NAME¶

mbcp - Copy files in parallel with MPI-Bash

SYNOPSIS¶

mbcp [-r] [-v] [-n] [-P] [-p] [-a] source file|source directory... target file|target directory

DESCRIPTION¶

mbcp is a parallel version of the standard cp command, designed to copy large files or large numbers of files—or even large numbers of large files—by distributing the work across a set of computers that share a filesystem. It can copy

•: a single file to a single file,
•: a single directory to a single directory, or
•: one or more files and/or directories to a directory.

mbcp is a bash script that uses MPI-Bash functions for communication. It should therefore be run like any other MPI program (typically with mpirun or mpiexec).

OPTIONS¶

mbcp accepts only a subset of cp's options:

-r: Copy directories recursively.
-v: Explain what is being done.
-n: Do not overwrite an existing file.
-P: Never follow symbolic links in source.
-p: Preserve file mode, ownership, and timestamps.
-a: Same as -P -p -r.

In addition, at least one source file or directory and exactly one target file or directory must be specified on the command line.

EXAMPLES¶

Use 16 processes to copy one file to another:

mpirun -np 16 mbcp my_giant_file their_giant_file.BAK

Use 128 processes to copy one directory to another, with verbose reporting:

mpirun -np 128 mbcp -r -v /scratch/me/big /usr/projects/myproj/big

BUGS¶

I sometimes see segmentation faults when MPI-Bash programs exit. I don't know why.

NOTES¶

mbcp requires Circle-Bash (MPI-Bash with Libcircle extensions) to operate.

AUTHOR¶

Scott Pakin, pakin@lanl.gov