This is a README for running CMBFAST in parallel. It uses LAM/MPI and can run on clusters with distributed memory (beowulf clusters). So far it was only tested on PGI fortran and does not work with g77. The strategy is to distribute individual k modes onto separate nodes. Since the running time for each k mode increases with amplitude of k the efficiency decreases with number of nodes. Recommended number is 8 nodes.
WARNING: this option is provided as is and there is no software support from the authors.
- Instructions
in RUN/Makefile edit NUMHDD=number of machines with local disks -1, NUMPROC=number of processors, LOCALDATADIR=local directory for input files
example:
NUMHDD=6
NUMPROC=14
LOCALDATADIR=/var/tmp/uros/1
or if individual processors have no local disks then
NUMHDD=0
NUMPROC=14
LOCALDATADIR=./d
In ~/.bashrc put
export LAMHF77=pgf90
an example to configure is in conf, run
./conf
cd RUN
generate jl.dat and if needed also
ujl.dat, eos.dat amd jlens.dat (note that the names must match)
cd ..
make
cd RUN
an example to run is
./cmb_mpi