ECMC 2021 Li

From Werner KRAUTH

(Difference between revisions)
Jump to: navigation, search
Revision as of 11:55, 10 May 2021
Werner (Talk | contribs)

← Previous diff
Current revision
Werner (Talk | contribs)

Line 1: Line 1:
==Progress in parallelization of ECMC in 2D hard-sphere system== ==Progress in parallelization of ECMC in 2D hard-sphere system==
-Botao Li+'''Botao Li'''
-LPENS, Ecole normale supérieure, Paris (France)+'''''LPENS, Ecole normale supérieure, Paris (France)'''''
We present a multithreaded version of the event-chain Monte Carlo (ECMC) for 2D hard-sphere systems (arxiv:2004.11040, DOI: 10.1016/j.cpc.2020.107702). The current implementation is based on the straight event-chain (SEC) algorithm. In the multithreaded ECMC, multiple spheres move simultaneously on different threads. The correctness of the multithreaded algorithm is guaranteed by comparing the algorithm with its singlethreaded version, which satisfies the global-balance condition explicitly. When ignoring synchronization, the multithreaded ECMC processes 10^12 collisions per hour on a x86 CPU with 20 physical cores. However, the correctness of this algorithm requires rather frequent synchronization. This prevents the multithreaded ECMC from reaching its full potential. Several attempts has been made to fix this problem. We present a multithreaded version of the event-chain Monte Carlo (ECMC) for 2D hard-sphere systems (arxiv:2004.11040, DOI: 10.1016/j.cpc.2020.107702). The current implementation is based on the straight event-chain (SEC) algorithm. In the multithreaded ECMC, multiple spheres move simultaneously on different threads. The correctness of the multithreaded algorithm is guaranteed by comparing the algorithm with its singlethreaded version, which satisfies the global-balance condition explicitly. When ignoring synchronization, the multithreaded ECMC processes 10^12 collisions per hour on a x86 CPU with 20 physical cores. However, the correctness of this algorithm requires rather frequent synchronization. This prevents the multithreaded ECMC from reaching its full potential. Several attempts has been made to fix this problem.
 +----
-'''Slides'''+'''Slides''' [http://www.lps.ens.fr/%7Ekrauth/images/c/cb/ECMC_2021_Li.pdf here]
'''Recording''' '''Recording'''
'''Further material''' '''Further material'''
 +
 +----
[[Workshop_ECMC_11_May_2021|back to 2021 ECMC workshop]] [[Workshop_ECMC_11_May_2021|back to 2021 ECMC workshop]]

Current revision

Progress in parallelization of ECMC in 2D hard-sphere system

Botao Li

LPENS, Ecole normale supérieure, Paris (France)

We present a multithreaded version of the event-chain Monte Carlo (ECMC) for 2D hard-sphere systems (arxiv:2004.11040, DOI: 10.1016/j.cpc.2020.107702). The current implementation is based on the straight event-chain (SEC) algorithm. In the multithreaded ECMC, multiple spheres move simultaneously on different threads. The correctness of the multithreaded algorithm is guaranteed by comparing the algorithm with its singlethreaded version, which satisfies the global-balance condition explicitly. When ignoring synchronization, the multithreaded ECMC processes 10^12 collisions per hour on a x86 CPU with 20 physical cores. However, the correctness of this algorithm requires rather frequent synchronization. This prevents the multithreaded ECMC from reaching its full potential. Several attempts has been made to fix this problem.


Slides here

Recording

Further material


back to 2021 ECMC workshop

Personal tools