TY - GEN

T1 - Bottlenecks in Implementation of the Mode Decomposition Algorithm Based on Phase-Only Spatial Light Modulator

AU - Smolyaninov, Nikolai N.

AU - Kharenko, Denis S.

N1 - Funding Information: Russian Science Foundation (Grant No. 21-72-30024). Publisher Copyright: © 2022 IEEE.

PY - 2022

Y1 - 2022

N2 - The article covers automation of a mode decomposition experiment made for a highly multimode laser beam and the issues of applied algorithm's software implementation which have occurred in the process. A short description of the experimental setup and the architecture of the resulting automation solution are given. It was established that the bottlenecks were pseudoparallel threads' execution implemented in Python language interpreter and ineffective direct data copying by the tools of this language from subsystems implemented on C# language. The article also shares the description of methods helping to minimize the negative impact of these 'bottlenecks'. Due to efficient use of NumPy library procedures and native Python threads we managed to partially get round the restrictions of global interpreter lock which allows to use real multithreading of multicore computation systems. With regard to the automation of the installation at hand, we succeeded in reducing the time for one measurement by an average factor of 4, namely from 20 to 5 minutes. This allowed to reduce the full measurement cycle, which would earlier run for over 5 hours, by the same factor.

AB - The article covers automation of a mode decomposition experiment made for a highly multimode laser beam and the issues of applied algorithm's software implementation which have occurred in the process. A short description of the experimental setup and the architecture of the resulting automation solution are given. It was established that the bottlenecks were pseudoparallel threads' execution implemented in Python language interpreter and ineffective direct data copying by the tools of this language from subsystems implemented on C# language. The article also shares the description of methods helping to minimize the negative impact of these 'bottlenecks'. Due to efficient use of NumPy library procedures and native Python threads we managed to partially get round the restrictions of global interpreter lock which allows to use real multithreading of multicore computation systems. With regard to the automation of the installation at hand, we succeeded in reducing the time for one measurement by an average factor of 4, namely from 20 to 5 minutes. This allowed to reduce the full measurement cycle, which would earlier run for over 5 hours, by the same factor.

KW - C#

KW - Experiment automation

KW - global interpreter lock (GIL)

KW - NumPy

KW - Python 3

UR - http://www.scopus.com/inward/record.url?scp=85137345689&partnerID=8YFLogxK

UR - https://www.mendeley.com/catalogue/20bc4064-2582-3690-a2ff-875f840fe6c7/

U2 - 10.1109/EDM55285.2022.9855062

DO - 10.1109/EDM55285.2022.9855062

M3 - Conference contribution

AN - SCOPUS:85137345689

SN - 9781665498043

T3 - International Conference of Young Specialists on Micro/Nanotechnologies and Electron Devices, EDM

SP - 693

EP - 697

BT - Proceedings of the 2022 IEEE 23rd International Conference of Young Professionals in Electron Devices and Materials, EDM 2022

PB - IEEE Computer Society

T2 - 23rd IEEE International Conference of Young Professionals in Electron Devices and Materials, EDM 2022

Y2 - 30 June 2022 through 4 July 2022

ER -