Parallel computing of building fire using a domain decomposition method based on load balancing. (November 2022)

• Record Type: Journal Article
• Authors: Wei, Zheng; Xin, Hailin; Yang, Peizhong
• Is Part Of: Advances in engineering software. Volume 173(2022)
• Publication Date: 2022-11
• Subjects: Parallel computing -- Building fire -- Dimensionless index -- Domain decomposition -- Load balancing -- Numerical simulation; Computer-aided engineering -- Periodicals; Engineering -- Computer programs -- Periodicals; Engineering -- Software -- Periodicals; Periodicals; 620.0028553
• DOI: 10.1016/j.advengsoft.2022.103201
• Languages: English
• ISSNs: 0965-9978

Abstract

Highlights: The domain decomposition method based on load balancing was proposed. Four dimensionless indices were presented for the evaluation of the load balance. A load description function was established based on the four indices. Parallel simulations were performed on an express logistics center fire. The DDMBLB method has a better performance than the RCB method. Abstract: Parallel computing can provide a practical way in performing building fire simulation which often contains an immense amount of grids. Four dimensionless indices were presented for the evaluation of the load balance. According to the four indices, an empirical load description function was established based on the orthogonal experimental method. A domain decomposition method based on load balancing (DDMBLB) was then proposed. A preprocessing tool based on the DDMBLB method was developed, which can decompose the domain automatically with load balancing. Parallel numerical simulations based on the DDMBLB method were performed on an express logistics center fire. The comparison of DDMBLB method and the commonly-used Recursive Coordinate Dichotomy (RCB) method was done to analyze the computing time, speedup ratio and parallel efficiency. The DDMBLB method can significantly reduce the fire simulation time compared to the RCB method. It has a better domain decomposition effect than the RCB method with a greater speedup ration and a higher parallel efficiency.