nBodySim is a small simulator programa capable of handling huge number of bodies (>1000). It's capable of using multiple cores of a modern many-core CPU. For this, nBodySim is programed using D Language 2, using the std.parallelism library.
This software was compiled with gdc 4.6.3 and dmd 2.0.65, and tested on some machines using Ubuntu x64 14.04, 12.04, 12.10 and Linux ROCKS 6.0 x64.
To build it, use DUB. So do a simple :
dub build=release
It will generate the program using parallel for. If you like to compare with the serial version, you can launch dub with --configure=nBodySim-serial
./nBodySim [OPTION]...
- -i, --input=FILE Input file with initial positions and velocity of every body
- -o, --output=FILE Output data file
- -d, --deltat=NUM Delta of Time used in each step. By default is 86400 [s]
- -n, --num=NUM Total number of steps. by default is 36500, that with default DeltaT, is 100 years of simulation
- --points=NUM Defines every few points that are written to output file. 1 means that are used all points, 2 that 1/2 of the total points not are written. By default is 4
- -e, --epsilon=NUM Softening factor used that avoid NaNs and Inf in calculations. By default is 1000 meters
- --rand=NUM If not are using a input file, then generate NUM random stellar mass bodies in a cube of 10 LY
- --seed=NUM Set a seed for the random generator
- --center=NUM Set the output coords relative to the entity NUM Useful to centre around a particular entity, like the Sun
- -h, --help Show this help
Note : If it is used without any option, will try to open ssolar.sim file
- Execute a simulation of 100 years with a DeltaT of a 1 day of our solar system fixing the Sun in the (0,0,0) position
./nBodySim --input ssolar.sim --center 0 --output ssolar.out - Execute a simulation of 10000 years with a DeltaT of 100 days of a ranmdon set of 500 stars
./nBodySim --rand 500 -d8640000 --output rand.out
For the moment, nBodySim it's outputing files that are compatible with GnuPlot, where in a row we can found the position of every object in the simulation for a determinated moment ofthe simulation. It's easy to plot hte data using 'splot' command of gnuplot, but be carefull about how many points (rows) are stored in the file, becasue GnuPlot can be easily swamped. I recomend using --points to skip points to be stored in the output file, and keep it easy to hnadle for the gnuplot at same tiem that allow to run long (big number of iterations).
View the output data of the simulation of our solar system
set zrange [-8e+12:8e12]
set title "Solar System : 100 years"
splot "ssolar.out" u 1:2:3 t "Sun", \
"ssolar.out" u 4:5:6 w dots t "Mercury", \
"ssolar.out" u 7:8:9 w dots t "Venus", \
"ssolar.out" u 10:11:12 w dots t "Earth", \
"ssolar.out" u 13:14:15 w dots t "Moon", \
"ssolar.out" u 16:17:18 w dots t "Mars", \
"ssolar.out" u 19:20:21 w dots t "Jupiter", \
"ssolar.out" u 22:23:24 w dots t "Saturn", \
"ssolar.out" u 25:26:27 w dots t "Uranus", \
"ssolar.out" u 28:29:30 w dots t "Neptune", \
"ssolar.out" u 31:32:33 w dots t "Ceres", \
"ssolar.out" u 34:35:36 w dots t "Pluto", \
"ssolar.out" u 37:38:39 w dots t "Haumea", \
"ssolar.out" u 40:41:42 w dots t "Makemake", \
"ssolar.out" u 43:44:45 w dots t "Eris"
- Make a animated viewer of generated data.
- Make a interactive mode that work with the viewer.
- Add an option to use relative coords around the barycenter of the all system.
- Implement collisions.
- Add other output format that can carry more usefull information.
- Add a way to do electroestatic simulations.
- Add a way to use a arbitrary precision calculations (GMP).
FX-4100 quad core times :
Bodies : 1000
Serial = 573 ms
ParallelFor = 192 ms -> SpeedUp = 2.98
Bodies : 3000
Serial = 5106 ms
ParallelFor = 1745 ms -> SpeedUp = 2.92
Bodies : 10000
Serial = 57726 ms
ParallelFor = 20023 ms -> SpeedUp = 2.88
~72% of max teorical speedup
Opteron 16-core times :
Bodies : 1000
Serial = 682 ms
ParallelFor = 67 ms -> SpeedUp = 10,12
Bodies : 3000
Serial = 6083 ms
ParallelFor = 474 ms -> SpeedUp = 12,83
Bodies : 10000
Serial = 69728: ms
ParallelFor = 5232 ms -> SpeedUp = 13,32
~83.25% of max teorical speedup
FX-8370E overcloked at 3.6Ghz . dmd 2.090
Bodies : 1000
Serial = 119ms
ParallelFor = 19ms -> SpeedUp = 6,26
Bodies : 3000
Serial = 1078ms
ParallelFor = 168ms -> SpeedUp = 6,41
Bodies : 10000
Serial = 11945ms
ParallelFor = 1841ms -> SpeedUp = 6,48
R7 3700x
Bodies : 1000
Serial = 43ms
ParallelFor = 5ms
Bodies : 3000
Serial = 387ms
ParallelFor = 35ms
Bodies : 10000
Serial = 4320ms
ParallelFor = 376ms -> SpeedUp = 11.48
~71% of max teorical speedup
i7-6700 4 core 8 threads
Bodies : 1000
Serial = 212ms
ParallelFor = 39ms
Bodies : 3000
Serial = 1868ms
ParallelFor = 354ms
Bodies : 10000
Serial = 20745ms
ParallelFor = 3761ms -> SpeedUp = 5,52
~69% of max teorical speedup
FX-4100 quad core times :
Bodies : 100
Serial = 571 ms
ParallelFor = 234 ms -> SpeedUp = 2,44
Bodies : 1000
Serial = 57373 ms
ParallelFor = 19987 ms -> SpeedUp = 2,87
~71.75% of max teorical speedup
Opteron 16-core times :
Bodies : 100
Serial = 676 ms
ParallelFor = 296 ms -> SpeedUp = 2,28
Bodies : 1000
Serial = 67505 ms
ParallelFor = 5921 ms -> SpeedUp = 11,40
~71.25% of max teorical speedup
FX-8370E overcloked at 3.6Ghz . dmd 2.090
Bodies : 100
Serial = 600ms
ParallelFor = 129ms -> SpeedUp = 4,651
Bodies : 1000
Serial = 60063ms
ParallelFor = 9314ms -> SpeedUp = 6,448
~80% of max teorical speedup
Bodies : 3000
ParallelFor = 83770ms
R7 3700x 8 core 16 threads
Bodies : 100
Serial = 220ms
ParallelFor = 42ms
Bodies : 1000
Serial = 21886ms
ParallelFor = 2087ms -> SpeedUp = 10.48
i7-6700 4 core 8 threads
Bodies : 100
Serial = 210ms
ParallelFor = 60ms
Bodies : 1000
Serial = 20706ms
ParallelFor = 3852ms -> SpeedUp = 5.37
~67% of max teorical speedup