readme.md

• Build
  • Edit/Add cmdstan/make/local
  • Build in cmdstan
• Run
• Results

Predator-Prey model based on https://mc-stan.org/users/documentation/case-studies/lotka-volterra-predator-prey.html

Build

Edit/Add cmdstan/make/local

TORSTEN_MPI = 1                                         # flag on torsten's MPI solvers
CXXFLAGS += -isystem /usr/local/include                 # path to MPI library's headers

Build in cmdstan

make ../example-models/lotka_volterra_ode_group_model/lv_group

Run

mpiexec -n 2 lv_group sample data file=lv_group.data.R

Results

Three binaries are built and used to run:

• sequential run using Stan's rk45 integrator.
• sequential run using Torsten's rk45 integrator.
• MPI run using Torsten's rk45 group integrator(with 2, 4, 8 processes, respectively).

The wall time of sequential runs and MPI runs(in seconds) with ODE group size 16.

run	wall time(s)
Sequential stan	2135
sequential pmx	2570
MPI pmx (nproc=1)	2107
MPI pmx (nproc=2)	1187
MPI pmx (nproc=4)	715
MPI pmx (nproc=8)	467