Start LAM/MPI once, before running
any MPI programs.
It is not necessary to start
MPI before compiling and linking MPI programs.
The LAM/MPI program "lamd" will run until halted.
[psiders@beowulf source] lamboot -v ~/lamhosts
LAM 6.5.4/MPI 2 C++/ROMIO - University of Notre Dame
Executing hboot on n0 (beowulf.d.umn.edu - 1 CPU)...
Executing hboot on n1 (node8.d.umn.edu - 1 CPU)...
Executing hboot on n2 (node7.d.umn.edu - 1 CPU)...
Executing hboot on n3 (node6.d.umn.edu - 1 CPU)...
Executing hboot on n4 (node5.d.umn.edu - 1 CPU)...
Executing hboot on n5 (node4.d.umn.edu - 1 CPU)...
Executing hboot on n6 (node2.d.umn.edu - 1 CPU)...
lamboot reads the list of available computers from
the file specified on the "lamboot" command
line. In the example above, that is lamhosts.
[psiders@beowulf source] cat ~/lamhosts.
beowulf.d.umn.edu
node8.d.umn.edu
node7.d.umn.edu
node6.d.umn.edu
node5.d.umn.edu
node4.d.umn.edu
node2.d.umn.edu
#node3.d.umn.edu machine is down
#node1.d.umn.edu login fails
If any of the computers in lamhosts are down or refuse to allow remote command execution, lamboot will write a verbose error message containing suggestions.
The lamnodes command shows node assignments
[psiders@beowulf source] lamnodes
| n0 | beowulf.d.umn.edu:1 | |
| n1 | node8.d.umn.edu:1 | |
| n2 | node7.d.umn.edu:1 | |
| n3 | node6.d.umn.edu:1 | |
| n4 | node5.d.umn.edu:1 | |
| n5 | node4.d.umn.edu:1 | |
| n6 | node2.d.umn.edu:1 |
[psiders@beowulf source] lamhalt
LAM 6.5.4/MPI 2 C++/ROMIO - University of Notre Dame
If lamhalt fails, one can find lamd processes with ps, then stop them with kill.