| S | D  | N :   ·  Sourceforge  ·  Red Hat  ·  Slashdot  ·  Freshmeat  ·  Themes  ·  Open Magazine

DKBF | Download | Contact | Forums | Bugs | Support | Mailing Lists | Tasks | Docs | Surveys | News


Sample PG file
FAQ
MPI
Statistics
Price:Performance Worksheet
Clustering Resources
Other Crackers

Sample PG File

A pg file is used in MPI to control the number of process and the location of those process started on a cluster for a particular job. The format of the file is simple. Name the file the same as the binary/executable and add the extention ".pg". The file consists of 3 columns delimited by tabs.

The first column is the name of a machine in your cluster. The second column is the number of processes to start on that machine. The third column is the path of the executable. This path should be the same on every machine to avoid problems.

For example, the following is the contents of the file dkbf.pg at /home/username/dkbf/bin/ on the machine Node1:

Node1   1   /home/username/dkbf-0.1.1/bin/dkbf -i dkbf.ini
Node2   2   /home/username/dkbf-0.1.1/bin/dkbf
Node3   2   /home/username/dkbf-0.1.1/bin/dkbf

This pg file creates 1 process on Node1, which will be the master process with a rank or id of 0. Node2 and Node3 will each get 2 processes. Node2 will get processes 1,2 and Node 3 will get processes 3,4.

Only the master node, Node1, needs a pg file. Put the command line options for the job in this file as shown above. This tells the master node to look at the file dkbf.ini for all job options.

The command line with options to start a job will look like this:

mpirun -pg_file dkbf.pg dkbf

The -i option is the only option currently supported by dkbf. This option indicates the path of the INI (initialization) file used by dkbf for a particular job. Options for configuring a job are located in this file.

This site hosted by SourceForge Logo