CTC: mcp on Velocity Cluster

What is it?

Utility for copying a file to local disk on every node.

How does it work?

It reads the file from disk on the primary node,
uses MPI_Bcast to send it to remote nodes, and writes it there
Where is it? h:\ctc tools\mcp.exe

How do I use it?

It takes two arguments on the primary node:
source path
destination path
No arguments are read on the secondary nodes.
Command line: (Assuming appropriate 'machines' file exists)
tcsh: mpirun -np 3 H:\\CTCTOO~1\\mcp.exe H:\\users\\zollweg\\mydata T:\\zollweg

CMD: mpirun -np 3 H:\CTCTOO~1\mcp.exe H:\users\zollweg\mydata T:\zollweg

Example batch file: (Does the same thing as the command line example above)
mcp.bat
ctc001 1 file://ctcfsrv1/tc/ctc tools\mcp.exe H:\users\zollweg\mydata T:\zollweg
ctc002 1 file://ctcfsrv1/tc/ctc tools\mcp.exe
ctc003 1 file://ctcfsrv1/tc/ctc tools\mcp.exe

Invocation:
mpirun -wd H:\users\zollweg -pg_file mcp.bat

Notes:

If the input file specification is incomplete, the first file that begins with the specified string will be copied. In such a case, mcp will tell you which file it copied. For example if my home directory contains mydata.dat and mydata.bin, mcp would copy mydata.bin and tell me so.

If the output file specification is a directory, the file is placed in that directory with the same name as it had on the primary node. Thus, in the examples above, nodes 1, 2, and 3 would have a file called T:\zollweg\mydata.bin.

If the output file specification is a filename (whether it exists or not) the file will be given that name. If the file had existed, it is overwritten.

If either the input or output file specification includes a directory that doesn't exist, the job fails. If the directory failure is on some, but not all, secondary nodes, some files may get written.

The copied files have the same permissions as the original.


Home Services Support

mcp on Velocity Cluster

What is it? Utility for copying a file to local disk on every node. How does it work? It reads the file from disk on the primary node, uses MPI_Bcast to send it to remote nodes, and writes it there Where is it? h:\ctc tools\mcp.exe How do I use it? It takes two arguments on the primary node: source path destination path No arguments are read on the secondary nodes. Command line: (Assuming appropriate 'machines' file exists) tcsh: mpirun -np 3 H:\\CTCTOO~1\\mcp.exe H:\\users\\zollweg\\mydata T:\\zollweg CMD: mpirun -np 3 H:\CTCTOO~1\mcp.exe H:\users\zollweg\mydata T:\zollweg Example batch file: (Does the same thing as the command line example above) mcp.bat ctc001 1 file://ctcfsrv1/tc/ctc tools\mcp.exe H:\users\zollweg\mydata T:\zollweg ctc002 1 file://ctcfsrv1/tc/ctc tools\mcp.exe ctc003 1 file://ctcfsrv1/tc/ctc tools\mcp.exe Invocation: mpirun -wd H:\users\zollweg -pg_file mcp.bat Notes: If the input file specification is incomplete, the first file that begins with the specified string will be copied. In such a case, mcp will tell you which file it copied. For example if my home directory contains mydata.dat and mydata.bin, mcp would copy mydata.bin and tell me so. If the output file specification is a directory, the file is placed in that directory with the same name as it had on the primary node. Thus, in the examples above, nodes 1, 2, and 3 would have a file called T:\zollweg\mydata.bin. If the output file specification is a filename (whether it exists or not) the file will be given that name. If the file had existed, it is overwritten. If either the input or output file specification includes a directory that doesn't exist, the job fails. If the directory failure is on some, but not all, secondary nodes, some files may get written. The copied files have the same permissions as the original.



© 2008 Cornell University \| Cornell University Center for Advanced Computing \| Copyright Statement