2020Unpublished venueRequires access

Research on Parallel Method of Moments On Domestic Supercomputer Platform

Yan Chen, Zongjing Gu, Yanyan Li, Xuehui Shao, Peng Sun, Xiaomin Qiang, Baisheng Yang

Open publisher page 0 citations

Abstract

The parallel method of moment is firstly studied based on many-core supercomputer platform with a peak computing speed of more than 100PFlops, which is the first supercomputer of China built by all use independent technology. For the matrix solution process with intensive computation and high CPU utilization, the matrix multiplication operation with high concurrency degree is executed at the slave with strong computing power by using the xMath mathematical library of "Sunway TaiHu Light" system. The parallel efficiency of solving matrix is up to 90% using one-way communication LU algorithm named CALU, shows better scalability than open ScaLAPACK mathematical library. The numerical results show that CALU algorithm has a strong advantage in solving method of moment equations based on domestically-made supercomputer platform, which greatly speeds up the process of solving matrix.

About this research paper

What this paper is about

The parallel method of moment is firstly studied based on many-core supercomputer platform with a peak computing speed of more than 100PFlops, which is the first supercomputer of China built by all use independent technology. For the matrix solution process with intensive computation and high CPU utilization, the matrix multiplication operation with high concurrency degree is executed at the slave with strong computing power by using the xMath mathematical library of "Sunway TaiHu Light" system. The parallel efficiency of solving matrix is up to 90% using one-way communication LU algorithm named CALU, shows better scalability than open ScaLAPACK mathematical library. The numerical results show that CALU algorithm has a strong advantage in solving method of moment equations based on domestically-made supercomputer platform, which greatly speeds up the process of solving matrix.

Why it matters

A significance statement is not available in the OpenAlex record.

Key contribution

A contribution statement is not available in the OpenAlex record.

Method / approach

Method details are not available in the OpenAlex metadata.

Main findings

Findings are not separately available in the OpenAlex metadata.

Limitations

Limitations are not available in the OpenAlex metadata.

Applications

Application details are not available in the OpenAlex metadata.

Available abstract

The parallel method of moment is firstly studied based on many-core supercomputer platform with a peak computing speed of more than 100PFlops, which is the first supercomputer of China built by all use independent technology. For the matrix solution process with intensive computation and high CPU utilization, the matrix multiplication operation with high concurrency degree is executed at the slave with strong computing power by using the xMath mathematical library of "Sunway TaiHu Light" system. The parallel efficiency of solving matrix is up to 90% using one-way communication LU algorithm named CALU, shows better scalability than open ScaLAPACK mathematical library. The numerical results show that CALU algorithm has a strong advantage in solving method of moment equations based on domestically-made supercomputer platform, which greatly speeds up the process of solving matrix.

Key concepts: Supercomputer, Computer science, Parallel computing, Scalability, Matrix multiplication, Moment (physics), Multiplication (music), Computation

Related papers

Back to paper searchBrowse research topicsOriginal source
Research on Parallel Method of Moments On Domestic Supercomputer Platform — Research Paper | ScholarLens