The acquisition frequency of aeronautical digital images represented by drone aerial data is getting faster and faster, and the amount of data is getting larger and larger. The traditional data processing method based on serial computing has been difficult to meet high-efficiency production demand and mass. The data is updated quickly, so parallel computing techniques must be used to improve the efficiency of data processing.
The so-called parallel processing technique is a technique that increases the number of operations in the same time interval. Computers designed for parallel processing are collectively referred to as parallel computers; solving problems on parallel computers is called parallel computing; algorithms for solving problems on parallel computers are called parallel algorithms.
Strictly speaking, parallelism has two meanings: one is simultaneity, which means that two or more events occur at the same time; the other is concurrency, which means that two or more events occur at the same time interval.
Currently popular high-performance computing systems can be roughly divided into two categories: one is a shared memory system (SMP), which is characterized by multiple processors having physically shared memory; the other is distributed storage systems (DMP). For example, MPP and cluster system, the system is composed of multiple physically distributed nodes, each node has its own memory, and the nodes are connected by high-speed Ethernet or dedicated high-speed network.
The cluster system uses a high-speed universal network to connect a set of high-performance workstations or high-end PCs according to a certain structure. Under the support of parallel programming and visual human-computer interaction integrated development environment, unified scheduling, coordinated processing, and efficient parallel processing. system. Hosts and networks in a clustered system can be isomorphic or heterogeneous. Compared with traditional parallel processing systems, cluster systems have the following obvious features:
(1) The system development cycle is short. Most of the cluster systems use commercial workstations and general-purpose LAN networks, which greatly saves development time.
(2) The system price is low. Workstations or high-end PCs that make up the cluster are mass-produced and therefore priced at a lower price.
(3) User investment risk is small. A cluster system is not only a parallel processing system, but each node is also a separate workstation. Even if the entire system is not efficient parallel to certain application problems, its nodes can still be used as a single workstation.
(4) The system has good scalability. Cluster systems mostly use a common network, and system expansion is easy.
(5) Saving system resources. The structure of the cluster system is flexible, and the workstations with different architectures and different performances can be connected together, so that the existing equipment can be fully utilized.
(6) User programming is convenient. In a cluster system, parallelization of programs is implemented in a familiar programming environment C, C++, or Fortran.
First, parallel computing technology
The so-called parallel processing technique is a technique that increases the number of operations in the same time interval. Computers designed for parallel processing are collectively referred to as parallel computers; solving problems on parallel computers is called parallel computing; algorithms for solving problems on parallel computers are called parallel algorithms.
Strictly speaking, parallelism has two meanings: one is simultaneity, which means that two or more events occur at the same time; the other is concurrency, which means that two or more events occur at the same time interval.
Currently popular high-performance computing systems can be roughly divided into two categories: one is a shared memory system (SMP), which is characterized by multiple processors having physically shared memory; the other is distributed storage systems (DMP). For example, MPP and cluster system, the system is composed of multiple physically distributed nodes, each node has its own memory, and the nodes are connected by high-speed Ethernet or dedicated high-speed network.
Second, the cluster system
The cluster system uses a high-speed universal network to connect a set of high-performance workstations or high-end PCs according to a certain structure. Under the support of parallel programming and visual human-computer interaction integrated development environment, unified scheduling, coordinated processing, and efficient parallel processing. system. Hosts and networks in a clustered system can be isomorphic or heterogeneous. Compared with traditional parallel processing systems, cluster systems have the following obvious features:
(1) The system development cycle is short. Most of the cluster systems use commercial workstations and general-purpose LAN networks, which greatly saves development time.
(2) The system price is low. Workstations or high-end PCs that make up the cluster are mass-produced and therefore priced at a lower price.
(3) User investment risk is small. A cluster system is not only a parallel processing system, but each node is also a separate workstation. Even if the entire system is not efficient parallel to certain application problems, its nodes can still be used as a single workstation.
(4) The system has good scalability. Cluster systems mostly use a common network, and system expansion is easy.
(5) Saving system resources. The structure of the cluster system is flexible, and the workstations with different architectures and different performances can be connected together, so that the existing equipment can be fully utilized.
(6) User programming is convenient. In a cluster system, parallelization of programs is implemented in a familiar programming environment C, C++, or Fortran.
Shenzhen Geyi Technology Co., Ltd. , https://www.gy-printer.cn