A data cross-process transmission method applied to CFD high-performance computing

Abstract

The invention discloses a data cross-process transmission method applied to CFD high-performance computing. First, a dynamic container composed of a plurality of basic fragments that are essentially character-type vector containers is sequentially linked, and the CFD numerical simulation is performed between different processes. The data that needs to be communicated is packaged into unified character data; then the MPI communication function is encapsulated in a dynamic container, and MPI communication is performed to transmit the data stored in the dynamic container. Among them, the upper limit of the standard length of the basic segment does not exceed the size of the MPI buffer, and it can be executed in segments during transmission. The present invention greatly improves the utilization rate of the buffer and reduces the number of data communications. At the same time, the upper limit of the standard length of the basic segment also effectively avoids buffer overflow. For large-scale data transmission, it can be executed through fragmentation , to get rid of the constraint of the buffer size and avoid the occurrence of communication congestion.

Description

technical field [0001] The invention belongs to the field of data cross-process transmission, and in particular relates to a data cross-process transmission method applied to CFD high-performance computing. Background technique [0002] Computational Fluid Dynamics (CFD for short) is an emerging interdisciplinary subject that uses computers and numerical methods to numerically simulate and analyze fluid dynamics problems. for the development of high-performance computers. At present, with the development of computer hardware and parallel computing technology, in order to meet the needs of large-scale numerical simulation in aerospace and other fields, the development of high-performance CFD software has attracted more and more attention. So far, parallel programming environments represented by MPI and OpenMP have been widely used in the development of high-performance CFD software. [0003] At present, there are already many parallel computing frameworks for grid discrete-...

AI Technical Summary

Problems solved by technology

Obviously, this method increases the total length of the communication data, which not only wastes the buffer, but also generates additional communication time

At the same time, on a given operating system, the contiguous memory that can be allocated by a single double-precision floating-point array is limited, and the size of the buffer is also limited, which means that the single array packaging technology has a limited carrying capacity for massive data. of

Although MPI provides users with a structured data packaging transmission interface, this communication method still results in a waste of cache and an increase in communication costs due to the constraints of the principle of memory alignment.

Images