bypass the operating system's file directory system, directly read the small block of image data, and perform corresponding processing. Based on the instruction content, the result, and the tile matching, they determine whether to return the data to the master node or directly to the requester.

[0031] The master node determines the next operation based on the returned result. For example, if the return message indicates that the tile has been sent to the request, it continues to send the next instruction. If the return message indicates that multiple small image data blocks have been sent, the tiles are assembled and returned to the request, and the next instruction is sent. If the return message indicates that the data import is complete, the next data import instruction and related data are sent.

[0032] Furthermore, the in-memory computing system architecture of the present invention is as follows: Figure 2 As shown, this is a two-layer architecture with a simple structure. During database construction, the master node is only responsible for task assignment and data distribution calculation, and files are distributed to take into account spatial adjacency. All data processing tasks, such as coordinate transformation, projection processing, and pyramid construction, are completed in parallel by each storage sub-node, without the need for file block processing, resulting in high efficiency. When a tile request is made, the master node calculates the file address based on the requested spatial range and issues the task to the storage sub-node storing that file. Each storage sub-node completes local reading and nearest-neighbor calculation processing according to the master node's instructions. This approach is well-suited for parallel reading and processing of small files in image tiling applications. When multiple files are involved, with the support of spatial non-adjacency distribution, each request only requires one parallel reading and parallel processing, resulting in high data reading and processing efficiency. When no multiple files are involved, the storage sub-node independently completes the entire tile generation and returns it directly to the request, reducing latency caused by large-scale data transmission and resulting in high overall system efficiency.

[0033] As another aspect of the present invention, a spatial image storage and computing device based on spatial anti-affinity distribution and in-memory computing is provided, comprising system equipment and software system. The system equipment specifically refers to a main control unit and a storage unit based on an in-memory computing architecture. The main control unit stores main control node software, and the storage unit stores both data and data read / write and processing software. The software system executes the method described in any of the preceding claims.

[0034] The technical effects of the present invention are verified in more detail as follows: First, an image database is built using a 1:2000 DOM, and then a tile-free image service is published according to the Tianditu standard, with a tile size of 256. The image pyramid is 256 pixels in size and follows a 4-level division rule, continuing until the last level, which is larger than the tile size. Table 1 shows the probability of needing to read multiple images for a single tile when generating Tianditu tiles of levels 14-19.

[0035] Table 1. Calculation of the probability of tile spanning multiple images for grades 15-19.

[0036] As can be seen from Table 1, more than 63% of the tiles at level 17 and above do not need to be transmitted to the master node, which can reduce a lot of data transmission time.

[0037] Secondly, using 185 2-meter resolution images of a certain province in 2025 (data volume 0.98T, covering an area of ​​approximately 500,000 square kilometers), a comparative test was conducted between a storage-compute cluster based on spatial anti-affinity distribution and a hyperconverged cloud storage based on object storage. The results are shown in Table 2.

[0038] Table 2 Comparison Results with Hyperconverged Object Storage Test

[0039] As can be seen from Table 2, the storage-computing cluster based on spatial non-adjacent distribution in the present invention achieves an efficiency improvement of 54%-118% at a lower cost compared to a certain hyperconverged cloud storage, and the advantage becomes more obvious as concurrency increases.

[0040] In summary, the present invention, by taking into account the characteristics of spatial image slicing-free applications, fully leverages the advantages of distributed storage and local parallel reading and parallel computing, significantly improving storage reading and computing efficiency.

[0041] The units described in the embodiments of the present invention can be implemented in software or hardware, and the described units can also be located in a processor. The names of these units do not necessarily limit the specific unit itself.

Claims

1. A spatial image storage and computation method based on spatial anti-affinity distribution and integrated storage and computation, characterized in that, Includes the following steps: First, a system architecture integrating storage and computing is constructed. This architecture adopts a two-layer structure combining a master control node and distributed child nodes. The master control node uniformly controls the parallel operation and data interaction of each distributed child node. Next, image data is stored on each distributed sub-node, and computing software is deployed synchronously, so that each sub-node undertakes both data reading and data processing tasks, thereby building a system cluster that integrates storage and computing. Finally, based on this integrated storage and computing system cluster, parallel processing of image files is performed during data import; when a tile is requested, direct parallel reading and local parallel computing of small blocks of data within the image files to be imported are performed on each storage node.

2. The spatial image storage and computation method based on spatial anti-affinity distribution and integrated storage and computation according to claim 1, characterized in that, The process involves parallel processing of image files during data storage; and direct parallel reading and local parallel computation of small blocks of data within the image files stored on each storage node during tile requests. This is specifically implemented based on a distributed image storage model with anti-affinity storage address space. The storage model uses files as storage units and achieves anti-affinity distribution of storage address space by ensuring that spatially adjacent images are not distributed to the same storage address.

3. The spatial image storage and computation method based on spatial anti-affinity distribution and integrated storage and computation according to claim 2, characterized in that, The master control node is only responsible for task assignment and calculation of distributed storage addresses for spatially anti-affinity image files; data processing and writing tasks are completed in parallel by each storage sub-node.

4. The spatial image storage and computation method based on spatial anti-affinity distribution and integrated storage and computation according to claim 2, characterized in that, The system employs a two-layer structure combining a master node and distributed child nodes. The master node centrally controls the parallel operations and data interactions of all distributed child nodes, specifically including the following sub-steps: After the master node software deploys sub-node software on interconnected storage servers as needed, the master node software performs image data database creation, image database and metadata database management, service construction and publishing, sub-node management and data communication management; the sub-node software performs operation command reception, data processing, local or remote data reading and writing and result return.

5. The spatial image storage and computation method based on spatial anti-affinity distribution and integrated storage and computation according to claim 2, characterized in that, The process of performing direct parallel reading and local parallel computation of small blocks of data within the image files stored on each storage node specifically includes the following sub-steps: The master node calculates the file address based on the request and issues the task to one or more storage child nodes storing the file; Each storage sub-node performs parallel reading and computation processing of small blocks of data within the local file according to the master node's instructions. When a tile involves multiple files, based on spatial anti-affinity distribution, for clusters with more than four sub-nodes, each request performs only one parallel reading and computation processing. When no multiple files are involved, the storage sub-node receiving the task independently completes one reading and computation processing of the entire tile and returns it to the master node or directly to the requester.

6. The spatial image storage and computation method based on spatial anti-affinity distribution and integrated storage and computation according to claim 2, characterized in that, The process of performing parallel image file processing and data import during data import, and performing direct parallel reading and local parallel computation of small blocks of data within the image files imported on each storage node during tile requests, specifically includes the following sub-steps: The master node sends the data processing task and storage address to the storage server based on the spatial range of the image file to be imported and the storage address that is not compatible with the computational space of the metadata database. After the child node completes the task, it notifies the master node. The master node receives application requests, performs spatial awareness based on the requested spatial range, and obtains the file storage addresses on one or more child nodes. If the request involves only a single file, it sends data reading and processing instructions to the corresponding child node. If the request involves multiple files, it simultaneously sends data processing, data storage, or data reading and processing instructions to the relevant child nodes. After receiving the instructions, the child nodes, based on the tile range, storage address, and internal index of the image file, bypass the operating system's file directory system, directly read and process small blocks of image data, and determine whether to return the data to the master node or directly to the requester based on the instruction content, the result, and the tile matching. The master node determines the next operation based on the returned result: if the returned tile has been sent to the request, it continues to send the next instruction; if the returned multiple small image data blocks, it assembles the tiles and returns them to the request, and then continues to send the next instruction according to the request; if the returned data entry is complete, it sends the next data entry task.

7. The spatial image storage and computation method based on spatial anti-affinity distribution and integrated storage and computation according to claim 6, characterized in that, The method bypasses the operating system's file directory system and directly reads small blocks of image data from the local disk or the corresponding virtual disk in the cloud, based on the storage address and the internal index of the image file.

8. The spatial image storage and computation method based on spatial anti-affinity distribution and integrated storage and computation according to claim 6, characterized in that, The process of directly reading and processing small image data blocks specifically includes: after directly reading the small image data blocks, performing projection calculations, coordinate transformations, constructing pyramids, converting image data formats, constructing internal indexes for image files, and processing data storage.

9. The spatial image storage and computation method based on spatial anti-affinity distribution and integrated storage and computation according to claim 6, characterized in that, The computational processing instructions specifically include data resampling, projection calculation, coordinate transformation, and clipping instructions.

10. A spatial image storage and computing device based on spatial anti-affinity distribution and integrated storage and computing, characterized in that, The system includes system equipment and software system. The system equipment consists of a main control unit and a storage unit based on an in-memory computing architecture. The main control unit stores the main control node software, and the storage unit stores both data and data read / write and processing software. The software system executes the spatial image storage and computing method based on spatial anti-affinity distribution and in-memory computing as described in any one of claims 1 to 9.

Images