HAL Id: cea-02557621
https://hal-cea.archives-ouvertes.fr/cea-02557621
Submitted on 28 Apr 2020
HAL is a multi-disciplinary open access
archive for the deposit and dissemination of
sci-entific research documents, whether they are
pub-lished or not. The documents may come from
teaching and research institutions in France or
abroad, or from public or private research centers.
L’archive ouverte pluridisciplinaire HAL, est
destinée au dépôt et à la diffusion de documents
scientifiques de niveau recherche, publiés ou non,
émanant des établissements d’enseignement et de
recherche français ou étrangers, des laboratoires
publics ou privés.
Optimization of Monte Carlo codes PENELOPE 2006
and PENFAST by parallelization and reduction variance
implementation
François Tola, Bénédicte Poumarède, Mehdi Gmar, Bouchra Habib
To cite this version:
François Tola, Bénédicte Poumarède, Mehdi Gmar, Bouchra Habib. Optimization of Monte Carlo
codes PENELOPE 2006 and PENFAST by parallelization and reduction variance implementation.
Second European Workshop on Monte carlo Treatment Planning (MCTP - 2009), Oct 2009, Cardiff,
United Kingdom. �cea-02557621�
OPTIMIZATION OF MONTE CARLO CODES PENELOPE
2006 AND PENFAST BY PARALLELIZATION AND
REDUCTION VARIANCE IMPLEMENTATION
F. TOLA, B. POUMAREDE, B. HABIB, M. GMAR
CEA, LIST, Department of technology for sensors and signal processing, F-91191 Gif-sur-Yvette, France.
franç[email protected]
MCTP 2009 – CARDIFF, October 19 - 21, 2009 Conventional TPS are fast but not enough accurate in presence of
heterogeneities
Introduction of MC methods allows better accuracy.
Recently, a fast MC dose calculation code, named PENFAST has been developed by Salvat et al (2008). PENFAST is an optimized version of the MC PENELOPE code, adapted to CT voxelized
geometries.
Reduction of calculation time thanks to two methods :
Introduction of variance reduction techniques in PENELOPE
2006
Parallelization of Monte Carlo codes using MPI (Message Passing Interface) or equivalent
CONTEX AND OBJECTIVES
CONCLUSIONS
Monte Carlo methods lead to decrease uncertainties in dose calculation but are too time expensive
Thanks to implementation of variance reduction techniques in the PENELOPE 2006 code, simulation time is reduced by a factor greater than 100 with unbiaised results
The use of variance reduction coupled with parallelization allow to achieve the reasonable computation time (10 min) with the required uncertainty (2 %)
Parallelization of PENELOPE 2006 and PENFAST codes
Achieved using MPI – Message Passing Interface – version mpich2 1.0.6
Implemented in fortran 77, excepted for the random generator and PSF management
Generator RP Brent : period of 101230
A processor master in charge of PSF read/write operations and NP-1 slaves dedicated to the Monte Carlo calculation
PARALLELIZATION OF MONTE CARLO CODES
MATERIALS AND METHODS
Step 1
Detailed simulation of the particle transport through the
accelerator head
Step 2
Simulation of the three-dimensional dose distribution in a
voxelised phantom
Phase space
MC calculations are split into two parts
PENELOPE
PENFAST
The implemented methods increase the simulation efficiency by a
factor greater than 100 compared with the analogue simulation.
VARIANCE REDUCTION TECHNIQUES
0 1 2 3 0 50 100 150 200 250 Depth (cm) D o s e ( e V /g .é le c tr o n p ri m a ir e ) -20 -10 0 10 20 (% ) analogue simulation
simulation with variance reduction relative différence (%) statistical uncertainty (+3σ) statistical uncertainty (-3σ) 0 0,5 1 1,5 2 2,5 3 3,5 4 4,5 0 5 10 15 20 25 30 depth (cm) a b s o lu te d o s e ( e V /g ) analogue MC
MC with variance reduction
0 20 40 60 80 100 120 0 20 40 60 80 100 120
Number of usesful processors
T im e c a lc u la ti o n g a in Real time CPU time PENELOPE
SATURNE 43 accelerator - 18 MeV electrons Simulated showers numbers : 6 x 106 Calculation time for one processor : 21 hours Particules numbers in PSF : 1.62 x 106
